threads
listlengths 1
2.99k
|
|---|
[
{
"msg_contents": "I saw the posting regarding the NaN issue and I \nthought you might be interested in the following:\n\nSun Microsystem's Numerical Computation Guide\nwhich covers the IEEE 754 standard, states:\n\n\"NaN's are often represented as floating-point numbers\nwith the exponent e max + 1 and non-zero significands.\"\n\nand\n\nSoftFloat, a free software implementation of the\nstandard,\nhttp://HTTP.CS.Berkeley.EDU/~jhauser/arithmetic/softfloat.html,\ndefines the following:\n\ndefault 32-bit NaN: 0xFFC00000\ndefault 64-bit NaN: 0xFFF8000000000000\n\nAssigning these values to int32_t and int64_t, and then\nmemcpying them to a float and a double representation\nyeilded a TRUE result when testing with isnan().\nIf they are NaN under Linux/Intel implementation \n(2.0.36) using gcc 2.7.2.3 and math library 2.0.7 \nof the IEEE 754 standard, is it not fair to say it \nwould be true for all implementations?\n\nFor what its worth...\n\nMarcus Mascari ([email protected])\n\n\n\n\n\n\n\n\n_________________________________________________________\nDO YOU YAHOO!?\nGet your free @yahoo.com address at http://mail.yahoo.com\n\n",
"msg_date": "Tue, 5 Jan 1999 00:56:56 -0800 (PST)",
"msg_from": "Marcus Mascari <[email protected]>",
"msg_from_op": true,
"msg_subject": "NaN format"
},
{
"msg_contents": "Marcus Mascari wrote:\n\n> Sun Microsystem's Numerical Computation Guide\n> which covers the IEEE 754 standard, states:\n>\n> \"NaN's are often represented as floating-point numbers\n> with the exponent e max + 1 and non-zero significands.\"\n>\n> and\n>\n> SoftFloat, a free software implementation of the\n> standard,\n> http://HTTP.CS.Berkeley.EDU/~jhauser/arithmetic/softfloat.html,\n> defines the following:\n>\n> default 32-bit NaN: 0xFFC00000\n> default 64-bit NaN: 0xFFF8000000000000\n\n My i386 Linux's nan.h defines them as\n\n Signal double NaN 0x7FF00000000F0000\n Quiet double NaN 0x7FF8000000000000\n Signal float NaN 0x7F800F00\n Quiet float NaN 0x7FC00000\n\n>\n> Assigning these values to int32_t and int64_t, and then\n> memcpying them to a float and a double representation\n> yeilded a TRUE result when testing with isnan().\n> If they are NaN under Linux/Intel implementation\n> (2.0.36) using gcc 2.7.2.3 and math library 2.0.7\n> of the IEEE 754 standard, is it not fair to say it\n> would be true for all implementations?\n\n Even if your ones work with Linux too, I expect problems on\n other implementations or at least on different hardware.\n\n If we add a little check to the regression, I think we can\n figure out most of the platform specific definitions for NaN\n until v6.5 and put them into .../include/port/*.h with having\n a fallback to a function that elog()'s why this test failed.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Tue, 5 Jan 1999 11:17:39 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: NaN format"
}
] |
[
{
"msg_contents": "Hello!\n\n I have a problem running configure (6.4 and 6.4.2) on\nsparc-solaris2.5.1 - it reports:\nconftest.sh: not found\n The problem is in configure line 743:\n. conftest.sh\n I replaced it with\n. ./conftest.sh\n and all goes well.\n\nOleg.\n---- \n Oleg Broytmann http://members.tripod.com/~phd2/ [email protected]\n Programmers don't die, they just GOSUB without RETURN.\n\n",
"msg_date": "Tue, 5 Jan 1999 13:48:24 +0300 (MSK)",
"msg_from": "Oleg Broytmann <[email protected]>",
"msg_from_op": true,
"msg_subject": "./configure: conftest.sh not found"
},
{
"msg_contents": "> Hello!\n> \n> I have a problem running configure (6.4 and 6.4.2) on\n> sparc-solaris2.5.1 - it reports:\n> conftest.sh: not found\n> The problem is in configure line 743:\n> . conftest.sh\n> I replaced it with\n> . ./conftest.sh\n> and all goes well.\n\nThis is fixed in 6.5.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 11:02:00 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] ./configure: conftest.sh not found"
}
] |
[
{
"msg_contents": "> Trying to use select statement using CURRENT_DATE\n> Which works fine like so:\n> \n> select * from headlines where dateof = CURRENT_DATE order by dateof desc\n> \n> But I'm also wanting to do something similar to:\n> \n> select * from headlines where dateof = CURRENT_DATE - INTERVAL '1' DAY\n> order by dateof desc\n> \n> Basically just trying to subtract 1 day from the CURRENT_DATE\n\nKevin,\n\n (1) Handy hint: try '\\df' once in the psql utility.\n\tThis gives a nice list of functions, their return types, and\n\tbrief descriptions of what they accomplish.\n\n (2) The function for which you seek is probably \"timemi\" which \n\ttakes a time as its first argument and a time interval \n\t(like 1 day) as its second argument.\n\n\tFor example, same time yesterday:\n\n\t select timemi('now'::datetime, '1 day'::timespan);\n\n- Tim\n",
"msg_date": "Tue, 05 Jan 1999 11:06:18 CST",
"msg_from": "Tim Williams <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [GENERAL] select using date"
},
{
"msg_contents": "\nTrying to use select statement using CURRENT_DATE\nWhich works fine like so:\n\nselect * from headlines where dateof = CURRENT_DATE order by dateof desc\n\n\nBut I'm also wanting to do something similar to:\n\n\nselect * from headlines where dateof = CURRENT_DATE - INTERVAL '1' DAY\norder by dateof desc\n\nBasically just trying to subtract 1 day from the CURRENT_DATE\n\nWhen I try the above I get\n\nERROR: parser: parse error at or near \"day\"\n\nAny suggestions would be appreciated.\n\n\nKevin\n\n\n\n--------------------------------------------------------------------\nKevin Heflin | ShreveNet, Inc. | Ph:318.222.2638 x103\nVP/Mac Tech | 333 Texas St #619 | FAX:318.221.6612\[email protected] | Shreveport, LA 71101 | http://www.shreve.net\n--------------------------------------------------------------------\n\n",
"msg_date": "Tue, 5 Jan 1999 10:55:53 -0600 (CST)",
"msg_from": "Kevin Heflin <[email protected]>",
"msg_from_op": false,
"msg_subject": "select using date"
},
{
"msg_contents": "How can I see a view in psql after having created it? If I enter \n\\d <viewname> I can see the fields, but how can I see what makes up the\nview? Best if the original CREATE statement could be listed somehow.\n\n",
"msg_date": "Tue, 5 Jan 1999 20:04:10 +0100 (CET)",
"msg_from": "Kaare Rasmussen <[email protected]>",
"msg_from_op": false,
"msg_subject": "Views"
},
{
"msg_contents": "Kaare Rasmussen wrote:\n> \n> How can I see a view in psql after having created it? If I enter\n> \\d <viewname> I can see the fields, but how can I see what makes up the\n> view? Best if the original CREATE statement could be listed somehow.\n\nWe could use EXPLAIN VIEW for this...\nNew item for TODO?\n\nEXPLAIN TABLE, EXPLAIN INDEX, etc\n\nVadim\n",
"msg_date": "Wed, 06 Jan 1999 08:58:09 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [GENERAL] Views"
},
{
"msg_contents": "> Kaare Rasmussen wrote:\n> > \n> > How can I see a view in psql after having created it? If I enter\n> > \\d <viewname> I can see the fields, but how can I see what makes up the\n> > view? Best if the original CREATE statement could be listed somehow.\n> \n> We could use EXPLAIN VIEW for this...\n> New item for TODO?\n> \n> EXPLAIN TABLE, EXPLAIN INDEX, etc\n\nHow about:\n\n\tSELECT * from pg_views\n\tWHERE viewname = 'fred';\n\nOR\n\n\tSELECT pg_get_viewdef('fred');\n\nI have added this to the TODO list:\n\n\t* have psql \\d on a view show the view definition.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 21:21:39 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [GENERAL] Views"
},
{
"msg_contents": "> \n> Kaare Rasmussen wrote:\n> > \n> > How can I see a view in psql after having created it? If I enter\n> > \\d <viewname> I can see the fields, but how can I see what makes up the\n> > view? Best if the original CREATE statement could be listed somehow.\n> \n> We could use EXPLAIN VIEW for this...\n> New item for TODO?\n> \n> EXPLAIN TABLE, EXPLAIN INDEX, etc\n> \n> Vadim\n> \n> \n\n SELECT * FROM pg_views;\n\n\nJan\n\n-- \n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n",
"msg_date": "Wed, 6 Jan 1999 11:03:33 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [GENERAL] Views"
},
{
"msg_contents": "Try:\n\nselect current_date, CURRENT_DATE - INTERVAL '1 DAY';\n ?column?|?column?\n----------+----------------------\n1999-01-08|1999-01-07 00:00:00+01\n(1 row)\n\nPostgreSQL has a syntax sligth different than SQL92. You have to enclose\n'1 DAY' instead of '1' DAY.\n\n-Jose'-\n\nKevin Heflin wrote:\n> \n> Trying to use select statement using CURRENT_DATE\n> Which works fine like so:\n> \n> select * from headlines where dateof = CURRENT_DATE order by dateof desc\n> \n> But I'm also wanting to do something similar to:\n> \n> select * from headlines where dateof = CURRENT_DATE - INTERVAL '1' DAY\n> order by dateof desc\n> \n> Basically just trying to subtract 1 day from the CURRENT_DATE\n> \n> When I try the above I get\n> \n> ERROR: parser: parse error at or near \"day\"\n> \n> Any suggestions would be appreciated.\n> \n> Kevin\n> \n> --------------------------------------------------------------------\n> Kevin Heflin | ShreveNet, Inc. | Ph:318.222.2638 x103\n> VP/Mac Tech | 333 Texas St #619 | FAX:318.221.6612\n> [email protected] | Shreveport, LA 71101 | http://www.shreve.net\n> --------------------------------------------------------------------\n",
"msg_date": "Fri, 08 Jan 1999 13:37:57 +0100",
"msg_from": "\"Jose' Soares\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [GENERAL] select using date"
},
{
"msg_contents": "On Fri, 8 Jan 1999, Jose' Soares wrote:\n\n> Try:\n> \n> select current_date, CURRENT_DATE - INTERVAL '1 DAY';\n> ?column?|?column?\n> ----------+----------------------\n> 1999-01-08|1999-01-07 00:00:00+01\n> (1 row)\n> \n> PostgreSQL has a syntax sligth different than SQL92. You have to enclose\n> '1 DAY' instead of '1' DAY.\n\n\nWell, I tried:\n\nselect current_date, CURRENT_DATE - INTERVAL '1 DAY';\n\nbut receive the following:\n\nERROR: There is no operator '-' for types 'date' and 'timespan'\n You will either have to retype this query using an explicit cast,\n or you will have to define the operator using CREATE OPERATOR\n\n\nAfter trying somethings, I was able to get this to work:\n\nselect CURRENT_DATE -1;\n\nwill subtract 1 day from the current date... which gives me what I need\nfor now.\n\n\nThanks\n\nKevin\n\n\n\n\n\n\n\n--------------------------------------------------------------------\nKevin Heflin | ShreveNet, Inc. | Ph:318.222.2638 x103\nVP/Mac Tech | 333 Texas St #619 | FAX:318.221.6612\[email protected] | Shreveport, LA 71101 | http://www.shreve.net\n--------------------------------------------------------------------\n\n",
"msg_date": "Fri, 8 Jan 1999 08:52:38 -0600 (CST)",
"msg_from": "Kevin Heflin <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [GENERAL] select using date"
}
] |
[
{
"msg_contents": "Hi!\n\nIn the INSTALL file, it says (under \"Unsupported platforms\"):\n Windows x86 v6.3 1998-03-01 not library compatible; \n client side maybe; use \n ODBC/JDBC\n\n\n Note that Windows ports of the frontend are \n apparently possible using third-party Posix porting \n tools and libraries.\n\nI believe the client libraries (at least libpq) are available in native\nWin32 since version 6.4.\n\n//Magnus\n\n\n",
"msg_date": "Tue, 5 Jan 1999 14:35:21 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "Postgresql INSTALL-file & Windows"
},
{
"msg_contents": "> In the INSTALL file, it says (under \"Unsupported platforms\"):\n> Windows x86 v6.3 1998-03-01 not library compatible;\n> client side maybe; use ODBC/JDBC\n> Note that Windows ports of the frontend are\n> apparently possible using third-party Posix porting\n> tools and libraries.\n> I believe the client libraries (at least libpq) are available in \n> native Win32 since version 6.4.\n\nThanks for pointing this out. So how should I phrase it? As above? If\nso, where do we tell people how to build it? Or is it obvious to anyone\nwho is trying? \n\nIf it could use instructions, would someone like to write it up? Or was\nsomething already posted on the mailing list which tells all?\n\nSo many questions, so little time :) As you can tell, I haven't tried\nmyself...\n\n - Tom\n",
"msg_date": "Wed, 06 Jan 1999 14:35:02 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [DOCS] Postgresql INSTALL-file & Windows"
}
] |
[
{
"msg_contents": "Ok, in multi-version systems readers never lock\nselected rows and so never block writers. Nice but\ncause problems in some cases: if you want\nto implement referential integrity at the\napplication level then you'll have to use\nLOCK TABLE IN SHARE MODE or SELECT FOR UPDATE to\nprevent updation of primary keys etc. Not so good...\n\nSo, what about FOR SHARE LOCK clause in SELECTs?\nLike FOR UPDATE clause, FOR SHARE LOCK clause\nwill lock selected rows, but in _share_ mode.\nThis is some kind of read-locking on demand.\n\nDisadvantages:\n\n1. FOR SHARE LOCK will use lock manager to\n lock rows, but lmgr can't handle \"too many\" locks.\n (note that UPDATE, DELETE & SELECT FOR UPDATE use only\n one entry in lmgr table for ALL updated/marked_for_update\n rows).\n\n2. UPDATE/DELETE will go slower than now: server will have to\n exclusively lock (using lmgr) each row being updated, update/delete\n it and release lmgr' lock after that (note that currently\n lmgr locking is not used for rows). On the other hand, \n SELECT FOR SHARE LOCK could mark locked rows in buffer\n (t_infomask |= HEAP_MARKED_FOR_SHARE) and so UPDATE/DELETE\n could acquire lmgr lock only if row is marked...\n\nComments ?\n\nVadim\n",
"msg_date": "Tue, 05 Jan 1999 20:37:58 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> So, what about FOR SHARE LOCK clause in SELECTs?\n> Like FOR UPDATE clause, FOR SHARE LOCK clause\n> will lock selected rows, but in _share_ mode.\n> This is some kind of read-locking on demand.\n> 1. FOR SHARE LOCK will use lock manager to\n> lock rows, but lmgr can't handle \"too many\" locks.\n> (note that UPDATE, DELETE & SELECT FOR UPDATE use only\n> one entry in lmgr table for ALL updated/marked_for_update\n> rows).\n> 2. UPDATE/DELETE will go slower than now: server will have to\n> exclusively lock (using lmgr) each row being updated, update/delete\n> it and release lmgr' lock after that (note that currently\n> lmgr locking is not used for rows). On the other hand,\n> SELECT FOR SHARE LOCK could mark locked rows in buffer\n> (t_infomask |= HEAP_MARKED_FOR_SHARE) and so UPDATE/DELETE\n> could acquire lmgr lock only if row is marked...\n\nIf one did not specify FOR SHARE LOCK then the performance would be\nsimilar to the current performance? Then if you want to use the feature,\nyou pay the performance penalty. Or do you think that the checking\nshould/will happen irrespective of any query settings?\n\nAlso, my commercial Ingres system would escalate row-level locks to\npage- and table-level locks when the number of row-level locks exceeded\na (settable) threshold. That seems like a nice compromise between\nfeatures and performance, but it can lead to deadlock during the lock\nescalation...\n\n - Tom\n",
"msg_date": "Tue, 05 Jan 1999 14:55:40 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "\"Thomas G. Lockhart\" wrote:\n> \n> > So, what about FOR SHARE LOCK clause in SELECTs?\n> > Like FOR UPDATE clause, FOR SHARE LOCK clause\n> > will lock selected rows, but in _share_ mode.\n> > This is some kind of read-locking on demand.\n> > 1. FOR SHARE LOCK will use lock manager to\n> > lock rows, but lmgr can't handle \"too many\" locks.\n> > (note that UPDATE, DELETE & SELECT FOR UPDATE use only\n> > one entry in lmgr table for ALL updated/marked_for_update\n> > rows).\n> > 2. UPDATE/DELETE will go slower than now: server will have to\n> > exclusively lock (using lmgr) each row being updated, update/delete\n> > it and release lmgr' lock after that (note that currently\n> > lmgr locking is not used for rows). On the other hand,\n> > SELECT FOR SHARE LOCK could mark locked rows in buffer\n> > (t_infomask |= HEAP_MARKED_FOR_SHARE) and so UPDATE/DELETE\n> > could acquire lmgr lock only if row is marked...\n> \n> If one did not specify FOR SHARE LOCK then the performance would be\n> similar to the current performance? Then if you want to use the feature,\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nYes, if FOR SHARE LOCK will mark locked row in buffer\n(and so - force buffer write)...\n\n> you pay the performance penalty. Or do you think that the checking\n> should/will happen irrespective of any query settings?\n> \n> Also, my commercial Ingres system would escalate row-level locks to\n> page- and table-level locks when the number of row-level locks exceeded\n> a (settable) threshold. That seems like a nice compromise between\n> features and performance, but it can lead to deadlock during the lock\n> escalation...\n\nAs you can see in old lock.c postgres supposed to do\nrow --> table lock escalation if #rows locked in\ntable equal 10 - very simple and bad -:)\n\nI wouldn't like to care about # of SHARE LOCKed rows,\nat least in 6.5. Actually, this number shouldn't be\ntoo big. In the case of referential integrity, FOR SHARE LOCK\nshould be used only on primary table and only if user \ninserts/updates foreign table, but primary key is unique...\n\nVadim\n",
"msg_date": "Tue, 05 Jan 1999 22:34:19 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> Ok, in multi-version systems readers never lock\n> selected rows and so never block writers. Nice but\n> cause problems in some cases: if you want\n> to implement referential integrity at the\n> application level then you'll have to use\n> LOCK TABLE IN SHARE MODE or SELECT FOR UPDATE to\n> prevent updation of primary keys etc. Not so good...\n\n\nI was wondering if there is another perspective\nto look upon this problem.\n\nAssertion:\n\nThe primary difference between \"primary key\" and\n\"unique key\" is that primary keys are _never_\nupdated (during transaction processing) where \nunique keys may be. \n\nQuestion:\n\nIt seems that the techinical solution here may\nnot be better locking, but rather a mechinism\nto prevent updates on primary keys unless the\nentire table is locked, or some other dramatic\ngesture for non-transaction processing.\n\nThis leaves the issue of delete looming, so \nmabye the suggestion won't help.\n\nClark\n",
"msg_date": "Tue, 05 Jan 1999 15:54:29 +0000",
"msg_from": "Clark Evans <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "Let me see if I understand what MVCC(Multi-level concurrency control\nis). I looked in \"Transaction Processing: Concepts and Techniques\",\nand saw some mention on pages 435-437, but not much more.\n\nIn MVCC, SELECT's do not share lock the table, allowing UPDATE's during\nthe SELECT. This is done by having the SELECT sequential scan look at\nrows that are committed with transaction ids less than their own, or\nsuperseded rows that have a superseded id greater than their own. The\nonly lock a SELECT does it to prevent a vacuum during its table scan.\n\nMy assumption is table writes still require an exclusive lock, but\nbecause SELECT does not need a lock, both can occur at the same time.\n(Sounds like my deadlock and lock queue code may need tweaking.)\n\nYour stated problem is that someone in a transaction doing a SELECT is\nnot getting a shared lock on the rows he is selecting, so they could\nchange while inside the transaction. This is a valid concern.\n\nUsually, doing the SELECT FOR UPDATE, even though you are not going to\nupdate the table is used. You are suggesting SELECT FOR SHARE LOCK, but\nbecause SELECT's don't need a lock anymore, isn't that the same as a FOR\nUPDATE in an MVCC system? Is the problem that SHARE LOCK does not\nmodify the tuple, so it is harder to lock the rows?\n\n\n> As you can see in old lock.c postgres supposed to do\n> row --> table lock escalation if #rows locked in\n> table equal 10 - very simple and bad -:)\n> \n> I wouldn't like to care about # of SHARE LOCKed rows,\n> at least in 6.5. Actually, this number shouldn't be\n> too big. In the case of referential integrity, FOR SHARE LOCK\n> should be used only on primary table and only if user \n> inserts/updates foreign table, but primary key is unique...\n\nI think lock escalation is nice. Locking every row makes for lock\nresource problems. I would recommend locking a single row, and if a\nsecond row needs to be locked, just escalate to lock the whole table...\nif that can be done. This would seem to be the most reasonable and\neasiest to do.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 14:32:24 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "On Tue, 5 Jan 1999, Bruce Momjian wrote:\n\n> Let me see if I understand what MVCC(Multi-level concurrency control\n> is). I looked in \"Transaction Processing: Concepts and Techniques\",\n> and saw some mention on pages 435-437, but not much more.\n> \n> In MVCC, SELECT's do not share lock the table, allowing UPDATE's during\n> the SELECT. This is done by having the SELECT sequential scan look at\n> rows that are committed with transaction ids less than their own, or\n> superseded rows that have a superseded id greater than their own. The\n> only lock a SELECT does it to prevent a vacuum during its table scan.\n> \n> My assumption is table writes still require an exclusive lock, but\n> because SELECT does not need a lock, both can occur at the same time.\n> (Sounds like my deadlock and lock queue code may need tweaking.)\n> \n> Your stated problem is that someone in a transaction doing a SELECT is\n> not getting a shared lock on the rows he is selecting, so they could\n> change while inside the transaction. This is a valid concern.\n> \n> Usually, doing the SELECT FOR UPDATE, even though you are not going to\n> update the table is used. You are suggesting SELECT FOR SHARE LOCK, but\n> because SELECT's don't need a lock anymore, isn't that the same as a FOR\n> UPDATE in an MVCC system? Is the problem that SHARE LOCK does not\n> modify the tuple, so it is harder to lock the rows?\n> \n> \n> > As you can see in old lock.c postgres supposed to do\n> > row --> table lock escalation if #rows locked in\n> > table equal 10 - very simple and bad -:)\n> > \n> > I wouldn't like to care about # of SHARE LOCKed rows,\n> > at least in 6.5. Actually, this number shouldn't be\n> > too big. In the case of referential integrity, FOR SHARE LOCK\n> > should be used only on primary table and only if user \n> > inserts/updates foreign table, but primary key is unique...\n> \n> I think lock escalation is nice. Locking every row makes for lock\n> resource problems. I would recommend locking a single row, and if a\n> second row needs to be locked, just escalate to lock the whole table...\n> if that can be done. This would seem to be the most reasonable and\n> easiest to do.\n\nMaking two assumption here...first is that your explanation of MVCC is\ncorrect, second one being that my understanding of your explannation is\ncorrect...\n\nIf you are going to set the 'table lock' at 2...why not just do the table\nlock period? From what youexplain above, a table lock won't affect a\nread, only other writes...?\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Tue, 5 Jan 1999 16:10:50 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> Let me see if I understand what MVCC(Multi-level concurrency control\n> is). I looked in \"Transaction Processing: Concepts and Techniques\",\n> and saw some mention on pages 435-437, but not much more.\n> \n> In MVCC, SELECT's do not share lock the table, allowing UPDATE's during\n> the SELECT. This is done by having the SELECT sequential scan look at\n> rows that are committed with transaction ids less than their own, or\n> superseded rows that have a superseded id greater than their own. The\n> only lock a SELECT does it to prevent a vacuum during its table scan.\n> \n> My assumption is table writes still require an exclusive lock, but\n> because SELECT does not need a lock, both can occur at the same time.\n> (Sounds like my deadlock and lock queue code may need tweaking.)\n> \n> Your stated problem is that someone in a transaction doing a SELECT is\n> not getting a shared lock on the rows he is selecting, so they could\n> change while inside the transaction. This is a valid concern.\n> \n> Usually, doing the SELECT FOR UPDATE, even though you are not going to\n> update the table is used. You are suggesting SELECT FOR SHARE LOCK, but\n> because SELECT's don't need a lock anymore, isn't that the same as a FOR\n> UPDATE in an MVCC system? Is the problem that SHARE LOCK does not\n> modify the tuple, so it is harder to lock the rows?\n\nI hate to reply to my own posting, but I must.\n\nI have just read the Date book, Introduction to Database Systems about\nMVCC, and it confirms my above posting. Date states the MVCC\nadvantages:\n\n\t* Reads are never delayed(in particular, they are not delayed by\n\t any concurrent long transaction)\n\n\t* Reads never delay updates(in particular, they do not delay any\n\t concurrent long transaction)\n\n\t* It is never necessary to roll back a read-only transaction\n\n\t* Deadlock is possible only between update transactions\n\n\nThis is an amazing leap forward.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 15:42:35 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> > I think lock escalation is nice. Locking every row makes for lock\n> > resource problems. I would recommend locking a single row, and if a\n> > second row needs to be locked, just escalate to lock the whole table...\n> > if that can be done. This would seem to be the most reasonable and\n> > easiest to do.\n> \n> Making two assumption here...first is that your explanation of MVCC is\n> correct, second one being that my understanding of your explannation is\n> correct...\n> \n> If you are going to set the 'table lock' at 2...why not just do the table\n> lock period? From what youexplain above, a table lock won't affect a\n> read, only other writes...?\n\nGood point. I am assuming he is doing some kind of row-level locking\nfor shared and write locks. I can only guess this from his statement\nthat shared locking of every row would be a problem.\n\nSounds like my explaination may be wrong, because it is saying he has\nsome kind of row-locking going, perhaps for writes. Maybe he is using\nthe fact that if a writer is going to update a row that has a\nsuperceeded transaction id that is marked 'in progress' the writer has\nto wait for the transaction to finish. If you do this, muliple writers\ncan update at the same time, making MVCC better than row-level locking\nsystems.\n\nReaders don't block writers, and multiple writers can write as long as\nthey are not touching the same rows.\n\nIn this scenario, shared locks are tricky, because the above system does\nnot work. You have to do some explicit locking, because reading does\nnot set anything on the row.\n\nVadim is sleeping now, so I assume we will hear something from him this\nevening.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 15:58:34 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?]"
},
{
"msg_contents": "OK, I have retrieved Vadim's original proposal for LLL from July. (No\nwonder I didn't remember it.)\n\nHe basically describes the MVCC system at the bottom, and at the top, I\nam discussing a way to avoid a mapping of SCN's to XID's. There are\nother postings that finally resolved the issue. At least Vadim didn't\nsay I was way off, so somehow he resolved it.\n\nThis is an great benefit. We don't have to apologize about table-level\nlocking anymore. I didn't think it could be done, and Vadim's changes\nkind of came as a surprise to me, probably because I didn't understand\nthe ramifications when discussing it in July.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n\nRe: [HACKERS] proposals for LLL, part 1\n\n ------------------------------------------------------------------------\n\n * From: Bruce Momjian <[email protected]>\n * To: [email protected] (Vadim Mikheev)\n * Subject: Re: [HACKERS] proposals for LLL, part 1\n * Date: Thu, 16 Jul 1998 12:14:51 -0400 (EDT)\n\n ------------------------------------------------------------------------\n\nI am retaining your entire message here for reference.\n\nI have a good solution for this. It will require only 4k of shared\nmemory, and will have no restrictions on the age or number of\ntransactions.\n\nFirst, I think we only want to implement \"read committed isolation\nlevel\", not serialized. Not sure why someone would want serialized.\n\nOK, when a backend is looking at a row that has been committed, it must\ndecide if the row was committed before or after my transaction started.\nIf the transaction commit id(xmin) is greater than our current xid, we\nknow we should not look at it because it is for a transaction that\nstarted after our own transaction.\n\nThe problem is for transactions started before our own (have xmin's less\nthan our own), and may have committed before or after our transaction.\n\nHere is my idea. We add a field to the shared memory Proc structure\nthat can contain up to 32 transaction ids. When a transaction starts,\nwe spin though all other open Proc structures, and record all\ncurrently-running transaction ids in our own Proc field used to store up\nto 32 transaction ids. While we do this, we remember the lowest of\nthese open transaction ids.\n\nThis is our snapshot of current transactions at the time our transaction\nstarts. While analyzing a row, if it is greater than our transaction\nid, then the transaction was not even started before our transaction.\nIf the xmin is lower than the min transaction id that we remembered from\nthe Proc structures, it was committed before our transaction started.\nIf it is greater than or equal to the min remembered transaction id, we\nmust spin through our stored transaction ids. If it is in the stored\nlist, we don't look at the row, because that transaction was not\ncommitted when we started our transaction. If it is not in the list, it\nmust have been committed before our transaction started. We know this\nbecause if any backend starting a transaction after ours would get a\ntransaction id higher than ours.\n\nComments?\n\n> Ok, I'm not sure that LLL will appear in 6.4 but it's good time to\n> discuss about it.\n>\n> First, PostgreSQL is multi-version system due to its\n> non-overwriting storage manager. And so, first proposal is use\n> this feature (multi-versioning) in LLL implementation.\n>\n> In multi-version systems access methods don't use locks to read\n> consistent data and so readers don't block writers, writers don't\n> block readers and only the same-row writers block writers. In such\n> systems access methods returns snapshot of data as they were in\n> _some_ point in time. For read committed isolation level this\n> moment is the time when statement began. For serialized isolation\n> level this is the time when current transaction began.\n>\n> Oracle uses rollback segments to reconstract blocks that were\n> changed after statement/transaction began and so statement sees\n> only data that were committed by then.\n>\n> In our case we have to analyze tuple xmin/xmax to determine _when_\n> corresponding transaction was committed in regard to the last\n> transaction (LCX) that was committed when statement/transaction\n> began.\n>\n> If xmin/xmax was committed before LCX then tuple\n> insertion/deletion is visible to statement, else - not visible.\n>\n> To achieve this, the second proposal is to use special SCN -\n> System Change Number (C) Oracle :) - that will be incremented by 1\n> by each transaction commit. Each commited transaction will have\n> corresponding SCN (4 bytes -> equal to sizeof XID).\n>\n> We have to keep XID --> SCN mapping as long as there is running\n> transaction that is \"interested\" in XID: when transaction begins\n> it will determine the first (the oldest) running transaction XID\n> and this will be the minimum XID whose SCN transaction would like\n> to know.\n>\n> Access methods will have to determine SCN for xmin/xmax only if\n> FRX <= xmin/xmax <= LSX, where FRX is XID of first (oldest)\n> running transactions and LSX is last started transaction - in the\n> moment when statement (for read committed) or transaction (for\n> serialized) began. For such xmin/xmax their SCNs will be compared\n> with SCN determined in the moment of statement/transaction\n> begin...\n>\n> Changes made by xmin/xmax < FRX are visible to\n> statement/transaction, and changes made by xmin/xmax > LSX are not\n> visible. Without xmin/xmax SCN lookup.\n>\n> For XID --> SCN mapping I propose to use the simplest schema:\n> ordered queue of SCNs (or something like this) - i.e. keep SCNs\n> for all transactions from the first one whose SCN could be\n> required by some running transaction to the last started.\n>\n> This queue must be shared!\n>\n> The size of this queue and average number of commits/aborts per\n> second will define how long transactions will be able to run. 30\n> xacts/sec and 400K of queue will enable 30 - 60 minuts running\n> transactions...\n>\n> Keeping queue in shared memmory may be unacceptable in some\n> cases... mmap or shared buffer pool could be used to access queue.\n> We'll see...\n>\n> Also note that Oracle has special READ ONLY transactions mode.\n> READ ONLY transactions are disallowed to change anything in the\n> database. This is good mode for pg_dump (etc) long running\n> applications. Because of no one will be \"interested\" in SCN of\n> READ ONLY transactions - such transactions can make private copy\n> of the queue part and after this queue could be truncated...\n>\n> Having 4 bytes per SCN enable to use special values to mark\n> corresponding transaction as running or aborted and avoid pg_log\n> lookup when we need in both SCN and state of transaction.\n>\n> ...Well, it's time to sleep :)\n>\n> To be continued...\n>\n> Comments ?\n>\n> Vadim\n>\n>\n\n--\nBruce Momjian | 830 Blythe Avenue\[email protected] | Drexel Hill, Pennsylvania 19026\n + If your life is a hard drive, | (610) 353-9879(w)\n + Christ can be your backup. | (610) 853-3000(h)\n\n ------------------------------------------------------------------------\n\n * Prev by Date: How about re-moderating pgsql-announce?\n * Next by Date: Re: [HACKERS] How about re-moderating pgsql-announce?\n * Prev by thread: proposals for LLL, part 1\n * Next by thread: Re: [HACKERS] proposals for LLL, part 1\n * Index(es):\n o Date\n o Thread\n\nHome | Main Index | Thread Index",
"msg_date": "Tue, 5 Jan 1999 20:30:03 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?]"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> Let me see if I understand what MVCC(Multi-level concurrency control\n> is). I looked in \"Transaction Processing: Concepts and Techniques\",\n> and saw some mention on pages 435-437, but not much more.\n\nYes, I use Oracle documentation to learn about MVCC.\nNevertheless, thank you for this book - there are so many\nideas there!\n\n> In MVCC, SELECT's do not share lock the table, allowing UPDATE's during\n> the SELECT. This is done by having the SELECT sequential scan look at\n> rows that are committed with transaction ids less than their own, or\n> superseded rows that have a superseded id greater than their own. The\n> only lock a SELECT does it to prevent a vacuum during its table scan.\n> \n> My assumption is table writes still require an exclusive lock, but\n> because SELECT does not need a lock, both can occur at the same time.\n> (Sounds like my deadlock and lock queue code may need tweaking.)\n> \n> Your stated problem is that someone in a transaction doing a SELECT is\n> not getting a shared lock on the rows he is selecting, so they could\n> change while inside the transaction. This is a valid concern.\n> \n> Usually, doing the SELECT FOR UPDATE, even though you are not going to\n> update the table is used. You are suggesting SELECT FOR SHARE LOCK, but\n> because SELECT's don't need a lock anymore, isn't that the same as a FOR\n> UPDATE in an MVCC system? Is the problem that SHARE LOCK does not\n\nOnly one running transaction can mark row for update, SHARE LOCK\ncould be acquired by many transactions and so, using\nreferential integrity for example, many transactions could\nwork with the same foreign key simultaneously.\n\n> modify the tuple, so it is harder to lock the rows?\n> \n> > As you can see in old lock.c postgres supposed to do\n> > row --> table lock escalation if #rows locked in\n> > table equal 10 - very simple and bad -:)\n> >\n> > I wouldn't like to care about # of SHARE LOCKed rows,\n> > at least in 6.5. Actually, this number shouldn't be\n> > too big. In the case of referential integrity, FOR SHARE LOCK\n> > should be used only on primary table and only if user\n> > inserts/updates foreign table, but primary key is unique...\n> \n> I think lock escalation is nice. Locking every row makes for lock\n> resource problems. I would recommend locking a single row, and if a\n> second row needs to be locked, just escalate to lock the whole table...\n> if that can be done. This would seem to be the most reasonable and\n> easiest to do.\n\nEasiest to do is don't worry about # of locks -:)\nLet's be on this way for 6.5\n\nVadim\n",
"msg_date": "Wed, 06 Jan 1999 10:47:47 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> > I think lock escalation is nice. Locking every row makes for lock\n> > resource problems. I would recommend locking a single row, and if a\n> > second row needs to be locked, just escalate to lock the whole table...\n> > if that can be done. This would seem to be the most reasonable and\n> > easiest to do.\n> \n> Easiest to do is don't worry about # of locks -:)\n> Let's be on this way for 6.5\n\nYou mean just share-lock the whole table. I agree. It is a pretty rare\nsituation.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 22:50:36 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> > > I think lock escalation is nice. Locking every row makes for lock\n> > > resource problems. I would recommend locking a single row, and if a\n> > > second row needs to be locked, just escalate to lock the whole table...\n> > > if that can be done. This would seem to be the most reasonable and\n> > > easiest to do.\n> >\n> > Easiest to do is don't worry about # of locks -:)\n> > Let's be on this way for 6.5\n> \n> You mean just share-lock the whole table. I agree. It is a pretty rare\n> situation.\n\nNo. User may use LOCK TABLE IN SHARE MODE for this.\nI propose SELECT FOR SHARE LOCK as alternative to\nLOCK TABLE IN SHARE MODE and SELECT FOR UPDATE and\nwould like to share lock each row selected with\nFOR SHARE LOCK clause in use. I don't know what's\nreal limitations of # locks, but I think that\na tens of locks is Ok.\n\nVadim\n",
"msg_date": "Wed, 06 Jan 1999 11:16:10 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> Bruce Momjian wrote:\n> > \n> > > > I think lock escalation is nice. Locking every row makes for lock\n> > > > resource problems. I would recommend locking a single row, and if a\n> > > > second row needs to be locked, just escalate to lock the whole table...\n> > > > if that can be done. This would seem to be the most reasonable and\n> > > > easiest to do.\n> > >\n> > > Easiest to do is don't worry about # of locks -:)\n> > > Let's be on this way for 6.5\n> > \n> > You mean just share-lock the whole table. I agree. It is a pretty rare\n> > situation.\n> \n> No. User may use LOCK TABLE IN SHARE MODE for this.\n> I propose SELECT FOR SHARE LOCK as alternative to\n> LOCK TABLE IN SHARE MODE and SELECT FOR UPDATE and\n> would like to share lock each row selected with\n> FOR SHARE LOCK clause in use. I don't know what's\n> real limitations of # locks, but I think that\n> a tens of locks is Ok.\n\nSo you are going to shared lock every row. And if a user does a\nsequential scan of the entire table using SELECT FOR SHARE LOCK, he\nshared locks every row. Isn't he going to run out of locks?\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 23:25:31 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> >\n> > If you are going to set the 'table lock' at 2...why not just do the table\n> > lock period? From what youexplain above, a table lock won't affect a\n> > read, only other writes...?\n> \n> Good point. I am assuming he is doing some kind of row-level locking\n> for shared and write locks. I can only guess this from his statement\n> that shared locking of every row would be a problem.\n> \n> Sounds like my explaination may be wrong, because it is saying he has\n> some kind of row-locking going, perhaps for writes. Maybe he is using\n> the fact that if a writer is going to update a row that has a\n> superceeded transaction id that is marked 'in progress' the writer has\n> to wait for the transaction to finish. If you do this, muliple writers\n> can update at the same time, making MVCC better than row-level locking\n> systems.\n\nYes. When transaction begins it places exclusive lock\nfor transaction ID in pseudo-table (no wait because of ID\nis unique). If other writer sees that t_xmax is valid \n(and not committed/aborted) transaction ID then it tries \nto place share lock for t_xmax in this pseudo-table.\nIf t_xmax is running then concurrent writer will wait for\nt_xmax commit/abort when exclusive lock for t_xmax\nis released. With this technique writers use single\nlmgr entry for all updated rows - just to let other\nthe-same-row writers know when transaction commits/aborts.\n\n> \n> Readers don't block writers, and multiple writers can write as long as\n> they are not touching the same rows.\n\nYes.\n\n> In this scenario, shared locks are tricky, because the above system does\n> not work. You have to do some explicit locking, because reading does\n> not set anything on the row.\n\nYes, but this will be user' decision to use FOR SHARE LOCK\n(to lock rows explicitly) or not.\n\nVadim\n",
"msg_date": "Wed, 06 Jan 1999 11:25:47 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?]"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> > > >\n> > > > Easiest to do is don't worry about # of locks -:)\n> > > > Let's be on this way for 6.5\n> > >\n> > > You mean just share-lock the whole table. I agree. It is a pretty rare\n> > > situation.\n> >\n> > No. User may use LOCK TABLE IN SHARE MODE for this.\n> > I propose SELECT FOR SHARE LOCK as alternative to\n> > LOCK TABLE IN SHARE MODE and SELECT FOR UPDATE and\n> > would like to share lock each row selected with\n> > FOR SHARE LOCK clause in use. I don't know what's\n> > real limitations of # locks, but I think that\n> > a tens of locks is Ok.\n> \n> So you are going to shared lock every row. And if a user does a\n> sequential scan of the entire table using SELECT FOR SHARE LOCK, he\n> shared locks every row. Isn't he going to run out of locks?\n\nI would like to work with this issue after 6.5 and writes\nsome notes about FOR SHARE LOCK limitations/problems.\n\nVadim\n",
"msg_date": "Wed, 06 Jan 1999 11:28:27 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "Clark Evans wrote:\n> \n> > Ok, in multi-version systems readers never lock\n> > selected rows and so never block writers. Nice but\n> > cause problems in some cases: if you want\n> > to implement referential integrity at the\n> > application level then you'll have to use\n> > LOCK TABLE IN SHARE MODE or SELECT FOR UPDATE to\n> > prevent updation of primary keys etc. Not so good...\n> \n> I was wondering if there is another perspective\n> to look upon this problem.\n> \n> Assertion:\n> \n> The primary difference between \"primary key\" and\n> \"unique key\" is that primary keys are _never_\n ^^^^^^^^^^^^^^^^^^^^^^^^\n> updated (during transaction processing) where\n ^^^^^^^\n> unique keys may be.\n\n1. Is this standard requirement?\n2. Note that foreign keys may reference unique key,\n not just primary one...\n3. I told about implementing referential\n integrity _at_the_application_level_, not by the\n DB system itself - it's up to the user decide\n what's allowed and what's not, in this case.\n\n> Question:\n> \n> It seems that the techinical solution here may\n> not be better locking, but rather a mechinism\n> to prevent updates on primary keys unless the\n> entire table is locked, or some other dramatic\n> gesture for non-transaction processing.\n> \n> This leaves the issue of delete looming, so\n> mabye the suggestion won't help.\n\nYes. \n\nVadim\n",
"msg_date": "Wed, 06 Jan 1999 11:40:53 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> > So you are going to shared lock every row. And if a user does a\n> > sequential scan of the entire table using SELECT FOR SHARE LOCK, he\n> > shared locks every row. Isn't he going to run out of locks?\n> \n> I would like to work with this issue after 6.5 and writes\n> some notes about FOR SHARE LOCK limitations/problems.\n\nOK, just checking. You can't use the xid, so I see the problem with\nshared-locking. No way to know which rows were seen by SELECT, so no\nway to shared-lock them without an explicit lock.\n\nWith Informix, UPDATE automatically locks every row, and you quickly run\nout of locks. You have to explicitly lock the table in EXCLUSIVE MODE\nin a transaction to prevent your UPDATE from running out of locks on a\nlarge transaction. And you can still run out of log space or generate a\n'large transaction' errror because the log got full before the\ntransaction finished. With MVCC, these are not problems, and FOR SHARED\nLOCK is just a special thing people can enable for special cases.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 23:45:35 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "Vadim Mikheev wrote:\n> Clark Evans wrote:\n> > > Ok, in multi-version systems readers never lock\n> > > selected rows and so never block writers. Nice but\n> > > cause problems in some cases: if you want\n> > > to implement referential integrity at the\n> > > application level then you'll have to use\n> > > LOCK TABLE IN SHARE MODE or SELECT FOR UPDATE to\n> > > prevent updation of primary keys etc. Not so good...\n> >\n> > I was wondering if there is another perspective\n> > to look upon this problem.\n> >\n> > Assertion:\n> >\n> > The primary difference between \"primary key\" and\n> > \"unique key\" is that primary keys are _never_\n> ^^^^^^^^^^^^^^^^^^^^^^^^\n> > updated (during transaction processing) where\n> ^^^^^^^\n> > unique keys may be.\n> \n> 1. Is this standard requirement?\n\nA DBA at Ford that I had many afternoon chats with\nwished he could make the Oracle database prevent\nupdates of primary keys, short of a full table \nlock -- which is a privilige. I'm sure that there\nare others that might disagree.\n\nThe other thing that the DBA said he'd love was\na \"garbage collector\". Where a delete on row would\nonly \"hide\" the row after all constraint checks \nsucceeded. In this way, data referenced by systems\nthat are not attached to the database have a chance\nto be checked for data integrety concerns.\n\nAlthough you may be able to find DBA's that would\nargue about these points, the idea that primary\nkeys are updated only on rare exception is a solid \nmodeling practice (it's one of Oracle's Ten Commandments).\n\n> 2. Note that foreign keys may reference unique key,\n> not just primary one...\n\nYou may be able to do it... but I'm not sure that\nit makes sence. Also, unique keys can have NULLS,\nprimary keys cannot have any NULL allowable columns\n\nThe idea of a primary key is to identify column(s)\nas a constant pointer to the object in question,\nso that other objects can use that pointer.\n\nThe idea of a unique index is to enforce \nconstraints that a combination of columns must\nbe unique. The mechinism does not necessarily\nmean that the columns are \"constant\", and thus\nmake good pointers. For instance, a full name \nmay very well be unique in a small company - if\nthis assumption is made in the reporting code, \nthen a unique key is warranted, with clear \ndocumentation explaining that the reporting\ncode assumes this fact. However, it would \nbe a bad idea to use the full name as a \npointer, as a marriage could cause havoc.\n\n> 3. I told about implementing referential\n> integrity _at_the_application_level_, not by the\n> DB system itself - it's up to the user decide\n> what's allowed and what's not, in this case.\n\nI think you are right here, although allowing the\nuser to make an application level decision and\nthen configure the database to automagically \nenforce this decision is golden.\n\nMore e-mail to follow...\n",
"msg_date": "Wed, 06 Jan 1999 06:49:00 +0000",
"msg_from": "Clark Evans <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "Vadim Mikheev wrote:\n> Clark Evans wrote:\n> > > Ok, in multi-version systems readers never lock\n> > > selected rows and so never block writers. Nice but\n> > > cause problems in some cases: if you want\n> > > to implement referential integrity at the\n> > > application level then you'll have to use\n> > > LOCK TABLE IN SHARE MODE or SELECT FOR UPDATE to\n> > > prevent updation of primary keys etc. Not so good...\n\nI'm trying to understand. Please excuse my ignorance\nof database implementation issues. From a application\ndevelopers's perspective:\n\nSuppose that I have three tables: ORDER, PRODUCT, \nand ORDER_LINE. ORDER_NO is the primary key of ORDER,\nPRODUCT_NO is the primary key for PRODUCT, and ORDER_NO\nand PRODUCT_NO are foreign keys in ORDER_LINE.\n\nNow picture a data entry person with an order entry\nscreen, with order information above, a grid with\na row for each order line, and a drop down list box\nfor each possible product.\n\nThe problem you identified above would happen if\nwhile Lucy was inserting order lines for ORDER_NO=100\nPRODUCT_NO=230, someone else changes the order_no\nfrom 100 to 101 in the order table _or_ if someone \nelse deletes from product where product_no = 230.\n\nThere are a number of solutions to this pattern.\n\na) Pessimistic - In this pattern, when a child is\nabout to be edited or added, the parent record is\nlocked. When the updates to the child table are\ndone, a commit frees the parent. This is coupled\nwith code to check to see if the row is locked \nwhen it is read, if so, then the parent and the\nchild are brought back \"read-only\" with a status\nof \"being updated...\" This solution is great when\nthe foreign key points to a \"natural\" parent, such\nas an order pointing to an order_line... where deleting\nthe order would cascade to the order lines. The \nsolution works wonderfully when the parent and child \nare on the same screen.\n\nb) Optimistic - In this pattern, when a child is \ninserted/updated and a parent key is not found, \nan exception is thrown by the database - \"foreign\nkey violation\". In this case, the client application\nverifies which foreign key it was, and informs the\nuser about the failure: \"Oval Vase (203) is no\nlonger a valid product. Order line addition\ncanceled. Updating product list...\" This solution\nworks well when the object being referenced is \nmaintained by a different group or is configured\nby a different part of the application. In this\ncase, you wouldn't want the product database to\nbe constantly locked up as people are ordering \nproducts... fixing a product name would be a pain!\nIn general, if deleting the parent would not cascade\nto the child, this pattern is the best one to use.\nThis solution works great if the parent/child are\non different screens.\n\nIn any case, I'm not really sure what this locking\nin share mode would get you... could you explain\nin the context of the above examples.\n\nThanks!\n\nClark\n",
"msg_date": "Wed, 06 Jan 1999 07:24:27 +0000",
"msg_from": "Clark Evans <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
},
{
"msg_contents": "> In any case, I'm not really sure what this locking\n> in share mode would get you... could you explain\n> in the context of the above examples.\n\nI will take a stab at it.\n\nLocking in shared mode would cause any row read by the SELECT to be\nremain unmodified until its transaction completes. Other SELECTs can\nread it, but writer have to wait and can not modify it.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 6 Jan 1999 02:43:48 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOR SHARE LOCK clause ?"
}
] |
[
{
"msg_contents": "\nHello there,\n\nI have a problem using pl/pgsql (distributed in version 6.4.1).\n\nWe are using PostgreSQL for an ordering-system as a part of a new data warehouse. At this moment we have about 10K rows and it works great!\n\nNow we are testing it for a much amount of data (sales report) of about 1.5GB in size.\n\nMy first question is: is this the right place to post such bug reports?\n\nOk, following pl/pgsql statements doesn�t work:\n\n--------\ncreate table t1 (i int, a int, b int);\nCREATE\ncreate table t2 (i int, x int, y int);\nCREATE\n\ninsert into t1 values(1, 1,10);\nINSERT 35140214 1\ninsert into t1 values(1, 2,10);\nINSERT 35140215 1\ninsert into t1 values(2, 3,10);\nINSERT 35140216 1\ninsert into t1 values(2, 4,10);\nINSERT 35140217 1\n\ncreate function func1()\nreturns int\nas '\n declare\n begin\n insert into t2\n select i,\n sum(a) as x,\n sum(b) as y\n from t1\n group by i;\n return 1;\n end;\n' language 'plpgsql';\nCREATE\n\n\nselect func1();\nfunc1\n-----\n 1\n(1 row)\n\n\nselect * from t2;\ni| x| y\n-+--+--\n1|20|20\n2|20|20\n(2 rows)\n\n-------------\n\nUsing the insert/select-statement directly (via psql) it works fine.\n\nAnd one additional (last) question:\n\nIn pl/sql writing select * into table2 from table1 ... seems to be interpreted that table2 is a pl/sql-variable and not a table. So it doesn�t work. Is it a bug or is there an other way to select directly into a (new) table. \n\n\nKind regards\n\nMichael Contzen\n\nDohle Handelsgruppe Systemberatung GmbH,\nGermany\n\n\n\n\n\n",
"msg_date": "Tue, 5 Jan 1999 16:14:57 -0000",
"msg_from": "Michael Contzen <[email protected]>",
"msg_from_op": true,
"msg_subject": "PL/PGSQL bug using aggregates"
}
] |
[
{
"msg_contents": "Hi,\n\n How do I drop a table under plpgsql ?\n When I invoke a function which contains eg: DROP TABLE TMP a get a\n backend crash no matter if this table exists or not.\n\n I can always do it the hard way and put DROP ... in an SQL function\n and then invoke DROP indirectly but do I have to ?\n\nPlease help me\nPawel Pierscionek\n\n\n",
"msg_date": "Tue, 5 Jan 1999 22:20:50 +0100",
"msg_from": "Pawel Pierscionek <[email protected]>",
"msg_from_op": true,
"msg_subject": "drop table in pgsql"
},
{
"msg_contents": ">\n> Hi,\n>\n> How do I drop a table under plpgsql ?\n> When I invoke a function which contains eg: DROP TABLE TMP a get a\n> backend crash no matter if this table exists or not.\n>\n> I can always do it the hard way and put DROP ... in an SQL function\n> and then invoke DROP indirectly but do I have to ?\n>\n> Please help me\n> Pawel Pierscionek\n\n Huh - intersting. Not that I think it's good to change the\n database schema from inside of functions, but if SQL\n functions can do it, PL/pgSQL should too. Will take a look at\n it.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Wed, 6 Jan 1999 11:19:38 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [SQL] drop table in pgsql"
},
{
"msg_contents": "On Wed, 6 Jan 1999, Jan Wieck wrote:\n\n> >\n> > Hi,\n> >\n> > How do I drop a table under plpgsql ?\n> > When I invoke a function which contains eg: DROP TABLE TMP a get a\n> > backend crash no matter if this table exists or not.\n> >\n> > I can always do it the hard way and put DROP ... in an SQL function\n> > and then invoke DROP indirectly but do I have to ?\n> >\n> > Please help me\n> > Pawel Pierscionek\n> \n> Huh - intersting. Not that I think it's good to change the\n> database schema from inside of functions, but if SQL\n> functions can do it, PL/pgSQL should too. Will take a look at\n> it.\n> \nI don't like changing db shema that way too, but there is no support for\ntemporary tables and I have to use drop before I go with select sth into\ntemp_db. \n\nPawel Pierscionek \n\n",
"msg_date": "Wed, 6 Jan 1999 15:43:55 +0100 (EET)",
"msg_from": "Pawel Pierscionek <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [SQL] drop table in pgsql"
},
{
"msg_contents": "Hi,\n\n there was recently discussion about schema changes (utility\n queries) inside of PL functions etc.\n\n I've thought about it some more and came to the conclusion\n that it doesn't make much sense to support something like\n that in PL/pgSQL.\n\n The reason for this is that PL/pgSQL does prepare and save\n ALL plans on the first call of the function. As it is now, a\n PL/pgSQL function used once in a connection will get broken\n if an object it uses gets dropped and recreated. The saved\n SPI plans still use the OLD object, which isn't available any\n more. So the function needs to be recompiled and that will\n only happen on a new backend connection.\n\n There is no syntax that could tell PL/pgSQL not to save a\n particular prepared plan (like in PL/Tcl where planning and\n saving is done separately, so the programmer has control\n which plans to save and which not).\n\n Since dropping an object in Postgres is a general problem\n that affects saved SPI plans (not only in PL, C language\n functions using the SPI saved plan feature will break too),\n I'll not add support for it to PL/pgSQL. I think this problem\n will also show up if any function using prepared plans will\n be used on the temp tables Bruce currently is working on.\n\n\n\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Mon, 1 Feb 1999 15:20:50 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "PL/pgSQL / SPI and UTILITY statements"
}
] |
[
{
"msg_contents": "\tI had a problme loading the data from my 6.3 (Not 6.3.1) DB inot my new\n\ttest 5.4.2 DB. Aparently some things related to users have changed\n\tbetween 6.3, and 6.3.1 which break this.\n\n\tI was able to get the data relaoded, by first doing a creatuse for all\n\tthe non-postgress users in the database. Question is this safe?\n\tQuestion 2 is there a better way to do it, such as using the pg_dumpall\n\tscript from 6.4.2 on the 6.3 DB?\n\n\tI reall don't want to do somehting here that is going to cause problems\n\tdown the reoad, and thier is really no big pressure to get upgraded, as\n\tthe 6.3 DB is working great.\n\n-- \nStan Brown [email protected] 843-745-3154\nWestvaco\nCharleston SC.\n-- \nWindows 98: n.\n\tuseless extension to a minor patch release for 32-bit extensions and\n\ta graphical shell for a 16-bit patch to an 8-bit operating system\n\toriginally coded for a 4-bit microprocessor, written by a 2-bit \n\tcompany that can't stand for 1 bit of competition.\n-\n(c) 1999 Stan Brown. Redistribution via the Microsoft Network is prohibited.\n",
"msg_date": "Tue, 5 Jan 1999 19:19:51 -0500 (EST)",
"msg_from": "\"Stan Brown\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "6.3 -> 6.4.2 upgrade question"
}
] |
[
{
"msg_contents": "OK, I give up :) How do I use the numeric type?\n\npostgres=> create table n1 (n numeric(10,5), d decimal(10,5));\nCREATE\npostgres=> insert into n1 values ('1.23456', '1.23456');\nERROR: overflow on numeric\n ABS(value) >= 10^0 for field with precision 2087 scale 31828\npostgres=> insert into n1 values ('12345.23456', '1.23456');\nERROR: overflow on numeric\n ABS(value) >= 10^4 for field with precision 2087 scale 50860\n\n - Tom\n",
"msg_date": "Wed, 06 Jan 1999 02:29:15 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Numeric type"
},
{
"msg_contents": "> OK, I give up :) How do I use the numeric type?\n> \n> postgres=> create table n1 (n numeric(10,5), d decimal(10,5));\n> CREATE\n> postgres=> insert into n1 values ('1.23456', '1.23456');\n> ERROR: overflow on numeric\n> ABS(value) >= 10^0 for field with precision 2087 scale 31828\n> postgres=> insert into n1 values ('12345.23456', '1.23456');\n> ERROR: overflow on numeric\n> ABS(value) >= 10^4 for field with precision 2087 scale 50860\n> \n\nWorks here. Don't you hate when that happens:\n\t\n\ttest=> create table n1 (n numeric(10,5), d decimal(10,5));\n\tCREATE\n\ttest=> insert into n1 values ('1.23456', '1.23456');\n\tINSERT 18602 1\n\ttest=> insert into n1 values ('12345.23456', '1.23456');\n\tINSERT 18603 1\n\ttest=> select * from n1;\n\t n| d\n\t-----------+-------\n\t 1.23456|1.23456\n\t12345.23456|1.23456\n\t(2 rows)\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 5 Jan 1999 21:33:28 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Numeric type"
},
{
"msg_contents": "> > OK, I give up :) How do I use the numeric type?\n> Works here. Don't you hate when that happens:\n\npostgres=> create table n1 (n numeric(10,5), d decimal(10,5));\nCREATE\npostgres=> insert into n1 values ('1.23456', '1.23456');\nERROR: overflow on numeric\n ABS(value) >= 10^0 for field with precision 2086 scale 53380\n\n*sigh* \n\nAny hints on where to look? I did a CVSup update, then a \"cvs -PdA\" on\nmy development tree, a \"make clean install\" and an initdb. I did not try\na full clean checkout, but would think that it wouldn't help.\n\nAnyone else running on Linux/i686/libc5 having success? If so, I'll go\nback to working on outer joins...\n\n - Tom\n",
"msg_date": "Wed, 06 Jan 1999 15:15:11 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Numeric type"
},
{
"msg_contents": "\"Thomas G. Lockhart\" <[email protected]> writes:\n>>>> OK, I give up :) How do I use the numeric type?\n>> Works here. Don't you hate when that happens:\n\n> Any hints on where to look? I did a CVSup update, then a \"cvs -PdA\" on\n> my development tree, a \"make clean install\" and an initdb.\n\nMaybe you need to repeat the \"configure\" step? If Jan made any changes\nin Makefile.in's, config.h.in, etc, then failing to reconfigure would\nmean building with old files.\n\nI tend to do \"make distclean\" before any CVS update and \"configure\"\nafterwards...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 06 Jan 1999 10:26:34 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Numeric type "
},
{
"msg_contents": ">\n> > > OK, I give up :) How do I use the numeric type?\n> > Works here. Don't you hate when that happens:\n>\n> postgres=> create table n1 (n numeric(10,5), d decimal(10,5));\n> CREATE\n> postgres=> insert into n1 values ('1.23456', '1.23456');\n> ERROR: overflow on numeric\n> ABS(value) >= 10^0 for field with precision 2086 scale 53380\n>\n> *sigh*\n>\n> Any hints on where to look? I did a CVSup update, then a \"cvs -PdA\" on\n> my development tree, a \"make clean install\" and an initdb. I did not try\n> a full clean checkout, but would think that it wouldn't help.\n\n Seems to me like the typmod value given to\n\n Numeric numeric(Numeric num, int dummy, int32 typmod)\n\n got mangled up somewhere. First you should check that the\n atttypmod in pg_attribute for n1.n and n1.d is correct. It\n should have the value\n\n ((precision << 16) | scale) + 4\n\n and must be 655369 for a numeric(10,5).\n\n If that isn't the case, you might take a look at\n include/nodes/parsenodes.h. The typmod field of struct\n TypeName should be of int32 (I changed that shortly before\n moving numeric into the backend).\n\n If that isn't the error, it would be good if you're a little\n familiar with gdb. If so, set a breakpoint on function\n numeric() and try to find where on it's way from\n pg_attribute.atttypmod to the call to numeric() it got\n mangled up.\n\n>\n> Anyone else running on Linux/i686/libc5 having success? If so, I'll go\n> back to working on outer joins...\n>\n> - Tom\n>\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Wed, 6 Jan 1999 16:48:50 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Numeric type"
},
{
"msg_contents": "> \n> *sigh* \n> \n> Any hints on where to look? I did a CVSup update, then a \"cvs -PdA\" on\n> my development tree, a \"make clean install\" and an initdb. I did not try\n> a full clean checkout, but would think that it wouldn't help.\n> \n> Anyone else running on Linux/i686/libc5 having success? If so, I'll go\n> back to working on outer joins...\n\nI assume it is some OS thing.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 6 Jan 1999 11:01:50 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Numeric type"
},
{
"msg_contents": "On Wed, 6 Jan 1999, Thomas G. Lockhart wrote:\n\n> Date: Wed, 06 Jan 1999 15:15:11 +0000\n> From: \"Thomas G. Lockhart\" <[email protected]>\n> To: Bruce Momjian <[email protected]>\n> Cc: [email protected], [email protected]\n> Subject: Re: [HACKERS] Numeric type\n> \n> > > OK, I give up :) How do I use the numeric type?\n> > Works here. Don't you hate when that happens:\n> \n> postgres=> create table n1 (n numeric(10,5), d decimal(10,5));\n> CREATE\n> postgres=> insert into n1 values ('1.23456', '1.23456');\n> ERROR: overflow on numeric\n> ABS(value) >= 10^0 for field with precision 2086 scale 53380\n> \n> *sigh* \n> \n> Any hints on where to look? I did a CVSup update, then a \"cvs -PdA\" on\n> my development tree, a \"make clean install\" and an initdb. I did not try\n> a full clean checkout, but would think that it wouldn't help.\n> \n> Anyone else running on Linux/i686/libc5 having success? If so, I'll go\n> back to working on outer joins...\n> \n\nOn my Linux box/i586/libc5, postgresql 6.4.2:\n\ntest=> create table n1 (n numeric(10,5), d decimal(10,5));\nERROR: NUMERIC precision 10 must be 9\ntest=>\n\nSomething weird ! I compiled postgres 6.4.2 after applying Jan's feature\npatch and configure --enable-locale\n\n\tOleg\n\n> - Tom\n> \n\n_____________________________________________________________\nOleg Bartunov, sci.researcher, hostmaster of AstroNet,\nSternberg Astronomical Institute, Moscow University (Russia)\nInternet: [email protected], http://www.sai.msu.su/~megera/\nphone: +007(095)939-16-83, +007(095)939-23-83\n\n",
"msg_date": "Wed, 6 Jan 1999 19:28:56 +0300 (MSK)",
"msg_from": "Oleg Bartunov <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Numeric type"
},
{
"msg_contents": "Oleg Bartunov wrote:\n\n> On my Linux box/i586/libc5, postgresql 6.4.2:\n>\n> test=> create table n1 (n numeric(10,5), d decimal(10,5));\n> ERROR: NUMERIC precision 10 must be 9\n> test=>\n>\n> Something weird ! I compiled postgres 6.4.2 after applying Jan's feature\n> patch and configure --enable-locale\n\n Not weird. The new NUMERIC datatype (arbitrary precision\n numeric format) isn't part of the v6.4 tree and will not be.\n\n It's only in the current development tree which will become\n v6.5.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Wed, 6 Jan 1999 18:09:48 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Numeric type"
}
] |
[
{
"msg_contents": "Hello,\n\ncase of your majordomo problems this message seems to be lost. So this repost ...\n\n\n\n\nHello there,\n\nI have a problem using pl/pgsql (distributed in version 6.4.1).\nWe are using PostgreSQL for an ordering-system as a part of a new data warehouse. At this moment we have about 10K rows and it works great!\nNow we are testing it for a much amount of data (sales report) of about 1.5GB in size.\nMy first question is: is this the right place to post such bug reports?\nOk, following pl/pgsql statements doesn�t work:\n\n--------\ncreate table t1 (i int, a int, b int);\nCREATE create table t2 (i int, x int, y int);\nCREATE\ninsert into t1 values(1, 1,10);\nINSERT 35140214 1 insert into t1 values(1, 2,10);\nINSERT 35140215 1 insert into t1 values(2, 3,10);\nINSERT 35140216 1 insert into t1 values(2, 4,10);\nINSERT 35140217 1\ncreate function func1()\nreturns int\nas '\ndeclare\nbegin\ninsert into t2\nselect i,\nsum(a) as x,\nsum(b) as y\nfrom t1\ngroup by i;\nreturn 1;\nend;\n' language 'plpgsql';\nCREATE\n\nselect func1();\nfunc1\n-----\n1\n(1 row)\n\nselect * from t2;\ni| x| y\n-+--+--\n1|20|20\n2|20|20\n(2 rows)\n\n\nThe result of the last column seems to be pasted in all other agregated columns.\n-------------\n\nUsing the insert/select-statement directly (via psql) it works fine.\nAnd one additional (last) question:\nIn pl/sql writing select * into table2 from table1 ... seems to be interpreted that table2 is a pl/sql-variable and not a table. So it doesn�t work. Is it a bug or is there an other way to select directly into a (new) table. \n\nKind regards\nMichael Contzen\nDohle Handelsgruppe Systemberatung GmbH,\nGermany\n\n\n\n\n\n",
"msg_date": "Wed, 6 Jan 1999 08:51:17 -0000",
"msg_from": "Michael Contzen <[email protected]>",
"msg_from_op": true,
"msg_subject": "PL/PGSQL bug when using aggregates (repost)"
}
] |
[
{
"msg_contents": "> > In the INSTALL file, it says (under \"Unsupported platforms\"):\n> > Windows x86 v6.3 1998-03-01 not library compatible;\n> > client side maybe; use ODBC/JDBC\n> > Note that Windows ports of the frontend are\n> > apparently possible using third-party Posix porting\n> > tools and libraries.\n> > I believe the client libraries (at least libpq) are available in \n> > native Win32 since version 6.4.\n> \n> Thanks for pointing this out. So how should I phrase it? As above? If\n> so, where do we tell people how to build it? Or is it obvious to anyone\n> who is trying? \n> \n> If it could use instructions, would someone like to write it up? Or was\n> something already posted on the mailing list which tells all?\n> \n> So many questions, so little time :) As you can tell, I haven't tried\n> myself...\n\nWell, I don't think there is any documentation on how to do it. It's pretty\nsimple:\n\nYou go into the src directory and run \"nmake /f win32.mak\". \nThat one will compile libpq.dll and the required import-library\n(libipqdll.lib) and the static library (libpq.lib) for those who don't want\nto use the DLL version. It will also compile psql.exe, linked against\nlibpq.dll.\nNote that the makefiles in question are for Microsoft Visual C++ - I don't\nknow if they work in other Windows compilers, but I would guess they don't.\n\nHmm. It should probably be noticed somewhere that crypt-pw authentication\ndoes not work in the win32 version - there is no crypt() function\navailable..\n\nSo I guess it should say that the client-side library libpq and the psql\nfrontend exists for Win32, and then perhaps reference to either another\nsection of the INSTALL file or a separate file (INSTALL.WIN32?) describing\nthe process of building and installing it? If that's a good way to go, I can\nwrite up a quick info-file on how to build and install it.\n\n\n//Magnus\n",
"msg_date": "Wed, 6 Jan 1999 15:42:28 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [DOCS] Postgresql INSTALL-file & Windows"
},
{
"msg_contents": "> perhaps reference to either another\n> section of the INSTALL file or a separate file (INSTALL.WIN32?) \n> describing the process of building and installing it? If that's a good \n> way to go, I can write up a quick info-file on how to build and \n> install it.\n\nI hope it wasn't too obvious that I was fishing for that offer :)\n\nYes, if you could write it up that would be great. If you have the time\nor inclination, you could use doc/src/sgml/install.sgml as a template\nexample which includes some sgml markup. Then send your resulting file\n(install32.sgml?) to me or the docs list and I'll incorporate it into\nthe full docs.\n\nIf you find the sgml tags incomprehensible then ask questions, and if\nyou just don't want to deal with it, then send me plain text and I'll do\nthe markup.\n\nTIA\n\n - Tom\n",
"msg_date": "Wed, 06 Jan 1999 14:59:27 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [DOCS] Postgresql INSTALL-file & Windows"
}
] |
[
{
"msg_contents": "\n> > In the INSTALL file, it says (under \"Unsupported platforms\"):\n> > Windows x86 v6.3 1998-03-01 not library compatible;\n> > client side maybe; use ODBC/JDBC\n> > Note that Windows ports of the frontend are\n> > apparently possible using third-party Posix porting\n> > tools and libraries.\n\nI think this message can be deleted.\n\n> > I believe the client libraries (at least libpq) are available in \n> > native Win32 since version 6.4.\n\nThere is a port of client libraries using MS development tools. We are\nworking on a port of the whole Postgres 6.4.x using the Cygwin tools. Joost\nhave already send to this list (on Dec 7 1998) a description how to succeed.\n\n\t\t\tDan\n",
"msg_date": "Wed, 6 Jan 1999 15:55:49 +0100 ",
"msg_from": "Horak Daniel <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Re: [DOCS] Postgresql INSTALL-file & Windows"
},
{
"msg_contents": "Horak Daniel wrote:\n\n> > > In the INSTALL file, it says (under \"Unsupported platforms\"):\n> > > Windows x86 v6.3 1998-03-01 not library compatible;\n> > > client side maybe; use ODBC/JDBC\n> > > Note that Windows ports of the frontend are\n> > > apparently possible using third-party Posix porting\n> > > tools and libraries.\n>\n> I think this message can be deleted.\n>\n> > > I believe the client libraries (at least libpq) are available in\n> > > native Win32 since version 6.4.\n>\n> There is a port of client libraries using MS development tools. We are\n> working on a port of the whole Postgres 6.4.x using the Cygwin tools. Joost\n> have already send to this list (on Dec 7 1998) a description how to succeed.\n\nHi, Dan,\n\nGood job! According to Joost's postgres installation procedures, and your help,\nI have already installed Postgres on WinNT successfully. I also wrote\ninstallation procedures for Chinese version which describes how I installed,\nand put it on my home page, see\n\nhttp://members.tripod.com/~kevlo/postgres/postgresinstall.txt\n\nSo if you are a chinese, maybe this is a good info.\n\nBest regards,\nKevin.\n\n",
"msg_date": "Thu, 07 Jan 1999 23:35:00 +0800",
"msg_from": "Kevin Lo <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [DOCS] Postgresql INSTALL-file & Windows"
},
{
"msg_contents": "> According to Joost's postgres installation procedures, and your help,\n> I have already installed Postgres on WinNT successfully. I also wrote\n> installation procedures for Chinese version which describes how I \n> installed, and put it on my home page, see\n> http://members.tripod.com/~kevlo/postgres/postgresinstall.txt\n> So if you are a chinese, maybe this is a good info.\n\nHal, is there a place on the Postgres web site for non-English\ndocumentation resources? Or just a related links area? It looks like\nthis would be a good one for that.\n\n - Tom\n",
"msg_date": "Thu, 07 Jan 1999 16:00:05 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [DOCS] Postgresql INSTALL-file & Windows"
}
] |
[
{
"msg_contents": "Hi!\n\nI was just poking around a bit in the libpq (server side this time) files,\nand noticed a few things.\n\nFor example, the StreamOpen() function is defined there (pqcomm.c), but it\nis _never_ used (at least as far as I can see). Is there any reason for it\nbeing there? Or is it some old function that is no longer used? Looks to me\nlike a function that is only used on the client-side?\n\nAlso, the comments on top of pqcomm.c names several functions that do not\nexist (such as pq_accept). They are commented as being used in both the\nbackend and the frontend - but I don't beleive the file is linked into the\nlibpq frontend any more?\n\nFinally - is there any special reason that the backend still uses the (FILE\n*) method to talk to the clients? Using the global Pfout and Pfin variables?\nWouldn't it be better to be consistent and use the same functions as in the\nrevised frontent library?\n\n\n\n//Magnus\n",
"msg_date": "Wed, 6 Jan 1999 16:39:42 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "Libpq functions"
},
{
"msg_contents": "Magnus Hagander <[email protected]> writes:\n> [ Why is the server-side libpq so crufty? ]\n\nApparently, that set of files was once used for both the frontend and\nbackend sides of the FE/BE protocol. It no longer is, but no one has\ngotten around to ripping out the now-dead parts of the code, nor to\nfixing the comments.\n\nI didn't bother to touch it when I rewrote the client-side libpq last\nsummer, because there wasn't any functional improvement to be had there.\nIt pretty much does everything the backend needs done.\n\nIf you have the time and energy to clean it up just in the name of\ncode beautification, step right up :-). One thing that would be good\nright off the bat is to change the name --- I think it's confusing to\ncall both the FE and BE modules libpq, when they are no longer the same\ncode or even very close.\n\n> Finally - is there any special reason that the backend still uses the (FILE\n> *) method to talk to the clients? Using the global Pfout and Pfin variables?\n> Wouldn't it be better to be consistent and use the same functions as in the\n> revised frontent library?\n\nThe main reason for rewriting the front end was to satisfy clients that\ndidn't want to block while awaiting backend I/O. The backend doesn't\nhave any comparable requirement: when it's waiting for the frontend, it\nhas nothing better to do (AFAIK anyway). And using stdio does have its\nadvantages in simplicity and just plain standard-ness. So I doubt it's\nworth making that kind of change.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 06 Jan 1999 20:16:52 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Libpq functions "
}
] |
[
{
"msg_contents": "Hi,\n\nI'm trying to convert some tables from version 6.2 to 6.4.2 but I have\nproblems with arrays. The old table definition was:\n\n create table a (\n\tx int4,\n\ty char8,\n\tz char8[]\n );\n\nNow, because the char8 datatype has been removed from postgres, I suppose\nI should use char(8) or varchar(8):\n\n create table a (\n\tx int4,\n\ty char(8),\n\tz char(8)[]\n );\n\nor\n\n create table a (\n\tx int4,\n\ty varchar(8),\n\tz varchar(8)[]\n );\n\nbut when I try the above definitions I get the following errors:\n\ndz=> create table a (\ndz-> \tx int4,\ndz-> \ty char(8),\ndz-> \tz char(8)[]\ndz-> );\nERROR: type name lookup of _bpchar failed\n\ndz=> create table a (\ndz-> \tx int4,\ndz-> \ty varchar(8),\ndz-> \tz varchar(8)[]\ndz-> );\nERROR: type name lookup of _varchar failed\n\nAlso trying to define the _varchar or _bpchar types fails, so it seems that\nthe conversion from 6.2 to 6.4 is not possible for many data types.\nAny suggestion?\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+\n",
"msg_date": "Wed, 6 Jan 1999 17:39:09 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": true,
"msg_subject": "problem with arrays"
},
{
"msg_contents": "> Also trying to define the _varchar or _bpchar types fails, so it seems \n> that the conversion from 6.2 to 6.4 is not possible for many data \n> types. Any suggestion?\n\nThat sucks. afaik the only data types affected are the fixed-maximum\nwith variable-length strings. There was a comment in gram.y from Jolly\nregarding the possibility for allowing bpchar and varchar into arrays,\nthough she explicitly disallowed it. I had thought that this code is\nstill in effect.\n\nI'm not certain how much you have looked at this, but the leading\nunderscore is a Postgres convention for array data types. The trick to\nallowing bpchar, varchar, and perhaps numeric types into arrays is\npassing along the element dimensions. I haven't looked at whether that\nalready happens.\n\nIn the meantime, you can try using the text type instead. I know it\ndoesn't have an 8-byte maximum length, but it is allowed in arrays...\n\n - Tom\n",
"msg_date": "Thu, 07 Jan 1999 02:46:44 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] problem with arrays"
}
] |
[
{
"msg_contents": "Hi,\n\ninsert a few row in a table (>50000) and do\ndelete from mytable;\nvacuum verbose analyze;\n\nWhy is this that slow?\n\n[I am not on [email protected]]\n---\n _ _\n _(_)(_)_ David Wetzel, Turbocat's Development,\n(_) __ (_) Buchhorster Strasse, D-16567 Muehlenbeck/Berlin, FRG,\n _/ \\_ Fax +49 33056 82835 NeXTmail [email protected]\n (______) http://www.turbocat.de/\n DEVELOPMENT * CONSULTING * ADMINISTRATION\n WATCH OUT FOR TURBOFAX for OPENSTEP!\n",
"msg_date": "Wed, 6 Jan 99 20:09:20 +0100",
"msg_from": "David Wetzel <[email protected]>",
"msg_from_op": true,
"msg_subject": "vacuum takes too long"
},
{
"msg_contents": "> Hi,\n> \n> insert a few row in a table (>50000) and do\n> delete from mytable;\n> vacuum verbose analyze;\n> \n> Why is this that slow?\n\nAnalyze checks every column in every row.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 6 Jan 1999 16:02:33 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "> From: Bruce Momjian <[email protected]>\n\n> > insert a few row in a table (>50000) and do\n> > delete from mytable;\n> > vacuum verbose analyze;\n> >\n> > Why is this that slow?\n>\n> Analyze checks every column in every row.\n\neven if you only type \"vacuum verbose\" it takes _very_ long.\n\nI deleted _all_ records with \"delete from mytable;\" before.\nA drop and a new create is faster. But what is when you delete (maybe \n100000) rows but keep 100 in the table?\n\nI use 6.4.2 on NetBSD/i486 (that box makes gets 12MBytes/sec via the \nfilesystem out of the drives)\n---\n _ _\n _(_)(_)_ David Wetzel, Turbocat's Development,\n(_) __ (_) Buchhorster Strasse, D-16567 Muehlenbeck/Berlin, FRG,\n _/ \\_ Fax +49 33056 82835 NeXTmail [email protected]\n (______) http://www.turbocat.de/\n DEVELOPMENT * CONSULTING * ADMINISTRATION\n",
"msg_date": "Wed, 6 Jan 99 23:38:29 +0100",
"msg_from": "David Wetzel <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "> > From: Bruce Momjian <[email protected]>\n> \n> > > insert a few row in a table (>50000) and do\n> > > delete from mytable;\n> > > vacuum verbose analyze;\n> > >\n> > > Why is this that slow?\n> >\n> > Analyze checks every column in every row.\n> \n> even if you only type \"vacuum verbose\" it takes _very_ long.\n> \n> I deleted _all_ records with \"delete from mytable;\" before.\n> A drop and a new create is faster. But what is when you delete (maybe \n> 100000) rows but keep 100 in the table?\n> \n> I use 6.4.2 on NetBSD/i486 (that box makes gets 12MBytes/sec via the \n> filesystem out of the drives)\n\nNot sure what to say. Vacuum does take a while, and it is often faster\nto drop and recreate.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 6 Jan 1999 19:50:37 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "On Wed, 6 Jan 1999, Bruce Momjian wrote:\n\n> > > From: Bruce Momjian <[email protected]>\n> > \n> > > > insert a few row in a table (>50000) and do\n> > > > delete from mytable;\n> > > > vacuum verbose analyze;\n> > > >\n> > > > Why is this that slow?\n> > >\n> > > Analyze checks every column in every row.\n> > \n> > even if you only type \"vacuum verbose\" it takes _very_ long.\n> > \n> > I deleted _all_ records with \"delete from mytable;\" before.\n> > A drop and a new create is faster. But what is when you delete (maybe \n> > 100000) rows but keep 100 in the table?\n> > \n> > I use 6.4.2 on NetBSD/i486 (that box makes gets 12MBytes/sec via the \n> > filesystem out of the drives)\n> \n> Not sure what to say. Vacuum does take a while, and it is often faster\n> to drop and recreate.\n\nLet's ignore the 'analyze' part first...take a simple 'vacuum'\ncommand...what takes the longest? My understanding is a vacuum\nsimplistically, takes and moves all rows \"up\" in the file to fill in any\nblanks resulting from updates and deletes, then truncates the end of the\nfile...\n\nIf so, is there no way of having vacuum running on its own? \n\nBasically, if my understanding is remotely correct, vaccum is\ndefragmenting the table...so why can't the defragmenting be performed\nduring idle time...or, at least some of it. \n\nStart at the top of the table, go to the first 'blank' section (a deleted\nrecord)...find the next good record that will fit in the space, move it\nthere...clear out the old space, etc...\n\nif dba issues a 'vacuum', lock the table and do all records at once, but\notherwise try and vacuum the table live...\n\nWith the new MVCC serialization, this concept should be less intrusive on\nreaders, no?\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Wed, 6 Jan 1999 22:30:57 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "> Start at the top of the table, go to the first 'blank' section (a deleted\n> record)...find the next good record that will fit in the space, move it\n> there...clear out the old space, etc...\n> \n> if dba issues a 'vacuum', lock the table and do all records at once, but\n> otherwise try and vacuum the table live...\n> \n> With the new MVCC serialization, this concept should be less intrusive on\n> readers, no?\n\nWish I knew the answer. I can guess, but that isn't going to help.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 6 Jan 1999 21:36:31 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "On Wed, 6 Jan 1999, Bruce Momjian wrote:\n\n> > Start at the top of the table, go to the first 'blank' section (a deleted\n> > record)...find the next good record that will fit in the space, move it\n> > there...clear out the old space, etc...\n> > \n> > if dba issues a 'vacuum', lock the table and do all records at once, but\n> > otherwise try and vacuum the table live...\n> > \n> > With the new MVCC serialization, this concept should be less intrusive on\n> > readers, no?\n> \n> Wish I knew the answer. I can guess, but that isn't going to help.\n\n\tGuess == throwing in ideas, even if they are incorrect...the way I\nfigure it, I through out alot of guesses...some of them spark ideas in\nothers and we see some really neat ideas come out of it :)\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Wed, 6 Jan 1999 23:08:40 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "> > Wish I knew the answer. I can guess, but that isn't going to help.\n> \n> \tGuess == throwing in ideas, even if they are incorrect...the way I\n> figure it, I through out alot of guesses...some of them spark ideas in\n> others and we see some really neat ideas come out of it :)\n\nYes. Good. I have been bugging Vadim about possible row reuse, but I\ndon't know enough to understand the options.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 6 Jan 1999 22:53:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "On Wed, 6 Jan 1999, Bruce Momjian wrote:\n\n> > > Wish I knew the answer. I can guess, but that isn't going to help.\n> > \n> > \tGuess == throwing in ideas, even if they are incorrect...the way I\n> > figure it, I through out alot of guesses...some of them spark ideas in\n> > others and we see some really neat ideas come out of it :)\n> \n> Yes. Good. I have been bugging Vadim about possible row reuse, but I\n> don't know enough to understand the options.\n\nI'm not sure about the row-reuse thing. What sort of performance hit will\nit have. As it is now, you add a row by zipping down to the end, add the\nrow...bang, finished. with row-reuse, you have to search for a good fit,\nwhich could take time...\n\nHrmmm...let's look at Oracle's \"model\"...bear in mind that I haven't dived\nvery deep into it, so I could be totally off base here, but, with Oracle,\nyou have a seperate \"group\" of processes started up for each 'instance',\nwhere, if I'm correct, an instance is the same as our database(?)...\n\nHow hard would it be for us to implement something similar? When you\nstart up the postmaster, it starts up 1 postgres \"master process\" for each\ndatabase that it knows about. The point of the master process is\neffectively the garbage collector for the database, as well as the central\n'traffic cop'...\n\nso, for example, I have 4 databases on my server...when you start up the\nsystem with your normal 'postmaster' process, it forks off 4 processes,\none for each database. When you connect to port #### for database XXXX,\nthe listening process (main postmaster) shunts the process over to the\nappropriate 'traffic cop' for handling...\n\nThe 'traffic cop' would keep track of the number of connections to the\ndatabase are currently open, and when zero, which woudl indicate idle\ntime, process through a table in the database to clean it up. As soon as\na new connection comes in, it would \"finish\" its cleanup by making sure\nthe table is in a 'sane state' (ie. finish up with its current record) and\nthen fork off the process, to wait quietly until its idle again...\n\nThen each database could effectively have their own shared memory pool\nthat could be adjusted on a per database basis. Maybe even add a 'change\nthreshold', where after X transactions (update, insert or delete), the\ntable gets auto-vacuum'd (no analyze, just vacuum)...the threshold could\nbe set on a per-table basis...the 'traffic cop' should be able to easily\nkeep track of those sort of stats internally...no?\n\nHell, the 'traffic cop' *should* be able to keep reasonably accurate stats\nto update the same tables that a 'vacuum analyze' maintains, adjusting\nthose values periodically to give a semi-accurate picture. Periodically,\na normal 'analyze' would have to be run...\n\nIts a thought...haven't got a clue as to the complexity of implementing,\nbut...*shrug*\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Thu, 7 Jan 1999 00:38:18 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "\nAdded to TODO:\n\n* Create a background process for each database that runs while \n database is idle, finding superceeded rows, gathering stats and vacuuming\n\n\n\n> On Wed, 6 Jan 1999, Bruce Momjian wrote:\n> \n> > > > Wish I knew the answer. I can guess, but that isn't going to help.\n> > > \n> > > \tGuess == throwing in ideas, even if they are incorrect...the way I\n> > > figure it, I through out alot of guesses...some of them spark ideas in\n> > > others and we see some really neat ideas come out of it :)\n> > \n> > Yes. Good. I have been bugging Vadim about possible row reuse, but I\n> > don't know enough to understand the options.\n> \n> I'm not sure about the row-reuse thing. What sort of performance hit will\n> it have. As it is now, you add a row by zipping down to the end, add the\n> row...bang, finished. with row-reuse, you have to search for a good fit,\n> which could take time...\n> \n> Hrmmm...let's look at Oracle's \"model\"...bear in mind that I haven't dived\n> very deep into it, so I could be totally off base here, but, with Oracle,\n> you have a seperate \"group\" of processes started up for each 'instance',\n> where, if I'm correct, an instance is the same as our database(?)...\n> \n> How hard would it be for us to implement something similar? When you\n> start up the postmaster, it starts up 1 postgres \"master process\" for each\n> database that it knows about. The point of the master process is\n> effectively the garbage collector for the database, as well as the central\n> 'traffic cop'...\n> \n> so, for example, I have 4 databases on my server...when you start up the\n> system with your normal 'postmaster' process, it forks off 4 processes,\n> one for each database. When you connect to port #### for database XXXX,\n> the listening process (main postmaster) shunts the process over to the\n> appropriate 'traffic cop' for handling...\n> \n> The 'traffic cop' would keep track of the number of connections to the\n> database are currently open, and when zero, which woudl indicate idle\n> time, process through a table in the database to clean it up. As soon as\n> a new connection comes in, it would \"finish\" its cleanup by making sure\n> the table is in a 'sane state' (ie. finish up with its current record) and\n> then fork off the process, to wait quietly until its idle again...\n> \n> Then each database could effectively have their own shared memory pool\n> that could be adjusted on a per database basis. Maybe even add a 'change\n> threshold', where after X transactions (update, insert or delete), the\n> table gets auto-vacuum'd (no analyze, just vacuum)...the threshold could\n> be set on a per-table basis...the 'traffic cop' should be able to easily\n> keep track of those sort of stats internally...no?\n> \n> Hell, the 'traffic cop' *should* be able to keep reasonably accurate stats\n> to update the same tables that a 'vacuum analyze' maintains, adjusting\n> those values periodically to give a semi-accurate picture. Periodically,\n> a normal 'analyze' would have to be run...\n> \n> Its a thought...haven't got a clue as to the complexity of implementing,\n> but...*shrug*\n> \n> Marc G. Fournier \n> Systems Administrator @ hub.org \n> primary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n> \n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 6 Jul 1999 20:31:16 -0400 (EDT)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
}
] |
[
{
"msg_contents": "Bruce, Marc, et al-\n\nBruce Momjian wrote:\n\n> I think it might be nice to get this out on the web page in a prominent\n> place until we release 6.5. Seems like we should be talking up these\n> features so people can get excited.\n\n>> In February, we start beta on 6.5...\n\n\n\nI have been working on some revisions for the main home page.\nA preliminary version, including the 6.5 blurb, can be found at:\n\n http://www.postgresql.org/mainx.html\n\nWhat do you think?\n\nNotes:\n\n- I deleted references to Vadim and Jan to make the item seem less\n less like a private communication intended for insiders. No slight\n is intended here.\n\n- The navigation bar is no longer a frame. I think Marc and I agree\n the old navigation-bar frame scheme should be revised. I'd like it\n to have explanatory onMouseOvers, but this isn't there yet - I'm\n still trying out several things. Changing how the navigation bars\n work will affect most pages on the site, giving it a different feel.\n\nHal\n",
"msg_date": "07 Jan 1999 03:34:02 -0600",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": true,
"msg_subject": "Upcoming Attractions, web site"
},
{
"msg_contents": "Looks nice. Might rephrase:\n\n Those of you who did not like\n our table-level locking will be pleasantly surprised.\n\nto be something like\n\n Those of you wanting alternatives to\n our table-level locking will be pleasantly surprised.\n\nThe new CASE statement might be of interest, and I'm looking forward to\ngetting far enough along on outer joins to get a mention of that too :)\n\n - Tom\n",
"msg_date": "Thu, 07 Jan 1999 16:16:51 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [DOCS] Upcoming Attractions, web site"
},
{
"msg_contents": "> Bruce, Marc, et al-\n> \n> Bruce Momjian wrote:\n> \n> > I think it might be nice to get this out on the web page in a prominent\n> > place until we release 6.5. Seems like we should be talking up these\n> > features so people can get excited.\n> \n> >> In February, we start beta on 6.5...\n> \n> \n> \n> I have been working on some revisions for the main home page.\n> A preliminary version, including the 6.5 blurb, can be found at:\n> \n> http://www.postgresql.org/mainx.html\n> \n> What do you think?\n\nLooks great.\n\n> \n> Notes:\n> \n> - I deleted references to Vadim and Jan to make the item seem less\n> less like a private communication intended for insiders. No slight\n> is intended here.\n\nOK.\n\n> - The navigation bar is no longer a frame. I think Marc and I agree\n> the old navigation-bar frame scheme should be revised. I'd like it\n> to have explanatory onMouseOvers, but this isn't there yet - I'm\n> still trying out several things. Changing how the navigation bars\n> work will affect most pages on the site, giving it a different feel.\n\nI like that too.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Thu, 7 Jan 1999 12:31:06 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [DOCS] Upcoming Attractions, web site"
},
{
"msg_contents": "> Looks nice. Might rephrase:\n> \n> Those of you who did not like\n> our table-level locking will be pleasantly surprised.\n> \n> to be something like\n> \n> Those of you wanting alternatives to\n> our table-level locking will be pleasantly surprised.\n> \n> The new CASE statement might be of interest, and I'm looking forward to\n> getting far enough along on outer joins to get a mention of that too :)\n\nYes, CASE would be good to add. You can say:\n\n\tAn SQL-standard CASE statement is planned too.\n\n\nThomas, do you need help on outer joins?\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Thu, 7 Jan 1999 12:47:34 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [DOCS] Upcoming Attractions, web site"
},
{
"msg_contents": "> Looks nice. Might rephrase:\n> \n> Those of you who did not like\n> our table-level locking will be pleasantly surprised.\n> \n> to be something like\n> \n> Those of you wanting alternatives to\n> our table-level locking will be pleasantly surprised.\n\nI don't like the new wording. It sounds like we are adding an optional\nfunctionality, while we are replacing table-level locking because MVCC\nis much better. We can say:\n\n Those of you need better locking than our current\n table-level locking will be pleasantly surprised.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Thu, 7 Jan 1999 12:54:11 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [DOCS] Upcoming Attractions, web site"
},
{
"msg_contents": "> Thomas, do you need help on outer joins?\n\nYes. I'm going slowly partly because I get distracted with other\nPostgres stuff like docs, and partly because I don't understand all of\nthe pieces I'm working with.\n\nI've identified the place in the MergeJoin code where the null filling\nfor outer joins needs to happen, and have the \"merge walk\" code done.\nBut I don't have the supporting code which actually would know how to\nnull-fill a result tuple from the left or right. I thought you might be\ninterested in that?\n\nI've done some work in the parser, and can now do things like:\n\npostgres=> select * from t1 join t2 using (i);\nNOTICE: JOIN not yet implemented\ni|j|i|k\n-+-+-+-\n1|2|1|3\n(1 row)\n\nBut this is just an inner join, and the result isn't quite right since\nthe second \"i\" column should probably be omitted. At the moment I\ntransform it from the syntax above into existing parse nodes, and\neverything from there on works.\n\nI don't yet pass an explicit join node into the planner/optimizer, and\nthat will be the hardest part I assume. Perhaps we can work on that\ntogether.\n\nSo, what I'll try to do (soon, in the next few days?) is put in\n\n #ifdef ENABLE_OUTER_JOINS\n\nconditional code into the parser area (already there for the executor)\nand commit everything to the development tree. Does that sound OK?\n\nOh, and if anyone is looking for something to do, I've got a couple of\nCASE statements in the case.sql regression test which are commented out\nbecause they crash the backend. They involve references to multiple\ntables within a single result column, and in other contexts that\nconstruct works. It would be great if someone had time to track it\ndown...\n\n - Tom\n",
"msg_date": "Thu, 07 Jan 1999 18:26:56 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Outer Joins (and need CASE help)"
}
] |
[
{
"msg_contents": "> (back on-list)\nOk :-)\n\n\n> > I'm going to try to whip up some precompiled binaries for it too, \n> > tomorrow. Perhaps we should put a blurb in about that?\n> \n> Yes. Here is the modified file back (note the minor name \n> change) if you\n> want to mention it in there. Byron has some compiled binaries \n> for Win32\n> ODBC, and I might post compiled binaries for Linux ODBC, so we should\n> figure out where to put these kinds of binaries. There probably should\n> be a win32 directory of some sort on the ftp site.\nThere is a /pub/non-unix on the ftp server, which now only contains \"os2\".\nPerhaps create a subdir for \"win32\" in there, or something?\nI'll be working on the binaries soon. Where should I put them once done?\nMail to you, or somebody? Can't find any incoming directory on the ftp...\nNo idea to put in anything in the docs before we can say where it is\nlocated...\n\n//Magnus\n",
"msg_date": "Thu, 7 Jan 1999 12:17:20 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: Postgresql INSTALL-file & Windows"
},
{
"msg_contents": "> > > I'm going to try to whip up some precompiled binaries for it too,\n> > > tomorrow. Perhaps we should put a blurb in about that?\n> > Byron has some compiled binaries for Win32\n> > ODBC, and I might post compiled binaries for Linux ODBC,\n> > There probably should\n> > be a win32 directory of some sort on the ftp site.\n> There is a /pub/non-unix on the ftp server, which now only contains \n> \"os2\". Perhaps create a subdir for \"win32\" in there, or something?\n\nOK, Marc, do you have a preferred layout for where various kinds of\nbinaries or other distribution packages should go?\n\nRight now there is:\n\n bindist - Solaris v6.3 binaries\n non-unix/os2 - OS2 v6.3 client-side binaries\n odbc - win32 odbc source and binaries\n\n> I'll be working on the binaries soon. Where should I put them once \n> done? Mail to you, or somebody? Can't find any incoming directory on \n> the ftp... No idea to put in anything in the docs before we can say \n> where it is located...\n\nMagnus is building some WIN32 client-side binaries and needs a place to\nput them.\n\nSo I just created:\n doc - tar files of daily builds of current html documentation\n .incoming - world-writable file dropoff\n\nI have no write privileges in /pub/non-unix, so could not add a win32\ndirectory. Also, I noticed that there are still some \".old\" directories\nin /home/projects/pgsql/ftp/www/html/docs, which I had renamed a long\ntime ago since I did not have delete privileges for the original\ndirectories and files but was able to rename them. These should be\ndeleted.\n\nAnyway, let me know what you would prefer, and if you want me to set it\nup perhaps you can chmod/chown the ftp and html/docs directories to\nallow me to make the changes.\n\nTIA\n\n - Tom\n",
"msg_date": "Thu, 07 Jan 1999 15:39:52 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: Postgresql INSTALL-file & Windows"
}
] |
[
{
"msg_contents": "> > > > Wish I knew the answer. I can guess, but that isn't \n> going to help.\n> > > \n> > > \tGuess == throwing in ideas, even if they are \n> incorrect...the way I\n> > > figure it, I through out alot of guesses...some of them \n> spark ideas in\n> > > others and we see some really neat ideas come out of it :)\n> > \n> > Yes. Good. I have been bugging Vadim about possible row \n> reuse, but I\n> > don't know enough to understand the options.\n> \n> I'm not sure about the row-reuse thing. What sort of \n> performance hit will\n> it have. As it is now, you add a row by zipping down to the \n> end, add the\n> row...bang, finished. with row-reuse, you have to search for \n> a good fit,\n> which could take time...\n> \n> Hrmmm...let's look at Oracle's \"model\"...bear in mind that I \n> haven't dived\n> very deep into it, so I could be totally off base here, but, \n> with Oracle,\n> you have a seperate \"group\" of processes started up for each \n> 'instance',\n> where, if I'm correct, an instance is the same as our database(?)...\n> \n> How hard would it be for us to implement something similar? When you\n> start up the postmaster, it starts up 1 postgres \"master \n> process\" for each\n> database that it knows about. The point of the master process is\n> effectively the garbage collector for the database, as well \n> as the central\n> 'traffic cop'...\n> \n> so, for example, I have 4 databases on my server...when you \n> start up the\n> system with your normal 'postmaster' process, it forks off 4 \n> processes,\n> one for each database. When you connect to port #### for \n> database XXXX,\n> the listening process (main postmaster) shunts the process over to the\n> appropriate 'traffic cop' for handling...\n> \n> The 'traffic cop' would keep track of the number of connections to the\n> database are currently open, and when zero, which woudl indicate idle\n> time, process through a table in the database to clean it up. \n> As soon as\n> a new connection comes in, it would \"finish\" its cleanup by \n> making sure\n> the table is in a 'sane state' (ie. finish up with its \n> current record) and\n> then fork off the process, to wait quietly until its idle again...\n> \n> Then each database could effectively have their own shared memory pool\n> that could be adjusted on a per database basis. Maybe even \n> add a 'change\n> threshold', where after X transactions (update, insert or delete), the\n> table gets auto-vacuum'd (no analyze, just vacuum)...the \n> threshold could\n> be set on a per-table basis...the 'traffic cop' should be \n> able to easily\n> keep track of those sort of stats internally...no?\n> \n> Hell, the 'traffic cop' *should* be able to keep reasonably \n> accurate stats\n> to update the same tables that a 'vacuum analyze' maintains, adjusting\n> those values periodically to give a semi-accurate picture. \n> Periodically,\n> a normal 'analyze' would have to be run...\n> \n> Its a thought...haven't got a clue as to the complexity of \n> implementing,\n> but...*shrug*\n\nWith MVCC an occasional 'vacuum analyze' should only be noticed from the\nperformance improvements. As far as I can tell most of the work done by\nan analyze is in reading the table data. If you make sure to write the\nnew information at the end of the transaction you only lock the indexes\nfor the amount of time it takes to write them.\n\nI see a 'vacuum analyze' being less of a problem than 'vacuum'.\nAny of you experts can contradict my assumptions.\n\tJust my guess,\n\t-DEJ\n",
"msg_date": "Thu, 7 Jan 1999 10:43:07 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "On Thu, 7 Jan 1999, Jackson, DeJuan wrote:\n\n> With MVCC an occasional 'vacuum analyze' should only be noticed from the\n> performance improvements. As far as I can tell most of the work done by\n> an analyze is in reading the table data. If you make sure to write the\n> new information at the end of the transaction you only lock the indexes\n> for the amount of time it takes to write them.\n> \n> I see a 'vacuum analyze' being less of a problem than 'vacuum'.\n> Any of you experts can contradict my assumptions.\n\nGood point...I seem to recall that at one point, there was a lock imposed\non one of hte pg_ tables when a vacuum is tarted, since it has to update a\ncouple of the rows in that table...has that lock been removed with MVCC?\nVadim?\n\n\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Thu, 7 Jan 1999 13:12:23 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "RE: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "> With MVCC an occasional 'vacuum analyze' should only be noticed from the\n> performance improvements. As far as I can tell most of the work done by\n> an analyze is in reading the table data. If you make sure to write the\n> new information at the end of the transaction you only lock the indexes\n> for the amount of time it takes to write them.\n> \n> I see a 'vacuum analyze' being less of a problem than 'vacuum'.\n> Any of you experts can contradict my assumptions.\n\nThe problem is that vacuum analyze does both vacuum and analyze. \nAnalyze takes so long, we figured we might as well vacuum too. Maybe we\nneed to change that.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Thu, 7 Jan 1999 12:49:57 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
},
{
"msg_contents": "On Thu, 7 Jan 1999, Bruce Momjian wrote:\n\n> > With MVCC an occasional 'vacuum analyze' should only be noticed from the\n> > performance improvements. As far as I can tell most of the work done by\n> > an analyze is in reading the table data. If you make sure to write the\n> > new information at the end of the transaction you only lock the indexes\n> > for the amount of time it takes to write them.\n> > \n> > I see a 'vacuum analyze' being less of a problem than 'vacuum'.\n> > Any of you experts can contradict my assumptions.\n> \n> The problem is that vacuum analyze does both vacuum and analyze. \n> Analyze takes so long, we figured we might as well vacuum too. Maybe we\n> need to change that.\n\nThere is, IMHO, no problem with them being combined...in the past, the\nproblem was that the whole system was effectively locked up while a vacuum\nanalyze was being run because one of the 'statistics' tables was being\nlocked during the whole thing, instead of when required...\n\nAs DeJuan points out, though, this should no longer be a problem with\nMVCC...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Fri, 8 Jan 1999 02:17:03 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] vacuum takes too long"
}
] |
[
{
"msg_contents": "> > With MVCC an occasional 'vacuum analyze' should only be \n> noticed from the\n> > performance improvements. As far as I can tell most of the \n> work done by\n> > an analyze is in reading the table data. If you make sure \n> to write the\n> > new information at the end of the transaction you only lock \n> the indexes\n> > for the amount of time it takes to write them.\n> > \n> > I see a 'vacuum analyze' being less of a problem than 'vacuum'.\n> > Any of you experts can contradict my assumptions.\n> \n> Good point...I seem to recall that at one point, there was a \n> lock imposed\n> on one of hte pg_ tables when a vacuum is tarted, since it \n> has to update a\n> couple of the rows in that table...has that lock been removed \n> with MVCC?\n> Vadim?\n\nWell, even if a vacuum locks whatever 'pg_'table-row that holds the\nindexing statistics for the table in question MVCC won't block the\noptimizer's reads. As long as there are no more vacuum analyzes run\nthere shouldn't even be a waiting transaction.\n\t-DEJ\n",
"msg_date": "Thu, 7 Jan 1999 11:31:56 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Re: [PORTS] vacuum takes too long"
}
] |
[
{
"msg_contents": "> > With MVCC an occasional 'vacuum analyze' should only be \n> noticed from the\n> > performance improvements. As far as I can tell most of the \n> work done by\n> > an analyze is in reading the table data. If you make sure \n> to write the\n> > new information at the end of the transaction you only lock \n> the indexes\n> > for the amount of time it takes to write them.\n> > \n> > I see a 'vacuum analyze' being less of a problem than 'vacuum'.\n> > Any of you experts can contradict my assumptions.\n> \n> The problem is that vacuum analyze does both vacuum and analyze. \n> Analyze takes so long, we figured we might as well vacuum \n> too. Maybe we\n> need to change that.\nIt seems that VACUUM would have to lock most of the rows in a table even\nwith MVCC; where as ANALYZE can benefit directly from MVCC never\nblocking a reader. I for one agree in the separation (hey I always\nthought they were separate).\nHow hard would it be to have VACUUM cheat on the table write? Have\nVACUUM take advantage of the fact that we are actually using a file\nsystem file as much as is possible in it's VACUUM. Therefore the actual\nmoving of the rows could be accomplished in the time it takes to select\nall the rows into a new table file and then change the file. There\nmight be some issues with file-node trouble in the system catalogs, but\nthose could be taken care of quickly as well. The only things that\nyou'd have to watch for is a write to the table in the middle of your\nreading of the rows. CLUSTER could also use the same system with a\norder by on an index. Let me know what you think.\nYou know what else... MVCC would allow us to ignore updating indexes on\na COPY if combined with the new quicker ANALYZE after the COPY.\n\tKeeping the ideas coming,\n\t-DEJ\n",
"msg_date": "Thu, 7 Jan 1999 12:17:09 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Re: [PORTS] vacuum takes too long"
}
] |
[
{
"msg_contents": "Sorry for posting it here again, but I haven't corrected my subscriptions\nyet. It's just a small patch to ecpg to keep it in sync with gram.y.\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!",
"msg_date": "Thu, 7 Jan 1999 19:40:33 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "ecpg patch"
},
{
"msg_contents": "> Sorry for posting it here again, but I haven't corrected my subscriptions\n> yet. It's just a small patch to ecpg to keep it in sync with gram.y.\n> \n> Michael\n> -- \n> Michael Meskes | Go SF 49ers!\n> Th.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\n> Tel.: (+49) 2431/72651 | Use Debian GNU/Linux!\n> Email: [email protected] | Use PostgreSQL!\n\nApplied.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Thu, 7 Jan 1999 14:10:40 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] ecpg patch"
}
] |
[
{
"msg_contents": "I think thequestions below are very good ones from the Windows side of\nthe fence. Six months ago, I would have said PHP is what the open\nsource community uses instead of ASP and ADO, but recently there seems\nto have been real Win32 progress with PostgreSQL, and I'm sure we can\nbe more specific.\n\nDoes anyone on [HACKERS] want to take a stab at answering G. Stangel?\n\nI'll try to work any detailed answers into the right place on the web\nsite.\n\nHal\n\n------- Start of forwarded message -------\nMessage-ID: <01BE3A31.6A0566E0@JAYSON>\nFrom: \"G. Jayson Stangel\" <[email protected]>\nTo: \"'[email protected]'\" <[email protected]>\nSubject: Questions?\nDate: Thu, 7 Jan 1999 11:32:29 -0700\nMIME-Version: 1.0\nContent-Type: text/plain; charset=\"iso-8859-1\"\nContent-Transfer-Encoding: quoted-printable\n\nIf you are not the person to answer these questions, please forward to =\nthe correct person.\n\nCurrently I am using Microsoft SQL 6.5, we are starting a large project =\nthat our licensee would not cover. We have just put up a linux machine =\nas an experiment. While researching linux, I ran across your site. I =\nbrowsed through the competition section, but could not find all the info =\nI need.\nI am trying to figure out how PostgreSQL differs from Microsoft SQL. I =\nknow the price issue is huge, but how about things like transactions per =\nsecond? Can I integrate PostgreSQL into my ASP pages? (with the ODBC =\ninterface I would assume I could.) Does it support ADO? Any other =\ndifferences positive or negative please.\n\n\nG. Jayson Stangel\[email protected]\nTEL: 485-8532 . FAX: 494-2938\n\n\n------- End of forwarded message -------\n",
"msg_date": "07 Jan 1999 13:06:39 -0600",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": true,
"msg_subject": "[\"G. Jayson Stangel\" <[email protected]>] Questions?"
}
] |
[
{
"msg_contents": "> > I'll be working on the binaries soon. Where should I put them once \n> > done? Mail to you, or somebody? Can't find any incoming \n> directory on \n> > the ftp... No idea to put in anything in the docs before we can say \n> > where it is located...\n> \n> Magnus is building some WIN32 client-side binaries and needs \n> a place to put them.\n> \n> So I just created:\n> doc - tar files of daily builds of current html documentation\n> .incoming - world-writable file dropoff\nOk. I've put the client binaries in the .incoming directory. When it's moved\nsomewhere permanently, whoever does that should add a blurb in the Win32\ninstall notes.\n\n\n//Magnus\n",
"msg_date": "Thu, 7 Jan 1999 20:22:48 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: Postgresql INSTALL-file & Windows"
}
] |
[
{
"msg_contents": "> I think thequestions below are very good ones from the Windows side of\n> the fence. Six months ago, I would have said PHP is what the open\n> source community uses instead of ASP and ADO, but recently there seems\n> to have been real Win32 progress with PostgreSQL, and I'm sure we can\n> be more specific.\n> \n> Does anyone on [HACKERS] want to take a stab at answering G. Stangel?\n\nSure, I'll take a shot.\n\n\n> ------- Start of forwarded message -------\n> Message-ID: <01BE3A31.6A0566E0@JAYSON>\n> From: \"G. Jayson Stangel\" <[email protected]>\n> To: \"'[email protected]'\" <[email protected]>\n> Subject: Questions?\n> Date: Thu, 7 Jan 1999 11:32:29 -0700\n> MIME-Version: 1.0\n> Content-Type: text/plain; charset=\"iso-8859-1\"\n> Content-Transfer-Encoding: quoted-printable\n> \n> If you are not the person to answer these questions, please \n> forward to = the correct person.\n> \n> Currently I am using Microsoft SQL 6.5, we are starting a \n> large project =\n> that our licensee would not cover. We have just put up a \n> linux machine =\n> as an experiment. While researching linux, I ran across your \n> site. I =\n> browsed through the competition section, but could not find \n> all the info =\n> I need.\n> I am trying to figure out how PostgreSQL differs from \n> Microsoft SQL. I =\n> know the price issue is huge, but how about things like \n> transactions per =\n> second? \nIn my experience MS SQL Server is quite a bit faster. Can't give you any\nhard numbers though, since I have not tried them on identical machines.\n\n> Can I integrate PostgreSQL into my ASP pages? (with \n> the ODBC =\n> interface I would assume I could.) Does it support ADO? Any other =\n> differences positive or negative please.\n\nYes, you can integrate it in your ASP pages using ODCB. ADO can work with\nany ODBC drivers, so it should work fine. (I haven't actually _tried_ this,\nbut it should work - since ADO can act just as a wrapper around ODBC, and\nthe ODBC driver is supposed to work).\nYou can also use the native libpq interface, but you can not interface\ndirectly into your ASP pages with that one. You will either have to create a\nCOM wrapper around it, or you will have to write it all as an ISAPI\nextension. All in all, the ODBC/ADO approach is probably much easier and\njust as good.\n\n//Magnus\n",
"msg_date": "Thu, 7 Jan 1999 20:32:56 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>] Question s?"
}
] |
[
{
"msg_contents": "> > Thomas, do you need help on outer joins?\n> \n> Yes. I'm going slowly partly because I get distracted with other\n> Postgres stuff like docs, and partly because I don't understand all of\n> the pieces I'm working with.\n> \n> I've identified the place in the MergeJoin code where the null filling\n> for outer joins needs to happen, and have the \"merge walk\" code done.\n> But I don't have the supporting code which actually would know how to\n> null-fill a result tuple from the left or right. I thought \n> you might be\n> interested in that?\n> \n> I've done some work in the parser, and can now do things like:\n> \n> postgres=> select * from t1 join t2 using (i);\n> NOTICE: JOIN not yet implemented\n> i|j|i|k\n> -+-+-+-\n> 1|2|1|3\n> (1 row)\n> \n> But this is just an inner join, and the result isn't quite right since\n> the second \"i\" column should probably be omitted. At the moment I\n> transform it from the syntax above into existing parse nodes, and\n> everything from there on works.\n\nI could be wrong (I don't have a copy of the standard), but I don't\nbelieve that the above syntax follows the standard. Let me know if I'm\nwrong, but my understanding of the syntax would be more like:\n SELECT * FROM t1 JOIN t2 ON (t1.i = t2.i);\nwith the same result set as you listed (t2.i isn't suppressed).\nThis would have a difference in approach from the above. If I wanted to\njoin on columns with different names I couldn't use your syntax (as one\nexample).\n\t-DEJ\n",
"msg_date": "Thu, 7 Jan 1999 14:26:03 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Outer Joins (and need CASE help)"
},
{
"msg_contents": "> I could be wrong (I don't have a copy of the standard), but I don't\n> believe that the above syntax follows the standard. Let me know if \n> I'm wrong, but my understanding of the syntax would be more like:\n> SELECT * FROM t1 JOIN t2 ON (t1.i = t2.i);\n> with the same result set as you listed (t2.i isn't suppressed).\n> This would have a difference in approach from the above. If I wanted \n> to join on columns with different names I couldn't use your syntax (as \n> one example).\n\nThe standard allows both syntaxes; USING is simpler to type, and ON is\nmore general, as you point out.\n\nIn fact, the standard is annoyingly helpful in allowing multiple ways to\nwrite the same query. Makes the parsing and parse tree transformation\nmore complicated :(\n\n - Tom\n",
"msg_date": "Fri, 08 Jan 1999 05:32:01 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Outer Joins (and need CASE help)"
}
] |
[
{
"msg_contents": "My experience with MS SQL 6.5(my first database):\n wrote an application or managing risk in natural-gas commodities\n[PowerBuilder] (~1 yr.)\n wrote several shopping cart and invoicing applications [ASP] (~6\nmonths)\n\nMy experience with PostgreSQL v6.3 - v6.4.2(my favorite database):\n wrote several web sites for the advertising firm I now work at [PHP]\n(~1 yr.)\n\nI also have no formal database training. Therefore everything that I've\nlearned was because I had to (you always learn those best).\n\nWhat MS SQL 6.5 server has over PostgreSQL v6.4.2:\n1. row-level-locking promotable to page and table\n (therefor faster concurrent transactions)\n2. referential integrity(RI)/foreign key implementation\n (not including triggers [Postgres has those too])\n3. stored procedures that can return multiple rows\n (a design flaw in earlier versions of Postgres that hasn't yet been\ncorrected.)\n4. fully implemented dynamic precision datatypes ie NUMERIC, DECIMAL\n5. a fuller implementation of the SQL92 standard\n (only slightly)\n6. multiple database access from any query\n (to tell the truth I haven't missed this one at all)\n7. server replication\n8. automagically clustered tables\n (this could as easily be considered a drawback)\n9. service-pack patches of the server\n (don't let the fact that it's larger than the software it's patching\nbother you)\n10. all those nifty Microsoft interface programs\n (yeah right)\n\nWhat PostgreSQL v6.4.2 server has over MS SQL 6.5\n1. PRICE!!!!!\n (this includes unlimited licenses)\n2. stability\n (anything in Windows is subject to crashes)\n3. support\n (the Postgres mailing lists outdo Microsoft support by a factor of\n1000)\n4. You have the code\n (if you want something added you can do it)\n4. clustering on any index\n5. multiple PL languages for functions (a.k.a. stored procedures)\n6. type extensibility with full operators and aggregates\n (I've never even seen a MS SQL 6.5 extended type)\n7. a larger and more versatile base-type set\n (have a look at interval)\n8. database size is only as big as it needs to be\n (you don't need to set up data/log blocks in Postgres)\n9. regular expression(regex) matching of strings\n (I hope you know what regex is)\n10. functional indexes\n (you can index upper(my_col), think about it)\n11. unlimited tables in a query (limited only by your ram)\n (MS SQL 6.5 is limited to 16 including those in self joins and\nsubselects)\n12. multiple index methods\n (one method might work better for certain data distributions and\ntypes)\n13. a rule system that by itself could make up for the lack of RI\n14. a developement/release cycle that is faster than Microsoft patches\nand contain more feature improvements than Microsoft releases.\n (no joke)\n15. cascading triggers and rules\n (try triggering another tables trigger from a trigger in MS SQL 6.5,\nit won't happen)\n16. 8k row limit (settable at compile time)\n (MS SQL 6.5 has a 2k row limit, just try it and see)\nI could go on but I have to get to the dentist.\n\nIn PostgreSQL v6.5, scheduled to beta in February 1999, the locking\nsystem will be better then MS SQL 6.5 (already implemented in CSV\ntree[see the web site] if you want to compare speed), therefore\nconcurrent transactions should far outstrip MS SQL 6.5. PostgreSQL v6.5\nwill also have a fuller set of the SQL92 standard (including NUMERIC AND\nDECIMAL) then MS SQL 6.5. RI should be fully implemented by then (no\npromises, I'm not the one doing the work). I'm hoping Jan would have\nrewritten the function system to allow for multiple row return as well.\n\nI replicate a PostgreSQL server using a copy. And there are so many\ncool PostgreSQL interfaces written by many people out there I'll let you\nreview them for yourself.\n\nOh well, off to the dentist now (pray for me),\n\t-DEJ\n",
"msg_date": "Thu, 7 Jan 1999 15:46:02 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>] Question s?"
}
] |
[
{
"msg_contents": "Hi,\n\nI have created a new patch for compiling PostgreSQL on Windows.\n\nIt requires:\n- Cygwin B20.1\n- a crypt library\n- the cygipc package\n- added the included headers (endian.h into Cygwin's include/, tcp.h into\ninclude/netinet and un.h into include/sys)\n\nNew features:\n- it creates only postgres.exe (a hybrid between .EXE and .DLL) - it is\nsimilar to executables on Unix created with gcc -export-dynamic\n- it should be possible to run it on non-binary mounts (thanks to Craig\nSetera)\n\nBugs:\n- dynamic loading still doesn't work (but it is possible to create the\nmodules, e.g. plpgsql.dll)\n- sometimes the backend survives only one connection from a client, it\ncrashes before the second connection is established. I have had not time to\ninvestigate this yet.\n\nThe binaries are also available, but they are a bit huge (~3 MB in *.tgz)\nbecause it is compiled with debug info.\n\nI think it should be possible to begin incorporating the patches to the main\nsources. Some of the changes are new files. And the rest is ifdef-ed (or\nifeq-ed in Makefiles), so it should not interfere with other ports, I hope\n:-)\n\n\t\t\tDan",
"msg_date": "Fri, 8 Jan 1999 14:26:54 +0100 ",
"msg_from": "Horak Daniel <[email protected]>",
"msg_from_op": true,
"msg_subject": "New patches for running PostgreSQL on Windows"
},
{
"msg_contents": "[Charset iso-8859-2 unsupported, filtering to ASCII...]\n> Hi,\n> \n> I have created a new patch for compiling PostgreSQL on Windows.\n> \n> It requires:\n> - Cygwin B20.1\n> - a crypt library\n> - the cygipc package\n> - added the included headers (endian.h into Cygwin's include/, tcp.h into\n> include/netinet and un.h into include/sys)\n> \n> New features:\n> - it creates only postgres.exe (a hybrid between .EXE and .DLL) - it is\n> similar to executables on Unix created with gcc -export-dynamic\n> - it should be possible to run it on non-binary mounts (thanks to Craig\n> Setera)\n> \n> Bugs:\n> - dynamic loading still doesn't work (but it is possible to create the\n> modules, e.g. plpgsql.dll)\n> - sometimes the backend survives only one connection from a client, it\n> crashes before the second connection is established. I have had not time to\n> investigate this yet.\n> \n> The binaries are also available, but they are a bit huge (~3 MB in *.tgz)\n> because it is compiled with debug info.\n> \n> I think it should be possible to begin incorporating the patches to the main\n> sources. Some of the changes are new files. And the rest is ifdef-ed (or\n> ifeq-ed in Makefiles), so it should not interfere with other ports, I hope\n> :-)\n> \n> \t\t\tDan\n> \n> \n\nI think it is good to merge this into the tree. Tracking our tree\nyourself is very difficult.\n\nApplied. The only problem is that this part of the patch did not apply\nbecause the function has changed. In fact, there is no call fdopen()\nleft in the entire code tree.\n\nDevelopers, be warned that all open()'s need a win32 test for 'binary'\nopening, and Makefiles not need a $(X) optional extension added.\n\n---------------------------------------------------------------------------\n\n***************\n*** 88,93 ****\n void\n pq_init(int fd)\n {\n \tPfin = fdopen(fd, \"r\");\n \tPfout = fdopen(dup(fd), \"w\");\n \tif (!Pfin || !Pfout)\n--- 88,102 ----\n void\n pq_init(int fd)\n {\n+ #ifdef __CYGWIN32__\n+ \t/* the fd file handle isn't readable :)\n+ \t it can probably be an error in Cygwin B19.3 */\n+ \tint flags, res;\n+ \t\n+ \tflags = O_RDWR;\n+ \tres = fcntl(fd, F_SETFL, flags);\n+ #endif\n+ \t\n \tPfin = fdopen(fd, \"r\");\n \tPfout = fdopen(dup(fd), \"w\");\n \tif (!Pfin || !Pfout)\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 17 Jan 1999 01:16:58 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] New patches for running PostgreSQL on Windows"
},
{
"msg_contents": "[Charset iso-8859-2 unsupported, filtering to ASCII...]\n> Hi,\n> \n> I have created a new patch for compiling PostgreSQL on Windows.\n> \n> It requires:\n> - Cygwin B20.1\n> - a crypt library\n> - the cygipc package\n> - added the included headers (endian.h into Cygwin's include/, tcp.h into\n> include/netinet and un.h into include/sys)\n\nI have created an include/win32 directory, and put these files in there\nwith a README.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 17 Jan 1999 01:26:52 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] New patches for running PostgreSQL on Windows"
}
] |
[
{
"msg_contents": "Hi,\n\nI found a bug in the notify handler of the libpgtcl interface.\nIf I have setup a notify handler and the backend dies, the tcl interpreter\nwill not remove the handler and loop forever with a bad fd in select().\nI tested the bug with Tcl7.5 and the following code:\n\n$ wish\n% set tcl_version\n7.5\n% set tk_version\n4.1\n% source pg_test.tcl\n% connect dz\npgsql4\n% sql select 123\n123\n% listen\n% notify\n% --> callback test\n\n# From another shell kill the backend:\n$ kill 3937\n\n# Now the wish shell is looping forever in the following syscall:\n$ strace -p 3934\nselect(5, [0 3 4], [], [], NULL) = -1 EBADF (Bad file number)\nselect(5, [0 3 4], [], [], NULL) = -1 EBADF (Bad file number)\nselect(5, [0 3 4], [], [], NULL) = -1 EBADF (Bad file number)\nselect(5, [0 3 4], [], [], NULL) = -1 EBADF (Bad file number)\nselect(5, [0 3 4], [], [], NULL) = -1 EBADF (Bad file number)\n...\n\n# The fd 4 doesn't exists anymore because it has been closed by pqReadData.\n$ ls -l /proc/3934/fd\ntotal 0\nlrwx------ 1 dz users 64 Jan 8 16:00 0 -> [0801]:2081\nlrwx------ 1 dz users 64 Jan 8 16:00 1 -> [0801]:2081\nlrwx------ 1 dz users 64 Jan 8 16:00 2 -> [0801]:2081\nlrwx------ 1 dz users 64 Jan 8 16:00 3 -> [0000]:12617\n\nThis is the test file loaded in the above example:\n\n# pg_test.tcl -- \n#\n# Test libpgtcl interface\n\nload libpgtcl.so.2\n\nproc connect {{db dz}} {\n global conn\n set conn [pg_connect $db]\n}\n\nproc sql {args} {\n global conn\n if {$args != {}} {\n set cmd $args\n } else {\n set cmd \"select 1\"\n }\n set res [pg_exec $conn $cmd]\n switch [pg_result $res -status] {\n PGRES_TUPLES_OK {\n for {set n 0} {$n < [pg_result $res -numTuples]} {incr n} {\n puts [pg_result $res -getTuple $n]\n }\n }\n PGRES_COMMAND_OK {\n\t}\n default {\n puts [pg_result $res -status]\n }\n }\n pg_result $res -clear\n}\n\nproc disconnect {} {\n global conn\n pg_disconnect $conn\n set conn {}\n}\n\nproc listen {{relname test}} {\n global conn\n pg_listen $conn $relname \"callback $relname\"\n}\n\nproc unlisten {{relname test}} {\n global conn\n pg_listen $conn $relname {}\n}\n\nproc notify {{relname test}} {\n global conn\n sql notify $relname\n}\n\nproc callback {args} {\n puts \"--> callback $args\"\n}\n\n# end of file\n\n\nThe following patch seems to fix the problem:\n\n*** src/interfaces/libpgtcl/pgtclId.c.orig\tMon Sep 21 03:02:03 1998\n--- src/interfaces/libpgtcl/pgtclId.c\tFri Jan 8 16:27:51 1999\n***************\n*** 638,649 ****\n Pg_Notify_FileHandler (ClientData clientData, int mask)\n {\n \tPg_ConnectionId *connid = (Pg_ConnectionId *) clientData;\n \n \t/*\n \t * Consume any data available from the SQL server (this just buffers\n \t * it internally to libpq; but it will clear the read-ready condition).\n \t */\n! \tPQconsumeInput(connid->conn);\n \n \t/* Transfer notify events from libpq to Tcl event queue. */\n \tPgNotifyTransferEvents(connid);\n--- 638,656 ----\n Pg_Notify_FileHandler (ClientData clientData, int mask)\n {\n \tPg_ConnectionId *connid = (Pg_ConnectionId *) clientData;\n+ \tint status;\n \n \t/*\n \t * Consume any data available from the SQL server (this just buffers\n \t * it internally to libpq; but it will clear the read-ready condition).\n+ \t *\n+ \t * Must check the connection status and stop the event source for\n+ \t * closed connection. -- dz\n \t */\n! \tstatus = PQconsumeInput(connid->conn);\n! \tif (status == 0) {\n! \t\tPgStopNotifyEventSource(connid);\n! \t}\n \n \t/* Transfer notify events from libpq to Tcl event queue. */\n \tPgNotifyTransferEvents(connid);\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+\n",
"msg_date": "Fri, 8 Jan 1999 16:41:48 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": true,
"msg_subject": "bug in libpgtcl listen"
},
{
"msg_contents": "Massimo Dal Zotto <[email protected]> writes:\n> I found a bug in the notify handler of the libpgtcl interface.\n> If I have setup a notify handler and the backend dies, the tcl interpreter\n> will not remove the handler and loop forever with a bad fd in select().\n\nGood point. I don't like your particular fix though, since I'm not sure\nthat Tcl will always call the file event handler in that situation ---\nother platforms and/or Tcl releases might behave differently than yours.\n\nI put the check into PgNotifyTransferEvents() instead, from which it\nwill get called in the same conditions as your fix, and it will also get\ncalled after every PQexec():\n\n*** src/interfaces/libpgtcl/pgtclId.c\tSun Sep 20 21:02:03 1998\n--- ./pgtclId.c\tSun Jan 17 15:30:53 1999\n***************\n*** 582,587 ****\n--- 582,596 ----\n \t\tTcl_QueueEvent((Tcl_Event *) event, TCL_QUEUE_TAIL);\n \t\tfree(notify);\n \t}\n+ \n+ \t/*\n+ \t * This is also a good place to check for unexpected closure of the\n+ \t * connection (ie, backend crash), in which case we must shut down the\n+ \t * notify event source to keep Tcl from trying to select() on the now-\n+ \t * closed socket descriptor.\n+ \t */\n+ \tif (PQsocket(connid->conn) < 0)\n+ \t\tPgStopNotifyEventSource(connid);\n }\n \n /*\n\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 17 Jan 1999 16:19:54 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] bug in libpgtcl listen "
}
] |
[
{
"msg_contents": "Hi,\n\nit seems that the year handling in pgsql dates is not very consistent:\n\ndz=> create table a (n int, x date);\nCREATE\ndz=> insert into a values (1,'02-02-0');\nINSERT 259246 1\ndz=> insert into a values (2,'02-02-00');\nINSERT 259247 1\ndz=> insert into a values (3,'02-02-000');\nINSERT 259248 1\ndz=> insert into a values (4,'02-02-0000');\nINSERT 259249 1\ndz=> insert into a values (5,'02-02-00000');\nINSERT 259250 1\ndz=> insert into a values (6,'02-02-1');\nINSERT 259251 1\ndz=> insert into a values (7,'02-02-01');\nINSERT 259252 1\ndz=> insert into a values (8,'02-02-001');\nINSERT 259253 1\ndz=> insert into a values (9,'02-02-0001');\nINSERT 259254 1\ndz=> insert into a values (10,'02-02-00001');\nINSERT 259255 1\ndz=> insert into a values (11,'02-02-69');\nINSERT 259256 1\ndz=> insert into a values (12,'02-02-069');\nINSERT 259257 1\ndz=> insert into a values (13,'02-02-0069');\nINSERT 259258 1\ndz=> insert into a values (14,'02-02-00069');\nINSERT 259259 1\ndz=> insert into a values (15,'02-02-71');\nINSERT 259260 1\ndz=> insert into a values (16,'02-02-071');\nINSERT 259261 1\ndz=> insert into a values (17,'02-02-0071');\nINSERT 259262 1\ndz=> insert into a values (18,'02-02-00071');\nINSERT 259263 1\ndz=> select * from a;\n n|x \n--+-------------\n 1|02-02-2000 \n 2|02-02-2000 \n 3|02-02-2000 \n 4|02-02-0001 BC\n 5|02-02-2000 \n 6|02-02-2001 \n 7|02-02-2001 \n 8|02-02-2001 \n 9|02-02-0001 \n10|02-02-2001 \n11|02-02-2069 \n12|02-02-2069 \n13|02-02-0069 \n14|02-02-2069 \n15|02-02-1971 \n16|02-02-1971 \n17|02-02-0071 \n18|02-02-1971 \n\nThe problem I see is that the same number is converted to a different year\ndepending on the number of digits and the number itself. I think that this\nkind of things are the most likely sources of Y2K troubles.\nA more consistent approach would be to treat the year literally and let \nany smart hack with dates entirely to the user under his responsability.\nOnly then we could declare pgsql as full Y2K compliant.\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+\n",
"msg_date": "Fri, 8 Jan 1999 16:53:55 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": true,
"msg_subject": "postgres and year 2000"
},
{
"msg_contents": "> it seems that the year handling in pgsql dates is not very consistent:\n> The problem I see is that the same number is converted to a different \n> year depending on the number of digits and the number itself. I think \n> that this kind of things are the most likely sources of Y2K troubles.\n> A more consistent approach would be to treat the year literally and \n> let any smart hack with dates entirely to the user under his \n> responsability.\n\nI agree that there are some cases in your examples which should be\ngiving a different result. Not *every* example you gave led to an\nincorrect or unpredictable result, so could you please identify those\nyou feel are a problem? In glancing at the examples, the ones with zero\nvalue (but lots of zeros) seem to be a problem, and perhaps some or all\nof the ones with an odd number of year digits. Any others?\n\nWe do need to handle two-digit years, and we currently do so using 1970\nas the break point. I've read recently that some industries are using a\n\"50/50 rule\" for 2 digit conversions, where 1950 is the break point.\nDon't know if we should try to use that (rather than the \"Unix rule\" :),\nsince it really doesn't offer a magic cure for all date possibilities.\n\n> Only then we could declare pgsql as full Y2K compliant.\n\nfwiw, the date candidates which are failing are outside the range of\nnormal usage or could be considered mal-formed. But I should be able to\nget a fix in for it, and can post patches. Let me know what cases you\nwould like tested and fixed (but let's not bog down in discussion on\ntwo-digit issues).\n\n - Tom\n",
"msg_date": "Fri, 08 Jan 1999 16:42:53 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "> it seems that the year handling in pgsql dates is not very consistent:\n> the same number is converted to a different year\n> depending on the number of digits and the number itself.\n> I think that this kind of things are the most likely sources of Y2K \n> troubles. A more consistent approach would be to treat the year \n> literally and let any smart hack with dates entirely to the user under \n> his responsability.\n\nOK, here is a patch which tightens up the date/time interpretation of\nrandom input :) \n\nFor numeric fields, anything greater than two digits in length is\nrequired to be either a \"concatenated date/time\" or an explicit date.\nAlso, now *only* two digit years are converted to a year in a nearby\ncentury.\n\n> Only then we could declare pgsql as full Y2K compliant.\n\nI agree that the interpretation of 3 and 5 digit years was not right for\nyears with leading zeros. This patch should fix that.\n\n\"Y2K compliant\" means to me that the behavior of well-formed dates must\nbe predictable and understandable. Dates *intended* to be in the 19th or\n20th century should end up being interpreted that way.\n\nThe Postgres date/time types allow year ranges from (roughly) Julian Day\nzero, which is Nov 23, 4714 BC (I wrote it as -4713 in my notes, so I\nmay be off by a year), to far into the future. So if you enter a three\ndigit year, that's what you are going to get (or will, with this patch).\nI don't see that as causing Y2K trouble; you could just as easily call\nit Y1K trouble :)\n\nAnyway, there is a tradeoff between flexibility in date specification\nand catching unintended typos. Dan suggested that all dates be required\nto have 4 digit years, which may be overly harsh for some users and\ndevelopers. Should we have a way to specify the range of valid dates and\ntimes so databases installed for specific applications can have more\nrestrictive constraints on input? I already parameterized the year range\ncheck with a #define. Perhaps we should have it be able to be overridden\nby ./configure or by a Makefile.custom??\n\nA patch is enclosed which fixes all of the problems I can see in your\nexamples. It basically does no century adjustments unless the year is\nexactly two digits, and interprets anything greater than two digits as\nan explicit year (or a concatenated date if 8 or 6 or 5 digits). It adds\n\"two digit year\" adjustments to concatentated dates, which I had\napparently omitted. It also does more rigorous checking on the usage of\n\"BC\", so that if you specify a negative or zero year along with BC it\nwill complain (before, it just flipped the sign on the year).\n\nPlease install and let me know what you think:\n cd src/backend/utils/adt\n patch < dt.c.patch\n cd ../../..\n make install\n\nI will apply this to the development cvs tree sometime soon, unless\nfolks find problems or need something different.\n\nMassimo, would you be interested in adding some of your test cases to\nthe datetime or horology regression test? Send me patches and I'll add\nthem in...\n\n - Tom",
"msg_date": "Sat, 09 Jan 1999 18:38:18 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "> it seems that the year handling in pgsql dates is not very consistent:\n> The problem I see is that the same number is converted to a different \n> year depending on the number of digits and the number itself.\n\nI've posted a patch intended for v6.4 and v6.4.2 at\n ftp://postgresql.org/pub/patches/dt.c.patch\n\nwhich addresses some date interpretation problems, mostly for dates\ncontaining leading zeros and/or an odd number of digits (I'm referring\nto this as the \"Y1K problem\" :). It also fixes problems with two-digit\nyears for \"concatenated dates\" like \"990110\" and adds a \"yydoy\" 5 digit\nconcatenated date, like \"99010\".\n\nI'll claim that none of these problems or fixes were related directly to\na \"Y2K\" problem, since it didn't work in \"Y1900\" either :/\n\ncvs commit notes are included below. Let me know if there are any\nproblems.\n\n - Tom\n\nBe more careful to check input string lengths as well as values\n when deciding whether a field is a year field. Assume *anything* longer\n than 2 digits (if it isn't a special-case doy) is a valid year.\n This should fix the \"Y1K\" and \"Y10K\" problems pointed out by Massimo \n recently.\nCheck usage of BC to require a positive-valued year; before just used\n to flip the sign of the year without checking. This led to problems\n near year zero.\nAllow a 5 digit \"concatenated date\" of 2 digit year plus day of year.\nDo 2->4 digit year correction for 6 and 5 digit \"concatenated dates\".\n Somehow forgot this originally. Guess not many folks use it...\n",
"msg_date": "Sun, 10 Jan 1999 18:08:16 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and dates (year 2000? not!)"
},
{
"msg_contents": "\"Thomas G. Lockhart\" <[email protected]> writes:\n\n> We do need to handle two-digit years, [...]\n\nIs it at all possible to get away with _not_ doing so? It is, after\nall, incredibly stupid to use two-digit years in anything but spoken\nconversation, so in a way, I'd prefer computer systems to blankly\nrefuse them. If they're allowed at all, I'd say parse them so that a\nyear specification of '99' means the actual year 99. _Not_ 1999.\n\nThen again, I also think computer systems should refuse to accept\nnon-ISO8601 date specifications, so I may be a bit too pedantic. :-)\n\n-tih\n-- \nPopularity is the hallmark of mediocrity. --Niles Crane, \"Frasier\"\n",
"msg_date": "11 Jan 1999 08:01:54 +0100",
"msg_from": "Tom Ivar Helbekkmo <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "On 11 Jan 1999, Tom Ivar Helbekkmo wrote:\n\n> \"Thomas G. Lockhart\" <[email protected]> writes:\n> \n> > We do need to handle two-digit years, [...]\n> \n> Is it at all possible to get away with _not_ doing so? It is, after\n> all, incredibly stupid to use two-digit years in anything but spoken\n> conversation, so in a way, I'd prefer computer systems to blankly\n> refuse them. If they're allowed at all, I'd say parse them so that a\n> year specification of '99' means the actual year 99. _Not_ 1999.\n\nFalling back to a Unix standard...type 'cal 99' and see which year you\nget :)\n\nI agree with Tom on this...if someone types a year of 99, we should\npresume that whomever entered it knew what they were entering, and/or that\nthe programmer of the interface had enough sense to program checks into\nit...\n\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 11 Jan 1999 03:26:51 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "> > \"Thomas G. Lockhart\" <[email protected]> writes:\n> > > We do need to handle two-digit years, [...]\n> > Is it at all possible to get away with _not_ doing so? It is, after\n> > all, incredibly stupid to use two-digit years in anything but spoken\n> > conversation, so in a way, I'd prefer computer systems to blankly\n> > refuse them. If they're allowed at all, I'd say parse them so that \n> > a year specification of '99' means the actual year 99. _Not_ 1999.\n> Falling back to a Unix standard...type 'cal 99' and see which year you\n> get :)\n> I agree with Tom on this...if someone types a year of 99, we should\n> presume that whomever entered it knew what they were entering, and/or \n> that the programmer of the interface had enough sense to program \n> checks into it...\n\nWell, we're all Toms here, but I'll assume you are trying to agree with\nTom H.\n\nYou may think that ISO-8601 is the way to go, and I may think that\nISO-8601 is the way to go, but it isn't yet a universal usage, so why\nshould we presume that it is? That timezone stuff is pretty annoying and\nconfusing too, so perhaps we should revert Postgres back to its pre-v3\nGMT-only date support? :)\n\nIf we don't accept a reasonably wide range of common date and time\nspecifications, then each app will have to, or may have to, do that.\nistm that it is *much* easier to write an app which restricts data entry\nto a particular style, which is then forwarded to our backend, than to\naccept *only* a particular style in the backend and expect every client\napp, in every supported language, on every platform, to learn how to\naccept general date values and then convert them to a single specific\none.\n\nI should also point out that in a recent review of our database the\ndate/time I/O was mentioned (with 2 or 3 other features) as being\nparticularly noteworthy. So *someone* thought it was a nice feature :)\n\nI suppose we could consider a compile-time or run-time option to\nconstrain dates to a single style. Seems a bit of overkill, and\npersonally I'd like to work on outer joins, but...\n\n - Tom\n",
"msg_date": "Tue, 12 Jan 1999 02:48:23 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "\"Thomas G. Lockhart\" <[email protected]> writes:\n> If we don't accept a reasonably wide range of common date and time\n> specifications, then each app will have to, or may have to, do that.\n\nJust to throw another Tom's opinion into the mix ;-) ...\n\nI agree with Tom Lockhart on this one. If we don't provide date\ninterpretation in the backend, that doesn't make the problem go away.\nIt just means that every frontend application has to re-invent that\nsame wheel. And, no doubt, cope with bugs in their re-invention.\nGetting it right *once* is the whole idea of re-using software --- else\nwhy not expect everyone to write their own whole DBMS?\n\nA frontend programmer who has his own strong ideas about how to\ninterpret dates is certainly free to do so, and then to pass his results\nto the backend in some unambiguous format like ISO. But not many people\nwill really want to do that --- they'd much rather have a flexible and\nrobust solution provided for them.\n\nDate handling is inherently messy because there are so many different\nconventions. But we can't make that go away by decree. Guess what:\npeople will keep writing two-digit years, even after the turn of the\ncentury, and will expect their computers to understand what's meant.\n\n> I suppose we could consider a compile-time or run-time option to\n> constrain dates to a single style.\n\nI see no need to do that. A particular frontend programmer who wants\nthat behavior can make it happen himself --- should you break other\napps that may be talking to the same database server in order to do\nit for him?\n\n\t\t\tregards, tom lane\n",
"msg_date": "Tue, 12 Jan 1999 09:33:11 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000 "
},
{
"msg_contents": "Tom Lane <[email protected]> writes:\n\n> I agree with Tom Lockhart on this one.\n\nSo do I, actually. However:\n\n> > I suppose we could consider a compile-time or run-time option to\n> > constrain dates to a single style.\n> \n> I see no need to do that.\n\nNot compile-time, no. But I think it would be a good thing to have\nseveral run-time options (of which PostgreSQL already has a few), to\nspecify exactly which behavior is wanted. For two digit years, it\nmight be useful to be able to specify to the backend that they should\nbe handled as, say, 1920-2019, or as the chronologically nearest year\nthat ends in the two given digits, or maybe even as being in the\ncurrent century. When using a four digit year mode, though, I think\nit's a good idea to handle '99' as the year 99, and not e.g. 1999. It\nmay be that even this should be an option, and the dangerous mixture,\nwhere there are two years between between the starts of year '99' and\nyear '2001', should be available on front-end application request.\n\nI would suggest that the defaults be safe, though, probably ISO 8601.\n\n-tih\n-- \nPopularity is the hallmark of mediocrity. --Niles Crane, \"Frasier\"\n",
"msg_date": "12 Jan 1999 20:55:19 +0100",
"msg_from": "Tom Ivar Helbekkmo <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "> \n> \"Thomas G. Lockhart\" <[email protected]> writes:\n> > If we don't accept a reasonably wide range of common date and time\n> > specifications, then each app will have to, or may have to, do that.\n> \n> Just to throw another Tom's opinion into the mix ;-) ...\n> \n> I agree with Tom Lockhart on this one. If we don't provide date\n> interpretation in the backend, that doesn't make the problem go away.\n> It just means that every frontend application has to re-invent that\n> same wheel. And, no doubt, cope with bugs in their re-invention.\n> Getting it right *once* is the whole idea of re-using software --- else\n> why not expect everyone to write their own whole DBMS?\n> \n> A frontend programmer who has his own strong ideas about how to\n> interpret dates is certainly free to do so, and then to pass his results\n> to the backend in some unambiguous format like ISO. But not many people\n> will really want to do that --- they'd much rather have a flexible and\n> robust solution provided for them.\n> \n> Date handling is inherently messy because there are so many different\n> conventions. But we can't make that go away by decree. Guess what:\n> people will keep writing two-digit years, even after the turn of the\n> century, and will expect their computers to understand what's meant.\n> \n> > I suppose we could consider a compile-time or run-time option to\n> > constrain dates to a single style.\n> \n> I see no need to do that. A particular frontend programmer who wants\n> that behavior can make it happen himself --- should you break other\n> apps that may be talking to the same database server in order to do\n> it for him?\n> \n> \t\t\tregards, tom lane\n\nIt is nice to provide smart date interpretation in the backend but in\norder to be really Y2K compliant we *MUST* forbid any ambiguous date\nformat in the backend. If the user insists in wanting two-digits years\nin his interface he will write his own not-Y2K-compliant conversion code.\n\nSomeone mentioned the ISO-8601 standard. Could you post a summary of\nthis standard ?\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+\n",
"msg_date": "Wed, 13 Jan 1999 10:23:00 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "Massimo Dal Zotto <[email protected]> writes:\n\n> Someone mentioned the ISO-8601 standard. Could you post a summary of\n> this standard ?\n\nVery quick summary: Today is '1998-01-13', and I'm posting this at\napproximately '1998-01-13 13:02:00Z' (the Z is for Zulu, which is the\nmilitary designation for UT), or '1998-01-13 14:02:00' my local time.\n\nSee http://www.iso.ch/markete/moreend.htm for more information, and a\ndownloadable PDF version of the complete standard.\n\n-tih\n-- \nPopularity is the hallmark of mediocrity. --Niles Crane, \"Frasier\"\n",
"msg_date": "13 Jan 1999 14:04:41 +0100",
"msg_from": "Tom Ivar Helbekkmo <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "Massimo Dal Zotto <[email protected]> writes:\n> It is nice to provide smart date interpretation in the backend but in\n> order to be really Y2K compliant we *MUST* forbid any ambiguous date\n> format in the backend.\n\nWhy? Has some bureaucrat somewhere defined \"Y2K-compliant\" as meaning\n\"thou shalt stop accepting 2-digit years\"? I have not heard of any\nsuch definition and I am entirely prepared to ignore it if it exists...\nit has nothing to do with reality.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 13 Jan 1999 10:16:27 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000 "
},
{
"msg_contents": "Tom Ivar Helbekkmo <[email protected]> writes:\n> Not compile-time, no. But I think it would be a good thing to have\n> several run-time options (of which PostgreSQL already has a few), to\n> specify exactly which behavior is wanted. For two digit years, it\n> might be useful to be able to specify to the backend that they should\n> be handled as, say, 1920-2019, or as the chronologically nearest year\n> that ends in the two given digits, or maybe even as being in the\n> current century. When using a four digit year mode, though, I think\n> it's a good idea to handle '99' as the year 99, and not e.g. 1999.\n\nIIRC, we already have both behaviors (99->1999AD and 99->99AD)\navailable, but it's controlled by a combination of the DATESTYLE setting\nand the actual formatting of the particular input string.\n\nThere doesn't seem to be anything in the documentation about exactly\nhow ambiguous inputs are parsed. Thomas, maybe some text needs to\nbe added?\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 13 Jan 1999 10:22:46 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000 "
},
{
"msg_contents": "> It is nice to provide smart date interpretation in the backend but in\n> order to be really Y2K compliant we *MUST* forbid any ambiguous date\n> format in the backend. If the user insists in wanting two-digits years\n> in his interface he will write his own not-Y2K-compliant conversion \n> code.\n\nI've never heard that interpretation of Y2K compliance. Remember that\nonce a date has been interpreted, it is *never* ambiguous since it is\nstored as a full value. The rules for interpretation are not ambiguous,\nand follow common usage. \n\nThe cases you found which did not \"do the right thing\" allowed me to fix\nbugs, but imho did not illustrate fundamental problems in dealing with\ndate/time. The *only* bothersome case really is that for a Roman from\nthe year 88 sitting down in front of Postgres, entering dates, and upon\nseeing \"1988\" in a result saying: \"What the ...?\".\n\nSaaayyyyy, you're not one of those research types bringing old bodies\nback to life are you? I've seen movies about that... ;)\n\nBut back to the Roman, the modern calendar wasn't adopted until the 17th\nor 18th century, so he'd be confused anyway.\n\nRegards.\n\n - Tom ;)\n",
"msg_date": "Wed, 13 Jan 1999 15:24:11 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "> There doesn't seem to be anything in the documentation about exactly\n> how ambiguous inputs are parsed. Thomas, maybe some text needs to\n> be added?\n\nI'll bet I won't get away with \"see the file dt.c for a complete\ndescription of date/time parsing\". Darn. Will try writing something\nup...\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 15:39:03 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
},
{
"msg_contents": "The FAQ_CVS does not discuss how to retrieve \nand follow particular branches. Could someone\ngive me a quick lesson, (or better yet, update the FAQ?)\n\nI need to be able to fetch and build the current\nproduction release. Right now I'm fetching\nand building 6.5\n\n:) Clark\n",
"msg_date": "Wed, 13 Jan 1999 16:15:26 +0000",
"msg_from": "Clark Evans <[email protected]>",
"msg_from_op": false,
"msg_subject": "CVS Branch Retrieval?"
},
{
"msg_contents": "> The FAQ_CVS does not discuss how to retrieve\n> and follow particular branches. Could someone\n> give me a quick lesson, (or better yet, update the FAQ?)\n\nDon't put any new work into the FAQ! I've merged it into cvs.sgml, but\nhave not finished marking up cvs.sgml (there was some other stuff to\nmerge in, and...).\n\n> I need to be able to fetch and build the current\n> production release. Right now I'm fetching\n> and building 6.5\n\nLook at doc/src/sgml/cvs.sgml, which aside from the tags has some words\n(taken from one of scrappy's e-mails on the subject, maybe not on one of\nmain lists) on how to retrieve tagged versions.\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 17:04:13 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CVS Branch Retrieval?"
},
{
"msg_contents": "> > Not compile-time, no. But I think it would be a good thing to have\n> > several run-time options (of which PostgreSQL already has a few), to\n> > specify exactly which behavior is wanted. For two digit years, it\n> > might be useful to be able to specify to the backend that they should\n> > be handled as, say, 1920-2019, or as the chronologically nearest year\n> > that ends in the two given digits, or maybe even as being in the\n> > current century. When using a four digit year mode, though, I think\n> > it's a good idea to handle '99' as the year 99, and not e.g. 1999.\n> \n> IIRC, we already have both behaviors (99->1999AD and 99->99AD)\n> available, but it's controlled by a combination of the DATESTYLE setting\n> and the actual formatting of the particular input string.\n\nWhy not just try to parse the date according to the DATESTYLE setting\nand cough up an error if the date-parsing code doesn't find what it\nis looking for?\n\nI believe Oracle does this, but also has the to_date(string, format)\nfunction to tell the backend just what format it is getting. The\nfunction also exists without the format arg in which case it will\nuse the Oracle default.\n\nIt seems to me that either ...\n\nPostgres needs a to_date function to be told what format to use\ninstead of being expected to blindly guess what the user is giving\nit. Then overload the function s.t. calling it without the format\nwill use the current DATESTYLE.\n\nor\n\nPostgres needs a way to set the DATESTYLE to the actual format\nstring to be used to parse the input for a date field instead of\nbeing used to indicate a \"style\". This then eliminates the need\nfor the to_date function.\n\nThe latter would seem to be the easiest solution and allow the\nuser to enter many different date styles per connection simply\nby issing a \"SET DATESTYLE = 'abc-xyz'\".\n\nThomas, how hard would it be to parse an arg to \"SET DATESTYLE\"\nand use it to parse dates?\n\nJust my thoughts...\n\ndarrenk\n\n",
"msg_date": "Wed, 13 Jan 1999 18:54:46 -0500",
"msg_from": "\"Stupor Genius\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "RE: [HACKERS] postgres and year 2000 "
},
{
"msg_contents": "> Why not just try to parse the date according to the DATESTYLE setting\n> and cough up an error if the date-parsing code doesn't find what it\n> is looking for?\n\nPostgres does use the DateStyle setting to resolve input ambiguities.\n\n> I believe Oracle does this, but also has the to_date(string, format)\n> function to tell the backend just what format it is getting. The\n> function also exists without the format arg in which case it will\n> use the Oracle default.\n> It seems to me that either ...\n> Postgres needs a to_date function to be told what format to use\n> instead of being expected to blindly guess what the user is giving\n> it. Then overload the function s.t. calling it without the format\n> will use the current DATESTYLE.\n> or\n> Postgres needs a way to set the DATESTYLE to the actual format\n> string to be used to parse the input for a date field instead of\n> being used to indicate a \"style\". This then eliminates the need\n> for the to_date function.\n\nSo what problem are we trying to solve again? Stripping out\nfunctionality should buy us something useful...\n\n> Thomas, how hard would it be to parse an arg to \"SET DATESTYLE\"\n> and use it to parse dates?\n\nNot sure. It would just be a different set of parsing code.\n\n - Tom\n",
"msg_date": "Thu, 14 Jan 1999 06:59:08 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres and year 2000"
}
] |
[
{
"msg_contents": "Hi,\n\nit seems that the CREATE USER command doesn't handle the group value at all:\n\nINSERT INTO pg_group VALUES ('todos',100);\nCREATE USER miriam IN GROUP todos;\nselect * from pg_group;\ngroname|grosysid|grolist\n-------+--------+-------\ntodos | 100| \n\nI would like to have the usesysid of a new user to the grolist of all his\ngroups. I would also like to have a WITH UID clause in order to be able to\nassign an explict uid to the user.\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+\n",
"msg_date": "Fri, 8 Jan 1999 16:58:34 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": true,
"msg_subject": "bug in CREATE USER"
}
] |
[
{
"msg_contents": "Hi,\n\nI have some patches for version 6.4.2:\n\n config.patch\t\tadded some configuration options\n lock.patch\t\t\tfixes to lock debugging macros\n pgtclid.patch\t\tfix for notify bug\n tclarrays.patch\t\tadded missing code for tcl arrays\n trace.patch\t\t\tadded a missing variable\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+",
"msg_date": "Fri, 8 Jan 1999 17:07:02 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": true,
"msg_subject": "some patches for 6.4.2"
},
{
"msg_contents": "Massimo Dal Zotto <[email protected]> writes:\n> I have some patches for version 6.4.2:\n\n> config.patch\t\tadded some configuration options\n> lock.patch\t\tfixes to lock debugging macros\n> pgtclid.patch\t\tfix for notify bug\n> tclarrays.patch\t\tadded missing code for tcl arrays\n> trace.patch\t\tadded a missing variable\n\nApplied to current source tree, except pgtclid.patch modified per\nmy previous message.\n\nI still think TCL_ARRAYS is a kluge ... we need a better solution\nthat can be turned on all the time without breaking other things...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 17 Jan 1999 16:22:22 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] some patches for 6.4.2 "
}
] |
[
{
"msg_contents": "I just received this via e-mail. I am boucing the whole posting to the\npatches list. I am going to need help understanding the issues here.\n\n\n\n> Hi!\n> \n> INTERSECT and EXCEPT is available for postgresql-v6.4!\n> \n> The patch against v6.4 is included at the end of the current text\n> (in uuencoded form!)\n> \n> I also included the text of my Master's Thesis. (a postscript\n> version). I hope that you find something of it useful and would be\n> happy if parts of it find their way into the PostgreSQL documentation\n> project (If so, tell me, then I send the sources of the document!)\n> \n> The contents of the document are:\n> -) The first chapter might be of less interest as it gives only an \n> overview on SQL.\n> \n> -) The second chapter gives a description on much of PostgreSQL's\n> features (like user defined types etc. and how to use these features)\n> \n> -) The third chapter starts with an overview of PostgreSQL's internal\n> structure with focus on the stages a query has to pass (i.e. parser,\n> planner/optimizer, executor). Then a detailed description of the\n> implementation of the Having clause and the Intersect/Except logic is \n> given.\n> \n> Originally I worked on v6.3.2 but never found time enough to prepare\n> and post a patch. Now I applied the changes to v6.4 to get Intersect\n> and Except working with the new version. Chapter 3 of my documentation\n> deals with the changes against v6.3.2, so keep that in mind when\n> comparing the parts of the code printed there with the patched sources\n> of v6.4.\n> \n> Here are some remarks on the patch. There are some things that have\n> still to be done but at the moment I don't have time to do them\n> myself. (I'm doing my military service at the moment) Sorry for that\n> :-(\n> \n> -) I used a rewrite technique for the implementation of the Except/Intersect\n> logic which rewrites the query to a semantically equivalent query before\n> it is handed to the rewrite system (for views, rules etc.), planner, \n> executor etc.\n> \n> -) In v6.3.2 the types of the attributes of two select statements\n> connected by the UNION keyword had to match 100%. In v6.4 the types\n> only need to be familiar (i.e. int and float can be mixed). Since this\n> feature did not exist when I worked on Intersect/Except it \n> does not work correctly for Except/Intersect queries WHEN USED IN\n> COMBINATION WITH UNIONS! (i.e. sometimes the wrong type is used for the \n> resulting table. This is because until now the types of the attributes of \n> the first select statement have been used for the resulting table. \n> When Intersects and/or Excepts are used in combination with Unions it \n> might happen, that the first select statement of the original query\n> appears at another position in the query which will be executed. The reason\n> for this is the technique used for the implementation of \n> Except/Intersect which does a query rewrite!)\n> NOTE: It is NOT broken for pure UNION queries and pure INTERSECT/EXCEPT \n> queries!!!\n> \n> -) I had to add the field intersect_clause to some data structures \n> but did not find time to implement printfuncs for the new field.\n> This does NOT break the debug modes but when an Except/Intersect\n> is used the query debug output will be the already rewritten query.\n> \n> -) Massive changes to the grammar rules for SELECT and INSERT statements\n> have been necessary (see comments in gram.y and documentation for\n> deatails) in order to be able to use mixed queries like \n> (SELECT ... UNION (SELECT ... EXCEPT SELECT)) INTERSECT SELECT...; \n> \n> -) When using UNION/EXCEPT/INTERSECT you will get:\n> NOTICE: equal: \"Don't know if nodes of type xxx are equal\".\n> I did not have time to add comparsion support for all the needed nodes,\n> but the default behaviour of the function equal met my requirements.\n> I did not dare to supress this message!\n> \n> That's the reason why the regression test for union will fail: These\n> messages are also included in the union.out file!\n> \n> -) Somebody of you changed the union_planner() function for v6.4\n> (I copied the targetlist to new_tlist and that was removed and\n> replaced by a cleanup of the original targetlist). These chnages\n> violated some having queries executed against views so I changed\n> it back again. I did not have time to examine the differences between the\n> two versions but now it works :-)\n> If you want to find out, try the file queries/view_having.sql on \n> both versions and compare the results . Two queries won't produce a \n> correct result with your version.\n> \n> -) I included some queries in the patch (intersect.sql shows which \n> Intersect/Except/Union queries are possible now!)\n> \n> Create a new database dummy and populate it:\n> \n> $ createdb dummy\n> $ psql dummy\n> dummy=> \\i queries/create_insert.sql\n> .\n> .\n> .\n> dummy=> \\i queries/intersect.sql\n> \n> \n> So, that's it, hope my work will be useful!\n> \n> regards\n> \n> Stefan\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Fri, 8 Jan 1999 11:27:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: EXCEPT/INTERSECT for v6.4"
},
{
"msg_contents": "Thomas,\n\nThe documentation Stefan wrote for his Master's thesis\nis great stuff. Perhaps some of could be incorporated?\n\nStefan Simkovics wrote:\n> > I also included the text of my Master's Thesis. (a postscript\n> > version). I hope that you find something of it useful and would be\n> > happy if parts of it find their way into the PostgreSQL documentation\n> > project (If so, tell me, then I send the sources of the document!)\n\nI took the liberty of making a PDF file and posting it on my server:\n\nhttp://www.manhattanproject.com/stefan_simkovics_masters_thesis.pdf\n\nIf this is a problem, I'll remove it. Also, once the paper has\na permanent home, I'll put in a re-director for it.\n\n> > The contents of the document are:\n> > -) The first chapter might be of less interest as it gives only an\n> > overview on SQL.\n> >\n> > -) The second chapter gives a description on much of PostgreSQL's\n> > features (like user defined types etc. and how to use these features)\n\nThis second chapter is a great introduction. (***)\n\n> > -) The third chapter starts with an overview of PostgreSQL's internal\n> > structure with focus on the stages a query has to pass (i.e. parser,\n> > planner/optimizer, executor). Then a detailed description of the\n> > implementation of the Having clause and the Intersect/Except logic is\n> > given.\n\n\nAnd this third chapter is wonderful, it gives a very good low level\ndescription of the rewrite process and internal query representation\nin a graphic format. (****)\n\nBest,\n\nClark Evans\n",
"msg_date": "Wed, 13 Jan 1999 06:04:15 +0000",
"msg_from": "Clark Evans <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: EXCEPT/INTERSECT for v6.4"
},
{
"msg_contents": "> The documentation Stefan wrote for his Master's thesis\n> is great stuff. Perhaps some of could be incorporated?\n\nI'm sure all can be incorporated. Stefan, do we have your sources yet? I\nhaven't looked at your text, but I'm sure it can become an integral part\nof the Postgres docs. We are in sore need of both your introductory\nmaterial and your descriptive material on query processing.\n\nI would be happy to integrate it, or perhaps others would like to help.\nEither way, could you send on your sources to me and I'll put them into\nour documentation source tree as a starting point?\n\nRegards.\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 06:34:42 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: EXCEPT/INTERSECT for v6.4"
},
{
"msg_contents": "> > The documentation Stefan wrote for his Master's thesis\n> > is great stuff. Perhaps some of could be incorporated?\n> \n> I'm sure all can be incorporated. Stefan, do we have your sources yet? I\n> haven't looked at your text, but I'm sure it can become an integral part\n> of the Postgres docs. We are in sore need of both your introductory\n> material and your descriptive material on query processing.\n> \n> I would be happy to integrate it, or perhaps others would like to help.\n> Either way, could you send on your sources to me and I'll put them into\n> our documentation source tree as a starting point?\n> \n> Regards.\n> \n> - Tom\n> \n\nI just bounced his message to the patches list again. I have not\napplied it because he is looking for comments on it, and I haven't heard\nanything yet from the group.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 13 Jan 1999 02:21:08 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: EXCEPT/INTERSECT for v6.4"
},
{
"msg_contents": "> I just bounced his message to the patches list again. I have not\n> applied it because he is looking for comments on it, and I haven't \n> heard anything yet from the group.\n\nIt's something I could look at, and it would make a nice addition to\nv6.5. But I'm having a *lot* of trouble getting time to look at outer\njoins, and am nowhere close to seeing a path for completion.\n\n>From his message, I gathered that we do not yet have the sources for the\ndocs?\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 15:26:44 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: EXCEPT/INTERSECT for v6.4"
},
{
"msg_contents": "Applied. I am printing your masters thesis to read. I will add it to\nthe docs directory until I can figure out how to move it into the web\npage.\n\nI will now go through your comments and see if I can address them.\n\nStay tuned.\n\n> Hi!\n> \n> INTERSECT and EXCEPT is available for postgresql-v6.4!\n> \n> The patch against v6.4 is included at the end of the current text\n> (in uuencoded form!)\n> \n> I also included the text of my Master's Thesis. (a postscript\n> version). I hope that you find something of it useful and would be\n> happy if parts of it find their way into the PostgreSQL documentation\n> project (If so, tell me, then I send the sources of the document!)\n> \n> The contents of the document are:\n> -) The first chapter might be of less interest as it gives only an \n> overview on SQL.\n> \n> -) The second chapter gives a description on much of PostgreSQL's\n> features (like user defined types etc. and how to use these features)\n> \n> -) The third chapter starts with an overview of PostgreSQL's internal\n> structure with focus on the stages a query has to pass (i.e. parser,\n> planner/optimizer, executor). Then a detailed description of the\n> implementation of the Having clause and the Intersect/Except logic is \n> given.\n> \n> Originally I worked on v6.3.2 but never found time enough to prepare\n> and post a patch. Now I applied the changes to v6.4 to get Intersect\n> and Except working with the new version. Chapter 3 of my documentation\n> deals with the changes against v6.3.2, so keep that in mind when\n> comparing the parts of the code printed there with the patched sources\n> of v6.4.\n> \n> Here are some remarks on the patch. There are some things that have\n> still to be done but at the moment I don't have time to do them\n> myself. (I'm doing my military service at the moment) Sorry for that\n> :-(\n> \n> -) I used a rewrite technique for the implementation of the Except/Intersect\n> logic which rewrites the query to a semantically equivalent query before\n> it is handed to the rewrite system (for views, rules etc.), planner, \n> executor etc.\n> \n> -) In v6.3.2 the types of the attributes of two select statements\n> connected by the UNION keyword had to match 100%. In v6.4 the types\n> only need to be familiar (i.e. int and float can be mixed). Since this\n> feature did not exist when I worked on Intersect/Except it \n> does not work correctly for Except/Intersect queries WHEN USED IN\n> COMBINATION WITH UNIONS! (i.e. sometimes the wrong type is used for the \n> resulting table. This is because until now the types of the attributes of \n> the first select statement have been used for the resulting table. \n> When Intersects and/or Excepts are used in combination with Unions it \n> might happen, that the first select statement of the original query\n> appears at another position in the query which will be executed. The reason\n> for this is the technique used for the implementation of \n> Except/Intersect which does a query rewrite!)\n> NOTE: It is NOT broken for pure UNION queries and pure INTERSECT/EXCEPT \n> queries!!!\n> \n> -) I had to add the field intersect_clause to some data structures \n> but did not find time to implement printfuncs for the new field.\n> This does NOT break the debug modes but when an Except/Intersect\n> is used the query debug output will be the already rewritten query.\n> \n> -) Massive changes to the grammar rules for SELECT and INSERT statements\n> have been necessary (see comments in gram.y and documentation for\n> deatails) in order to be able to use mixed queries like \n> (SELECT ... UNION (SELECT ... EXCEPT SELECT)) INTERSECT SELECT...; \n> \n> -) When using UNION/EXCEPT/INTERSECT you will get:\n> NOTICE: equal: \"Don't know if nodes of type xxx are equal\".\n> I did not have time to add comparsion support for all the needed nodes,\n> but the default behaviour of the function equal met my requirements.\n> I did not dare to supress this message!\n> \n> That's the reason why the regression test for union will fail: These\n> messages are also included in the union.out file!\n> \n> -) Somebody of you changed the union_planner() function for v6.4\n> (I copied the targetlist to new_tlist and that was removed and\n> replaced by a cleanup of the original targetlist). These chnages\n> violated some having queries executed against views so I changed\n> it back again. I did not have time to examine the differences between the\n> two versions but now it works :-)\n> If you want to find out, try the file queries/view_having.sql on \n> both versions and compare the results . Two queries won't produce a \n> correct result with your version.\n> \n> -) I included some queries in the patch (intersect.sql shows which \n> Intersect/Except/Union queries are possible now!)\n> \n> Create a new database dummy and populate it:\n> \n> $ createdb dummy\n> $ psql dummy\n> dummy=> \\i queries/create_insert.sql\n> .\n> .\n> .\n> dummy=> \\i queries/intersect.sql\n> \n> \n> So, that's it, hope my work will be useful!\n> \n> regards\n> \n> Stefan\n> \n> ------------------------------cut here-----------------------------------\n> begin 644 patch_v6.4_to_except.tgz\n> M'XL(`.L/EC8``^S]>W_;QK$X#I]_C5>!Y+0UZ=`2+B1`RHV_E2DJ48\\CNY*<\n> MIC\\GCS\\0\"9*(*9(A2,NJZO?^S&TO`$&)DNTT/2=*3(+`8G=V=G9N.SL[3Y;]\n> M\\>Y_?=8_M^G%K9;[7Z[KQK%/WWZS2=_RY[EN%#5;4\"Z*0GCJARWOO]S6YP6+\n> M_U;Y,EFX[G_-9_ERM$CSS>72Q<:'_[E_<QI_U?E?)H_?13O--\\O9F_1]/YTO\n> MWV339;KHYVE_N3/(AL-[M>%[7B3CO3[^0;,9^$T>_S`./3^&>WXK;/Z7ZWWB\n> MOE;^_1\\?_T>/'KD[N_FBOWN>]-^FT\\%N^C[MKY:S!5WL7V0[_0=GJ]0]3><P\n> M;*[7W`O:>W[H^IU.VWG\\^+&[LU,FG]U;*SQ-ENY?DRF,M.M'>WZ\\UVIBA1WG\n> M4?$/?[M-O]&,7/J)[>'/C@M7CQW7_>]LVI^L!JG[I6YE.ANDQVF^G,QF\\YWQ\n> MEYL+?9ODXY]GV?360EC@*W<7`#A]=`2?N_!K0_%O%K/5G,MOKG-_-+JQ35W)\n> MQA(G:;Z:++'(#>.GFJH8OZ!UG_$S%?X]'?#X16X0[(7^7N#?,'Z=5J/E^3R\"\n> MKOO@@?O(W<_=R6PZ<I/<78Y3=Y$N%UGZ#FH=8=_=P2S-W>ELZ5X@>\\(BZL5?\n> M5LDD&V;]9)G-IKF;+=T,/D?3V0)>3J8#JFZ:OE]*3?!TF$(=Z4!JV'6^@&^L\n> MYLV\"L.A^[?:@AW^#.[5A]O[-;)X-\\AKV]?'3^229[F#9>L--^S/@A>^7]2=4\n> M4S9T:[-I>K::3](ZW7DP'R[2U-RD<A^89@D%346UOPD49,-R)]TOOG:/7SU_\n> M7K^F`J[[<5C\".C[0Y]T0MD[2V$R^.ULMAZMI/T>\"'J_<%_TE$)_KAWNM0.AO\n> M2X)>J\\YF1\\V]E@?_;R;GH-5I!%'3,\"2^H5G2@V0^AU9.83BGHZ/I<%;+EXN&\n> M*[@:)_GIZOQY-GV;N__/_7*Y6*5?NGONE\\-DDJ=?UAD'U35`N=44AKP[24`8\n> MN5]RV3?0CV.HFLMP(U8Q*`2LZ\"LH6.1?-S1\"(A\\EOM705QM:*I7EUFZJ?))=\n> M9,L7PV&>+F_J@57L%I10R>YL->7Z-A'/(DT&FZDGN#/UV/6MD8]W@S3SH[#A\n> M1VU#/G0C#C7Y+&?0#$RW23Z'RQI.R(;[ITDZ'2W']2</8!S=_&TV+Q(#C\"F\\\n> M.IGU8<*NT0!4QD(C&=20X`BC,%!K)+%UTV42D??M]LM%UF'XBH\"X4Y=MZEGO\n> MLOVTLLOKQ#&;+^&E?Z:+761B]#%-%R+A2&1V7\"_<:[;WO+MPF!NK-8(S`&%\\\n> M`Z<!H>'[324W:4#?2%6UOZW2Q97[:)X`CI&;7B,RGF?Y$MBV^V@Y@2O``3U^\n> M_!0TW%&ZQ*?$E6%H@/E2F3<31CRAND:W>+[IJA;)%-Y-SB>IJ6]!OZG<2X\"'\n> MRY&8(/B@8`WONX_J)$^>.(_=!T?30?K^P8,%FA1P1>-!_Y-@D(IMQH6/6!!M\n> MJ!I_OV&L_`+8R-*<JQ'1`T3SPP\\_@!QU+U$<`N*7,_RU'&?Y_W//?[YP_6#7\n> M[^QV8B41-=86Z7PQZZ=Y_H8QA_<9.0TJ*'^NPD=_=G$!@O?L:IY6%^`>G*03\n> MDMD;RA!2E1\"$CQ0D@HO(J2FDN5^SW'3=8;;(\\>88^KU\\`\\_3Z7)Q53.#5:^C\n> M-'_LUXE@D9*`TH3%?#)*PGXP#_F6>`C]=/NS^17I(+-%-LJFH%<PZG`<+K/)\n> M1`;#+@`J`%4&VLUPMJ!G^]]\\0U/8!9RE\"R@AHX1MH]WX:)I>OE'0X?5JFL\\6\n> MRW3PQAHN(86M\"9K9T78TS2RTFJQOI6G7O1=1N^Z=J!K*P__\\XEV(^P:ROIF@\n> M/S,I5\\I3/P)Y&ACKXB;$JE8*J&VX\"BA1[>_&.=S?)N<P[%,H_]&C=_,+]VNA\n> M7IZ\\T%56]6'6CM/^VS?CY!TH6&](\\8>9^.8=M.`ND[<I&RG\\&(T`;6A0YU0E\n> M8,M`MU<7,(0YE5B`Z;C(LW?IY,K-^\\ETK1J<[M_OG]!4SUU5SW(,*E6R2,D`\n> M0HLBR:8P#-E4R/@J7>ZX1T,WGS70_NF#-K:$UQ-D!\"[1CZX*H4B6RP3Z1[;2\n> M<F;`WG&?PZ-T0:P)E10C$Z&8!F8!>A[4+VT_=C.TUG*PB%,WZ2]7BL'M[.S(\n> M*\\K`TCP`M'XPG/,ZW>=1*3[6^%`FF`RT(JS-H[->1\\.U>=\\'-=2_$1H5:*S>\n> M\\R/3<]?@B\"FV1CS_$4[B9`(*7TW4E6+5,%(YJ%RSH12O\"P(>7*07H!'6H*Z&\n> MZS7<K5]VM?WZDD7E]\\GB9;)($*0)$&7.598>2JM%+H3\\YDI+7.8Z-IINP?3Z\n> MX\"@P%7K)^GZR\"5AT!M2JX<1Q0\"P7Z8X-<[BOF,H7)-5VB<,\"KF#J7=(TAAG@\n> M?G/RXM5+]]D_W#X).9B1TSQ=P/P5_P-)\\=IEMM1>#)R6<R#\"^2+#>85RFTKM\n> ML+;B!T'#;T?&6_);9^F\\B''#(!8*;)@NQ3(WLG3792VHS-*UL\"\\I9:0VG(W1\n> M392#)@6\\>)IBCQ+@DT``R0@LF1%S4.!IP&61%\\*7X-2=9&_3/:X(ANZT][S7\n> M/8/Y,G#UO<.3%]\\A[$M]1U,%E-,WO]W__NCX&_>[H^/:/!O4W:>N_\\1U\\1HA\n> M@U8'[OD*:\"%]!RI?GD[`?(1;P.J^^.*+NJ[E/.V328E\\'&UP1*/[L##''FJ9\n> M@(6(<)1B.4F'RR60MZYN-J3[#PGBATRN++N6J\\54$+8<SZ!%D!:+#\"\"$>@M2\n> MZ2'1P\\,=7>>9K=W*4_6*U/@0$//0!>)-198FRX>Y>SF^TE-K.=/U`;7\"'`;3\n> M4D28!<CE..N/M;#42$,\\XDU\"*@K.<6J&RPB+'5>1B$4\\[L/-\\N;AS>J`\"$3I\n> MLF[P?HJ!U4M=4TUK\"_4;U`2%\\FI%P8!E*0P;%`5B`P^UL?'0K2'KARL`5)&.\n> MKD_!IM%QTCL]@)G!]Y#\"L6O,\"K&VW-U[7'?=NAZ$#3H#<H9K=5$JM:XZ&`'J\n> MNK:5=\"<E0K]H2T)L_(/-:.SJJSFE+E%@=+?PRMLYY;I:8=/OMSBK<&%OF1<-\n> MS1H`*516!XH9I!KS_Q;[R(+Z!H5HG0`^@5[$%=U7-RK2PMWUH[+]>X.\"=)MP\n> MY>=K9,:W-^M)E5!O5I1X@,J:$D%F=\"5H5:E+#WXE?:G2)(Y#T)\\\\>\\%-=!I<\n> M9GF9+FB!T7@_A-,HWT51W1(=&GC^&P(/L5/[DZ#:=96:[Q;J+CR15NAM]H,4\n> M'ENM`:I)_^O$C<\"+[-6RK>'_BI0]H_WP;_<5\"&^;B0/*EVDR0`:N1(6TL47G\n> M+5K[]`BH'-!VJ^%W/&O)R?<;0>!K#`GP5=*#=,:OU'P',IXNW=ED\\(:G^M<P\n> MZ;$[ZM<30IC\"6I?$9=$=AWH%*E\\*;2AZ;<SB<B;Q39:<2E>[3*E2H.3!;/IP\n> MZ5XF``<I#(-!ACP2=0@C[FND(I'NHO0VV\\#-4WCQ?/8NK:M:05KGX]FEB^N=\n> M4UE0I77+JF$U)@0-+>B_PN!+8R'S8?=1]:AT`EKWLZ:9W3#1@](0\\M4YUIGO\n> M%.BY5H,Q0C#L9A\\_!7AR17O`G1$^QNZ;13H$R38%Z53YJG*JKWD?JMT+8F1K\n> MSRCSHR>%NT430^I&VN`+Z/*)N$*L4=]S\\6J1T?SZ'EU(,,HRE);JJ:O8S_/5\n> MQ1RI8,]RP3QU/>52T3X5!;-R`S^:\"+B\\J+'DY0Q/;LHC:#8%M<ZM`>5L0#DM\n> M81MYK]POJC7\"YE=?<?5?&QA5N0</SJ&)MPJ6#_+-\\F121_P=/9>G'WCN*T1J\n> M).3+U7\"(SO,,,;6873\"%(V6/0>1->![B7`*DFJFBWZ=79JLEF!5DO`')L3&0\n> MVRA<\\SU5*!NFWY_$_W)O!XR@B[[NK&*LX?>,3*T,O80:-8I7%'C+K&2`T=NB\n> M$Z)AL,I1]:=9G5Q8%BAP/C):\"M7K\"AY>8OL/2^1<Y-@J,D\"1U?K\\_=H]/7VC\n> M]%-H9X+%:[5CTF3K[MH+!52@B?0.-8V2J044!D]^IGZ\\0V-J<<$&V`48.]-A\n> M74-;!8]N&TIFPZM:K?=^OL#?FX%JN+(D6S%*W6327TV4U)G-A]D@9]+FW],9\n> M\"(>O!;UXJS];@\"VY---`AB/[email protected]\"%IF@V29+$%S5/C=KE\\FVJ6L3%:B^28&\n> M\\XM4O;&>:I2@V>:B2^1*5$8E5<KFLO'C8)]U!6394OP$OG6!XVT!Q_*FA!%A\n> MF36EHM[6JQ+?N%&PN5.@MG[MAC*6I5=MLV(A0_436D(26.O&3WH3T,H+I)8T\n> MUSRL#]RU\\;84N/H6_.D&K^L&MVO![_J%,KK7O*^',-;3V64#E\"L<)V97,+@8\n> M?:#]70U&'2E&*2V^+34#E'HPI/5Q/D_[V1!E#ID5Y%)!!VF.;YZGJ*ZE.Z[;\n> M>Y=.T;,SN6J0TY16<ED7#<)&T\"EHZY]<#7+=N^E!KJ4(E7RA&/>V-N^^^)I&\n> M099\\7.4\\83_KFFYD/5CSP-+#]8O=>\\OW@ES7]0&.\"R*>4:D?WR+?V1+X+:RQ\n> M&&@^Z.M[K+8(KFUT?Z2LJW(>F&J^+@F[&V1=4=0)</?212P=!.JJ_'MT-^5$\n> M*25VIV_22`ITLQD[,&(B3@P*[Z&VK%/&#;/KOGI#45^PY]<&U6%+`ME::RAU\n> MY39!^_46RH-\"R3WU!DM?*.&C6G6P,7(_O:%`5P#Y=V\"5`Z$N4I%9Z.*W@%R;\n> M*,88S=C5KQWW%ROR05\"D=0H]'6(LT6:=0NL_EH=$>-D-0@H=?P]LJ^[3J#^(\n> MD_NK/\\(1[JL#Z==OU8'TJ'W+[LW);/86Y)>LM2'N%-GQ@AH0&,=>'%DJR=18\n> M6KC-B(4[!@/;XTY\\\\#Q-<2UO-1WH93/W(@6=HD'7NAILD&CC/,7!S]5Z#G$\\\n> M-1]L,ZQ6LQQA6PQZ'3T!AD[J[K_^9<;-J@N+>75E<B@JQT(`V*C6.SEY<=+X\n> MDE9G^[/59&!U$W4OI/:D`/477WRIQ*Y+[D/$JG9I&4>#S3EOD:\\WJ*I*)JQI\n> MJUQMP?LM]UEK?2!+&P]^H[IKE4LOB&/098WGG&(A33P@MB(>1'(<JM_*>RAS\n> M!&^+GW?=/8<8*0T95[::9K^L4E65JEI7RH\\/)\\FH,BI`[;!8`T$(3YJ\\+]@/\n> MN\"F[*@V]559#YB`5H'40>LU&Z,?:.OC,&+TO2C\\&I[7[0+X1JU5HK1?Q>FL8\n> M.ZCB8\"'96QW<MNNW]X(._'__,'93K17&[K?VFM%>>$,8>[L3-7Q/;7F`CDA7\n> ME4OY`P^BPVK[-F&(-\\<>&&%6\"#MP2(NY3^Q!(<Q`-N3IF`K3&C=`$0?[&^)$\n> M2NU0K,MLN5Z57LN@54`)PT&E_Y&.Q2D$XG#3)@C'K8C`^=CPFQVN]8PX*]F4\n> MQG*U`X487:)>3^QH%A7\"4CE6!.&#&D>S4)CXN>CRT[2X$,3]*L2&U]US\"CDO\n> MA.CLL%',\"HRS29>B\\`=6I932RF\\8T&!NOQDF\"PZ._\\(.CA?]!V\\BHSC*][7F\n> M\"V*4A3X+_)-T\\F*^Q)U+#Y87\\S>+=&*]JVY9*Q#`$-<`J(C:_3;YPMW'M@@?\n> M7\\CM75-0K``6H63*H!\"=X/:EJW7:M-JW@E@9!C#QS]-%K8:-H;M7:?A*05S'\n> M%VH^EG<K&6!L_AL@*YR/M3])IZ&*0I5&J10^L8Z<)]HM]H456Y[E;]\"8XUJ4\n> M.@Q:&4ZQ-VMW14<\"\"EA_`8BZ=%>/9`G_*J@NT7;DCGMBL9<^:!VL<X#>O`14\n> M$Q5GLB:#_$BQIW+TL+*DEFA$*9^Z0`,:Z$)%$S]8P_O-D3YB+2SKC8H1>U+`\n> M>Q7]L4M2X]RF^'W2U&VBKP\"-8;X!/E*YH9Y\"=[FB3P_Q8I%<G:1#&V9U#^<W\n> M\\/PA^L;T/0U3Q534I`#Z.I:'=^](\"RF,<IKGY*ZA2#\\T=7&OA'EP`Y4@L!AX\n> M-%R!Z;Z@%^>+ST8C50U/9I>?O^$[UJB!0Z`^?:U)GH^F-U5]5_($)C:;WY'W\n> M3-T9C'BRG\"VV(;CS&:@4%M6A%X#5HFPT7G)5TT%>(#.4CR3QJ+\"U.`;]>8/W\n> M9O,U)O6D^\";77GJ5;FYX5XL?:O,+ZTU+IMR5OJ\"JS42`C3&8GZ8UJNM3$8;-\n> MM]@^!TEEW^S.IOG2YF2[&&N#PF\\,H'VAAO*6)D$5?TSF^F,K<`VC.3#HV!77\n> M+3G8IOB$PG24>B=&N\\18HVHHM5!(@CB!+\"97-N_)S3>]LD.\"0&')N9)=_*K$\n> MAOBHH>N/L>N/R\\)3*2BUFI0LB!8%*K_^J3G6XQLQ_ICR0XC)^9A-SM(D?S&;\n> MS?,O$-5@J)\"QH!>+DN47A4EJN9/<+S>#O.=VRS5EN6N5_,+]X^#'Z9?*\"XF*\n> MY5DRTMRH('&5M^F#L>,V&G'D_JP*'F^4=4]ET%G.1K'>RF'DE>:;;=\\5_=OD\n> M=$0!K>X`_V33B#V*G$M#HL`'3/,:(NV^9%O#TOAJ^\\1H2QY.97&BXVXV[:?3\n> M9,EUKGO00?W6>\\3WBMMSZUM9,>01KK)B9,7U3._;50]L1^ZZ86-<T(\\F?G%#\n> M.*[\"BH5\"H\\\\DRWO*M4'Q)IMRQ*3HT2HHR]H>NVXLZ1V+#&V/PO)5#[Z_H_WS\n> MX(%YD;Q+R_X8Z3\\;K)DP:SBJ5[34'>.&5!IEJ&4Z<_^$WT`J<#G,TLD@5QSN\n> MG4!#2W8SC+>4.\"*S,#Z875\",^-J[`@H+/?3-UPM=*H$.[$O#\\+5K^DMUFS;*\n> MXD8F;:7R\\1DM*.U3V]7ADO<WG]QU\\PG@WTEWJ,3^\\4'#13Z(MQF`_(MZ(>A@\n> M&_O*VHEH43VOWZ@;>L'DAHEI\"8GU^2CORU]A?47MO'#-V*EF*_965]IB&NGR\n> M/B^E2S'5](W+3#?98[?UIJ([-\\!LK#&\"&EV4[#OLZ-0%&WR'5HS'[W[$._@1\n> MW;(CT?U5/(GNK^!*!&C^`WR)5H31TD1ONVMS0WD3U3HY_QFGHHX`L9R+_,>,\n> MOO!P\"S>C?I?_.-Y;_JJDKGJT:S?WL=Y'_MOD@2P[(&_Q/\\KNJ=N<D,H`9>FH\n> M=O:9/MWHDU3KK%+^GG)5-7H'UV21+*P!L$?M8WR5NI*2O](>=_[;TF\\IP_')\n> M_4,?UD9J$WE_*([1!D^FPNP&6!5]?&*'YL=!;WLUBY2QO7-S.TJZCZM3U[*U\n> MN_-&(MOL]OS\\)+:5$_3S@_$Y7*(?5_.M;M$2>=V1VC<X2>_&!>_M-=55;?:<\n> MKE/L?3VHI;?O[$4MX>)F?^HG]*::YF[PJ'Y*?^K'452E=]6M=J\\6R:S*RWH7\n> M$'8_VLFZP;W*K3S:ULNJ(+_%Q5KNOF&!VM%J^UFKW*R_,>E?,:+V)+_5`;L^\n> MUS^'(Q:JK7+$%CHK?AWNWF?VQ[KN?Z)'5@']'^F2O8,KUKW5$?M@*T^LZ][@\n> MBK42;@B%5OICU=S`33P?YY&M:O#CW++K_EA=\\6U^69WT95>#]7'>V?(T+O*S\n> MM9Q.'V-<WM5I2^D#$&-VJH!&(8W`$R[TG^K?==V*B'^19VJ6V;MX;G<'W\\@6\n> MMG0(\\\\\"7?*@:!MY'9+%?H-47)RZC1V%%>-W%'.99CKM<X/%\"TM*H?(O*GVER\n> M-:@\\8.7-4Z1`J#T5R1)4D`2NT:LFRJ^ID#<B%3B[H0')R*02P@XSYO1*SA2V\n> M#5@P*J0KEUXVQ81.M&6%-G:0-Y`VU3%AS:P.;AX-:R<\".V?+&V)8'E**&FL^\n> M6G^/UB+W<4,9;4<:6#NRU&ZKXN8#U)*K-Q^4Z:]BQ\\&FC09*HU$T]-%;#:H4\n> M#;[[P?TU5BG<FR?4/5<IW*H9QCL6ME^GJ,P-XL4M7+AHF>P@M_LLF!VYOXGE\n> MIMM=&[\\E:&\\#X*;V2^Z06V\"H`.&6QK=IVO*7W-C^NC30*9NJYATE4_(Z<</W\n> MK2T8GY@2/\\G@5LBYST6+GQ_>CYH.=R#'#4!\\FN:W),E*&&X@R4IFZ7MXTD%;\n> MIU+2.V\\VN*JV'>@;1YJ<5[<N8%<A^\":OUFV@N;?3H+BY[KZV?@/:[Q5):`5L\n> ME'U<5DMB>>IX?#'.C/*/\"ONI[/F6;*5J63K)W=X/+R6?YXZKC/89:F-2BWGN\n> MUHR=DS<L:P5^:'LFKX-6B?I+LL(!(FV0:]K=*-#1.M@89?N\"L[]N<+>)KPX4\n> M*;7GD`:65YA5%98ZR65L=\\:ZYZ&6IZEZ=;,+HA@8='O`(_V0O66:D%RWRCA\"\n> M71Z^I&]KSKOUD!BI[U[34#/<`G57^$@JV:W90F-&#*G(&$5BY`L!2B;617IS\n> MAHU''\"6`41-B!0DR=CBUJ!6(H/RQR9+\\M'K,2V$?,`^H`Q)[H#+S?K%Y%&_8\n> M\\R$#<'NT'153M4[\\(A+YJ4K9L[;,;HE\"5L=KM6\\L_Y_9>#[QZP`=)=%M(#>@\n> M\\QK4:-K9;6X$V+?RZ?\"5E0*)#325'(7,/(5?CBA`=6`@#@(Y$&VJC4>LP>ZX\n> M/7S)X&<T9[7_B)V8RF]33HN$2-K8B1)\"UZ;'IOGQF2:(QN8#>YNI@J[@H_['\n> M;,56/2<@%6^J03&'S%BSQ.Q`+P<\"JR;Y<TT8?6%OLOW\"1*IK;OM)PYT1@9\\Q\n> MW)G4:S\\`]3KPVTJ]_E2ZR\\>I+AM4P]^`YG)7A?%6S:5RE>[NJLN#W8_66*PG\n> MMI_HT;T5E^V4%BLS].Y'Z2N6HE+V<SW:6G.Y3661Q3*=]0:_;UE$7%-=!\"@M\n> M([?17'1CRC%FD;I46CD5MS,6;8UJ*];LRC*=%2ZGW/SDU:6JM1J#P[>SL\\,=\n> MI,%+)OFL-()JE04Z*-6;AB0,\\6+^9KY<6`\"8*ST&7*;AWC@\"#U2VHEM'@&ZP\n> MUD()0CB+TSKR-Z!_JQ0X,A`:FA*[48J*=*VN=*GBL&TS<$KI?*`&3Y#'J4]D\n> M^#Y*_[053S-=[J2\"WJ)[6K5J+=2B'7LNE@>7L&9/),[XLIT>:JFA:UJH[:R^\n> M01&MUD,KU%\"EA<H`;\\A\\MQ'NK_VJ_&-REBTK+/=72==U<!B)[352>V/^+:JH\n> MC=6J\";=QRFV[=E6</EMR/HU)K93:<-H+%?=72[&F];6*#V::6K.W2D?EX`D-\n> M84'\"6O/G=FU5I)VZ_\"PZ:[7*6H$\"%1;R56&_!-!R;TK2X_:\\,BIWX/J9XIV]\n> MUEW.%+^A8ON(5'^/ZK[A;'&OU6@&3;/*\\N#&1(/&JB:_L7\"T)VM*%OL]==03\n> MH&B$VRN62H'$7::TMY,WRO).-$Y8=:%SUV$E+/^X'8XZT!*!\\C.J`V-4+2#9\n> M^36MO>$YX#ON<9JJ!.I6TL^U%5I5S25&CM'TF$W==UEZ\"56\\0!7U,LO3BA79\n> M7-9ADTQ7H8_=K($$X2/@:<]N=I%^47=/TB3';.UG%#\\!X@J(X2I?IA?4_UQ5\n> MDN/^CV3\"64]U,(4Y[X_B?J=,WOT$!X\"VK.#2)%\"_J@44>%!V]A[7Y`9M!9:<\n> M;9:'2X948TYG*WM,+A/:W5O6%%Q]8[TBTERTU/N:*WAPPPM:V>1]O[P8@,^`\n> MZ;[A)%',(-=?-@F>Z+AXI&I?-CVMSTI*+;783:;)Y.J?J4P9.M\"Z[?K1GA?M\n> MM9IWF(MKU=GG6;?VX'\\OO.$X]'8C#*V3*>\"GMH7_6S8YN5]*&_1%IS7OC+_<\n> M7$`%*-Q8B%=WL8CB9^H@ZPTO(**YN\"G03Y8)\\.3=^>@-IO<L/Z87^68!$)@+\n> MDWSW?)5-EMDT+X/)3R_Z[Y?\\I!)O8;O5\"#OFW%(Y-.SH^+1W<@9?9R\\HN$6V\n> M9E&<BUFR3]Q7QT<OCG'N],?99(#)S:02^]SOB^0*#00Z6YX?DR3Z97$EQ])B\n> ME,37;KZ\\6)H;3XI%NDJ36RZ2:8YQ#:_,[9KU:E<,4WK5G!.L[7O[,!R3SJ^\\\n> MC6U#'(8$PZKW.+5<QVLTO<`Z!.77PJ![*PX?B&%7.&*=?[M_195FS+$NH-B,\n> MQJQ(6JTR6RR<G[[#U0%X?Y>@&5$-,KWM3^?Z[;WOI_/EFR-5PYN3]'(!++HF\n> MR6LKAW=M))^8DNMGN7/ITOTG#S['4%=-GDZSU>BT?'ORW#9JMX_9?PK==UK0\n> M^2BRZ?Z3=/[_'KW^6N3J>WZS`1^QD9.^%W0:OA^8863(\\V?9%(_E5\"\"?T\\\\G\n> M9'(2^!BO/Y\\M0&Q)SG>S*1IJP#!,5/#79*+R;FJD@[(VSPLC2H=B?D4';*)@\n> M<XW)Q?KFF\\LW,TISZ2Y6$RE)88-0>I\"KN&Q,7@^C?4JOG$(?:/-*SN'<_!)R\n> M:=T$O436S10U7,H#\"P3RZ)$TBL]W^#T8'@HU9\"URK07R:6*8G2+JA^YPDHSX\n> MW9QW5V`V=O>1*O`&]TJ@>PT;S[GMAUB$,J)_Y;Z;X8[HK^38R#<61&)1D*.,\n> M?6LVO`WW?#:;K+>#IU5A-\"V%_*&=0>^;CM3K\\A@SHYKJ,$,OY^.N%5M!9]83\n> M<N/Q2Q1(:.'E41U+V-/^:_?LY%6O7H$\"?F+7QE1E[GR@3_K8?;0>QDL.9-OU\n> MM>/NN\\<OSDC!I<&!>3/!'U\\P=MG54ZOMOQ%EG(&E(EIY5\\!4C4'UJPVW@*.U\n> M,7A\"-6Y3W6+;ZO2,DW!TID.IC8D3:,_ZA=,F?=^7POO'!WSKQ4F^8Z;(@FPL\n> MO55=E<?)`D^N]I\"S9A=SWLF,^D[OY!0T':ZJ]T.W]Q*N94)?H94&/%5\"47.S\n> MSYZV!UTDTV6&HWCEIK^LLG?)!*O41=\")\"/Q29N$Q-8`C\"Y?&&MTI[><GP4#+\n> M(\\!79*\\=!QU3-372O_(-<->1!^!DQ]!E&%:.OK7=9%P+\"6>=]E\\!#(QRYAYD\n> MN3[FX9B/>3BD8QX.C@_K.^ZK*2!VN<(=(),KJ<QD([?.%1[040\\/+1E0\"&AW\n> M'_)9$`^Y#EXMFB]F@U4?((&VV!B_M`;0'F^5!QN1#X/#E:@J\"W2\"C5]H_KC4\n> M)ZPPQ6@.\"78;XQ`CO35R$68UTQ]I,:B(F%^N8H*(9F9U3,UOEC/ZMOD?\\[JU\n> M2@VSX[(,[1/#1/++;-D?UY3WBCSS,ML5%R2?P-D;;GM/L0)Z_$!-UH0V,5LS\n> ME[8BX#]NP*TE:A7B*W;ET1=5#2.QIPY'M$Y)Y+K$<D?#'7M`+H/Z$U/*VM+)\n> M)4LHVL2>UC%55>EBRTH7-U5JZET?'\\7Q31ESQ3XHM(A?@`3\\VGWVXL7S%T<'\n> M3\\HE9G,\\B14*O#AY@XN?:P7([\\4X1#%9$Q0(T(]]N^MR8)\"U,(VEK`)\\QEX)\n> MV`\\/7-IX8@;UQ<G_D3&]\\R@!N?_;ALD:(!`=GV\"$_MW(A%YLA\\Q/C\\4/Q-ZD\n> MA.:*'^1JD`X3J+O$+G6#PB'E'5V_J4&L\">'7&W/.BV]MM$@N=JZT\"])OHL^P\n> M%>TUP[N[(*4NV_\\8[X7^GM^Z806@U6A9)^/BSW#=_ZC\\?/`]32[6O(_LL</5\n> MF?*3I(]+\";OOD_YVKD;E154^0ZIL\"$/B?O?J^=G1LW^<]0KU7YRC[_$21#SY\n> M)BNMQR!J^*&UQO$2US<&JT6*2G[#?0\"J>#:7'R?IQ>Q=BIMA[-\\O0/S1;WJ?\n> M[QV\"`F.7L9\\CCM2S=[.WZAK,/KXZ0[4K(?V';WR?I9=\\]7R6#+@J7)OB$X0'\n> MY_SL(,V7B]F5^OE]TE^M+OBZ\"\\;$3&`^79VS$4.7Y';AVE[-!U`;%SJB0[/E\n> M!6WR-(`_+$%U4NU-4E6^.UF!R;LPH`&3F\"29@C]99*BXGJ;+THWQ[+)XYP2X\n> MQ](@BWOX*I>JW?T)M&)^'BQF<_6+R>&/R&/</P,FGCZ8S9=O\\&BI<V\"+?L.U\n> M?P8-=S+K\\_'Q;CKMSP:H'I+K(&@#/5@'[OW?H8<;J:#DI=B.+!@JBS;<_TBZ\n> MJ.0;L=?PX\\C:5\"%-H,*>)M.G#\\X6V6B4+L`L>C&G450_4<318I1Z!8W?IP\\X\n> MH$MOTC9VL:F;\"_X\\RZ:E!WB+MD(^E4A\"*D-WF*KC`*\"-]9K76G>\"*&P$L5DP\n> M>=!]<7QZ=K(/-@Z0RTEO_ZR'WR].3^'KU<E)C^[SQ9L#?BJ_SHZ^@U]<B77K\n> M]&S_NY>FU*O3W@G].GUQPF-ZL/^/-U#@H-<]^F[_.5T\\WS_IX<7A/K!WO'C>\n> M.Z,;IUWX/#H].SKNXOT7KYX]Q_LG+UX*^P%=#*`_!FA[/_2ZK_\"MW@_P`MV`\n> M)_`:-7JX__RTA\\T>]LZZW\\+7\\Q?[4.,AQA#`1^_HF^,&)>R#SU?/G_-+WT#-\n> M4(C2VS4D6Q]\\OWAU\\D;:/X*WCHZ/>WB#UDR.3X_.CK[O-60!I<'&X_?83UQ+\n> M@<]3KONO+_#=_^G]`R;W_O$WK_:_@9>>]_8/J(WGO4-X]_G1_^#-%]U]`>B[\n> M?0+^NZ-CZ\"DV^=V+X[-O\\>)X'S#/A8[WSXY>'&.+</7JA\"ZZW^Y#1X\\!)?#Y\n> M`O_A!724WGWU7>_DJ-O@I<>H\"002&K[XOXA`T*#_;1,*>AG^LVBFBLF$8:L1\n> M-O7\"TA]Q2?_!@X<[#ZT?KQ^Z#W^B&]/9-,GS6?_!V3]>]KK[IV>`,\"E%WBTL\n> M\\T=FO>C-/P?&_7:/KR]`M\\YX&3.,H$F]C/F135HN'Z::6V&H1$,38&J%ENC0\n> MY>UWZ<I]^.2AL/1K]P]_L)S5?P#1]8>@_D3BO?Y5>E&BQS:_Y.J7*MJ@4$`H\n> MC*%^JO`31F@S!N`[5D#^1P-?7!7;I46M_NR\"'*_HB</S\\VA51.=0P$\"6-+U0\n> M4=8+T,^F@V0JN4G1KZ=]A'[DYN-LN-S%0GW*3#.<9!C?0UE;N?Y+>[V#UE^H\n> MHJF)\"Q*O'U:M:>!TQSV=N9A8Z<@=9!14@_^R/J6`XBK>93.5#H+<Q*#[J80%\n> MV([$AD*?[SZ\"*MS^CH-8*?V]`+B[%^JS^4!C05R@A0VC\"C/\\6V`E#ZS*,>I-\n> M$]$+CC<CS0W>JH+!M8'XEXMS3MIXAG,&)Z!YZP^!H>J'M7+)^EI)[A@1HUUT\n> M3__\\3O,#[\"3V5!/P;Z.G:@H<;)@$68Y1F#;]:P)/F+[=7;=$X9)^6)-YU3JC\n> M(K\\M,:U(;@MD5Y,9\"(#`:P86Z[-?*M51QG:!Q&]AAC?5HR?6+;SQIN9O&/,G\n> M0FA-G%*MR.*4GZ&O_ZF\\\\U.SSH\\:;\\-)[SWDE>0>MJ-&@$%40NXN?WS*/SF8\n> MWECJ0%=VC)8*]GN#3@8R=?NSR>J\"[4<WH]?>+%*R)1]8.Q@?_*&%SMT)O0:S\n> M/WQB/X`ZT`G[AZ:Y^P?+V_J'UA.IKL\";K=8`2M!C7_5.B>O`'0DX)+`\"Y#P:\n> M'M=J0/NM37_K3TP)=>BH^[5.8:^?F-@AZ8[U#*/<=3`+Q7/3#&YZ(\"J:@58>\n> M/]_PK<UBBB:A:4MQQ@]+`_<09N[R$C,+/2P,,*]I/K00_9#GTFR1C3)<6;_:\n> M<?\\N$37`!H\"8!]S,;%A\\S;U,<I>71@>RWLPUJ4G_T,SZAUQ%;5A8$S7KS75+\n> M7@V4O!*>PI4JQD)<X<BE+1JZ<9GON?O0C'I%1VEW1Q]CG0R@%`6>+Y-ERMN%\n> MD&VA',4,G1LX&XW`EM-IP_P1PFINFD\"%J=+\\5:?*5X;H90ZO391/-85*E(Z_\n> M84KYH``T;7>[O&\\GO]3Y%O53#A8SCRM@MH+7K+>5.B!AJCB-I']E1,KN'.R'\n> MVE]%N;KDASPN[+^2+2.%`K83[P[C\\D4%_D'CLN^O8?^+#=@O<.7'3ZD7YF&K\n> M\\+\"(]C]$A8<EK/\\A+L-IX?P/[0HZ9BZ*.E^SXQO/T><:<B'NXAH2WG'IC?6X\n> MQ35\"*?)A\"@O$`,M$E*GA:H+A@9>X=6.HU1[-77C9\"764*B[H5\"2W>\\09&8D2\n> M,-I%FOV7I5)I(I+>V0%H4[7(BG]XRG<Q/.T/_I.[D5Z!)521Y%1=;Z;,J?U[\n> M,Y%.[=\\WT.NT<.,&VIT6;MQ$Q]/BG1N(>KH6`JL+KA/3M\"(\"]HL\"$T+^S4*<\n> M<`6<^U;._D5)*;O+,);X1XFO;^0LT>9!,VDHJH9JK=Z*\";YY5*J(KC3!]8[(\n> M\\C2]98;?.@K6;\"LAO#3;OM`@5,RVDC+\\L</T^UR[VUQ;DSZ5\"D@K\\!M!*VP;\n> MDPS?XE4&]_1L_ZSW7>_X[/1S:?M?6$NOH-C)\"H9KC#-Z*JL>N*X@ZN2_1X%Q\n> MT4]4?,P`%NYMF\"WV%+\"\"L.U)!.*5-AB_?'%RMO^<L:\"GM,Y0A)L2'_@M-UF-\n> MT\"[P.Z1+N#F(V]0-=SQ71;_0;DY<OWV+RYZN?E<2IURF[B#+^[-W@)0!A]UR\n> MJ'_.N4'$WU!X52I#ZV4Y@PF[XZKM`?@$MU5.T761SY.IBU$L^8[U<L+..2C_\n> M&&CUPAUF[PD*D/=0PT1L&\\XO@\".UPPI3*VX#B78LL_-7)M$B>[TCP=K>H#*Q\n> ME$VD->[*1TPP@UWCK]$37<5FRG$MVG'_CU%/:2CQ-U!3,P1J:D6%[(^(W\\=/\n> M&19EH`9/K$?G:I\\,&AJ/]6UCF\\3V[:(\"T;8?%2V3COVH9)CXGOVP9)CXOOVP\n> M;)CX00E(VS#Q\"SU`CF8]:Q8U9&653\\OBA\"=F)VX$D=H368UM]/A%D5<4+\\*\\\n> M/__<I?EC9@U,V\"WD!@L8B6#<*#[<V\\3'XYO%1\\D6WDY4N!645Z#332X)5A!F\n> M8@V[&\\BQ93\\JD6-D/RM18VP_*Q-C^P9:[-Q`BH;\\RZ3X6$CQ,9,B!@[IH*WM\n> MAOCQS:K!X[N/K0!UK6#>;CRK.$FPF9.$FSE)\\P9.TKJ!D40W\\9'XAJ%KWV5X\n> M;$S!\"''L@`P0[@ZT8KMNP6-1!DIZA#^$95RJ?:<6-GG5+BS\"]R\\0[KWO7I[]\n> M0](\\E%<YK3Y80.+\"OK5AKMP=%49ZM[X4A]#N0UCL0V&]IEDO=ZC<^J9UFN+L\n> MH5?V'OQ:<^>33Q0+7Q29])N:1-N+U\"@,&D$<MXNZ[J\\F+ROW\\6Y:WT@&R7QI\n> MG6Z+67;H@*\\59ZDJQP:1[X\\CE7@W+RU.Y'9B*_1%T$)&_2%O%<0FZ?\"\\Q8\"=\n> MC+BSUNSND_V+UBY&!(RW%*\"Z6-JU*Y0HBSUV1854@US=+5O.82\"/GN.>TNFL\n> M7!?IOVK?AVQKLY20!^Y:Z(&M&)#[4KDW>0`H`1MO!2\\VA7DD>'6(X>>>&3FJ\n> M/;X`)^X`QOEOS2M7->.ZM_`FGS=Q<-DU61T\\T?64:?RKLHMW[>]#95>+_934\n> M/[/94NW'5J<IJGWDU/HCW'@X2\"<9&R7G5Z!&T$Y*SIFE4ESQKD.@*FO/[(ZI\n> MXW2&Y?,T60!\"%7%CT-O%#/UC]CR0K9$*&%,'0H6-RZG0&5,YIUGD\"ESM*%>U\n> M\\*93JXZK.9]6+/OB`=D+W&;)1Z;06=<9V)1HKB`;PX.N];OFZDSV@UH$<CF>\n> M34H81+)_6*+PAPVK&D1;<<]]`1%ZV_%#:]KP7#9U&-QOQBB-A5O[>XIYF#()\n> M];E,KDPM*FZ#$#K#Q'<XI4UNJ5JZ,]IQ.0F/[(]QTV5_IVZJ2-_WY\\18;++G\n> M7(E3CKVPSNLD_'Q!!$B(*PVRS3C6,.A>SJ8/Z?BBY6Q%I<[M$2X`6:L3F,!W\n> MEMFDF++\"O'%#\\HI:-I5MU^FWE/QBL=.OJQW\"Z0)SBN06[,36<;<1VK62GL'L\n> MJ.8-ZX`@3\"#U1=TP$DI<9M@&9R,H9@^PUIZ*?(7(7S;!FQ+,*V9SBV^XS&_X\n> M.6WXK=I/2G)^?8^\\T0#T6A#G.4W>INNL@XX4G)K4IS;%TDJ\\0L3MZ2;TP>)0\n> M[8Y5AUW=O9-*Z!KLU!*%HJK__`IGA<QG4MZN@;-2Z+7`RN0$,A\"S></]4R$C\n> MP9\\J4Q+8>#Y)@6;Z:C=^*3%!B6'R$)Q?V>\"9TC<@$8;,J!>8\"D+O0S=;S_D@\n> M+?O%AVMD9)!52-EAD$-T6=KT25C9O,-3(0+`I'-\\L04^&<UN0R474*FA]I\\_\n> M+PRSE@X7V7L\\UVO.,6:%7DN7[?=XOJL$D>>I),(9X/EIB(Z+&55J]9!.]5N;\n> M7G_ZTWKR\"=..80!N,8FF[@MS44E+\"J/.O2A*=4J@J2OZ4$+?-^FR*\"=PMEBZ\n> MG(C69+6<72223\\+&Q\"(5N&^4O3(32?;:?,YU;6ZUR?\"TIL;:\\.\\;@5OF'?:4\n> MQU@B2]/<`,!:/B+3\\`WM;A#T94UV8YOK\"TWZM(_9O-SND..IRIJH=?Y<=2/;\n> MJ9)5L@,SK-(5JY`\"D&7!N,Y7Q>`Q$KBYHH*\\*E2AO*_!!\"R(L6';*#C/SB4F\n> M#>T5T\"JR(6MKVF(Q_&XM'XF:P+HUT.'^+B%9&ZRB2Q-^P3595;/\\YG/\"<==&\n> MEF`Z$CMC!VDQEK'!Z4>6FDF@9OE%I4TE:9<P)WU)-9(C!VTVOT-G0C)]ZVGH\n> M2#X7G8#EQ4FYB\"M96K\"S5,.+,TG=:8I(RI8=\"\\\"UP!/)5#Z;@H:#%BKR7L5*\n> MK\"/4[;39CLK)4NR(HQ*=K%MY*D=_07M7\\A\\;SS,\\`YAR9,](['-E99N4)D;)\n> M(-RC,2];B1@`4;`.=2\"Z/0W^99RA5:7]<NER*T*3I=L55:FIB1X;1EH-3TK%\n> MP(4&&)HJ#N?!`ZL`[HVB`EPJK\"L)L`F:#:[\"$E@/C)E+F:[^T\"PFNM(\\I2\"X\n> M;O?'6877?7+JX8=;D`)R4>&$W'4W]M<PGH\\>`/**VOS1K0JN-9[S;9Q_GVAM\n> MY%,N?GQEWR[Y`&]8&-D8'&<?S3%-+_<00RH%<KZ:XR(A37.*?[5]-I(+F902\n> M5'SR8DHKSN6W1!FDU`[KO=._/>\\$*)&F@V0QX+/-=6IBRK\"HSWC]PGKM[REM\n> M+$!+>5.T,>\\S4+L*;%`3/I5:FSCHYB-F_U`3A?_0$HU%B(?PUFJ1TGFVT/P%\n> M'^T.8WZ.L<F7.@^CRC.O[2IR)YC&$#R=<RHCY@V\\_IUD&7M!-H.-*#,:R2C!\n> MW'V[;O4JUU?NK[[0M96C]]&FY\"9OTZM+4!MR3(J\\FG*\"D[8;A'NAM^?'=T]P\n> M8M=G)3EI[@7MO?\"&),L=KZ&.DJ4,H/!3.Z0?7'^)YRI\\V7![^]UO/S3DEF0&\n> MP-O'W1>XS]8\\&M!=V5[\\@>1\":=9!*?)K?*DV'U,I>9^2@:=?ZOW(NN+W*%^_\n> M5-N3S6U*J$#W7S[?/SK^4+WWUO?Q)*7`RI$9>'##[BC&H8(41R7JR\\)F9-46\n> M!ZI^J?8F6[>7:3+@^V>]_8,-G=;J]9?6KF:KZ_3\\73+YTNQT-DTL9WS[A;Z%\n> MN#@ZY>Z6\"$S\\0KME_Y#)Y>UC+N_`WPONDLM[8[68KI_(K84Y=0)O+XAOR.G=\n> M;`0=*Z=WLV'2C=PQ93<IBUR&CTSBC-A;INQ6F<(W%K\"3@%>5*J3C,4^MXP=`\n> MCZU(\\%W\"XRE+F7+G-F#[YE+?)=-LOC'73RN.&GZK'12\"820-GB2-WV/NI6+:\n> MY.Z#1_GJG%0FG5H>L:=/UWKP()G/)U=O3HY.WN\")`#7%&A\\(6WQ4K_VI!G78\n> MAR9Q8@4`\"B9B*XX+9YQM`Y2[\"2@N94XE4DTV^\"<J]1\\+>35Z.R'TI-.VF$RK\n> MW6[XD95'G/Z6A4//'EP`,QUFZ<#<P<,E0\"9.\\C=@<7SE^K:W85V%L?N(S@(;\n> MV\"=80,K9IS`A$AI<5*<O9)59J<\"6-KSI8`-5D760@?U:X6PDZ_P\"4Z($^02W\n> M:VXJ_>&!_#&66+Q*GK*-\\:[-R&_`1]MLN&'##:CF;[email protected]=Q:>)<[\n> M#71DI$`GM!NK1O74U=D*C^U4R?(\"/OK`=\":<+HH:0;ME$</G`68#)+=F;5ZR\n> M3G<Y<YGS3<BB18^YMIQ9>TLQ>B]C7T>AZ()SJ@*S/E=)G/F,(O*,BG--K;'1\n> M[K3I\"H]A(F^9.5B*TX&GM\"16KI4=`3J!,I]S8P'Q)D'#1!<7#Q:JF%)*Y53N\n> MKQ8+Z)/<-0E%^2E2HCRRLHKRT8O3T7)<LVKDPZ#4TJ<\\+U6ODH_:WM,>.L[8\n> MXOV7TM[_92GU3!T7G\"']75J%-([DSW>^K#]1<XX:4E/=ZD;#7>^QZYITJ*[[\n> M(AO`C,H0ZT]*-V?%F_J\"[B,!U_A<IAZ>D&5Q\"!M+E\"XQ'\\PNZ+M8879+/58_\n> M-M3SH`094#<MO2R8]BBM+IK*@K+_QYG5![1@UY]-WX%FAZ<X2/)H5P9;^O>U\n> M>S0%S2P;`#:(54IF26ADOIC-DQ%;5&^GL\\LI42T.P\"58=@VL)7F79!-:9V.6\n> M+4YX*O?%6M78]/W0\"?C3@_??T`19Q5PE'3;UU5I6S'NW\\^KX?XY?_/VXF,/R\n> M?H-771NRJ_\\&AI<-\\<X'/2CZ)*7LII'Y8),!#/25#/1%@CE[,QXC(8'JP2^U\n> M`\\-$I&`/ODX6:B77%&QS\\M`M\\6&]+B]V9^FBG\\K+Y$=#'X\\KN481CH:`\\\\30\n> M$[^KCB5:'^O\"\\6\"OIBJ3MCZ!POTC&>-_S+\\DD0MO?/GC],?E;4Q*'#Z&2UFL\n> MFOO(;/S+AJH6?V8#S`)W#,R,$5S?])2[62_0Q8.M24WP:6'XWC2J.>\"##]7<\n> MAM@,RBJA9B&O%*AS1>P!T`N-W$QG7]N3P2:V>\\-]UOOAS$RL>T]XNQHSNZS7\n> MM/IF5Z48/+ZQZ>0(\"6G(;UB*J<@GGY03RK.V4<XJ?T-\">:>T[E%<Y<CR6T(]\n> MZ#A:<U9%.4Z(L\\W6U8K00S(9L>:'[BA[!W1!@0C\\KE*O=MQ#6K6\\1$\\LGT_I\n> MYC.0]SS'<K=6C/=JF&4A<9'2ZC9%O[@Z$JB8Q7YM=?#@^%`4M8&5TW[*.>V1\n> M+]0QB$4EC=]!/]>2SR`TF>15QA2NJ'M\\2%%826XGIA]P9GI:U<;4](`CK7B6\n> M%MSK7$^:+\"89H``J0?K\";/?)$,^-DRII&<K.=L\\.0CF;_2M>%*8D]G7@U9<4\n> M%6;[/EF]@L[)%%6A@OC,\"O%2D3\\ZM`L:0%A`UN]@TG\\*S;JB!2RK!9CX-,5L\n> M:'A)@U:'$>2A?O?%2?%5FU(1XT2\\7(=U0!_IQ9SRE.K\"!3:U@E;?L0[Z94BT\n> M#[H(2>ETE'-*LT2G!M`8\\U$`3(Q01BA1*,9:RP6BJZNC9_6L4=A<SN:RB&^(\n> M1X(U*%6.50*76$'H#V&8:8<3%M-SQYHZ*H!'A8?-AHY>TE,43S8-43AB(UD^\n> MQ/.V\"^<\\Z&E6\"J14`V]-*Q4M()8:7&T*SM+6G(9;&Q?$#K6[0N(EJ?0#5QUB\n> MT+#3Z;-#0YDE6\"CEP!RKC(G#LF_;UVJ/J%T/Q^@7WC4(>\"*@JL)HCK]A]>,1\n> M+:;J'VF.&ZAR$P>&Q[LF\"WP)XV8>,08/`8$,T(S+8)07ZBUR(A%H-?E+VHZE\n> M6NY>D`Y`R9R`/\"4M>O>[@S>\\;%7L#+P)^L8;=M\"J<'Z7!8XI5916`J^CCYRQ\n> M2.%$=4O8-].=X>&Y9?:RB'M$S#E/]>+SHE3%YH`458%95C:+/X5@4&0:8PX(\n> MX\"`7L?Y4!=B>>@.W'X$]/1WES`FTFF=/\"YQMZN6^1#[S8?,6G)Q+BY;8:7+!\n> M3$7#BMBA>IFD$[_)G::\\#@\"ELD\\YUI?8BD@W0+7=<96A1]ZG^%,=67*AF,DZ\n> MU@6[PD,H<(CXB'`8L\"\"XB.WG$/5+V<@RH_3H%@Y:6S.2\"_K3D@,02TJ5Z%2I\n> MCA,J*8O6I+'<.G0+1#I465#`\\'Y=9^(OGI,DYT1;L<4YRX_B*84UGC-P63=1\n> M,&;L=!Q,WK`#-G3$F>60*5\"1'C\\F)EX25!1$HZ\"B&XFTMUJNKLJE`B8*I\n> MO&SZHL-U\"\\1B/;2`2=2>L*D.N&`3R1KB`E]AQL*5Z4.AS)!O>FV-'>'?>@FN\n> M]TE5=<SQ3V1YXX%+1YJOT<NN.K*)EN'58HA;@^*$C:F$>1'S0QY;Y\\A2G+.Z\n> M$M0S2Z*9N:8.5UL;`,M,*3/8Z7*L'6%6=ZA4W7WL^H7Y)/>?/\"B>!6'H^!2)\n> M@E1=D)\\7.`L*ISU5<`)&!JEL`JXB@/+,3]<F?B$$S0+3DG]4SLBN8CESG\\O)\n> M(K15P@@ZZ\\P]^[G<5658]I7+%\"6BPM6ZK6*/:Q&(0C@?1KTN<9'^X2*Y?*C%\n> MEE#1)$V&ZN!DC%VEDJ5X^=R5M4[$+\"LYK&VJRDYG%%ONED_^XYFK/0U$:50?\n> M,Y]28/R.5*=J/2OH@#7+?C)Z((=TN3?&S/.T2\";`]@=XC\"7F@5,PI12$H9F(\n> MNHW094NQ+\"XI0$.V(TCDPXXK&T@0*C;U;.R\">I_37!52[,]F;Z$IHQDJ>C9.\n> M>-T1%1:!W1P,,MEM563!.YBKC4;D,BUNN%\"U/.0F'VI0*$S\\%AP:Y1*!M+8L\n> M*&:Z%BDJ$UH%C&ZH0`5);*;2AK6`I$HK'T5A!JKHV'IC_06#WB>%J5,Q(_#8\n> MP!F8IPN5L5-BZ#/2>7#/EW3YAHEEZ9O4C!V[2U$QYZFJ71]\\+@>^B0VV--N8\n> M]!\"LP];8!KCUO$$/2ORC,`VK_`%N3=M%!?N:1#=-&VW4-XSTMNTK9:^#+LHS\n> MG#T%=:WKJN]],NAI9B23\";DX)UD?^L<-+U).?,JSDP*JT<XFH*<#K?@2P\\+3\n> M\"J]8?J@@S?)`$;(WDC##J!<YUP@3X_B9H!0JC?5OF]C5S1)6IHALEMC:/6#K\n> M/Y915E)2+;NN5+U641^4=(;M_`OKKHCC%V>VUF\"[3<IN-*)(<3*\\!\\[47^I]\n> M(#HD55=D\"0QCVS-+)X]`?S9=`IN5&%H`1/LT\\H(6,[UR+]\"04*8_I2;![6K5\n> M-H[UKET-;SXDM_4YF0L@`%$/FJ:B?2)($B\\]I=-X!_;KO*Q)>S$!1K6/C,&M\n> MBZ)*;P\\,#]!FH%V1D\"J-.:_:`K';JU`5=OX3TRG9S5/8QB/\\4K]9+^Q?V625\n> MJ.5@,NX8XCYJ9&8PK1%LT#B3IF]WQN@/MF`E_<AR7*VF`YKT@I\\OJKM+J%M?\n> M&M_8D35$L3MZJXYW*8U,9F_W<JT`<[7_\\,85Z\\*@<DW]V0+TU/EL2O&4=O&%\n> MWA1&=9*3O;3T:/O14<3=NM)=6.,NXK%X=OB#8O^+;I'-+QKUG7#68P6;,_`@\n> M`RGVUHX+L'I6(#<MB1!9*O+_`G@`J5TEJB$!,!@H*\\!&-WM.DDDVS/H2KE_8\n> M<F<1K@W+Y3@#M)7HY@N:877WFA01_%<,46;G75V?`PK]^):W,;LCWDD*3\"\"W\n> MU]586I?9+B\\%RU'-U;3=8+CK=5E6O\"X=V/?I\"/?!1],MU7`W$C4C>,-?[9;Y\n> M7R^2=@$Y=Z50'?:$W[CK70_6U^YF[D.E:=_@EU]_^>3!6A5(+OJ(S'^\\.7WU\n> M[/G1\\?\\\\*12#<9KA0N40E',5S?!A(W4I+W\\EF2D9O49JM-(HA/2_@X*J\"6B-\n> M8K1&MY&\"*':M1$F?B8Y*,GI+V$IT]N>G7SY9;\\,FM.?/;R8TU&5M.A-:>[E(\n> MY_;N*;(H\\VQ@MC\"#T0O-Y'ONONAYJ,=FVLU-53ZR4*(UK-K1<1TU.OHMFQQD\n> M9:O.Y%\"BUY)[]@:!5B)JB<M0A[VN>6.IT!9NW,*9J18.[8!%<=]:=Y4'-Z6C\n> M975(B(UELS)G`NMPR]@<??QHG2L?Z7EQ@<_@-^=AH#6'*[WD0!-*MI071D_O\n> M=-7K..[7=D^HEV9=AT)7U\"2JV01<G\"1:!I:&2K>R863,`<C58[-V\"J\\4,Q#>\n> M:1SYI&']!DXXJGTV?U,YQ')?I/Z;V9QV@,NIQ?;@5M`&C6&!+CC05NHT[Q2P\n> MS9KJ>O=LHCG:I-\"03VLT6I\"-0J:O;%%\"6VDPT`:9I4J)SXD/9\"/-B>-0.'&`\n> M*%1__[9WTN.'6E<I3T)H>W^$!Q*+Z$*U22-D?S#X&[Q=>JEA=LK433B-R3-!\n> M[WU+>VVV>MNU<\"2K_\"@I988HCEP!N&ARN>*&,E6W-\".T)L9_'TJ_R^UM2,VL\n> MBA<68Q7QF)ME[E>(JREZANT5?[77_(U:Z)^M%I4>?G->=$ETFM>ER=U']H9O\n> M#`^JW-&NJBR[HQB7=JT&>MG(KLB5'3E6Z($8RN?I*)N2@;#!NZ]:+GCYUY>V\n> M+0\"-D]_R[)>*B8_?./;U`^.\\UQ[[4@&]`E!T^^ME'1.J8*WAE\".0\"-^\\;2Z1\n> MX.KU]<^U%:\\;EKGT.E?!;Z2@EK6A<N\"Y>M`HK@C5K9/\"=<'\"VA;6Q.M%Q8HL\n> M@UP]**Z@6;^>:#@_%(F^O-9NCGLI++N;95*4F`J#50$KU6]5KQLKL#_SHG'E\n> MNK`E&P4%17SJ9686,.4RA2/9K3/9K6`]MUO.B,-VNM[CFF:TW&L[]Z`0*G8H\n> M:+$!Y1;B[$T<H9!8+Z@`!1W?`S-]OEP8?V`BP7VVAVZJG'*EBI3/;7WV<!UK\n> M[,,X6QVUI<!RJ;&P`6C4KH%2W$UYVP`,:6&[`%GW^+I8\\FN4H5+N61U:@U]6\n> M-`WH-4KO6+=]&>OQ0)P\"DD`O`VW3\"([]$XL%?*@$'\\>R`GHHH2T8*%9'=4?Q\n> M&B\"!-[T?7IZ4XN7M7K(YR`S&3$!<N:T)VZEKBZJB@\\*92D\\:2#R6YFVZJ+0-\n> M_E,A\\0*:FM4P@J!D%#M%-IC2W*R0:IM29!Y\"T7K#W>0F^'`K5-4#8J;CF5XZ\n> MU7/E(8>5FE5;6H%$YP\"M_3UDC]GFE593J:FSD`UG;=64EQ3)^<]Q]85@CD=F\n> MQ117N82]).Y\\1FA1PXR6?#9;Y<\"D9554K5\":GEE1LU(SKW9*\"#7/S@UKC?P0\n> M;S6DH-TI=6\\]1D\\>&)U(%$ZML-TPT[FU4BJ*XI0A1G`&Y;B:7_\"5:J<T:/RR\n> ME*;\":`HLLC!$C&3]MDHVQZ',9LU>$L-54$+1^TP=(:W>:J1.E*_QHALPT)>V\n> M:5PKO;RX\\L_D.=4I!-4\"M2U\".)G/@U\\8CA*PS-ED9(^>U_GX99YE:\\/V\"R^+\n> MTE?^VOO)TE0^E-'^H6(T-W`^E(RHJBY2S/(/8S*YLE;1])*Z4L]*GN=J9D-D\n> M:'$;X3069RGXYPO47F`_14(OKH87>(W@BD;Q04D$%+80%K0\"V=-XZ[YSWHQ<\n> MM>L\\NO^N<U6IM>?<Z^PUO;UF<_.><[\\5X,9G>V,NW(BL0[DV;#AVM]H%_55I\n> MPS';_(6]QT\\<O:.63]4J.'HHMRY1RKMDD2644H.R3,$\\+VSGE2HD0Z?>OJM.\n> M\\'SPX,5P\"(/]?;*@A$9KNYGKI:W,9LOQC%[<L+,Y0FS%%OJB5L-OA\\5=S<4F\n> MC,]HO8W[;7,&O'V'21WSU<):2RW,-8,0$C:Y4?T$:^0B`QRCFZR`\"3[^@A8T\n> M+S->5,9%4//F8C5ENTSG/&&):KE>:G`]G?'B*7LXK8;3);N-^\\O)E95X[O_*\n> MWNT[;-H.PZ`1AM:6[8^:FU^H??Y=BEBYQ\\Q8+,'*':3O&W(`?+L1MEJ?B'.4\n> M4Q54<@XN].FZ48GT9K,1-JV\\'&'D-\\*X4YSBI@YU!ZSE\\JW*66V1\\N\\S>_#D\n> M_^\"L[L2-IA?J68W>*7(NZRVKG/\"`S!_.7/:@/YM?D0MH?O7B_&=`,+W!:0YF\n> MDT$Q7A8?/5$UTU.[XE)!J973\\#<#`,TDH=D*M-(9?+N/W$W`RI%>\\IBA9*;T\n> MB;I0A>VF#UT*BM@>:Z?ZW7!NWMN`>5-@*^!M..Q1:'F-9JM3&(7M0\"Z-A?3`\n> MAFK3\"-E=XW'Z3!VL'*,@`K+S#<MM@IK5;-LGQ>H,V;02!;./]]_5\\\"<*C5IE\n> MJPUJM.$^]CDER`?6/=&Y!5_[@\\'Q;&FML)1WJRG3V$*.V,8/],H<]XJ5M4T#\n> M]94U4%RRQ/G-D-BI#=\\`@Y5NNHJ]50\\9C]E7>L2+V#+5J%H*A,P0%=>:+!1@\n> ME51*,K@0\\EQ!WHUHHUE?/>!AJ]%L^O=A@9NZ5N8V=9782\"V^K7<)Z8&HK0W0\n> M=,)/P_5N0_TZG$6V>-O[FN_?WK<JS,=1W(CCL'#L-LD[K=!0C,J;Z>JBX>IT\n> M3.YY,@\"CM.%Z9AH1A\\`]5%E?T01J<IR=Z_LLO3Q9393OZ1$^>;E<*-5(^XYD\n> M#86&H>UW&NV@5=2S[@G:?V?#04K.]S>O3GL'GQ#2*J1VO':CXUM682?R&AUS\n> MF,VZ&DR>*16[4-\"&795.#$NK#;$XETKYD=2ZKF5?KIO?E?F^OE)*M#12@0N=\n> M\\<O]$V=48D%;5XM=#2NNQ:B[I<$R0U2A![M%Z_;;1P>6%KP=1!83JX;K*_>N\n> M4-VE>6N5Z],U[^C<G;MJ>KIW<I*8ES1]:\"56'\"5\\%VFAFIB#5J,3-FT.@;'O\n> MJ/#?W3XP%9RG1?M`/U''O,Z'P+]JF]3Y`MPF4DV9A'=[S?U:'2A;+F;I]E]4\n> M*?0/'@!;25:3I5BWN)4&^BO&&IE6%&A!3*`=-7ROF/'N5\\$DYF*Z%S)%\\?O\"\n> MFIA*PFTPY(I88S>-%*ZTZ*QF53G;K-.W[FC76>]M,NRL(MM8=KJX2;8GY\\C>\n> MER0JYUH8-3I-G862ZB8)]KA\"@HGH*G&F->6+H:B4L$\"-323)CCZXM.2V7H+V\n> MH)W0._T'AXN,L_%&KA?M>:T]/[R#O[I<V]EXQ8[JV`W\\/;^S%P2;'=7-)BAE\n> MI=-JB?/*0HB^IH5%O:R1_:036J(:=Y\">KT8O%S#5>-F9'H@LKGCN/G7#.M/_\n> MM9E,]-Q*2L#I!-\"]DE'R/[X[P*I<8NE8C9F.9R]/CH[/#FMG)_O=WIN_O>J=\n> M_*/A?DDO[6&2J097\\(:K5?F^A<QH;58T\"-)5$2U1LY`B='NT?(7%B[8'W5\"I\n> MAUPZ#;'AZBVCG'J(]I'2JE3&&4_X)*692HY#;ZGGPLDVI1MR5:O9L*:!70\\+\n> M4YL/I#6*S5'7KFOW>.,N5UV(5_E5QOI/01LTP>0@=/>C*<1U[THB[D8:,7-:\n> M$N/N<A[IV6(7U9IO,*WWSIC6C$[3N>OZKM?<\"UM[K6T6HFZL<CW3]@W+4$'0\n> M\"\"(K]7&`F9\"%IH$;IPL2CP]ZT%!WMIHN3R>S94XMU>C3?231?[HX!:Q@^=YT\n> ML%W!D_2TGTS7RC8H7*<[FV*.,HZ=[B[?+QONRPGT#AFM.O.H.)$^>?4\\O3GU\n> MWP.DLN,7![UO3EZ\\>OGF6Q$HI:%>SP)-/)=6'&%4PKU6L%V>ZQLJO-,PM]J-\n> M*#3#C#^CDGV'G.39/V393GM5+#\\3\\:?\"GB;T@J-JP0<M\\*MFP;\\T+`\\DY\\UZ\n> M$J`'P@QT`2M\"\\\\D#M7N9(W-0;\\'#J&2_GHG9*':$YWFI+Y/L(EOR:B15^]\\(\n> MOZPX+U?SB61S>YO--ZD)K4ZK$7G&7:(AMK+T4]6,D/,KZ^RA=:Q:Y>58T#[N\n> MW^!=]8^1(-,\\ES#26]%C;58`,DDNTJ6H.\\#9*+71`W4X!0\\DOW0I9_U(5BDZ\n> M`(1:^X\"[^-,%A8R(((>7'E?^,4EU@*8\\0U3_2:AQ7?<.R/FBDL#-75+W^&^-\n> MU`F1=\\)M%1E&H*U&3<M-\"S\\:4<O3R-?8M$Z6>+(V??,R]C[[4&WB#=N@;D,Q\n> M/K?!+O,1!&&_:<5G$TW81XG53.SI*=PFJ?+LZB''/[JX,3;A#1USBK[&$%>N\n> MA#@6!V=C.F:HLVX=@%(I2\\J9_<LJ0[P7M.X@2=:JNY,<\\5L-OV/'K#2\"8.VD\n> MA#]7G&GPE.E<9',IXJP82R7W)B\")\\3X%.RMY;.[<(O@KCI1<#YPMG\"M)IWP^\n> MJCQ<4BKF\"DJ9(:U:N8:J8QG+\"L3^\\?[S?_Q_O4WJPZ9S%LI#W]X#6;3]T-]8\n> MK3HL(]IK1C?:@V&S82(A\"J/*6*6Q5:K_6MI!66BT<''2^_O)T5GOV_WC@^>]\n> M$T8)1S4TFH%E7MV]G2)[MDG/?;316*FLJ8*RMHO%ME[^I)&A:T!]JISS3VX?\n> MG%OI54[\\J`BRZ]R?6E6E-KMJ[37C/2^ZP;II-4+/LF[@IR8IQ-N#TFK;%NN.\n> M3\\SB&[V]U=K;$UGK=*V_^RQZBC7\"C1]F[X_3RYH0[M%T.$/R&LX:;@4-:Z#%\n> M;75R=+*OMD!7NZ^*;BOY9;O5U^E@M<PF^2YF]5[G59T];QN75755!7\\5\\\"?_\n> MIL-\\.HVPI?C3?P_2839-W?UG+T[.?GC@O6]ZGL?.I.0<!?I\\,<,CJ\"6-+S1'\n> M.7Q)BDNN=U[&DGIZSU]\\\\^:[_1^>]X[=IM>)-&LBU0(W!J7B?S`'@V=#]VJV\n> MXL/.KG\"_&_;JH3HC<IE=I/DRN9ASCA4BTTZCJ76Y3]6#(C/$6.XI<K+<G'EF\n> M9;=#'0?'YR+)L=+'`S?D-,G(42;I>]1V'[EX.(U+X&5R\\+F</5=*M@(58@AR\n> M@S[=\"9*02I-H#D#E*M%3!?BB!C-92UGSYG`$,Z43-<>GXJ'+R%P'[/82\"'.Z\n> M.91,R:\",DN<+;&A,!7>17LR@5I32BW?IX`M6[JN&V@^\"=ON3C_5__?[WJ__-\n> M\\;B%7<G)OOMYV@#.$+=:[G\\!DX]CG[[]9I.^Y<]SW2AJMII>%/L1/`V](/PO\n> MM_5YP\"G^K8``%Z[[7XK];BX'>LJO`=\"O^U<<?Q:M.R\"#/F4;ON=%,MX5XQ_X\n> MS3C6XQ^\"\\0:*##S_+]?[E$!L^OL_/OXH!L%FOY#L?*D^GH$72\\2#05)2/`0Y\n> MZ$P.)8>#&Y/<T4X1>.!HC\\=JNJP]JKM/7?^)<Y<W+Y+WM7DVH#=#Z]5T!PHU\n> MW'R']L<VN-Q.?X;[.[G*U7Q.9Q'D#3[0==[@9N#3(;</OIP-ON:J2+(Z3(%P\n> M!YK\\>HZ?%E#2%)=7\\.6K\"]4N@@BZ!]5D@>-^#:9:RP/8#6XI&9[H\"AK#!:3^\n> M-OOGN)O^:@(XP>D+H!M+NZ4N^`9V7W7'O^%UZ9\\O'?3+/=S\\AX6I\\_2UY3L^\n> MH^9K01'E3]JF(8/^NL.GS2.>Z<835^CAR,H?0\"JD(/)RMGA;->O$%<9[_9>@\n> M+O%5#3$`E30;&R%CV-\\E\"_2^U0*O?D/1U<4ME<$X4PDPG>P,\"!J@C81L$=3O\n> M5%W\\^\\^CZB=JF!])>E`U_$^<P6(V+U.IYH':O8PIFJG7XC@F^V:J'>62D)>F\n> M`IU-,YG<S\"9_%>JR,*5P!S\"KORVH25ZZC9(DH>_%K21SP4`SH_F=]'\\ETF=:\n> M'E!8G#L%&D\\'UL*)H6(FW]\\([?Y.'MN3!Z5ZOJUPX>\\.,[SJ;]M97_57X`1,\n> MHDBC3*0<]W2JDOG:2H7$VNJ@)XM\"B6`J2)''2XB..>*?P8(SI,649FOPMH;L\n> M.KS(>(]FGF[7S)_=%BCB]V_FZ[LT<X-1\\N^21E^[#Q=9?YP]_'VV_TK:/9':\n> M;Y$.0\"H-D^GOA/`K:P5DY0]GB_-L,$BGU2:^YD9(0+>X07+B2D'!=[+]VY83\n> MI>E*AM&/J*ZV5M]%-M4W*&5LC5MPT3Z\\M8UU*(-\"N[]YM\\_O\\^7N\\T5-E]\\@\n> MS_Q==OZ:+%.W8K=8VYH>ZC=KZ;4[T]=FP&^GO,WOWDR3-R&KDEIO>&%=]F\\N\n> MO(G$-[]Q1^+?]'?'2;'I[R,GRZ:_^T^B37_WF5P;Z[K/I+L)L++)^!MDR;^K\n> ML;^N/2.WK5<_&4?^M#SYX[CR_?GR/3CS'7GS?;CS[_SYGG__,?SY(V<F9L+Z\n> M?79N>&7-`KBI].^3<\\/?_^')^>0_/V2Q&/\\EL?R\\PO_)PL!NCO]RFU$ST/%?\n> M7M\"$IU$SBGZ/__HU_@KA)9HYW!Y=LGUP23];7MD%T>0HMDI<\"%V!-S2I`*+0\n> M_?)?L3YB9;7Y5@#J1BLY#')::97;*38D*W>W(6L-518&*/D.X_T-5,/)>%R'\n> MMN3)6#B\\?'?.Q[\"JMM[,YOFF>K\"]+6HBL!`F/ZJN#<?ES;P(%8U5L:+YC2!1\n> M)?FME=S8+QS1FP`9)_EX'8Y\"H%*!M!NNJ#A(FT51^BZ9K##'3=1X>)%,5P\\;\n> M#\\\\7:;9,I\\/58OFP%/UTUTJ#AF@:G[365N-A.LB68ZCU8I8.)H\"1CZTRA\"K?\n> M)9\\63+^AC*#&P\\LLG7YL?4VH)IDLT\\4GJB]N/!RO1K,2%HM5:D[5<&VSN=C/\n> M9B,J@[+5>WXCN-=[02.\\UWM1P[]G>^4HP.W>\"QNM>^+E?G\"V\")^%%RW1T'#I\n> M]75E6KW_\\#R;#'*8K8N+APVHK-T,O1(<=ZH.-*W%;)I\"74$C\\#ZJJF&:+^>3\n> M9+G$VJ(&AAM_3'47R2J'BF!\\/@ZL98(;D%8P(6$.Q!\\'T]OD/)U`13ZFA2H/\n> MHR5X96)O8CG:\\6+)%-JF]>Z-Y@^8C;H0U:B>/\"F]01U8*XUWRR690-<KQMOE\n> MLM*9M<)\\_W.;&$7]?P\"$M9A=?>(-(+?H_YA6N*3_MT+/_UW__S7^*&RW4@N5\n> MD-X-FF7Q:5')[email protected]*,<0\\JZP;9E+:-WE\"67?4K*%;UKPMWBE59L_:\n> MXBU5W7^^H;_AKSC_L==O/ODFL%OF?ZL9AF7[__?]7[_2'T4OJ01Y=((QB``K\n> M7!1,+22*[7<KE?W&&\"ED+QN*&)TW;!&<-[30_&QA.^AZQ6185=[46T/S;5A]\n> M&UCZ\\;]C27!SE(:%QC_CEC?OGAL[-#%]LAT>91JR:.NWN\\]#ZWR?=:?')EK_\n> MG=3_W5L[/BG1_DX55>_\\6W=T;#6_J_YNW-/1M:Q&X)P/!=\"=G1U%G7#Y4$@T\n> M&P*SG5R5-W4[2+:4C^0B&:1\\`FB^U&<J4]UFSRG]1'U`VOP]*O=_:?B/\"<NU\n> M]G72H'-?,(J@M%;^Q+7L)%5<;W+^%MYZ\"`)^2EG&%17BL=`%MKEC0L3OJ\"^Z\n> MC?M)^`)%X>]U2GKJAJUFYW^QR5?X*]I_.AO;)_4`W6S_^6$S-.N_043^GR@*\n> M?K?_?HT_,Q5E?80R=]!DJ>#E:LHI-FA//3QWP4P^M;&.9]\\:(W<TI3G_+@@*\n> MJ4JX.4E7\\F\\#B7.W_@8Q\\OL&T;MM$+U[3-WONUS^=ZE4LD/TMT@(O\\?6_[J4\n> M4.+JOR%*^)TE_)J$4*]7:#V_(6KXG2_\\VD;W;U(\\;,<4BDJ<]Y\\O[LH=^D_G\n> MVJ7^_\"]A/J5>?<0<^MPK1[\\)\\?J?Z;7_]^G=OP)1_`:D[/]!JO@H;OX[J_A?\n> M212?0\"3^SB_^=Y(&ZA7_;L_\\K_/'ZS]\\KL6;_`V>(Y'E._-/NM)QT_I/$#:;\n> M7MSD]9\\P#KT@QO6?((Y_7__Y-?[^^,7+T\\?[@]EY^CC8\\9P__I'\"#&:+/7?P\n> M+IOG;]W63JL]=+NS^=4B&XV7KH^GA^-I,TWW)!ED_63B]JY2]W0V7%XFBQ1J\n> M.,N6DQ3>S^:39'&^`_7`S9?)*,WW7-^/Y<<+S(^]Y^[G_13FYQ__^&RVF@Z`\n> MDSV;O=]S/?BOU8G<=C.`1P>S_@K/%CF<39=0QQF>&O+X9':13`&NU0+Y$M][\n> M-IL,Y/)HF4RROO7RRV2>+DZS?R(4^TUXT)L.NGQF2>[\\\\>#[HY>G^!.8Q?-L\n> MFDKWBYV@0B_U`7E09IY]'48>KM!P1>XBO9B]2P=2]A2@ZT-=[EGZ@SM;+>>K\n> M)1WYL^/[.T%GS_<#'WN>CK+IR\\6L?YHN]]QE^GYGOI@YN_#.08:G1;4\\``,N\n> M\\-1M=?,<WW%WCZ_AY@=\\L/OL^CP#]HB_C]W=TVO0>,8?CIW='ZY/W>,/S]S=\n> MLY-K.H1FDBQ3+)+EP,.'R21/7?CU+@?4P/\"X<>!>K\"9X:TRWVCLMN.?(O0G@\n> M9SY9Y1WOFE[EMO\\\"I)%-K['&Z]>>:Q7RW<?^A^O'ONM_<+(AG@=,8^O]U)]-\n> M^\\GR0S;$!D_2?#99T0FJ@^P=WOF^=&N:@H@#6DL=:L1JP2X)+V)A[@M\"[+F$\n> MAFNKS&,NY(Q-H0\\\"V=D)PF.5M>OF2N5MUW>3P0#?=LY.7E\\DRT7VWI63P_C7\n> M]0\"D-?Y;(%UC,)F;G.>NMX/AD[X[65[3`VCP@S.<+9+)1$HBP-;E3WDJ->*8\n> M_04[#GB8I]<\\?LO%*L7A=7;_@ML5D\\DP30?7M`TG)Z+9-;>Y,-`@4L-?^K-Y\n> MEN;7N__-%^X/6,GA=\\GRM4]#])B^?H)&#Y\\]>^W)L,'/Z12NX/NHQ]_]Y8(O\n> M!L/'RR2;7#M8I,TT>^S\"M1`K3M^SJWGJAE@:?WW'B!M.EXOWZMXS8``N-.D>\n> M.W*\\T>XYGLKT@YLL%LD5$'H&[\\'4_`$N5]EDT!TGBVO\\H%_I`O'4F_9G`SKL\n> MNP?UI#@O5O/KW>$,]XXMA]#,AT#J@^[#QSML?S)+!M`5@!@GJNK8:T378`C=\n> MRH<P[E`[PXNX@E_2:RZ47R-Q8,$?I-AKN$;4P1?2,.+0*;S4NY[/YM@F$0P=\n> MM83P:3BA2'_\\^#(;+,?7<#%(E@D6=2;I=+0<NRTBK5&J\"HY3Y-)V25=*-DLE\n> MW\\^&T\"D_:+L5A4-5&+_EA2M\\H:)L@&4<+.N#\\#;E!^^K2OLE,+(+$`:JH(,%\n> MET@CNSSX=#V%YS`6V`(-\"O[C\"7B,\\P?JR09,@HM+_'9V%WW^/9H+C<KW-_Q%\n> M8P+?%M5<Y\\D[I$P?YA[,Q5/\",5,>MAO(CA6\\/C4TR#\\9,.GJ#RZ.IYX5A`?'\n> M<P7AKN#1U4-%Z+\">TDA3YZ0HF\";+?M(?IX/T7=9/31E3!4YL,VT?[_BZ0JF<\n> MZMDAU/]TK9#^@3Y!]7N+1WZ!S$\\1E0?7N_T^]&+32/R$C!SIU2`'RF-W<=+@\n> M_2)R\\,FI2U,'WA:NZ!2(`6\\J!#,VH3A.).15T)B>C?;``ZA'UPXTS3<.\\,;Y\n> M;'Z->AIS/OCU>#R;O77?3F>7TVOU$VO</3UR:<\"/799=#N(.1?=RYNY^[U9Q\n> M=`+3)_#P\"B4'\\6K/ON4@+#N=#C+XW;]]_^%Z]^1[)5Z(O:\"\\1`*!?\\BH4H`8\n> M8!'\\NPIX!Q_8P*<6\\/EX=CG'\\4.!@-KKTNHT_::B#K]I;A`JD39ML09\\JO\"+\n> M3K0A)@8:#-\"'^X.#G#YH1<(OCZ'#/OQN7</M4QQ!YM*(``]/J>.=73!D_7?$\n> M2#^`>'.B5MSV=^*(JV99^H-;O,L2^0?LUOP:.XE7)T6)Y)`(>G8P@RD9>:IM\n> MG/DK/%K'4Y=7?/GN6G[]H$K\\X'Z/E/*]<XU?)]^[EJP4.07:UV\"%)[*@V4K<\n> M6=VQV%@,HW[MN3'R/?(BO`-%&)[7#K(<5,:K>OJ+(H^F:Y6I':<_G.'#V>(#\n> MU<UJE\"A''T2K$HH!P75]/2)\"/3NA>0W_PY6/:$=MR.5.<1_I50<1=HT(LJ;W\n> M2,WO#Y75V95PM=\"`PZKAS=4)G(#)OWTO-4]3($Y`D#[PT*G4:]Q,%U#S3@WA\n> M8@(R$'Y[`A+(34?=XS(L2=6]808,>V2QL.2::R29G$Z`*1,QH,1UB`/)W1^P\n> M$E/T.]\"4403C(YS&2D*?TWWUJW_]N.E^]\\'!NJ\\?AW`)5^GUXX\"OAJCLTM4(\n> MR(,NQM=R)[N60C]?RWMOKYTF7UU>%V\"97,]=:.<2KR_P.N3K*5X'?#W#:Y^O\n> M?[EVD#G1]>(:V2E=YG`I;R[A4BI\\#VV=VJU=`3P!\"[ZYDS`G!=7P]%28)'41\n> M[L`-@V8VV\\\"*$M-%63*HQ!6,-.>/[E_^,L]3T<E@N-*_7,--UTU6R_%LX7[M\n> M?GFZ`];D>/G/AOMRQ_TNZ?]/<M5P]T';=?^:#H>+%'X]VW&_G2VF#?=_=MQG\n> MZ6)Q]66#*GD'IACR+JC%VXGDY@#CU^%.$+A_74U3M+K4(VQ?G&M?MA<[`)8\\\n> M`.+.)GCW[?E?_H+7.\\O5:&>V&,ESH`D\\NAU+^&'+[:8XI=UOD?I.!CON[J[[\n> M<G)U`6;>N.%^M^]Z`9B&\\BKT/.7]J?#RT>F+W?W3[M&1>HAG%>,)3%BQWR%_\n> M5!0%^-5L!DW5HUE?N!T4T_HM[M@[`^E/>C5>X[?OHN:H#MFD,T/II\"DP'G>^\n> MQ-'X`!_P_]$@35P^ZW2,XXQ&\"&X%P)PP29_.@)_\"_(3;5ZZ<0TN;NNQ&H)KD\n> M'>\"*.D?[!^$A:\"RXAW#'/<.C5.%_&,.TO\\S>I5!3J06HC]P/4-'I$JR;9#%P\n> M=?>^<@%9[O,$:H,&Z>#Q16*_3?[)YZM^-DAVN%-=]9\"0[B9YGHVF;)Q?IGKC\n> M0RYGE.9[_%;-KQ-@?P?Q!7?=_6.0RE8[N&4BF5Q@/B#$$H`#I;-%H3Q4,Y_E\n> M?/1K`S#?ISUO.6:&5\\7(B>3VDRE&U*=)GDWPN%6:&7CN+\"*!L-K@_7/HIGV;\n> MXBB`A`04@C+2IUUT$E9_E2_3BQS1G/)V#VD'JBEW`!LT`T4GTHYG=-AMPJU@\n> M(S!'<QI%J,`:1_?Q8V\"\\YT#9F/TC!A;9!V$PJ3?@G?-%FKS-YPG`5?,C#[=_\n> MSH;+\\=5\\G$X)KW%8WW%_6<V6*:;JF?#.P-&\"Z`V'`]TE[L^K?*DQEBT?YJ`7\n> MXL&Y6$6V6&0@HW&*8R?$WPT(N\"!:3'\\P6!<2J`5UL.)A!D\\IG+4TBD00T&:R\n> M.$<9M,`QP)/,QJBA`8G0VC;4PJ?JNB\"E1BF,!YX5C=5Y#=\"1J!,^*CLPF'AB\n> M%6X\\)RWPXIP0@*XP!4U8=__QIW\\(E;K/V(TF\\REG^F\"Z7L*7!>X3(3.H8SR#\n> MP>\"QXJ%_\"91XVE]D\\Z4BF'FZ&*/6S7MW!C07YZMS<H+!5)@\"@YDCDAK`\"XB\"\n> MJ7W\"8P,_KP@[>IJ3VDCGQ2Q=[[T?J,XT;=1JQ(-6-%RZ(-O)?AU050@`#>?4\n> MK2'\\PQ7PD3K4L9H#=@:/S6'L,`7>93E/&]1O`2,#T#3YT,3BZ(W2:8K.DBN;\n> M[RC86G67\"<]-YO,T@5>6EV@Q(22JM6R\"Z<B('9X#*EUBQ#2<,G>H,G2^/8.!\n> M]34G>@VWO?>@&]=XM@[6N!!Q,</#)/.FFOGACH_K++L[<)]\\9H/9,NGW\\<3D\n> MW6$&_R;X=(@U(J7OCE?34;)874P2-'Z>YQ-,<;4[H6\\L.1O-INE;$/;D(9%:\n> M\\0%,2)A:NQ?9=)6;Y@#Z,SHBFF>^L(OEY0P]EC*:(\"?7NH;3&=C#%>[^^B-O\n> M?+0I9P=;_/_ZR0)!_B=_$Z)\\SP%5B>]#3R<@-S,:HQ6`M0)5&>>M/;::-65`\n> MH>G.:*=1G#\"`/%71SX\"G(?Y#M`TGKF-A=>/%EH7^R&=RPWQ9PM!DLQ7.0N)'\n> MF+\\FFP(1VUR(^!K1/E(0_2)HL2*8.)V(&46SQ7/WDOD=#<!R#!58`G9(JO&2\n> MELJ`8J8DE5/D(J!*$KO<M;DH83D`<F0\")M,E1R0Q+]X%'7N27,@[@_,)J(XK\n> MW).(S(]0.9DD\"[!UT@73(![\"O<@1VEV[&[O`#-,I]9`OF7OMTGI1!K;`+KIA\n> M8:C1=0Y4R],/J\\UF4!63*X(:(D'\\,UV`>8W;P':1]':78\\S]MCO$@\\%A%F`?\n> M<S\"T=W,@8822O1N[4SQ^')J8(#7EZ44FET@,4.@7I\"?`$?*]!8*/&0QF4VJV\n> MB<T\"YG?WP6IT=[M@UKJ[/7?W$'U!N]^ZNT?N[E_=W?^!*08&@+L+9L(+>K&%\n> M+[X$NP:,1'?WU-T]<W=?H6M@]^_P`,S)?P#AXVQ+^F_3)2/H'*YEHLI]C:U^\n> MEO6S11^4O%WDDHN\\/\\.U$F@HPH8,'>T\"%L^ALS!$T#7`\"O0,\\`K&`Y@-8#!`\n> MKV%R`TX`C_1^C.^#2?,+\\`+`#B`5F@#+%TP*>`#(O`*\\,T\"I`G,AOVWPEN@&\n> MLV;)'T7D/Q$6\">I^3@VVO<JY9(B3_$+`T&;`FQQ#@7P;.$@VST$EW!TDHQ&.\n> M%W]#`>08A*,AVGA(0Z\"BK)@F3X67C%;9A!K!KL`++WHW36<$MU,-;A%N:%TJ\n> M5#\\%-><@M-\"[![VG<4TO!L2`=PE=\\+4`=?(B6<\"XY&40U12:I57M_F.0P>P&\n> M3!\"4^UX!];9>Y:C)3%QAEZ<K3C],48<X(R=5'X;Y\"J?>^6+V-IW2&&-@`TX$\n> MP+&TA;-(+>3MSA:#(4PFGET(]22]F\"T%#Y/9\"-?V`\")'3^L_DAY32R;`#&%.\n> M-5OU)TJ^/;35/J!$8$<(8CIP<`6BOY#I>(#='*0CFO;(.41,`3?(5\\0T%L(5\n> MU$^<OOW5$B7:BEA0`LQP/E8LID]&˝P?0A83XBZD,.GD!]+V:%'NIQV:*\n> M[&-!C`.NQ\\ED*`#([=PP%,0_]:*[email protected];M+LOY+%O$+_/DGH?\"+8K\n> MH$(M/:FE)[7T[%IZFD2`3TG!(REX9!<\\*M`2(;D'NLWNL8#Q0EY^02_CM+'?\n> M?J%*&6`O5I-E-@<M\"^X*1WLE=;P2`%Z9*J#4*_/N/Z3`&9CV.!W2!2ZJGD\\8\n> MN!XQ9*DKD:*)#4ZB.%%BZDP8>0F*`8.\\5*I)I9K4KB:UD)=)P4P*9G;!K(\"\\\n> M0X0O1>1-!2TS>7FFD#>SWYZI4@;80?8NHP[,!'4KJ6$ES:\\*J%N9-Z^DP))1\n> M=Z4A^\\DE=H;>%G2O5\"T:SW-:-BXM$\"^&UZ\"1#VA%J;`(@PXNX_2\\]L5ONWMX\n> M=.#\"3`D<^?UJF@']\\TT81EIOOE8.;'%:\\OKE:U6'YT;HET)WDI.^3_NR\"NNV\n> MU-W\"NFIIH1?=ZFJA;/<[E*#]W*'7D3-:;EJEF5_S0[U:@8Y6H#FU7\"&@\\DJ#\n> MPT[J\"=W^</U:=>#:6HP4=SZ`(JU;=[#S3J'3NI\"\"5:T9$*0I+\\B_5HL;UC-'\n> MGJGEOE`C!XUQNK4K2X\"T</`3+E'*8K^S^P+L.AB6TTDR79)'$P0.KK7,<K5:\n> MCBX[>HQ=NVX*`(A=&'FNI/<>=%RBIVNK=5G6X(7]DY30G%(ALZJJQH/JV>0'\n> MU,29SV%B)I,J\\CSE\"`?/4ZO%I_;3OYSRFP?I,`&&E%_OCG,W\\@-<X7N7NW$G\n> MH#\"%F?CZY7N<]WF9]IVZ2-!LI_6^YT<OU4)A=IJ\"J-+Q#XLQW3`!$2#V@?U?\n> MXQH$\",&4+I(F?4W243*!*UIV)X?Y:IHM:2<K+LF,Z=:U]426-_JXR`X%WI4+\n> M.+(JH@K02@CU]@<!VF<G[%_>\\9-WZT_ZH%==.WPOD'OCV7`(1'2-.)*FU9UW\n> MZDZ\"YL0UX8AO`&JN'5\\6@>`'K9TPNG2X`18;9]>ZU)A+C8NE`#N3R?OK7?B0\n> MNL'0P9]7\\G.U@*?PH7]>X4]Y>I&,..*#JD-RQ\"FS^Y?O$B2K5_9*W\\5`ELD*\n> MMW#6(AMP=O$6,83T%U.$B0S**9[H<4S'`&,CY=X(,'-!]ZQ%R\\`SW).B!W`F\n> M9`P>E!3BK:(?(1SG&.7%]?+]'#UPN+AUL:3)C52N%AO-\\L@I*%O)%'0?P+&L\n> MKC@XVK3,@KKGM2Y[O;:J\\N%#]5->._G@?+B.<$&!YKYYNN6]:R>SG\\OC;>\\!\n> M-Q0(KZ$_@`KLT(</^\"G,5:TET8[TY8ST;%Z%3'B]='?4=]\\/>%D31B/L.!XM\n> M/<)8#O\\)>!WFT-CA[K_>T)\\*%AM.T34#AOYR.$E&UR_!KE\\^FX#!AM$$P%MX\n> M8*[G[W-W?I7+DMA\\OG`36L)ED;*XGLZ&:,N0B#DEH76*JVDL;1[[#KV'A/$>\n> MR[E7](DRM/2\"W_8`(4M<L<#WK/9\"I&$'KWC=F:*@5N?TC7<#NHM7WMIS#F:R\n> MFG;(!;S>/,%K0-#MNM7M>@[7S-WA^G7-A3>*<-%;'S32SDX<)9J#N*+_57@3\n> M8/%?Q[SBK*-L`^BWH,Q90YGN6&7+&YIS-*;<S9BR*K2Z7P#061]3CT8,\\2@+\n> MK]3O\\ZM.=%TDSZ:6Y\\\"I'`Y9T[$MR+SX5L#\\\"X#9Z42D0S$BF384A=\"LP!@:\n> M1VM3*@3!G5_H%<%C9(`,&S,]=&RQI\"N$Z)&PJXC:TR%^KN+\\UQ*\"H`%#;'H8\n> M&04*]16QT7D^!+W@M!A)<(SMDP/[+'U_F(VNL=@?3D$''YP2GB7T0T=;&(9O\n> MQP$H,%PM@*P^*5UN/@/F`4ARJ`T47@22*S_?VS]1\\!5^%I[*,ZY(/2FT0D_Z\n> M%,Q`5^_554[%^25LU%'54W0-8D3*41O<DH*5H#+EK.KR*T?F%3<FCR@PQZI?\n> M@)+ZU5.L'9\\B9A2UD`Q,%PER>OF%]*=_X%HUHEF+>+PYF+%F8V%,(5+AEQHK\n> M8$K+&\"5!FDSI33!L6,WED`-MJ`\"A2S3!^H6636Z?IK\"(8)&OQ^0<$D)#%:!$\n> M:V9Z[/Z%O;2LU%Z3JNLH19AOXGN*.B6<X@:2*^O'+\"?10*5O((X!7]$GDC9,\n> MBOY;GA]_@6DDH#C7HDF\">J009L5#C?-\"4`;HGNH'/\"F$8CA%5.%T!<T35$W@\n> M,*C^HH;+74%U4YB>Z);7J&KB@,(@.XH>1:.\\1@43!]FBU0]H]Q3X(+Y)\\&`I\n> M%CF@WMQ0A<7QUE\\U7-81G=K\"#A;'HH(A!OM*$0S^I*;X)Q:>XE.5$TVI83\n> M1%$>#1_Z8X62>)!EU*%O9-TNTGF:K`W]P!1\"PTV54M$SKDV'ZAZ6PQ9A:A9H\n> ME_DDL>&_#%,-#1;'\");L6G1,#$F97\"^L7_WKA5+^,*8%[19L\\8<2N\\-[_P`R\n> M/D;K/5UB!?U$3T!F\\M>8L/XM`([O._R&F9G#;')-X4%<??DQ.Z-!QT?W+BVK\n> M_.\"PX)!?[N[5`D0J?+^7;YR4HJ)7!@D>(ZUC88?>9.KR*&;.JEBWERSZKE6E\n> M'>R]9E<7;>>PX[?:`>AMK5;8]CH@ORET+XSX7\\W:N5!W.$S0W3U,,+3+`S$Y\n> M_+/7]7KP3_Y\"K]7NP'\\A7$5>U(3_X-J+_2`^Z/C>P5/?:SD'+BZ=G`,;[5`5\n> ML:?_>H=M=>D?WG;M;%/H]XI^K\\B0YV%LKGOJVJ$?/?4@[H7FNAL>JNO#KF^N\n> MV[YJ`2^D!<?#TH?R=GS(_ZA\\#QMO<_DN?O2X\"-XB8`%,OH,U./PX/*0_KIRO\n> M5=7PIX`TMTW'0Z\\9Q0?QH1/&3SN^>_#G,E:LE^S*K$;LQJ$B!(M+6>#:W;\"[\n> M9W?;1@<\\<Z1RST:@C5@;X?9`V`.$(\\<56<.(U_90;TM+SC:%?J_H]XK@VA95\n> M:H[!%#OZ,]?C=\\WST+HVM+OY6L]'Q_Y1O-X&R%#S.Z<\\.\\QU;/.[PC0SUX>:\n> M]SG,\\/0#<^T7&\"$^6K\\.>>)S15B]`L.^1G`42'@OK+C&2@0DQV,FI`#2USY?\n> MMQ40\"@S[FNM@D!RNG,&PKQF8>,-3ON8JN#IG0YL:EFIX/6:=5@>=#5C0V-F`\n> M0;Q6(T\\H=ZK'91/?W<R#'?4COI$A*S`V\\V-=T9TG6*?I'-Q=3ZR4^TYI3FR2\n> M^[?J`$X9?[9\"L&E.6`.D!\\XICZ@UTE7S(.X624:3DK-&8X;V*N:!;\\T#KD\\1\n> MMV/_4*\"HSTTSQIY6JAEGO4T#UZ8Y;$]TU7%G'0L6IC9Q%8OUJ*%P*L;%C-TF\n> M/F<Q0T4<COU#@62H:1/G#2VRW$H`;9(%A6MJU[%_5%S;$DDDE2W#'+3*<*VS\n> MV2'#+O(.\\3^_!_]UPV[8QO]B+T;#CFR^IN>UP;#K=0+OV=-F6^RZ@:FA[7=]\n> M>2_TU)M=M!8C[R#NM0_L]]!MCWO4AW]FT/PV@QZVN9]QFW%RV#[``OY!NXL%\n> MPDZ[BXPA]ML.M`6LP6]#J]!!^`8PPS9^8^/X'4*E^.U#'?B-1&!_\\WW'%%0O\n> M2D6J8M60M*P@49`I2!T%NNJ*ZII_X'?!(NX%S:>`*T;`T`UBUP\\\\!U#`N(\\\\\n> MP'$S[@+VND];;5#C^3Z`1/\\%\\%\\3AZ'I/9-21W_F8?68'A3+C(6<A*S\\TG<8\n> M%[\\C^>[QM^-U^<)O,R'[;::E4&@J*GTKNT&^_3;?#QTAPE\"F,730Q^\\>?Z,=\n> M@6\\<XLR#;W_?[P!YM?W>TZ@%`H!,%2Q/)@LQ-9GKAU+QH50HW_2OR]^A?,?$\n> M)7M4!F\\2(*'4![_I?2Q\\*'C;^`\\K.O36_H4]`Y@RIY0EY3\\#%/8Z,?2(1@IZ\n> MTR0]#9HG^-M(@0RGU12!VA;0VCQY0X-+&ASX=O0P>]M]$UEVS?<A#Y;GJ(M8\n> MB`DKC@0OU(]#A(,[QI0C['W?WP>ZAEY@_V*G8KP4>O2_V-.XI@[&5D?5MR,7\n> MU.-X\\W>LB,+Z5LA$K!TZI<$]%\"QL&J\\#'J_`I_Z(I0S?#G?,HC>ON_Y-S_@[\n> M+'W3S`1(G(C1IRW^GCQ@R=KVNGZ[^*W&(Y1V(OYVU(5?[H_Z;EOT1^/#LD@L\n> M9::=F$6V^F&(9>V3FJGX%,[`EK[^4:K3^HP5PK&&0,`+GL:A3/BV)Y^?%)!N\n> M^1.P>^C)9TB?,=8/A,Z#CI]`A7XD$#Y[&D<N04B3$#[YBC[IBCX.>O:5_XRO\n> M<,83/^KTA.G25:2OB-_Z*/FZ/-2'_(NN0D]?.?HR$EHX!$)AUAJW]16R4+YZ\n> MIF@2K@@;7501PJ[#EUUFD_1%T^009?<A\\65TZN)5(\"@((IX3;7(&8;_AVQ'M\n> MK<V.GU`ZCM_=0_JF?D0D2]K%;V@P(IF\"R@68B@@JUH6#0-_>(>L.WF&'^5&L\n> MO@_X.SR@X63E431C1TU(7[Z)F\"+S'7C2'T_-<1Y'86V*FSBVC+&^;=E2_J:Q\n> MZVI1P^S,,;+&,W-A\\_=&)B[ZGG+>J6_@P4RBG:=MSQ6I*3--/'A22:%KBCU;\n> MH!>^95Z)=*3B%K=B=N;(@QLXM$+A.J>.]#?--\"!?ZE+8-G@`]M41\\=D.R*L)\n> M&DR3)AUA]Y#:Q_^I0S%:>_3%,SK4?(/9#]L\\AZ(`$\"#,]0\\MXX:9B>?TB(A[\n> M1*I=^8Q5,R&1/4Y<JH$-:.`5+`S;0%B^L&)'70`B#@F3Q&;:PFYH*LBWCRJH\n> MURU]*X@<0T&;OK>C+&<3B84R`]2W)BSL#PT-\\47^]LEHC`6Z:/U;N,W&;^9X\n> M(H%$UA6^F=-5?T>E;R8RAR=J+-2MA*KV#(B;FJCJ@'@Z\"\\4V,C/U3<R1-2TF\n> M`>Q1Q/5X,O/$FCI\\)G:MLETM@YT0Z5@VJEW@0``14R>4`8B$4%E,$9&(QN>H\n> M:41>'U*9Z1^KS3VMM+3;Q`\"`38(<04X+W!5LN!`F6>RA$H(3C2X`-]\"+7DC_\n> M_%X4A;TF>^U`_LM>='3SLQ5H8M]D*/*XO%(*3YP<Q!5![YYZ--0/8BV1H'\n> M4EGPM-,I5`;<'ZN$L@31(<^FTC^N+\"Q7%CX%8PBU!5_8+8YV:`QH^P]'*@\"\"\n> M:,8H6OF?:H&X.4I\"O]LD6XW^`>[\"CI\"Q3`\"OA^N8S']87#+;+#3IF#91(L\"@\n> M12%J;_R/&*;]K]1;5:_?92K&=P_Q'Y3M13*%>_YA?`ATZQ%_:3-6-E;LT`7:\n> M\\C%T$&BV?>@?0,7[+\"_#9VSU0N]\\^!?*OZB'R[AH_>+XX\"QI.=!<2(ZP*$*X\n> MFDA:1&!=Q*=/!@59R(7/6%@F%@,>@ICNX1<2`)2((K$3_1;TL@?CVGGJ^RC`\n> MPF[<C:)N#.T<0OWX+[)QY900Q@,B7(/HMT<DY_M-17.H1+=Y1B#585]\\%#HT\n> M=<-#HK-#^=<3MP7Q=$-S@'+DQU!10),*>H#?T*[3C?E\"?9,7@:5'Y3?B0\"2*\n> M^6ZRZM+S<*0C/R;J\"GM$AWY'.G7P%%,]L/]@Y#;=*\"#'B[!%6DM$(;A&%/?X\n> MAT+`(4G0]6,@A,AGW\\6?D4FA!(A)\"T4W3=A5_V*>CS(><5N1I*/'*/28:Z.!\n> M`U6#,M9%=?]0_B$QPI3#QHZH)9]4@D-\"*8@*8$#R7VA5S[9>6TB$C%AB#0B.\n> M!1XRKS:\"'?+8LU-'-W<H5I/ZWN;/L7^4*X!AVT>WEW_X-/)EU,:N'X:O@]@)\n> M/1?^;X;X&888'AZ$Y!%Z':I[+<_U8[SP7_,W<$LL1]>!W'$[KR.J!_^GU_`\"\n> M'L;X\"=>MR('[P>N`;LL3]8;_&C[:KQF8XO\\(8?#:;R%H^/D:FX8',0(8O+YV\n> MUW:[V_L^W`^M-MBZE+C0\"]Q=*]7F8@AXR`@/8>\"&$?[?H@N$L^D$;1?N`V2>\n> MW(NH80*`[P5-+('7@=SIO(X#QW\\-U33;>`LN6O!N$S_]UU$;;T,%;?S-#^`&\n> M%I*W``616_F_C\\WZ;1P#J.4U-MUY?>W<VOO0C?V=#OQ%';OW#O7^9^A]YW78\n> MQ%YB$ZW8C?$CP)\\A`A1`CUJ`]]<,)':L<SO.@7_Z7KB#,R\"(G#6DOZ4A!T)`\n> MIA';O@(RYEA/;8<T02-2$E\"F.Y%<1!+XH[Y[H@=U^=O?\\!VVY=N1BYC-4_1K\n> MBGVBOTD1$F>=\\E2!'MH\"N=?TNT\\#<@LJ\"\\>7>1K*=RQNRD/6`Y%?B_E2^&:'\n> MD[/N@;(LOO*W-EL\\X8[RW6/%7'0!E\"WT)QJ@WP'(>PPY*E,HT/@_LH&CN-MN\n> M@7(!,T4_ZQWVXGWR]+*OMPGRDRY!AGI2GIV^C\"#/,_K_H8JVT+_O_=/9\\%A_\n> MZ\\6%6VXX]WBGZD;7P86)N-O9#W#]_?XU.9\\\"'/QVK`<?A7'GQL<ZV,[J?Q1(\n> MB(\\5@>!WM3`R\"S_FRJQ(':Y?^=:57ED*]5I2J)>\\(GW57;\\\"Y:A\\Y;![AR'0\n> M5Y&^ZMI7RBC2:_OZ*CQTE.-9^VE(<3I4&\"I?D2K(UJRYPON.FKG6$CU=H6'K\n> MB_CNRE4`O#+N[0=!2Q8DM(-5\"7N-6PF<4@8NK_/WU)5OKGB26FU[\\IBNPHJK\n> MN*M^LN._ISI$P0S<2-@S7@X;K%\"#)2Q78\\BX6'G9UKI9^/\"5Z\\B&M_C!2RI$\n> M.!:LZHH7G?3RK0DS.]2!:$$S\".+#?9]=C%3/H1`^^JR9N:*)2%8>MAFRWZFK\n> MJC=`*H\\7=]W1GG^^R3A2;@.;B\")MIINKKIDWCCV%K,G4ONN5H^:U=KM47$5%\n> M4,4KH#K\"5Q2`I!V+RE/!;QBO)0$K4BDT080J8B\\(0.&*#X#$2>%>H^^VZJY%\n> MWS8M*^@./16D(Z)277$9ZZI;?64B\"ZQP!/NFNI)PG;:Z.MQPQ4OMOAYTZZJG\n> MKLQ4\"KOKH%I==#PU(36I&TKGUXKTW?9L7`:1T'?',5C6\"T\\R7XOXL.9C7''%\n> MRTAT&:G.Z:MHXQ79PDP+<N6PGZNK\"$2[RI0%%'>5;Z7JZK`MB\\\\4$!6*)\\Z^\n> MLGFQQ96]]:N8P7*D)SSD&M_L4\"8,=S7>S%7;XB(M#M`[U`S&7)6(R?!EK^J*\n> MB<E1E[[22DM7W=)5V%.^H<)5Z.A+\\<C&AW:$KZ=P&Y>O#MLRJ^7*J;JIP;K]\n> M2LMNQUQ:5UKP&SM8\"Q4O\"`'+/<`R+9\\@X#W&M9X4UB2L^!1*6/ODYATCSNQQ\n> MJOJ\\N7YG0P.%F@ED@ISA]V.F(#^2]4C=[X)0M;\\58]_PS2M2%O++WPKKF[[M\n> M]ISRC;4&N\\5OI4&5OYU026[%S.1;]`D5E2;KH.QL%C.Q)Y,L>!I'@B)&J6)Z\n> M]B\\52*9HFWZ)]YO72CVM<-)/_UG/>EC\\I=106C%%XZGBE\\,^]:YZ6/@5\\I*O\n> M\"H+A):(-OYQ#XX_GQ=9(#1G]ZMJ__+;YA1Y&]0L=^8[1@/%G\\5=4^*6U8?3\\\n> MM[O:BFR31]\\I_K1^/=/4$,JXJ5Y8>J$*(%%AF8?L?)69<X@!'BHJAGZ9&!F?\n> M)WRH)CP)30[M/^2-#FK7@*<M#K[FA@^$TOBZK:^[ZIH6HVGI1;J$O8DJKK$W\n> MZ]<AF0Y*F`D?Y@?8/8T9?8U+;FV_K:^[YCKF:X=_].0!R0P%=?':/UR_9C''\n> M_(SY_Z$>U^*U<%IZS0QA\\5K9T=8B,L=0^'KHU'6P+Y.S+7JTFGA>V%7KFCC'\n> M8T]5S$.DPF7$WZ(_U.JOYY75:&/8<PN';;5PBG15<24]JKBBGDDXT:%WTY6F\n> M9:)JKW@5=QUS:8%EA8X8^:`-*;DBJ:V55UYMZZDVU6J?0@`ZM(R$Y&^P%)ND\n> M1D>6&NW(^!3L1*U1WV(GVE?.#=:A=;5N>UD=-1:CL02-Q=C3`'JEJVIKQEF[\n> M>6?=0RS&TDU;]^B5KTCW.-2ZA\\$L!6@IU56[$DP7BZ9!N.F*I76L-;RJJ\\,B\n> M1@U\"26@J\\O0EA+F@UX4:0-_@VR+4=O658]WDE2OOUJNB9R\"D928*/.?+F*'N\n> M,F[-E7>H%S7Q/<!W2+I>]+0=D+!'CLJ<ZY\"8($:VD=[>U1/+5URA9_[949^>\n> MHV#KEKXM0E#1UZ&@SQ<:\\)5&JN=77(Z'T3O05!U5JI?`+M].Z497C)$NCZB]\n> MEH5#1ITYE$@&T73:U\"Q.U`/F!/ZAC@(Y/#3?'OTOZ(I#9I&Q2.I8]+#($1]$\n> M%'&+47?#MV=F8.6W4]U]^WL[#=3R?VRI`AO50MM@-/J!Q_H$Q21RK`D^H9G\"\n> MPURRM#FH@R9@VUPY53=5ZZ%WVY6QF)VB\\5R^8C1T;[TB^5,&IO*J`JQ8&?)T\n> MY9A+3WL[>&*&1DOJB8CRO$,3`:0W\"9!+ER:(GL2!,=@(Z^:*](30\\#.JC8($\n> MN%%4AF.I5K5J5\"NC3.&V\"4]=B7$*/5>R)-9[E&0GK&JF?&7+#5-.746':EL'\n> M:XU4KJVWVK0U?6IATJMT,],53WWC`_2ULXVO?.W&8/],5S0N913)5=!VE(M'\n> M@G\"@/\"WM1X=*0XUDS'!=BD0`+TAQF'.L/[email protected][-)GUCA(06RT\">`CD8%\n> M1EBTH]BGT8`&,&Z#EI[V)8Z'YY..7O64\\+LQ6%>T!DN'DD\\5>,M.E\"Z[49F+\n> M<7]*_W./G+C4UXI.2K0BTUDLL8@4/2`Q'I&GM6&@:?+/^^SQX:$EI9W7'#D<\n> MH4<25T6[JJARF6G;A#X3[3?IL\\LAK1(&*\"N<,?GT_4`6^Y036YSF/?U9':Y<\n> M^'1N'[email protected][TI1C$55;T+W/G(0HJDWQZ<2>R-F%8!!^R(@EBY0F\n> MEH3!VFCMW?#9)G5!!5N&K#U0<(W#`M[W`=T'@&[>CAM[1DLZU#7=]$=4[?NX\n> MRM,%X[X+HMS\\\\^RM&/*/=YCX%&\\7=G$O#/Z+,.:T*;N*H.\"!N)!\\'3/&#G'J\n> M\\N$:;5DDX6Q!'Q9)T/^T9ZF-/3CTFC2^;11$A[A6[!T&9*IHU=637?P]#M+A\n> MR&6)7\\81[]+HMRF>D#X=^L*6>N0?B0*)[/0C30GPF'IY2(H7$;K:BJ-C@9RN\n> M4:;L?Q9R#8(KWM>O.^I==I\"@MJK^H0X5A\\*PNPR>=\\AA6\\08PI#W<47A0=O!\n> M4A'KZ;2`R=_=PO<V8<U..:[Y$)?JZ3N274==\"420;Y\\M94`K,Z>`13O9G/XS\n> MB73H22B<X6ELJ[/M'BO>2#@)K4_!$..(/QFWYK,GF%;A>1(^':OZ_2[WD0(\\\n> M_9;P-'(H((&0HA(3M7NX)3CRV+^&U;69*HO?3@E!5=^^6/$5WSIRHRO;.I25\n> M&XH=QA$9SY2B)`Q;=(7*F'JG?$--.?.]%F/?0;L'\\-!A9H.Q:[[3C>([RO9J\n> M-NR4A5\\LG[[F\\QBFAO!T;I0CS@:1;FVK\\9O^H9924#_\"$W:`@44AAC='%''H\n> MX/[/GJ<__>XM&L(F!<'9M$U'XF5\"H??(\"!3>\\.232LDB/V8BBL4&\"SQ/NY59\n> M^K)FP])6K.2V[)/C_P\\YW!ZID<\"+Z'^M4U@\"I:<%2N]&?N0H?H;:'VF`=^-G\n> M.%\\I1!M9++N%T1FLA`A'^F(0*T=!`BET:/\\`;GIH1SYMER#ZDD_9BA%VI=^E\n> M3^ZS_F06Q6T1V!QO[].\\)P3SUE5>-&C)(+6!*;4E`-3'2,U.E^1'2.3N'S)8\n> MZ$[M$:N.^0*W3!-Z8J'\\2`A]_9L#P;K%;T==T&8M^YLLGQ[\"\"U3!+HYV,Y:%\n> M1X0]\\!AT6MRFM2[L=AA2_#E&1J,KMPM2$B..]:<R]S=1NV,1>IOD8C?N(%B'\n> M^A/NAKCB\"F+2(R0A-40Z&(P'%3J&*&0$^KV(T-BU/VWU>?U_K2@[^DK]SPS4\n> M?'+L32CI`RJ5\"FR%C?Q#^L2W2(F.)&ZWYVF;*\";R%*G@T3:)D-1_,.O08^&0\n> M+XX40=:8/>6KMR;_YMM-/4,=J27LRG)$3\"I,1#.6YQPP-#UM*<)VXOA-M].D\n> M&$>B9:C.$P-:11_C/_$\\40\"Y.'G4/PZ!=M3%VL3>^M]:178KM$S0E=[W+,A\"\n> MA8RV0`^]>!8&H-CM-_WP:1,U`^N1*5^N3&WG+;3:)='EF>CY\"DZWU;]R178K\n> MH;1N(=KKFC&PAH6[=L!=(\\-).1/N]>U\\;`7BJ3MTU,6_'2)RF!UTNDX0/VV&\n> M9,M+-`@Y(D63,V8WV6`QJ9GKG\\B[@5\\<DI_V8S]U19M:LUT`#\"/#*[!S\"%XS\n> M\"!S0UD'$/6VRZT0(*I9]3L]*=L`-_YQM\"V[ZIX??![LP[A)4M$N,:8/H7JD7\n> M7;&&V%G+2V`4.VA].HR<0^-E,JL%ZM.$-!4L8Q4]30_9X]4SQ,%6<:089DB3\n> M*289'A&S[,G\"SJ'U\";CV%:Z/%!T=BLUVR&X-14HQ+4?;GV%7`EIDY6;==63@\n> M5NXZUI]E7:1K^BPJ(<M3C1Y&51F'/5E*9&RS4D,6KU+00Z8CAQS+TC>-)^M3\n> M4*Q41)1D9$6P^Y4^FS$Y%V5#-GW2:[S=A%05=`6US39RK6<%],F;M5@Y9![2\n> M5:%*&[6*M4\\=VM0%2]N/#QWI%8YN4PP+GW@3>\\QYBZ-O,:OJ3\\?\\()0]D\\TS\n> M9+U1I5I!+X^2^<3\"&UE(3_(=\\%C%GG`D&C%>&1(7I%><^>2_H]8XI%_9'$J%\n> M\\<F'ZRLZ]+LEZK/2&!Q2WPX[].Z^[-4-*6LC4HTB*/DL=\\$Q/^+29T]R5+!M\n> MQR#P3;&MPT.+J]G46`C\\7_LK;@L`=:;;::IM`8>EO_6WM_FSMBF5*PR0V`\\Z\n> M/H:58^9(IM5-_SF;'VWW'_6A'0(7ZNYC)SNJ3=KI>8__G-N+J#;C0VX29\"@E\n> M/[IP_:B%>Y]\\W/V#.Z-XTX_GXA:DD'8:M>5FLTF[KV@C5J2>T-8CVI\\%W^W7\n> M0<NM^M_Q7X-&ZN.6)OAH4A6XEV?'CV)7G[9SV]:>L.FTRMNIZ*!LWMHS=7VO\n> M11#$KX-`VD8P6[B?\"IKT.R[>;]/_+?QL^B[`%=#6IJ;<AH)4RI&R<,=36Z-\"\n> MA8T.];SC\"M)\"O,:2'0=O=F3WFMH^!1\\1]GD#=@C\"EFMAB)#,_X<1[MG:8K-9\n> MU%K'CMGW-,-.@7V/-H&0?5=]PP0&S;K=!?KW8+Y9(HH6MCGFG!W])0<<KS'X\n> M9\"]Z/?9#B8N#-U!TS2UZS$75TH37L]QXO&_4[W*31BL`ZD9C->YT8@`P4-J1\n> MIP1K5W\\VX1/A0T@/V^22ZW:)S6'PG*/6O;A_[%S#!UR(7^\"7N2*[:KLQAY8Y\n> M<#LH\\+77*&;5SNOR\\ONF;RE?V#EY^Y\\,5T]]FR?;5W0+I$XUJ$]#GR8)$FD+\n> MR+KMDC[(0'65!`[5DBH:]OQUJ-8(M_SDM_B3JW-\\M79N9&;@-P']+=PM+5O2\n> M#[6VP`&:#+L2-#W!D=HII)\"@%FI5RK\"H]*TPW\"U\\.^4;Z]_JQ7*%JB&=K$Q!\n> MXI??5*\"KKNA(>M55^^^.=$3UR]17WWXOP%63SB%,,?9P%7H4RO?AEC^=.Y6V\n> M?^IO<>>NW;GGC;6-86M%2DC=]'<#LM?Y*B*5<!IHQOK1'7+N\\4[A6R'^YDUW\n> M=_CIK(_BEL@L_57@]@:<JCV@D54\\M.L)K0=VIHI\"U@H;TK6:M`?$=HM8C%?/\n> M:*K6KNK0:B/L63VSM%_?JLE.WA%9C76M:^5**??-!MN\"85.7-^*(ZPF>4:+&\n> M3A@\\;?J:O3(4`G+!TG4,^%Q9;%7&GV5!753!U7UCWA=>CBTXN1G?EL@,BZ\\6\n> MJY3,.`2=@?>*^D];GF)HH?5I@VD!Z%N.6&=3MS:4WUC<J2Y_RUL5+SEW++^Q\n> MN%,U*IC;(^[L0]FG+4P<TA5TJ6_!N!Z>6%442B7ZN\\U`$`1^VWQW!:\"N@-03\n> MH\"B_8(\\O#M7-GA3JVM]2B:Y4[<,RC3NAH$/U5M%1K]0=^>Y(YSOMXK<.U2J+\n> MIH)8][M!L^T!F1T\\;45BR%$B#XG2+/D</NZ3*N*]_G&AA7M5=',ANR-^*P@I\n> MHK.KDX2PRN889>3COBNU*G_?.R##W-?[W;O>S=_.;06V_?Z4%2F?8\"3XY=1Q\n> M;8FWE^E*6J[*=W6H]WH3D>_[A_&!TSD`1,C9%A(]J`K*BZHB5;%J2#7,@/PF\n> M<;3M\"XP+0@4I``6J<8Q05=\\ZC#HL?L<B<2,1C3J3!]7D<$8BQEID?X=J^(2G\n> MJE1KD<P<]2W+0YR+T1H:54@-46A7%FX:,G([;41$QS!K\\WP3T3GE%C45*O`5\n> M-0KDJBOE+CJZ[[HR::6,,X5+#9U\"=J26A]1HE$=)![\"71K4X[`H5CB(*L_O#\n> M7E-8C\\38\\.EL7[3BT_+*.F8;M?(>%Q8\"?B6(/-.RH_!\"$:%!Z`-;QY@>WT//\n> MRO:`?AQ`&ROR\"DUK0+<9S,\\$D36:A<$TJ$/1]%D<Q86*/&QPG[RW>`Z9F>_X\n> M=XO6S)]R-($MY.TZ[E\"3LZXNW!T:_'0^!33,0FZHR29SC\"FD&(DY9=_R.^3F\n> M#<BU2SYC],)VR%5L.6C1B]LD#VJ3+I2_5MS,RMG<>0T6$[O$V7,L[G%QDCM-\n> M7]SJOG&B2ZZQ2%*9X?^8%2M$1W;T6EI2WG/,EE]V#K>V<@Y[Z\\[AHV4RR?KL\n> M'OX%L-%\\':'[N=W!=&$=](TWVTZKQ;_@$7SZ'@`)W48(^2:4HB+PP3=\\#VIP\n> M5\"7^ZT[DQIC0JQVZK==P%6`#\\*.-&\"9\\;9-+#%<H@N8.!DQ`M]>2B2U<W^_0\n> M,H97F<BM(J_;7?_WH6+'?UWXO>6[]VGN$[\\5WDXB'=\\F$77D-Z$W)^J@9'C4\n> M<US*X'Q]K[G%)E$C$R:5HQ1]5!KS^5&*/9^3YDEVOM!I2>J]$#\\\"_1$+7EOR\n> M3?GW?/QH-K<A%-]=2[E':T1,*$NW'1.4'5XB*BT460P@-@LS^+3#\\UXS`&NE\n> MJ&4^0\\:`'SN^+!EML:1&W`5;=-:7U;B\\CV#ZK]4ZE0;`AM#^O\\`MU**6+V-$\n> M#`,:\"SJO92GL=MIH!ZZF#6=]Y6U%N?PP+R`,8H>'E?^/8LQ4\"(TV';X!CT/L\n> MAOZ`X<?+B-(6^LR\"6D@+6!*NG!#O1:W;@022;$<[(28<;%<`^8Z!1/H-FJX:\n> M'V@L4JD5'<K&&-`31!PF'HS:6Z0<])VU1(>Z8?LTTLEL!-?/\\!C2TW2YFL./\n> MPS19KA;IGOO(.M@[C+S!/\"N>6PIE7R;S='&:_1-*[S<=KE75\\S(9P6W?]1W?\n> M]=QS.F\\]!B;;<1/W\\%WM()M/`*+%>9HMZT$`Y?R6[UZYAZM:;SI.IOWT(IWB\n> M$^>\\-AO6A[7E.*V/:_O'IT?PX_1OS^NCVM'%?$+%$@02GD+!=AN7A4\"Z7-5>\n> MSO+E:)%BV0B)H-D)L(%E+5GEHW187]9^#%I>_3$T?EY;C1?+NH^M)1=8]31?\n> M9LO5$B[+!1=U'[^/ID,\\T!E;SO.K?`F`U.,(>MN\"0;JJ#=(%`'16OZBE_?$T\n> MR_OC%\"%\\-<WJL]H[^)<N<NAYKJKVL.4E5_WW^K260?%V!+01^5C=:KJD\"K':\n> M46U_.@&TK:8C0`76!9V/@\\CQXYA0F-=>[*B&)K4=&.?ASL$\"0-FIPX0[KWV3\n> MSA;P<%0?U+Z9+9>3V7D=-#@`O1WZC*'5=`\"=`<IO=Z1&J[[]/-^IOZU]EXQV\n> M`);_P5-NI_5^[32%`E>+.FY.`O;1\\@39DUR@3!?)='DY6RQ!R(^A'X/;L)$G\n> M>9X!9J=+:,#W4%E!=0F0NUKTQ_5V\"Y?/@<*AG47M=)D.DVD]Z,\"KI]G%V]F[\n> MK)_7V\\B:0AA]((=D-0$,ODP!D_D(JD[K:2V,ZC&09!\"\"BG!5VZ]GM<?0)_C9\n> M!*\"?+0#+Z72X`M((@#,&<;/#V#B>*;@OSJ$G?A,:#3H->-6'N5Z?NUP\\0G8+\n> MW,U]!VH)+N?[1/W+V@',L8OZW/&;N)`9HW1!Q@3E_%8'\"WJ>FP!Z$-3^>)$-\n> MEU!G.C/S\"KB!`\\-#\\\\J^#]S9`5Y%]_U.VXV``4*+O]3VS_/E(ND#A<'=MIH)\n> MUA2A/F0Y8`\"P.'0/Y[79^<]I?_EX`=283HC0DPD^6=8&R3(Y!_X#Z`2F`@W3\n> M7!TY-9X(@(;E.,&YLUA-<?1I<B:3\"V@-;D+]0\"@$2*?E`]I?'1_]`/>QQ@&T\n> M/P.F`LT!=:>J0G@;*'*$\\(UJ`Z\")178.E:R6\\,*PEN3UL5/K/LY!1>BG4`(Y\n> MX4X]B'\":0H];+O5L7)O\">\"+E4?>&U#$DVLN$+A$BQ_=\\F0/3F12<K\\Z!8G'R\n> M`</*IHR\"?#9<PGO83.AP,TW5#)[TO<A&8X0NK9U?09^0>XW*]'Z%J!D2[*#X\n> M`CN99@G<XIY!\\8OD\"FH[)\\!`D<3)75L!CAI0ZP7P_1]!^%$3B)S,*2,F1<1D\n> MM<F,^$22T[#@F$%OTFF>T6V@L`M\\HGDLU*/AARD*7422Z/?3.=0)O8U!2_*1\n> M3]3V58W#VC2YP/XEDT6:#*ZP1_EJ-$KS9=[`*6>H+,4.Y#2>4!Z1_QYFV(\"&\n> M,5^=Y^F2V;CP>P<Y?2?`A\\MD^I@1X2-/`AI<#)B=42\\9CE]6Z0*Q.DFFHQ70\n> MY4X]1):^#Z/3AI(\\\"DN0AX+YVN4B6PHBQC1V@UE_1=2L)`[,F170/1\"W&K?9\n> ME/%%P`3(YPKS\"#GX)9)D[1T(_YT`>>41\\\"_?.5>-%&IBR#,ERU*D\\WFR6'(C\n> M(Y)U`%>&*)K.EHS<^1SX**)E=B'D$1)GA&IF>#3Y$C`]9$F.HY1-^Y,58GD,\n> ME]+F2\"%W1,A%?\"*H9\\#B+V?`[email protected][J%3NTQ!76#\":/K$]6>U\n> M'\\$JJH-B=XYS!2L!*7:.DVF<O$-\\PQR=@!A.\\340F5[GAM=T=T<V3$PJ<#U;\n> MI$N8G@\"(,`X0R/41MI41_P0>AC^7S$+P/I$=$C3/^(A;7^(D3FGHU$1)5LOQ\n> M;/$0)\"=RL662O\\5',Z3TP8!;QW8-?#\"'-8`@ZN\"EBRRGF999*.,JN`F`Q:F]\n> MAXG+5\"@LC$8T0M&F>!GBX5D*]:?,:&0$LZ(*A*QO-4$*.D]Y7&6\\D*<3/F#>\n> M*Q8$M_/E\"J8KHDXC%A'$+!_1]`Y)*2-ZS!FJF(0IE!](/V6&.HJ*X%UH42-<\n> M2I4`'9H6QXX:$L'\"2%\"%0.1<$W$$:Y(-@8X&/'HQ*'Q7T%U2KZ9I/\\WS!'@`\n> M*3L=0-093CD_A\"L]M4>URVPR$;H&),YA5/C^;#JYTC,?[JXFRYP94BHWU_HA\n> MW-HAY+2;\"`RH/#.ACG'M[11UA<M).ABAPM&?329`D30HE^-LPL,$HT`]=PQ.\n> MX5-Q,1%_0T7YMIBS)P7AH^,Q\",R)AP7,FGDKC+P38>'N.)FC>IG5?.0N@HY<\n> MD8W6=1;O,A*72J+;XV\\S$5!%62*B`!YI>L-YE:#\\3O/^(ILO-2,=:N@##ZSS\n> M*T.\"2NMX2RH'3B?H\\F2=NC0;8$[](XA.N(7:.=SYL8Y2..F_'2UFJ-AFW\";`\n> MO0/7QRR$\"D\"T`0@:`Z(;/0Q`Q<L\\G>#++\"NH(^<H%OVV@Z26\\ANBZ-C,&&O+\n> M$$=+HAY\"[@()0K`MLW10=Q!)2$P85`9@3&8HAFF,\"=($\"5\"]/K0Y.S(?Z&BZ\n> M)%/!024>V!^.\"HS]#JM!\"&(?7M*V\"]1M1FA:_QGE@PA#0&1?*(/@\"3R$9PS\\\n> M\\><:*D(T56CBP0`D;^D;ZU6*1.U\\-GM+$N+U`8SF?OTQTL-9KQ/5?ZG]U&#>\n> M7GK4Q$?\\A,?W]:OGS]OM^B].[2>F6=#X8XMF2:XC^97I;%R@LY0HUJF-:3(J\n> M+KXB%=;PE1T`_3!;\\.`ME;Z9\\TO8/7G=T?,-C8N(*18H#!X\"&D#S0OK\\$434\n> M@N<-4\"#,UF3*VISARSC6-+&EWC'-C_YL\"GQ,V,MP,;N0_O0GF2@FBKL,I0FG\n> M1L9AGY2J<P58&P`#;(\"J/LGR,5+I4%-[GS4'4KI1MW6,3C\\LZ/0D]7)2&M\\2\n> M_P*^(ER:<\"QO-4AO5,^H_683VT^HBO,)UILN^\\*4$R7&[\"F$EMMA!M,=\"(OE\n> M/H*8K3'_QTC(+)2ICGR>]C.>^):,)7A)FL$HL^F,^O(T'3!X+0_!&ZZF?=88\n> MY@Z,6:,^J'P#<'XUAVKG+A42[!.?6C$+%SL%*Z%N_DQ][%O3>Z1LF,EDA;88\n> [email protected]%$AR&,6CB`9M+O,$?%.3RX80YGI..9R3NB^3$FO4;F[:@X\n> M;S69L<BR<,V01`%\"HN0GMTFJ-](X#M'KYR^Z_P.&[ZSVTX]UA9_4*8IL(\"O4\n> MI,>@38S$!D`,L,+%$T69-$.TFL:&(T=M!(#5,::-Y1B8^6@\\6RW%NY`LLMDJ\n> M9XW*XI(@8_MCTA=PFK#>,U(26I\"@>*@PF+C9LAC,N!;RG`2;B48MKSOCVFRJ\n> MU032,E%(@;6Z7*SZ2(\"L,U3@<^@4.041%+2+1@-S'JW=$>;;7A-@@3F,&*,9\n> M(I8UTHNR=L9D;2Q1W9PG>:Z4_-D\"O46B;8)T7)#T9G8ATY=5'+0.%ZF6S.<D\n> M359B(&(Q39@T&.T(8<*1&M%<DZ8-7>N.`QA3=E!-\"/-(_B3(T56@<`5D';:%\n> MK`,/;6B+LM-URAXR\"E!1T582X:H3(%R*ZE,8M[%0O8RST;U&HGN!;\"HIQ9,K\n> MI18KLY#T9%:YAJQRV18JH'\"2*M)\"4<^@M!$4T9I1(R!JL`7-B!73<;%]')FA\n> M\\(E!NDP`QH$1_,`;RTJ3!FW,H+%B#.BE+B,LH==$6(QF+,-%@H@8LV72*'U*\n> M62U`Z8I#)A-T6*!:G=%0I47-=E@;H<JUH,D[6\"U81-!((=&$N%`'&($I/ALP\n> M`L?*$!^3`4\\`HY%Y<0%3^9]4#]8GU.P4R?D\"^!8^0=I-0821]LH*T'PQ&RV2\n> MBPO0?9\"[@&AE`\"([email protected]\\FAHMH#@]N79B^X*-#!GH97)%UM:46408!%C9&8G@\n> M(5`5(4(LN%0$2@8CHC5GI;P[!9L26,\\@I<G+H&\\P,1A9QJAT:'S8&Q(&;02%\n> M34QV%5A\"$&N!.ECZG&MP1!8Y*':&Z_KX!IM-J;^$`R<,F^B8/A5-!:;;3DPR\n> M$QV^P'>SY7A3CU)L1'QBCO$)@,R>C4`9$9Y(^NI^#C+4]`('>S')4F5ET\\@V\n> MO8`U,*F'7N2JR%5'3DUXG3&IY@UBTA\"ZY0RQW\";\":QD;(GUXAF5(**GI&(#K\n> MM2,G9.=_6/+6-MVFTW1#\\LH&,1V>16[@9AU_!6%3B(E8!1(?JZ#,/T8%?YAB\n> M;T/-K)'+B5:+2$A&(_9F3G%&$H%J'6>$ZDFR8+8%$_J?1!C$4ZU)0#!U$\"8D\n> MDK26#)F?$C8R)!EG$\\V`]#6+-Z2'S^<P7(.BSX,YE6/HFMVL8&=-T\\7N#&J]\n> M`-`6!$C8:LHZ`L!\"/FK]`G'%M+]:PJAD--!$6@M>(Q&I1[W4TT<F/K1=9&M#\n> M)A)+UNRP+*)108ELD,C\\631F@K'IAR2DE4!C+Z<RZ(;(6>D7\\=V\"#8\\#RFR;\n> M&9/H)8#\"P41S0)\"/H)$TX]\":;J0*A#MM`)IGW,B><5GEC+/<+<8SIDWI8=$S\n> M1AZS(I*@JTXK1\"A8J1J(FU-9(J`C82_5E!J69Y2>4`Y-H8(<$@R=7VD,R71#\n> M@1,:[8A@Z!`GY]EB6_X;6=E8S!SFMV+7.<7A_IGUZNW\\AC21A,6P+X`0_IC(\n> M(6HA?.++S2Q%*D/4*Z6-)^F,/-`I#\\.2E@JA`#S^L9;MT.(\"6DO`TF\"X^<U1\n> M0:N2JJ=8017@N\\BQG`+L\"&+LRP*0=IEB%W*NQ89LB6XU[\"R-%:(N3T&/)WR3\n> M'9'^LE)83\"9LGI(/0#//P2S%J86N-B<5BQO&)$^Y,73C#@D,J('Q%\\?(0\\61\n> MFQ++XA4CH\"T8ZQ_KHB_O$U^BV2E6EE(Z+,-61MP1*@7V.67CXR(9B.HCO)5X\n> M4+)`TT\\/.S]\"HFN'@>68TN.F+3^2-.R2L;5HRV_G:\"HO\\Q*E%-LL1/EWM/VX\n> M`Q:J+5-:;LL!KK\"^TM>=H48I[MI.,T2PIS6_WD)^1DXMS/QX7NNP\"]#S0B8%\n> M?P?*!*[EQCJQEOUPR1)-@>_(`1?$4,$.L!549'?@-OH6BM=5]]:NG2U>;.(`\n> M^QX8\"3%`2UH<@`K`1KBB<HCS`*R%\"2^\\',],*(\"B!*L;8^[&4+K1Y&[@@L,=\n> M>M)$*00`M3#(R\"=-\\(!6!G,V3=_1\\C@U-*#E%'(=A*U-;6V%*>O:N1E3OJS1\n> MD0GB[P0\\J\"_,V@A%.63*@3NT\\3,J#7-`P+Z]R_#=/I0^#J7C!TT!D(:R2&Z$\n> MQ_W)*#U?)/4P=#>-DO,1:+3?<S9`Z?JAIZ`,*J`$1':327\\%EB_K\"W<FIK5N\n> MW(T*FTR%(9GD9\\\":5B\"&&'V%8=50-D.;##\\&=\\[F$6XR[IKD-D+<A=6X4R,\\\n> MDCEC/>_;(#-B(_JZ$7F;P.'%R3:#$\\*,L+WUO-CR7'GY@[:K>-O6_.MC:`^_\n> MG4K\\M9H\"L)HAIRGZ+Z`W&L)\"1<X=6]WTW-F>!A':%M-@Y+6<*PR](0HTX,(`\n> M!FX%S6ULY#:NLBUT3A$\\\\J;^=89L^N<[S(!M1-1'3I>((8QQ):&V#_8;ZA=D\n> MO-$L8=8]6X`,N161=^K&[4,;(V2XQEJ\"C$P0'&)RTWZSF*WF.8F)#9/T3N1Y\n> M.]>+&6%M6F'X-D']$U6_?A&`&^LN`W/?&>Q4(Z[#$'9HD>%T=8ZZ>X9:0$=X\n> MF5/)RWZ=4>WPJ'9H(>W5%`93%N..E/V`+OP>FPL*X@UZRR<;U(`U/%X!)J8G\n> M`I<T$@+O@-=C,B(_2P7]*'YW4P<JT1?XA#Y9)>[BFI=(7-3UR*/:;&^C\"GQ&\n> M\"<(PJB5D@T&MC<H2`;E[.H3(_GT1>!,MWH!`7$]NVPC$F('[email protected]+K<=`LLM\n> M;:-+W4Z,@8`8M4MC_'W]HB91$\"1%-B@EVR+DXR2'`C,,$,R#Q6Q>)$2<QG07\n> MJ%SA%H-4Z=[([DN%2G$OMLE@A3R50PK-P:D<LK%F\"/&[9)K-5ZSMU</H+IKS\n> M1NWD;B,<,NJ:&-*+X=XIQ7\\S$YSQZN:M*LL:\\[_7$&]B`0J7!\">%R-5>S0=(\n> MB6FM($(^!AGW5AH9BTV!CD+5#D#?PWG2J<+=1_&-CYPFK$\\'$86N(3TV19]F\n> MWXW8<<MD,AN!MNK=:&Y\\3EPRE!U/H&PQE+V+\\Y0\\1T2>Y,%1U.ELI5#?AL!-\n> M+',3,F5EVG/(LQ2P9^DE.F)P#0']J00D0CO$GS,,X&9?PRN89P_!3$4)1&]D\n> MTR6.'+EKOJ=]#Y?U.*:@GHB7XW$7!#FH`N6@VA>[;8Q!EVGMVRP'[9BBCDO+\n> MK+>JI-O2X^U8(6\"=H!/@X`7*Z[(_95!?%./G*!I)Q2P?FO4F!+\\0--'L;$N+\n> M-RIF%HBX-NT3B\"&#^'=:9.68\"EQ[^`8=LD4@[L2>MZ*LVR<#J_@A;OH``:P#\n> MTLGHH'4C&\\!VU72XMT95)@#&G@#D>SYI]!SQ2^!PT*\\%3]CY=5Q]`!.M1OL1\n> MCVB31_1;7@TDKD9#NJ*H+@O`K+:]-5XK><<^3NJVB<>%01`PQ-JS@1%QM'CK\n> MBP<Z(X`ODAR=_/3RKX%1!5];P2=&2%>'MM$6,7CK.)60O0,5./?V;A!6*@^W\n> M,QJ\"4*V^(X3B6^M*U\"#!*#H@#CYZNS6(8?OS>_^\"#N.0EN4)0I&VRGRC478X\n> MP!-'F?5`VNZC%%>0PJVM.-_=K;=.8:6^55JIC]S(B=S6VDI]5,<C-W!?PD6M\n> M^^+XK'=\\=@HP=GCQGJ53DP2V@VY.%6NE*9HTWZQVB&%VJ=\"W'I4[F+*WSK#0\n> MH]5B7K\\/`\")B\":>T321PUP*1ZH^1:4HP)0E5\":8LR\"6R%0\\H'JG>]#]&]V`8\n> M?4(=+>TCC(H)'$U!%F5+C!_0-L$63M_/QPX$4%[?1T\"%&Z`6PYS@0(>(CFJ'\n> M)LYB,^15@&QET51C$B=_1V(#_D;KPS3*STW(^H^H,5(`LH$O8#8PN..<7P/#\n> MT8AB.#@ZX*4$8!'Q>S8T/ULP^.U-+/VN8#AFO%HT7AP@@.,5%L;++8Z7=H9L\n> M!\\J=R.E6/A1&!\"K'\"B\"H3>8=/=H/I[FXY?LMN$1N;.>3N$#\"#@%(D0($H+)&\n> MB@#:[FKT]#LWS-J[P.3<BL&FQP#&\"L\"(,8BK8*`LCC#^SU_3G[>\"X?,XY9HA\n> M`4R1!`1PK%9U9)L`BPI[^HP<]LD.D[[,Y9=',)<#7A'+/K8_#!93(NT))K#:\n> MLEBW0CGLA[Q87(H0N]%;5`G'O4V!9H>B0O!T-PYE\".]B<+Y<8%2MB@AETS,S\n> MIF=:M#PC#^MH^2A#'=P3S[(]7`N-^#LJ*E=UB<U#?9\"Y+_L$_XW.?06['Z!-\n> M$\"I#5-L$EJ:(7%B%F?3,-A$PISZ_DM@*2$WQ_3:#&=KX=7AP.19GB+8!R([F\n> M5AZY7V$:`^@X7_R@Z3'H2GTQ,&^P`K:VHVY\\?F>5MT5\\Q_%#3P$L:@QQRF2:\n> MFW![Q8`P5IRMQ&I:OC-&;^1`K2830Q@Q@,TR,=@!VLB6X%5QW=UL)MPJFN_$\n> MB%IM'OAF(&\"J@;<\"IG1@V%!#&%3-IOL@<XU\\JI'9)E7,;\\8=\"@F1!\"2_K%*)\n> M_%&&H<Z>`DS0K`14R<Q/Q)G:S)E:88?0)Z;)2PG[?:'#?OU;-9U?2?UOL0KD\n> MMSH\"L1KP;]*I2E%!EBP0:)[APE]&W3$^Y#NPT+M/ZPY/ZZBEH%M;,#W5^Y&L\n> M301C@G$#77YB+M_AB1W[#&+$0]Y[CR*;`[?]2F%S[^GQL:Q>4Q;!'WD\"?]PF\n> M^..U`$D,+[<VY:0JA88)2AC6NISI0+&J>PBP6VDA$IG4#@50%1YIR_U\"9$G1\n> M7.7H\\R,K!T7`<6(_;XLW#8U#T!`+:G?:]L:&=9FNC?Q;%:9/CBX\"L-,R`#J;\n> MV/?69M<F$.Z&.[:D`\\PJ>%7!'`>UVPW53QM`I2&*$2)[[HZW).?/&VL3L=$?\n> M^&$LM\"]\\L.`'ITA@,RM':E:.>8]W@?0->N\\E=!BJB+'F=V)GNSEPLZ.SJOOW\n> M&E>G#&'0BNU9^EO082+688+0CZHG@;/%+-@&HJTY1ML3B.*H8A(X=V'J-\\!W\n> [email protected]:W;S.,:!*T;Y=4$OB5.2;LJZ[V(TCL5W-SY/4F3%9CCQ6IH-D1Z)0B\n> M94]1@@?GXCBQ,M3(MM`3M5T-4^@M,_1/?U]_C')#\\BA1+RG?T@:678FVCL_#\n> MVFJU?IN22@$8^:W?FJ3J!`):W/J-2\"H%41RV?I.2JL,.P\"#NM&0:5$@J*P93\n> MST.UCXBF\"-Q&[]-L@<FHR)@;I935+FS>Q?FJ((HPXW2;EF).TZ7M=5:Z(<5J\n> M+&;S69Y)<'Y:>TY[WS*U_68UP3W<%E?>6DYV2'H[0<=O\"D[$9Z\\ELEY#,VQJ\n> M7&13P!-XCYTC!@[C\"33O9)Z<9Q,`.RV$)ZN6B5HZ<1.3A=[FO-E:NGVTE\"Z8\n> M(!T6A:$7JAT>!8]-SJ1=,>2?C93)/^![+8&K$_Z&M`>!+2JMZL9N[,1NM+Y7\n> M[GF6+SFJGORXA]E(I1W#P(IV)W`V;(W#E&^X#YTF\",U-M5E;Y]$)HU]IPY>G\n> M-O*Q*]5?CP*A[$%V5B':LIK:/M^\"\\+B1+#]*:+0\"`3:RW=.XK_*,\\_I0_CA@\n> M:#0RR(P<V8I)^S,/ZQ>T`7-V0<]YDS*E3N#-L>-RHBRX@Q@I;KZ['^2RBT@Y\n> MUD,-.490(=&(18'NJS)<*K3+`FQ$@-UU)6>;F<VNEY;:!NAKGZI3=/E*1.%-\n> MR%?[8&G3^'V1+ZO-Z8W@<G\":=JJW;@3W4C\"..388Z:D\"9E3`N*17K,#XO1P;\n> M+0ZXTI[T2-,`N=20,]R.%(>QLKWGNO+Z;L:H[`D-*2^V\\F,AW+1JM91,N(IR\n> M*:5A-D''\\`KCQ(<SE;\\D4]O7G;5NC6@VA]OO?%\"D7`UQ*'OVQ-?>+F!:`PRZ\n> ML$*XRAM0'GZ;%@,3R[A5],K=1'6D]AG&[.WL,\"-F=RP)#<L=BRD>*`H!$Q.B\n> MX:<37@VMC>@8_RD9HNY@@]VN=`D75JYX$B'.)D+&P*65:%2,W>9F#_WGC^;N\n> M\"&DHKSQM>&;:>,%I!!)R+*.=I`A%\\RI`[J3V8PUSS+RM[>/]6>TUWGU;>U9?\n> MT,,ZIH*;U:98QI3MUM]2@MH?,47$V]I!_1<I'.@A&FRBCOM*3$KRX7LR>96?\n> MWP]X_7T;NF*R<C`].NW%/%U>+)E74MXZ%3B[9:B[4X\"-5`\\A).7@]\\.;8$,=\n> MRRG@?\\SX'Y7P[RK\\\"]Z7DH.&<R@7ERPEB=/!\\2&'!6RE<3D\"OO!SY=^GG<*W\n> M@*_!!E(8,H28E<\\`V3<I/I0RC,9)T[O).KLC27`0HRM!C'%)W6V[;:<-2F\\Y\n> M.7S'[3@=>*K58%#\"0`W^6>>Z(^!\\3AP145SC+[S?`!/+!G'$T1!T!U'->>?.\n> MT_Z,$VI/\"]EH1RH-ZIPB/#$/\"29+,WO2E8!4^55)EZ:\\O%;>+!7'<3E;3#`_\n> MM<,Z.&6\\)E\\7I1[/:C\\&01,#W>\"K+GD[*1MF<GZ^X%!,E4V+M`349]3J&+((\n> M=\"S]322'#E>;6ZP7,WBCT/#!7$1RPSZ@V@^HNY\"\\Y\"&9]>?I8D+Q,FRBSM`T\n> M)[M`1YZ-53(2*['L:>]OKWK/6<G]L009V+G3$>KI=15E`IQ%`0G394\"TQYGG\n> M_\"9YLHZ>?5='IUF>)HO^6.>-Q$S(.(5TPT-)X:-R3A7SD&-V63D/@&H<8VZU\n> MY0Q386)R1TP!0QEN&'C!0HL\\53^<8'%*R(08>QRW)*$7XM%C/$:/XQA'\"&J0\n> M(<HI:[V5'5VT)\\'.2#6)V;KHUBZF%^($DL7TGGZ+O%*DNXI3@`A%TDTZD@A&\n> M9R'BS&F4@1[(<KJ<<*H[9KRTC6T?5$:3/LH@[email protected])&M1,+*IU8<#9GOGRE\n> ML.EP7\"U@(A:^H/8CX376(4F%..\\VI9;6(T-I&1,*;J)Q+\";/U]B!H:UQFC[N\n> M)`6\",E8P,'RZHO,K1J+.F>PXF-,=9W!\"R:F6,X:A33LV>0>!Y/A+56(RP`@>\n> M%['#^1.*/1D3-I\"<P.3D$>0$)#8,8B)1[M51+<^6:=X`G)_##*)TIY)EDP&A\n> MK<N(Q0&7K.M<6YA!\"]-+(D^\"$4TEX^.PUE_E2[BI*]?YP9/I6PE^UCG*L()5\n> M7Z)@)`L:'8%`L<ECG?V_0YN5I;,#&C/)&#?+4]V;L<#(XV[2_TJ2L@*-X\"C,\n> M,%2-H[`)KX-5?\\F)MI)E?528H@P%[41F*F=\"MB8XGZI13%DJ``\\)8,HZA2<!\n> M&9(/?#)P9->2YN=`_@YE#Q[5KH\"Q($C,:(2@$/5,F3*F*4<=$`^L2X[7Z0\"C\n> M62GS\\72V`G0B'C@%A^XJI[^F:\"`8AC$G>29&!05W#TZ1^_R(;7.RV8-GR`;D\n> M!D#PXF0?.&?W>0_E0NW%(NE/<,]D=[:8[T`!.A+BX)M=DF-8\\X\\J*Q#'=4PD\n> MW369'NJEM';ZCV>8JOJTQZV?7DFVY*-I?X=R4B'B0HJ`YAE\"3A:F1Y9%@(P9\n> ML5H:VU1GS$?DR22E$0B$3_)F_7905UG,]H$NLC[QXV,M1QU<HJ!J:\"$3#RYZ\n> M+!F@*:1SN5IR[\"B>G40]@6D.]+?@XR\"6='C*@3G1I0ODAXF_$&4_A-\\&2L,J\n> MDVG\"(S:?Y;1=`,6JPUEE=;<X#S3I*);DQUEG#NCP(T>))M0I=(TC$4WHB-8'\n> MG1#3@%JCNLJFW)]-Z8P@$JLP7]%A+*G)=-K$$9$IYU2F<.8!%$$_\\L^%3/\"`\n> M<A@3G$V$<!H^.6_*&BCFZVI8Y5R4NKC)<6@<Q:NTL[R(CQ8%+&MS;\\1^[:G)\n> MM/6\"N7HR-0MV[-E0XVSN`\\V?OD#B&S(*H4.+;)1-U\\C+DJF<T9F\"D*'W;V7N\n> M97S.P9)6VQ;I,%U0*E=0568-RKM+&O>$Y&*CSF?N**5KMQWAU0XKSTBY$4GX\n> M=D=3K@&KC$L&AXYTL8^#Z3,_=]1`*-%K7K=AQK629$2'\\ZR!ZICNC\"73;0'V\n> M#L,.>CV.*:TT9%/.=!SCD634D08GOF0RIO&VX%`*`\"IO>B(-C8X&[^D]O)RD\n> ME;+%56H(G%Q5<BQ32F3B!J<O-)L5DNRK:5I:,E69=!U4F=^RB![2$MB%3J/'\n> MW'MDBP11GYAU.))+V<ZCB!-[A+O@)OI4B3G;FJ7S;.QLI17#WF!51X\\)I?(C\n> M!F,-'(T1`T)IS&2P`E+O:?<S>7S4&'8\"B_Z8EZ#/T0(+ST7HRR%%1:92H*C'\n> M='H$U%6>DZ?V%):4RC`HNT>]+LC]CA>W]OQ.)X`&&YQOT=\"!(0,^!8*L\"NK(\n> M&-J31!P3CLG\".LC(*?8,CVQ0I]*,BMB>SCAQHI5DF5$(XYTF?-C)F+)?S\\$4\n> MG'#JUR'G<YW)-#\":@SG%9<`8)05NGV>%E9!Z/<_F&LB$*$ZZ3./AVD>!\\#R%\n> M1O7Q#XXY&4(2?MJ'-@TY5[1HM\\JD%BX],B>[`,DE]5$%LRA0%^=-IL1]F@S9\n> MM3##:8XU2F=\"VYZD\\S3JP^*,E>[email protected]!)GLY7$MV/.5ZW,##H3@=3G\n> MQS\"+YY(T`<:/<SDKV8C'?%$2T#&9Q33EYRMUF(+**(1\"(-T9B2!=I/W5PCKB\n> MAV2J%\"7=GTZ^H6,HAF\"8<]AJ00C[W`.6[TQJ@C'><LH04Y:6,1^))*NQ9-4N\n> M^(@BE?A9NK>6CK6*F]/Y&(Z<-74N9R,!DU$9V\\_E]\",$A?>6PFSI%!,O=TJ>\n> M$!_WC3CPV5G;T>E[H/,#?^<MG=_NOSSKG8C*M<-+%CH7,F_QS&@9,-8!WVVW\n> M*D/F2&?(])RP*1Z3?5%N[$36W)V&3+R1P;FE*8$,MQ0E/K\\%#7VN*Z'E>7-.\n> M\"6OZ-E?G26E52-F*/3K,`8']F<_3(2(?VHFG5VI=D%/R.K7>SB$>_0C3HSL;\n> M6)I8[.%2\"L8CX)D41%Q#:YZ+B<<S#TED0X)>@HL2M\"WL9)URV@_O/Y+$ZM@K\n> MSO\\I*X*T<B4I0''X)`6H5?NP(@4HP,(#-=(K'DL+!V@!$U0MRF%XR8?XD/90\n> M\"$]*T%$]F<W>LN4G&>GI-`W*^$L9:XG9+5?`M/2;./EP_]%8F*?*#$'ICILM\n> MMT6Y\"O?)1%T_#LE:9:9,Q&QI(&LR:CP9B[+2.P=S,M6-R_$:E,Z]3H?@+,@&\n> M8,66UQ_DJ`2-2)HYE(YPR?NSU1+!CV)WP[0=RXZU=)*3&L(X3>4-U%))<N^#\n> M-%O*P0.BO>LSMA8SWH.`T%\"U`,?J8LH'>;!+(DWZZ`CCDIPOF%)T+E)];A7-\n> M(U3[^K/%0.M-\\B+72*I<\\:`K'$!S?\"+[>T<JHH3KR@FB*8C!J3I+3V<(=F-2\n> MV:AGJ$U2+$'*<@6I-5/G\"(EK<EA<G-+[R*PA%.P(6D6O6J24MIAPND>TTF97\n> M;>GTO--7+U\\^/T*>%CGGAD(8\\VBCS2S_3:8L=DP\\?7K\\`NUJM>+'?C,R>8]1\n> M6'_7LYQ(DE49TX.C^Y/5QA]KW:.S?T`IZCRY`,@LE)@)3>2^%U1!_I(WI>^?\n> M0,$S9G::.2K\"4>`K]=4\"_R6\";_PX!/Z8[I>@5^_.P:9F\"_GER5$7GT-70&B0\n> MYBG0\\Q%X.VK36B7\">\\^?GW*\\!4(G&^?MHTR2<SRN1IU),^;S2VR0:93RV80,\n> M'[(0U?$U&G=C-4`[A'`_P`3/J'H,]7%3.9_BE?(L``JGZ4CG%4@0B#'+R#.3\n> MTF$(8S6UT4J2D#(:J0@D:,CG`\"\\,6;7)S7?\\HA[B`/VK'E%ZZ6-`,(#R+_0C\n> M``W`C\"=)2JCI>/(&+IE1$>AW/:(M561H/<:_KQZK/WU5#RGNTKH)8IA<S62>\n> M<=/_0@/Y`G/@8_,$U//9=`\"(;Y)V9A7SY74Z4!7]&WCSKR\"7<_/NRV0!H]%L\n> MEEX-^-6(#D`-Y=7]07)AO?I]EDZGB30;F'>;\\B[E=VK*N\\\\FK!G*NR?HMFS&\n> M^&9H`^PC=_%C.FQ4-1O*W;9O5QCP@+5C.I=T_^2LCGGVS@G7]E\"]M(:*Z%X`\n> M;IIF0QZ<#J7-VC0X,#JM5O$]Z2B8S30\\#-=I?Y%>4GMM7GO#0D[@4?XU\"TO'\n> M*TRKTY%?;:HJ\\\"G[F8619[/)$BIR!#\\M*49IR\"Q(NLF%KHR:#:!DT'$X8Q[.\n> M6L6F`U+\\_#TY!R#=)L++(>VPV52;2/0,&A(/<PT/HRS^5(F>/T-SO*\\SJLG&\n> MG,5`^9+XP`%0%REVT9Q#('-I3&_+Z0F+E!U&]#;!U**C^M@E91_IR3&#XVS.\n> M'HCS='G)1R,EZCREK$]K=2DSI[&U5F`]':[Q<B=HD4M^'^,L+U-E:`SI?+`Q\n> MF;<(`?II6-WE@P=2XX5@S#ERT)SBS30I1[6?5Z27V@>RZ5.2Q2Z0`P=0[PXB\n> M<LM?TG(/@I'*0:<JY18=&J-BJMI<F'=8\\=QB9<4P)R,:B=K:0CJRT\\Z'1\n> M.;.\"\\\\-2.E.5`$8;!VX@8GMIO#SVH94H`>C0,3QM@DW/2>%(JL)2*BO*(^,E\n> MP$.\"T^5C?3H&+OH7B$!.>0D$!G6BA^@)([6>-')L-;J+?N,\\HX,,Q5DO>A*O\n> M1V6Y_LUG9>=R`+/X9'!!O0#5T(+JL8!$-C\"C$W7;XKF.:F%9NDOS40Z?_A':\n> MK,N9KHBZ(<:%L7WD\\.$HY:'FB5`X^P[AZ>&LNWK,)IQ'\\O7$?H.L#`:!?3EU\n> M.2Q&$P7Z&$*)NY1#P]59DKAX<R7*#\"&8T<7G9ZSH8%9SK`$35T%/E.-CU%(?\n> M'9'#;A_T.B,3098UD!-.$CT2:*:1@[.OS@U3C=.A&'@T/*]4+5&ORK7Z.TFG\n> MHR4J*X&G-6DF!CQ;1VM>=)3%(A5XHM`I'%RH)H8F\"BI&W@Y,Z%#(ZSLD;1D#\n> MM-3<B%IL`QGJ(W:H\"%`6F:1NC';!\\!R_U<8F.C1U28)B;@?5YMO:DSI2(Y3#\n> MO.!8SJ%R09V6>^URH]J>_C>&WT-\\J]WT]%O#VMLZ\"4GUUE(?G4+])+\\)!PV%\n> MH+'`[XXC0*$ZJ9[STA\">9(0EFR$]Z;@\"5C/J.*628RRY!V/-L/VL'XRXBBCN\n> MZ,80QK@5Z\\;(9FSI@Z1M:D!?$U7]MK[`\"?AXN:(Q3.UPJ'>`X\"\"R(`0$MYK4\n> M'L5,*P1CN4ZQ)W[DMRVD%!$\\HM_P5MP*'/46(3B.#?04)=22:$5:(;2XOSH/\n> M#('$-%CP6\\,8@@:DGL\\0%JJ#\\.W326*Z9+/INZ8FU9DF)S;3?6D%H5-=7^3[\n> MA9)1V\"S6!Y.\\3%?OZC$H,.HM['7;BTOP.E)_&[%JE<3$DZHD!JV4_&`^_._@\n> MIV>B?\\@1=D'G,[!*_N)E[T0=Q7OTXOA4\\N$>`XAGW^)BZDGO^7[=T070GGK.\n> M[-`^Q1?7*/CN=R\\.>L\\!23Z'<?N^K$NB8:4X7%3@<`VV])$-D!9A.5M&#M%D\n> M0'&@D]K74\"10![+V&2FH<09-7U.D!\\)'O'?X`LV0(<?U>4S'-+#XV^>HLQ$!\n> MI&+.92+Y8:MMZL51\\'S7JC<@+R)5C3`F3\"WG2,E]5:T`2AZ0)750ZHX[D6/#\n> MW`Y#5;=,9PDX!!83%:#H1\"U'IY_+9\"6M(V\\.5?BBW5&0!W,3XTASKU0(U\"E3\n> MQ-U0I(]O5Q3Q39%$QT<6B]1_P5+.IK;&I8J&C$`3E3G2IP-\"9RDX>$V`_$R'\n> MWBBK?%]MF)OH0PS%WX-+85K/L<^CMP3-4$L9Y8#!(+>,5Y_$M2>\"9X]!:GL\\\n> MPT^0D>,`!@&>847,'Y@7,A#%B^)(%9>!%I)H@B8-Y91`:GE>J=SH%L8?^YXC\n> M#;%L:KNMMM\")3+MB:+0FWL)`]VURV40+Y]O0PMP.[.0#&L5M.2BXO-'II/S9\n> MLC`V3'A;JADJTB4$I:!IM-Q6Q\\R>*(I47TNS)P9Y`26UF(_;%E:,8YXX(RWO\n> M6>=E[G#L'[G]V??..H]._V34<72[+&AC`B\\*@[J^DK.;L8H>.B&I#OV\"D)X$\n> M+%E*CJ8]7!N*O5`I$4&(:=FT8H-Q)_*T2`8AJ&&Z7%`/.W&IW&8%@K$;@?21\n> M\"I\".$\"VQ.BV,SDO,DTRBF(8%3HV',2^O%,?.S&EJ6<[VBRP3L8O<Z![K6@>'\n> MU;!K%2/P+*SI\\]2'RGO.)ZYI_Z_R%?[(]R[Y@$506P=H.0^R08,B#WDP`0OJ\n> M1.,ZK_10C.@9GVVXFG-(39O\\F^(7^%DL<O8+F+/0^2AQBD323FL=T(H\\A[JF\n> M+`+MAE;1@;)\"/4.;'E-%<YI4NP;E%F>[email protected]`#4VG)D4X\\=6@)\n> M#BZGVM3`V@D\"\"8#[)E-)$<C:0&N*XB,T!3L%QWA.':.54G'^BHN>CYM#8\":`\n> MXES1A;AA!S:+UI\"471-H@.#9V&`&\"QJ9T_M>7,7J4SK-R6+UM%^L6./(JG',\n> MG'DD!,QL1>HK:,&C4I]YH\\!0A:W[Z%_Q*2&&TGU!@4(@E>[;IU+MME6*-%]?\n> M=Z7:M-A'1M5QU%NL^<((J[=L\"<CK@2DMM[+X!PTO1+>?;7<J<P+@HPP96&E6\n> M;V)&/2IL-D[Q`DFQY_S^?KT=M0KOM]N!?E_46=:B_)AU(1!\\37XY8^29&V\\E\n> M_$5`UT$-#EO`=>MX[5%I&(;\"7/G8>J(,OPT&#MGP&\"6O5B(YDFFD3VDLL>B\\\n> M3_3`B_T.$<6\\N#,$.H<8ZT1ZB/THH-9L#1Y)+O\"M<C#(>*Z=7:YRD,-.X*BW\n> M:)\";?JS?4K)24W_0$4OY>+9,]\\A#(`3?6EL>+!!\\)I2>L@>-'7EO,XY,Y^6M\n> M%%0`#CZA`=G@,]0G*7.,&T?RZEA89!$2`]\\*K6,RJR:UQ-1':AE=]AW9Y6F#\n> MBEX:S\"2Q$ENA8[WBR/S?\"!,'8WR@M!$&%*66F64O7)&B&'G*YXN.HGZ-SMLE\n> M1E9F'!50<V1Y*Q!6)!X'/KWN9W;Q]1TKKRP6CYHQ3S(FS9\\U:2Z3R5L&DI>%\n> ML'WEQ!C)<<;*8S5)R/\\I2^\"R)>$D10[->H6$!AM_#XD7M*9*<7M$!QQ0[F%F\n> MFH1\\M-KXUWXH0C3&E30$/ZEQ.HT*OB;<\"ZABA/E<Z;H*ZN9%WXX$!5*$=;JX\n> MT#/<+'4Z<HRJX\"R.M/]6A20BMH1HD5P&?$SEDO#,:^!`D2!W5$>TJ^DM[L^=\n> M8G!\":GF016M3*L(EB<K\"RKXY9GZHB+M-;N5!VL]R-C1XY(;B/5T7&EQ)-NU/\n> M5H-4`GOW!P-)HX!>0K22RX8N[@_AN/!^1N'W4CU/7!-\\WJ93V08<NDSB;#23\n> ML*@95D*J$2U%&KW(9NUIP>MH')AL,16V=O+8&N*0=')J01CU=,YLON3-.8[?\n> MH;/9%'ZM50]RR(\\XP,<X'H?*\\=A02_V\\30.7\"X<J:(GZPL$4BOHD0HWG5M\\>\n> M7@H-HQ!\"CPYAHWT729YGHZDZCWFL*6E$E,1,CI0G\"190^'(084A+R++Z4G@-\n> M![+D05W[OOXXID7L[K?[)S_6`@^83X=BS15@;N!%D43'*4><BA\"3!GBKDH6F\n> MD?;/CKAU<<\\.;0F[I-;4O!UR96.Q7Q&A(VL01K6CX[/>-[V3'8FZIWB7OV,^\n> M7*@WDY%GO%&(G6-RWI8FHHAU*L_AT06;0&_%$/5<^-10=N\\FM-7(HD_A<FJ[\n> M5^\"W\\7Q)><ODLM?XS[1U(2QM;/DS;5)S-!*EA\"#1J\"EC1F*CKA8^-$^5)E_0\n> M[AT9QH#C8JK:M3DE_!\"NN,-TV?0[*JXLX+@R^Y#>V+WED%X=P<11Y:$89D>2\n> MQ56'2\\FRU:C(`*UCJ0O+3K2`LA:OM[#CW&1M[\"UH(Q2#TZ<Q=VIO90\\#PT-[\n> M#SCB!3@EBS$AECY9MS9[&LDIZ12_.2RN;<D$56L^LNXW9O$PP,G*(8.J>5SM\n> M86G(7'0F(>84(\"CK0!Q/N:1CR#.S:DLO+.3T!D>':J%I+M%-A<4B*PJK=I7B\n> MUJ>FM733+\"_=8#7GLP%'4P:<6+T_6TT&I(>B=#/[V)VQ&&+*5\")BLM>D1>+#\n> MK)IH-4$\"2A[B6CN9CP]E.]:9UI+8`7F)EMDBH1@/AB;F?1&X0V4(9:=+#MY!\n> MD9,KJ0]XUEF!AL:)A/9KSMB9V8F#.%9/+[:5?004ICXV!+DGQSW%+6<]N(:X\n> M0).5W4+0)1\\AC(*)U0`56\\/F'>TQXT.&<;51]Z\"AU%UXK@)IAY(%DS<Y4:>`\n> M,L,`CTPBCS5+$!B!*^YLSD=E4%HRU?790HE]IS1UQ)41>F&G(GA()99QROW3\n> MIR\"/S3+MT\"S33C`/DYJVRH?F;.C;V!HOY6W`_H4>)>H<<B(GBC<;R`ZZGW4+\n> M.\"N1;^J<H4*`3'E:.]<A<:@-3?-+6E2\\6-'\"<`YUYT/>9L8QV*P;C^44<%I9\n> MM\\^(;EIG1)L09+UF[MU,#V-'-EF0QD\"^-.-CZ>U0]BJ,:4TYIE5@]SMQL%,/\n> M54`W@E2,4F3-4>O,5CSRK'3&.G%1/?Y[Y349,\"?!\"$2C<CTX.=@F.)DR_H5V\n> M+-I``M%ZW3.>Z<J@_=$+0W:L_5C'&!?>:H,2,<?%D[D:0$Z4P+H<;\\S5-C.-\n> M=^;H460\\%&/Z+1)!;H@Q6)A3+.)5JQ,*69,EBB4+`%L+EY!5B;!<\"\\MT6&ME\n> M^US9D3$ON%'_P!;'(\\-\\-JJIS*CV-?J12(4/`=?2N+'T3[1SFC:B?(WUZY6<\n> M);O&Q:-ZPAK\"S_QK6?88J'K&N#1%ZT#*W]TW[HXAOXIXH.-O\\,D@A;F;LB]<\n> M7'-J^=VXJR0%T@WMCNRZE.*BW2I:D%0X=XBUL,^%/9/2#OO?D,J:L5\\5J'D\"\n> MJM>+OU81FV^(C:P>F]Y(8R?F*5MP2)$LF4[TMO(MDLM1)*_MR1$_`M&8$,()\n> MXG[)FFU\">&Z12TP3,CG?B=+&AM)&50'`B/T?ZDLQ_2--:.@+:D50KWC#?L`0\n> MLR:WHX;&*9#6$$G\"\\B\"6QI`:X3%T;`(;%8BON)\"8%@EJ[E36-[9I@I!!P8=F\n> M'Y%%&B7'WY!K8KP2>?`9@04*<16%.+$7;Z20@U=((8&F$%EEF\"L*R53&H8%1\n> MX*O60=3>,@I6M>2J\\S.10&IY=X6WF(!A4OP0`^Q;X`6*C!<H6L`VXKBI/'VM\n> MJ,FE]+P#\"V')?3YEE]IYRN&UJGH),!JJA0]V<_H!$)NN&%V#86AJWJ$<VPPS\n> M1]/;RS>GG/>$?'>E$`W1!NT1)R^Q9W?\";_LMU=:D]A4C!XJU.Z%C@]2QBRFW\n> MHPKY0'RQT^.2-K1;B2(8<WA\\43O4C6)H*Y5G!YXL/(BZJ\":O7OV8KC&XD;2C\n> M'%X\"<]2,3\"L4H1,X-S0SE(G?YFWQPD]5<Z?U7%,M)LVK\"N5^=8S1#>2PXLF,\n> M26HDUFN/X]@,Q?;-^-B1;9(R251GTKAE28IBO/3\"\"XM.WX3BB#4BLWZL8A&&\n> M3!44].U%8NS]R.D14M:P2GM3!\"R*CG.8,M$MH$2#P,?-I)R(9\\R-_+R1]!RM\n> M*\"AIS8IE)@$-)SP_P#@@0/U`5BU/F:-3/BD$6=!/Z=[6T(\\>B9-3$BNH1FFF\n> M\\:,9@[$U`*FQO:T!8--[H<Q@9V0&8J3'@41181P\\&0<=RTM[`2CA&_?O9[<X\n> M(#GSXXR?#JR$1J?BW):%[K5`*ZTS*^?M0J;%D*?%V$P+M9\"CB9?<=>SU(M)1\n> MFY\\7Z2^K;)&R^9$LU?#28(3J(%U*:&&-M$4U&I>T:T*HQFS8:,957/[@Z/\"P\n> M=]([[O88G@%EX/+DY\\Y8D=;IM%J%1B(:HPY9HG]=ST$XGL[4PATCTM[3H\n> MUQAS#*]FU!`&6I0QZ)Q<>!8=*/&@S/K\"E*'.V[.$>C4L3I0*MY,:7<>:&^/\"\n> M7C..5%T;6D(T,-8**_BO+XZ.<:&:)D3*,1:,3<I/+;L[<CZ\\H=1#LGH`RFQ1\n> MY\\WCL[+&`SBE[0Z(TZ!9P*FH4TK>$3)C6D4AQ-%NZZKU.E8P.?;HF81N\\0CU\n> M);F78F]]DK%C2\\8F:NL8.8\"MX2DM\"3I4T2\\FHK-C6J'V>QJS!'8L\"6L7>&)K\n> MQHZ*=%W],3@BAS#Q6>.9RAOP,H(SJ%`:NTII;&JE-\"0!CT,VTI2#H6,$4CL,\n> M>=V8E,LF\\$R*+R7E\\F1O_\\G)WK/ZH@9?W2>G>P?8U=.]7KT=1`Y%BQ7L&3*$\n> MVG&QBBZF5B-3\"'^?PF_?@Z%>?_M$)I-H(JDF3-0-S*(D'5CZ=]J*BG3'C(O=\n> MBS-VKB-Z_Q_OQQ<NA05-FJE^,GG,K+5-\"81Q7\\'2K#141F!I>E3A4@5Q3M;F\n> M&2[U4%.7QN5<U_EXU*0<:GW&6G9FGQ8-2(=\\A4OR]5=.%68P*C?=`LD`&\"\\Y\n> M(`H#`0*PB=5UK)%P2B,1=&)I4063DS%U;)W<0C[UT6Q9L)>=*H,9828CVR/W\n> M8G&]1V6K8Z&$_($\\Q;BU7N4Y&G);%$Y\"L=6T6I&Q[TM2/$FJ;5SZ4C$@`\\!K\n> MUN<19)N-9UO@MV3ENJ?7DNSP;F`U<O8L`E(,(2+OUNPM]MGR,^DM:M1U1]2/\n> MN?+NI`BIB9:!D55'\"PMK7Z9SENE!P*O`*69Y2NCH/I/N\\N=9IGH0$+]X3C&\"\n> M7,=P-IG0X&#%3EIDM!F;'I9;9*\\>@4$1-#F0:@$D3/N6]M5NK!%P1MDQA]&@\n> MN$>.M\\IU50F,.!]3R5X]!L$0-&ESE^S\"PG_`\\9'YV7?B=N@&K8\"W8(6>H_:O\n> M<5-A/>A@4Z%I(Y$VSN75=L?:5=;B]YQ:)\"]&YL6^O#C@%R/:T17+BVUIL,-(\n> MC]N1VG>%C(\"&_5+\"Y80'#&_B`4Z)!]BV$M+O`-,7`\\9!K0\\ZK'PBQJG_[W'Y\n> MBU&[;Z!_)M\"?['0)9\"@#5WSO0(`'9NN'CFQ-L9!L?Q+2VTWTIE*>!%]J\"ZBN\n> MR*,=?7PKI%L$02(MG/.KOG?+JY%YM2^O#N35*-#C-83QTJ]&M[;*I['?\\.KF\n> M5@-*+QU+N;9YM7-KJV'SEE<K6RWY4#&IL8.?P;\\SKCWD90):)=B73*7:CF`A\n> MI&TMD@MQZ*N@[VKYW/%:)L+<2.?4*1%^;CMZ6$P@_<<Q[CZ(F?SWU?3?EMAC\n> M@*R)_ASG1F+'1IKD3;DKK>.;N'WD[D2'>(YX!64I)QE*WI.+F24\\RE)IK+?]\n> M$]KV1/&\\1,$M4JU.H<?C&D=T6.FY\"NP^-4[^/03&B3H2Z$6Z&_KL8R^Z376+\n> M6TWU8EES\\PH55!!&W%I_U5@F@666BLZFS4BAC7;'<]I^8&A#2R'FV%U!]H'\\\n> M!N87^6X[#IP\"[S,\\#SAJ)_2MK;YJ2W4H-27R^YS:[E\"N+R-8^#N2LGWY/9#T\n> M]27_C\",&Y_='!SV>9)9+L6GLH;<DL0?B$73639B1\\MA5VA1BM@QJ-'J$%*4V\n> M:>/EP`0V#]:,EY*#$\"G%]RF*^;(05['>;-<29MS,+\\X&!7=HXETTR7,PKR1/\n> M37(5+%AR,C85_^\"M*Q-KW8,]NT-Q1`-?ECNT_<3W8V%:.>T_<;E3[*>F3@:\\\n> MHQ3K[N/MCGJ_\"9S3#T)Y?5%OMCR72SNVESLM;^D25W/<:A5>C^-8-V:MAF#D\n> MK4DD32&R7RL?N=JF\\__C/8*81Z=3A,F'[KCKM8IUB5DC];86J3H5Y(1-SZH)\n> MTZ-@M\"C4Y$C0KA7R*N#XK997;!TS`RJ4F-;?(QU>F3[)MC?EU1<R(R\\GCP!I\n> M*>PP85*WG`_B:M(KFFC\\4D(W$^SAV\"Y5@O6]H73J+P'##FS>^YSR4JVLJDC4\n> MT4JVEA`YD1,(6T(&#M@BFT\\!,;0\"?@NMV^M3VN7VC+TN\\,K`K.J8(%XR]+&1\n> M)EM=1=/#;`PJ>/U\"1\\5?%;B]WI8>89!O*_:5/JF52*>LP8]$EN(F#=3D*>M`\n> MUW!9W)WC.Y)`HLQ*T1346CR)89\\S(R:FJ?,U1HF[TU!&ZC8&_\"IG+S3,5S3X\n> M5&X,Y<74O#CD%]N45^-<RBGA:]Z#0BK!1*JA*$-%-77\"JD*F)H=+=;QU0&L#\n> M^9VR<N6K#3`G%)FB&9G-=<D3-T@7A51'F$D0H^?!X(N<HM!#A0@D##ZC=-$*\n> M\\UP^I&/$V\"S\"?`,4WQ]PEN]4[JJ<GIY*)R'!/P'I.R/>(S:R4N>54V\"Q?[=J\n> M=\\1Z6@(5RL+)C884-DG.J=>OGC]OMW]\"[47B>5H11G2\\MM777J?)V?4H^\"QJ\n> MEPQS6NVPMU.H\\.+`U>'%EVS^<I@(6MN.1-<I_TH!6A,1(V!28*1X]\"I#K3@4\n> MB9C#@/,E6&&EF0ZW$O>W<1XT20-<KA935?-L5=[CQCS&CHJCW.',D3G:!?V`\n> M8\\KU)1O!@C;MK(<:!SJ\\3&<<6@O/<JSXK*%BANOQ62,5GT7A61GG1J!\\?CE'\n> MVCLF0&XD,30:'C+R)%9N`VV0I]>I\\E44E%<T^@(4X49[!3(&(S<.6\"\"=@NKS\n> MJH[IF%#4O82*GQ_5<U`'@0'NX<-C>+C_'7STZE$K=E1E)8TV)E-<58K58$Z:\n> MO+:'EU85XA>#:8X!5[S`-:AYY*ETBU#U%T#(Z24NP^/)<599/X[<=3A`&X:>\n> MC*#Q3/4#6\\Q))W+$+SM6)?$!I5CA^R1S7M8I<QD`/M)Z-65L]**V'=U/UF%?\n> M!>MQ\"/&T;L7%B0[*1Q]0LF?DZT=#GGZ7*D9+^ZZ,.\\3:CBB!75<H^W%Q(U^J\n> M,P4L?S\\MA7(23LQU8W:WI;R[S97=;6.;:IU-6]I&8E_J^.W9^3*Q-B@4Q2:R\n> MIWPU`6/#]RB]0B#F!N>'@C'&F^U`I14BEANU),&!I''B5SFRBC,LE6Q_T-V:\n> M#GZ&Z_%3S3LF=^00M*`B!`W%MXK!H_0\"82D$K1@<VU=E91/K6E)&S)2J$RW.\n> M%@,*#*[email protected]*/\",G9OH7Y18_4663%4<FF-2)U)V+AL.8MH*%,X2U^RT\n> M*RPI%7=(J8M5D'-U`&+&\"\\X')UW.#B=[&QT-F0314DI/9-WD3;6T.=JZLV2W\n> M?XL.'2[N*DS-&?4CO6L60<=-O1M`U^>77=18W]P`_(B!/V/@QRI\"<E@$7K(Z\n> MLY#3:WB<KB0F`UAMXKGDU+H<89SE_56NO.\\ZB&H#76`@_I12P&!&8CF\"6C(U\n> MJMG$)ZFL)?U!\\FM3XFXZ6:5?%`#Z&!59UE:JB\">J2$$'89`E0`'''(:U/M)[\n> MJRA^?\\SH0JQ0SL%2KM[F3X27-M&>TGE$$V&4P3R/6,DXXY1^A!@J5PAGM^<7\n> M[BH,S02S0UA7.6=YIX5N``V7+(B?6>I227U')HA,R.^`7=2,K=0(:-5L\"*1[\n> MRU%^Q/@/ZS'O;-366'DS.\\$<1C[#S&3UEDN.99/[B+>@\"5E8%%<P<'1!>P\\!\n> M^P<<O:%(+VH>ZG5.E;0W,;H9JQ6(&N;#FBE;N9_\\)AUD6-QFIW8MJ,7KBN\"R\n> M4L\"A\"D-70:&\"-/?<CH1;VDCC!40/-](U*U70II[B(!<[2BS:,TX\"B]6@:]5I\n> MI#*OVL?^C4FK53L,1D0YLGM`J[(,4.1CWB+6-4N$1/(6&\"D6BX-6F9\"<DK9Q\n> M3#K3/A[VT=,^VI;X'WY65C5G<]'.#U936-E:5U0\"Y;R(VRW'CRD_K#A7KIAD\n> MS7JMUF%&E@OFGTB/6*JDT;3:;H'\"E:NAV!V8O2_*SH]AQ30RO7^!CBO+98&P\n> MT]9+O]T4K?.?Q:88IF,@6^O=B)H:5$!6**XA<PI-=;SF+4VI02JVY\\<8A='F\n> MU:H^JI88-V\"JPVYJ-;2#<=QV67*52=DRKP@\\7Y;`>H452KWO0YT>IE4[)5;4\n> MTJ?81%JAHQT^O'_TC&-H-J0NM/0YAH2W6J@0F;YLW!];&_<E$$CX2%W.?2`M\n> MP3*NAIB2.91T>4'H&8TJK-:H5%#_D+;.TJ%F=I&PI'2!EJ;\"_;\"'/U=;/J*]\n> MCM<M9TO\\9O8J<Z&3E.]!-DC0&<!Y]@XYP7R&O(;\\DE8V=H8*L[18F=C5>CSM\n> M92,7L&4LCZKL-H19P%?[HF3OW'DJ2KRRM>7U3*L\"01,E6OED%:-HB\"[_CK/;\n> M<O*(/%D[<86/H;\"2?R]FQGUB=DN,9;?$D'=+4.\"2,Y+M$F^Y/EI,H=W:66XG\n> MHF0E^&U-/!XD9D:S!6CY%ZB-/,8$Y%@KF_,X=+C:*^:2*HF%'JNE&;7-AP]<\n> M21;GV7*14'(-:TN2\"K]D2\")U2-(FTG#&MN%-,)@$_IRO[P*$$.Z&X3,P?EEE\n> MVE;C!!LL=W3KLG6]E(`]B&CI2\\8==XZ+JL8;`#BMD%,X;J&DK@V5V&4OD$C%\n> MLGZ&WA[Q1!$A:-V,]HWPJ69'2W50$6URR]BGP7E='$7/YH\"^@CVCUAP*D]$F\n> M/S8/<-S&V6A,[JJ).3N!LLYS@DW*H$KF`M[4/K/^)$&KF=X`%2B50S,`JNEF\n> MJ(8;769&WY90-<V$?D0&6<=C-#@S)QUSQN<(`?/.K,T*1\"6BI\"_X.`C2/JVM\n> M/T2<$V.X6Q8A\\%?N/AY5Q'I_:L@+325XD>I<U\"5^U8(J:@.S`=I:JM,[2*=/\n> M%Q?95+66L64`M&F.FI+4+P;:L5.*CZ&-%&A&>LVVVI$:EDXZH--!S%$F9`G[\n> M*@QY7U10:1RW:BZFY7EFQDO.P\"&Z<RHLY\"H+RE\"69%+:SRE!%19Z+/DDZ40/\n> MDE/E?#GJ4&\"U[T--<V=8Q:%5%)H*'=]$X#]27*NSQ@@DV8))<EE)DK15R&+Y\n> MCB0SL@`::IY/1U2-Q#=`0=>XOYNWUYJN4H9=!UG:/#G/)I([&*I];&6BK%W-\n> MK\"R>>1](MX)1C1UK?M#Q00;_Y/7XEKP4*,S%<)'\\KV+%T^Y!#*K4Z0ZP:VF\"\n> M9ZFR;!#7CHB;;&#.)9`]\\\":^?.1(L*PX4&G*;?!^TM0QKH^Y4_`981\"P@Y_-\n> MBGB14.)%,\"B$=!`A^OWC;UX1:YW6OL%59PQ3(\"]32PY]J]I#@,>H);AYEC@P\n> M^Q-`GTL7O$DS0P8O1Z/(?(=I0TRF;^4%8Y5-MHZ&53'!^R0<*>1=O#XR]F\"N\n> M^'(>5U]0JG*KX;&\"<O9+?:R25$#OD(IF$Z4;):;FTL[>A\";N);JV:?\\HA7\"2\n> M&H$\\##.G4X3LF`^FX;#?MW3PPY]IS25U:$_*H!9J@]#7*XDMM=XPY)TK*H!<\n> MSE)E>:\\A&Q>6&9+Y/`5.CD`U6QS>ERTYQ3T=95H\\6\\5H7E-AN)MY#@Y`JQ54\n> M#8#)C<\"A^\"1L7RXRXKN*#/98RSU:6G.F.`88Y<;OV\"GQZKQ82P?\"*GV,CMFC\n> M2!@*\\F3F-I!$8503YVNPY,J*54A=*[7HE/(B#6E10U*N^54[_O9'(\\E43B.%\n> MN]U7T[YL;Z6-?G8*`]KG)U[PG+-M)._XA.Z1I/=JL\",Z7UVHC6U4[\")YKW[C\n> MZ<=(QGP&,A\\31*@\\I^!KLG0D\"1VB)([9J\\$YR?HZT9B-4-Z>K3W8M)T7V*P<\n> MJY10^F>]2;GCA\\J<\"6F/,CJ/.:Z8,Y-YD;%,:,*)%%1Y?MAM17'$D@1O2IE-\n> M>$:JO0-CM6V7CD]<TOF>:M.MN))7N>KHL'+ES.=\\(LPP2%;@Z:A7TV7RGH0M\n> MC!`F)/.C%B^#2H/$\\E[O/W_^KX.CT[.CX^[93_4FQB0'E`KD&EI^)$NR?TY!\n> MMWWC/X6?K_=/D0?\\N?\\FF61)#C=_`G!?8TJ]G9V=>A1C(`4Y*5[CEB-^\\ZUZ\n> M<V2]^?;I3S_]](%A\"RE_W.')\"S3*_TP6[QND26ESJ1H#\",/8\\9MT\\HMN=(B7\n> MX\\)[TZ<(E7IO\"BUA0_`FQ>*\\_ONWO9,>#<Z4Q=1/#$>+\\OR]_N;DQ2OTPSS[\n> M!U9+%<Z&;]`I]R9[6G=:;<!2BP3\\ZX8-0*'DS]`_&(C7W^Y_?W3\\#6C[IC$!\n> M)J(5S]?7O.V,46WV0*7P.ZOU?NCV7IY]J`\\-G4\"+4@%E9:R]?G%R0\"EVJL%-\n> M$?/=?QWT3KL_U0%T>\"UBT+DN7$?&Z5:&'BT'_2+\\/>$$/VWE\\E2!ZBT5=2\\K\n> M1/\"Y`KYEPOI']MEAN%K*I%E8[I;.9=8D&*I(+7&<SE9Y71^>\\QB/<J>#>'U*\n> MS$C>^K927\"G7A7+MT_21G#K2FIVBW&`AE2!&)H!TM'+L^=NV\\D`5'BYRA\n> MD*MC>7V9CIP54C824RX;7U;?3S/V?`::JZBD,F0X]RP?9DMS85:]`KWS@E.'\n> M.IQ</JT/\"_T:BP[)`=[K3&:/XP%\"LDDYHI+\">A2'T:X26>E7V__$TZJBZ8J'\n> M]E\"LFEIB846SN'U\"CAP9LR\"D<R7%H0OL'318/)V-4D53CJM+\\0T;?ZV*YZ]R\n> MT]*<\")KMEKO&YA[1]`Y:E+%'V`R?D8*W(Z^)2SS$$Q2$H]I3Q+[W1((35/Y3\n> M'&X:@E%:<M'MU>,H<L\"(E#7/E^;$FPT'K\\1A1$O[S@U'X,0=7-D.^-@9?MD^\n> M`V5,9Z!0F3:'5[;4(3OV&[email protected]+?L!7Q&!\\`9M-+%5/&5HG54YLM/1\n> MJ5GJ%&25S'M0:S-CD2BZ*6=&4^3#$6=R:M;/Z/&73&WD[]='CX=^DW?Q%XAR\n> MK);25,*14B/JR*>Z.AU<49T**Y'#L$HG3DE2-O2XI7ILK0D3T`D!81OC](KT\n> M16/;T,3#!2FY2HG8G!\"#KS836VFI&X1IY.!G:WVI.[KC4C>GB.*H'II]_82S\n> M(%D+1J1R[-4[`0+:472,F&RIDWD8XC:.34>?#U0@67R;@O`4I;KZM!YXY+8P\n> M_57E\"3WH7(B\"0(F796H%(6EMB4Z1T>+\"SG^;&O^UG(\"H-B[-IG+4(276*G@G\n> M;%M@J9Q^RM,:4:@\"OJN#^<47]2.MC9;BBE1XQ4G!SVMEIC''AN`\"*IG!%$V`\n> M\"C-N([/,P-R>>5`+$PQT>)+P^29]\\:<:I[%D[)EPHF#:$DH:J#@W'!TVP-%$\n> M;W$^75W2\\7;DQ7EQ@@K5_O%!72T?``?#-=QI;8\\.E^I$\\3IK+9$^;C;V/<H=\n> M8),^G4WE$5/7I'_,V?.^!A`?(J4\\A!:Q]19E8*4\\@C_J\"8*E/.S$BQ-K/O^9\n> MR.K'^A->R`M:PMF()1'Q3#@WY-*D^]0Q*DB,,(5:MY\"YW_1;S@8ZQYW1K2I\"\n> MYX7.N,WK++:MCIRX8%3KXZ!&2M.WQAUM)\"(]\\K&8<%R3(FY4T\\HLD0:=0:A\"\n> M`(KY''FQ$_E7+1OR>4UCB]^.=*R;\\C\"A3V_%&2[)A7!)6<M0GZ/UJ8P.Y>4]\n> M\\WQV$^V(FV>XMZYN)4WB_&2.VF/(E:QR!1$Q2MY4NV'YE4XV\\\\);J6\\$DGY,\n> M^Q#0V*@=O'CU['E/7B>EEVARJ&D2%<BP@AVK6I\"^6MX37G;Q@KBD`YB--T@+\n> M@1\\&E;0D8+11W_([?A739&H\"RPN?\\I%I&`7+'#+PB*<&(<7G'M/I5T1G$;\\5\n> MDNUKTR\"_%WJR7A2W;)[:+/-4Y`%X5!6#^7,M&8JKF<CK+9F;Q\"@47K-<$Y\\^\n> MT1>,YXG.-J1V0*')4S=;0_G@M%;(9YS3!M-EVA]/,QCGNI797<\\$CB6R?;KH\n> MQ)ZPYF'GBW:+$X5CWOG,=_*(<(2%`(N*!T$K9*_X8M#J\\(FB*G8B,YM:`U>M\n> MG:'B5>KUD-/BL^6?FL.E5=@\"[;)%2Q0J\"AW+NTXWV1VLSS^6SO.*48O.%ATO\n> M9JO1>+:2Y'PJP5Q>Q($=74C+*QV)L_@KL@/<)TS4T*:H:]O\"B-;2T97C1XOY\n> M#,><HSZW^`3B_^>93F',]@GF!*8L>9)B#63WGIRN14<76E8'OJO5S.)AIW63\n> M1K91;7#(\"7MR0-[,5D)+B0^,/NJPKLC+'Q32OK6Q86F$#O*0T(\\K3([3G5/-\n> MFW:8(5#A($1I(K)1Y\\>%2S3'B2\\!>VJPR0;]P7[)BQW;2L'J7XA</.WM<+Y<\n> M)3U#RC%&[5IEX,>3LE\\\\AO\\=_(P^A5\\\\ML[YJ324JG99U,?ZP$(CF]MM/#S*\n> MLZ-']6X%1F:[%8$&ZAEFJGAIW<&7FW1D)D>4V@=1XK.67WZ&/!6>.\"VBA+_*\n> MZ8/J)$E\\)Z(=U1R/6JH/U,3R,^;%\\%I,FZ+YL,]B4VVO_$2_U*8@?OL1:<L@\n> M/B3$V61[90+]_[/WKUUN7$>:,/JY\\U?4^/0T`3=90EZ1H&U-TQ0E<T82V20M\n> M3Q_32PL%9`))H@`80*E4MOJ_OSN>B-B7S$05BF2[WW4.>[FI1%;N^]ZQX_K$\n> MUX#;H@UEV<-KQI2Q\\(<6%),(3F23Y[%](DQ3X6E,U70**0$^=^R::HZ3:F!&\n> MT!9>`A>;*@7.;PN\\_ZW`#[.;AF:]G9I\"\\$]3>N8`,SF?-^N7,KUH^7J8<V;D\n> MJV9/^$=DXH):BK/V:EI+L<#RH:^/]$M\\'#0U7T0V5F+`?W:VP#;Z@TY;B$N\"\n> M.>4T#-!C<N9DM]D=/@'CM!_$E`9J2S$,0OU$*I.`>.$$]9+>2\\O3&[Y1%II7\n> M>BW?12&5)O1?R2(`;2D'WBL<45]&QRJZRQEM$<;?JU._^K/A+P)]XGNCQ8GU\n> M6%R$83HD^BN:<.2$>@HJX:YS;NTK228HWH;L3<,4Q2I5H4]JK&9\\[?/12TLV\n> M[?E@G9_%IB:)#EX^5J<@FX>S&X!)/+:[N_#Q%Z@!R\"`N260CT._6X=\\%(&_`\n> MT@E@M9@P@'@<,6(0$7N+=@W\"BEY!`_V2PWS.5>?!<3YQ'N=\\^9/IAOK&IAMG\n> MJB'#&2<JT!@FHN[O/;LLQP7XEI\\0[?:MF/G@\\CA\\1$K7'\\R7WSQD3Z.G+_[X\n> M/=WPN,O^^!VIF+][_CW=A]\\]^;^<3$:09)G.T;RCZP7('P[17&0(_^86F!YF\n> MJB`#T_@65W0CGTL:WWG40CFTT13]PQ&GA/#P;>QFVDEZ`@14T#(U'B9^H,<.\n> MX\"U;\"*:U!;8$^?,8Y>OE9F49'C&*(C;$3./\"\\'Y;ZU;3:J\"1#3HN582J/)\\'\n> M#NJ4(;&ACX\\@!SQI!*CCEW04;G`NS<>\">X)UY^2^:P;W=%H0SOI`&O!13CJR\n> M=<!E>CAOM;4:'),_17?L5H`LATL5/.&TP![+ZN'(9UV\"S%IZO?&9A$WST9>-\n> M=H9`YIN0)P%_+5(G=,.C(N[R=D]^^,8EC1>9R+S[T6K_C)P'E:VG`A%%<A)[\n> M#HTQ*]S%Q1*:-TJIG,8BZ;HZ\":[1R'FI)SMR]`SA163D(#\"(L_-<@@`3`4IX\n> M7BLGX'M16\\!TY=Y)!R%Y\\2([H4(@%:?=F]]E)\\MUY,VOQ1\\,]:;L_!9,I-*(\n> MM\\(1\\&=EWCMOXS+QA>_H@N5NGH!R7'+=3',F8T(;AI[##S,R[R;D2I6)9CD=\n> M=2@D'Q)X&GY#[W#TH#1/DM1129'=>-O8,$<DRI8J_&`63Q/BDI,9UHR4E7RZ\n> MC5@H3E=K#@#BDQP0*]_A(6(4;=!_#BN]J(3;]L/%&K!,E<1?BL7<DWE-=X2Z\n> MU'K]*;B9I9Y!@L2@0[6CGI$(P.N-!]<]6!GV#D)8FZHJN5DZ//YA`,E5'R.I\n> M#L_#Y0'O5AAU\")BC7^<^</>;Y94+WPFOALB-3O0=87P>Z^`$=F%^A:N=V6GW\n> M)='A2-N%-W@@>U$JEXC^'7>5^^7)RGU%U88BF7:1W882^MF.K0*Q/`\"65%9?\n> M!HAK05RQU&69I\\97%2\\&WT#:^.-+/H:__P\\6HB%Q^BSO@L\\1\\+R:O<.P;7%/\n> M8`*B91O28F&/!@[;\\]K3N`4Y<\\RI9W,]*#.9P-GG*\"X-%9#X'`;&G8P5>_M8\n> M(MXGAE7+;\"G*ST5#RV+-[DS(@>U3[J4Z@($!I(#[3KQ.`&FA2);*SW(84[-V\n> M89SB=L]7;44:5I6:<%NC.[@%&CT@G,YTNB\"5\"6+,AC:89]F%&A$4]4F1:&Y1\n> MFIN)D4=1+4_-O\"]S&85G21$D+HOSDKK\"\"F[6.!6C46]@3QG<_'%/9$_4O91J\n> M>RDMY4Y:6+/WBB8$T'2!NZ<+Q!XN^=H+`WV*PH%@>YKE,4.9=-4W+V@!/3V.\n> MW%:L2C'GF0IS9L@CZIR6\"J<L?5VSI\\&)5(5#7TW@%<)[6GPZI#.U=H85TO$H\n> M*[I&Z<=LM(ASY5N<&7HIFA3!K,!=&9&R`8#5+2.T5:?`S)%`,1*K3ANH%*Q`\n> MJ:UY60PJ@'Y(W)?F/ZQ0X2]-P5B^A/8F#;YD)4J[SBP?19YM&U^R>L1^F6IL\n> M61IZIQ\")6BE@P#(,`H'O*D,-@C1=>TY72W(I7$L`'>D^'FNL6&*%::54(KY[\n> M=Y:-V7$)L#1(A??$H@>SPJS<B&Z#6*ULY[QV8NB7M9?5XTTX'(^I!*'SMM5@\n> MOJ6*7,CC$B\"XL9H(X)/06L)XB-U03OC+6J&Y.BN-SR9Y@96NM;;6,F?XC..>\n> MQ'FAZ%GC6#Z#(2UUC;8^2_FS&.`<F:O-WP>1]BV))[=\\YFI+<O^SJ/.9!\"K`\n> M9H]=)2*/@QKMN0ZLJTS`XS'%,@6N#A+WU7*L.6S,Z15'5_%79ZHN+O)`HD5W\n> MLAAR:$#8YY:>,`ZRIS/@?66D@B3<5V>W[ZLHR?U(_2/[BG`?DW!?=3>,N7V3\n> MLB`.AX0BA@JCG9N,Z::-?)&F]7_<1`EAZK:=AMK*,@[EH][*)MGHUHTFG4U'\n> M=)L)GAE5GX[H^KBSLVE,V<EOW:%<?URZ^OD-80?QF\\SSVU=U+TE9!&_BNWD9\n> M'@*\"UP)>\\C7_V6D,8'RS%^LUZ[4X\\F\"[=0XTMVA%^)*`S8999@@/H;Y8]0)D\n> M:K=Y\"&TLQ>:@283@7Q+Y9%)5%7XRF468@+O%,T_.**1[0LSS)[!73#Q[!=M&\n> M8V<;=0Z#2^<=8TW:W/6HMIRRY8KA-JH`/T<D)DD6M91$DX9QJ=:B\\P1#SSH*\n> M8\"RWV&5,=F7YGTM)[XJX_4C]X'WW@)9EU,N?,WA\"3A*DJJNO=N(M'S;7B%/,\n> M5-)@,Y-/32!7E.<:%K7&OVP[JECL@^'2:5U%U+:.BH@L;&W&6AS-2;7W@M5=\n> MG,>511.)#3#W->;[9G.EFUP]>E@I<:16R)H=0'3XE&UUG`L_'Y\\3?B:EL.)8\n> M;G6^(&5!,1+X@S^`VX!>B3;J9.*I?DBJ^8.J3<E!F)(&Z*PU9)-_OT>\"!\\@/\n> MJ^:]DXQ:OD\"U)@M`OWUO<03Y,]((YI]Z,89GUD%PL*T$Z*G6K46FY9(T#.\"-\n> MK<X-TX*11#H2X)+)A!`\"NAMQ$*7)JI4R%JVJ\"HXXNIY_2SA+Z9DW2]C^V.WK\n> M*(ABZ>@J[(*+.OZ9TB\\88,+@415_D69]C!A%FCTCAZRF35L5#2WY147*CKWJ\n> M__O]<G#F:`OGUG`7[%XO/J_J\"6E%4`([[[email protected]^P>MA$)`JK=7!Z4_\n> M?:N7>*O')KZ\\\".4Z4N=.(-1%O4)=K1.F^&\"^PS0A8_&QTHA7./\\N!?U$<AFB\n> M,M\\/TU/*&N&F)Y3A%(D-7O-)?)K(AJ82RIESJ\\S&`0RX0VX5VKCQ%.1=8@*Z\n> M':P'7YI5BG\\C\":LGO4K/DF`&D-'I!`%O/\"$/TORX?$>&]'@\\FH3R74LX8,X:\n> M7Q:3NZ4V?%DFD[NE-E'$Q^(6?W6A4?6<PC?OL7<[`LBN<\"%]2\"P5==G0>!\\Q\n> M$)QK`5Z<5Q=\\&W&N,<ZO\"%(422IQBWE@\"08[\\N,F7$UGI!,22\"3HNCR[$Q-D\n> M/L6X\\,2R_=SF_1&CD[?-O>*2VX>-B6UT0UR<';C[VAHWDGA,[B(RW!O!C3!2\n> MJYK(R,N9C\"J+GGEA*:<BIUIQ&ZU]8N29.[=,M&T\\-.)5L70V=[U/-.)1H`H:\n> M=U5!+I_K,@I,/)KDB$D'ZW,Y>@`!=#.>PJGS(E*G0S;;,J.'_`X.AL^C6[[I\n> MPI&:*,GR45<YI.$$T)ITP@GR\\>A(.,%;K<5YL/H.B>.TB)(\"^I/`2?9WZ/'U\n> M`_)S96-(4OI&I*AE1!H7%%>@]I:3XA\"BA+.*'Q7U.,8@<S+ZD3B$E`PRI\\4A\n> MQ!I3@43M+HX(*IF%G_VA'P/O6NR;:[\"4#M7#::N'-FKJ'52O<\\YY(0E[,7G4\n> MJF)@L%D(?J)L(;5=@&.>N5:;]9JY?6M2]H!;'JG64?*!>R`-`O4%F_7FBKTK\n> MK0_[(X>_)SPM&A?;KH2W^VX,'G.MB6$OX*2YD.R90W)UW+-?@\\Y6V#N%D*AM\n> M[S2ZUPP1[Z5[*YNS%<I+BU@1<BH:N&]E.TJ[=<6A/7:<;-3B^KW7RZE8T\"2#\n> MNM0<J:%%[.&\"KY\\*$^E38[$WM<$-F(N(FA:$YT+(1:BPDP$PL5CZ-((#;-]%\n> M+&GS=ILMJQFI#1MG&^-*ZU9J6#$MS0Y7'%MOL]+YX4_DL1G*N0FA])CKCP3>\n> MMFTH&=W;-D2T5@\"UFEZ#>RU>#0$DI-@LYALH$;RDTX%>?EBK(PB0CTVU@+;L\n> MP7SUD/J=Q>B`7-Q85A]%E<.$^.YQ`+\"XHY5<0WFO&4I#L_*O\\:<\\C?OY//[K\n> MQ*>SW[^@X+AG__?YZS>OAT4^B9AC\\@CVK\\U7J\"O@$0E-91R[FCQG3.81Z8?&\n> M)Y1J0WXN9E[2URQ]=8Y#_Q5RCL-*8!D(6[S<'FYP3-0KSS\\)E6_]H)7:*Q%=\n> M5<BJ&'D<MH2@X$RWPN_%8X`O1&9XJR#]>=0]VN0/9^C5_!B9\\IR4^?R31FG\"\n> MH)^.SLP)W*/V+-IU0/FM05AY&I_#\\BVSJKL)_,;M$0^@A64DD480`KT0)Q?F\n> M<*[email protected]=IKDJ6DDX_CJ>GH#QTKR9O(0Z<0()P0@R/(KPVE-)DU$Q(\"(\"F3^\n> MQ[6D+G^NF0U)4GGV\\ZS:'B3P!7I49@CV`B+J!9E47IH=97*NN-+&I4L42D<!\n> MI>JR*SD5*S]#?$@ZK0(;%JD4NL^05X4\\R5GH$![B.$UV!H^SHD>\\C.,[G=*]\n> M3=!$C+?K8Y/))D!P-=,*CMEN$XL^N?'U^=/G;_Z#\"W&4]A$RHC'8OH#HA35!\n> M3_V`Z^&H:X[UYHB4Q-O_]9W]**&-[_0C$GH6\\W$ZTA%(:`]^(_AON>\"_.77.\n> MGF>8KROKA,UR1.++F5%;S@1/B5ZR125-CDF:'!EH!(/$*O-;UD'#D!*(B$2U\n> MY'$H:`:J^N7@AZ9:KZ=B$2GDIO-\"I-<A:97]$'D1]A+71CFO>IR\\;E\\/PZAG\n> MMUTO8NMH*PX0Y\\GE8?OP>H-2)0[0B7N\"F/;1;7U@4T=?'Y+?2\"\"&VDM/V`II\n> M/![=8RND``2\\;2L8GG)RUU9PB$&!O\".^BPXPBD#@53!N-&NC2HX])@?&H/+Y\n> M;2=>UFU`2TZ.B&PI@T3SH[<XMMC\\+Z)_1Y_`,I'$GF7\">@[+MF[AVW.,%.-$\n> M8$LCS#A+[K6C27`UL]P?F<)_+8[O9VHQAXU.ND$%<*A.VLI4_/AIHKK&HV,;\n> M.658BE+16.[:QT8JGL39B=N8!-H)%(!'=G&9%Q&0Z^_8Q=\"'C0\"7=\"=!8\\=W\n> M!BUF]!EQIOB*4B,`?.<K/VT$^X)GS@V,G8KV8JCR$@_/%)I*S02Z>23I0I\"-\n> M@B2UV>IJKN%;>MUCYT86U4U4Q%E9\\AW^%*Z4;&1YPVZBD`+8*;W,O7-,VQF8\n> M.3#'F(FHUF0KK\"QDSE*N<'1O&77ZMQ>ADH&SK3_Y#&*5A!394+W(CVMY.Y!Y\n> M+J#1T+A`]ML@-^;WZ.),L4I\\@7\\Q>/KJF4VNP8RSC/0)Q4]R+@`90J7X.X91\n> MR.5(4G%3D`D\"B@28-U\\RB`Y2=46FIR$N\"P!Q?GNXV0:O*'587-)-#52<J%4H\n> M(128L%`BA2:I%@*NS%]$QV1DSZ(_44CL4NE('C989':5#_BB_B^1AZHB1LD6\n> MJLH1`&^@)'DP$^K]Q:I-:$1H!XMOL213\\J8V<E-KB0E-J^$^1*XS9Q\\>\"W0!\n> M?T.AA68Z(L-TX/YE?3W]^8<GKXR,_>JM$;L)09[FC%-Z,H7(.]_0!$:L!\\3]\n> MU[?:T/:A.YQM$Q[0K$KSNZ,Y-5_>WIWQB/3LJF?\\ZMG3Y]\\]^?8M,JV1G)!P\n> MG]\"E$A$,TJ7:31$$6M,AZGAIYV?9ZLT$>)T<4RA_T>V2CMA4L18JA7!9LSW,\n> MEA/O?%Y35K&5GB'4AE=&RM?C7),.RW<;9:P;)@7F($J]?BYZ]O[;[`XNW$H\"\n> M3BE?Y23O@36303QFGH(\"A5EZO%JM)\";ZPMR]NQN1DD@#\"%'M[2\"EG7O18%]N\n> M=]4,F=>(<$1H3AB65G.OS;I\\:]K$+8TIDB:)^5C5:'(>'6TSSH^TV>)$$O._\n> MB/Z-N[JCA'1'V1'=42ZZ(^#$!N\"&\".J5G<72I0)J;QD$_<_\\BW-Q_Y4QM__B\n> MTO+!]\"F#1?'M=/8><STWX[ADW=CZZO+\"6A*X;L1QSYM%@VQA=M\"DRF3;U'3/\n> M>278:V)_=2GV8RK^5YLP#-(^2>.[9K$,;)\\+VP%@(S9KB<UCS'`SXW.>@R1G\n> MKG\"K,7#FQ9(;64;ZHJ)2(P_MWL;Q!!UJ2\"#?7#8'UE4@^Z1Y%U&&*5:*3?<\\\n> M$J<670Q&C-^;97WI/+[^]H6?L!3[*SWSMO1\\<W4!!\"Y1BQ/*SX9M1@L9.#E,\n> M\\`I<2C;Z/$O[`\"Y!?QC@?^V6&+L(X9-JHEI5ZP46:;:<[J8SUG+L#SNF\\`Q(\n> M*]]@N=<VC1OE_DC+GK9_8#NU$D'>='XGZ!9FO#\"X,LRT@:7?B4A[(;3E<OIS\n> M<WEUZ;J\\#+J#+\"`%H#Q\"_N;Y>E[][!!56+,W;V;6M]N\"`&(9]]N*$UE<;:$3\n> MI@34#-Q8!4B%WL9!=/O0#\\GE?'<XCTL.FV:W\\SG#JPE,=!`K!UU;$K-W]Q/>\n> M@P<RN=0NKK9N0RPM6ADA.)V#M!L(2H<P$8#+$Y&0*5:R!4PI*^1!XED;AI8G\n> M3(,5Z2VWNPW1;07?#\"!X&-9*-J;`.JI/?9\"L`JD&=U,X5\"S4$LB0BVAP$3;C\n> M(?T0\"N3?*H_NO%(1CS9B/()DPKS.4AF=>B#UVWFW5YQ:<L$E1I9-A.'W^Z^>\n> M_5_'B*L=@#:)9'\\/N$ZVYD%7%'<X'%N=81ZI&S\\+\\PA=4`PNY\\7W`%ST64NZ\n> M1X8MD,#FPNR2+X?S@<7124=IU,O[I6W>+_%YO]96C&S$[',(V<-.#.=KA],%\n> M_MKCT\"W7!L!9S;;\"6=\"<M?PX,@-F+>MCN776GHM.$`[=-%6^V/F6>4!FYN#W\n> M':@@8COL>7CZ&D;T,>34_#_],1I,KPZ;RZG\"QQ-`](*5RPWY#:P]%.?*%UB6\n> MOJT(=CG\"$);V`MB2N\"C\\-%!&=O$&PAO<]C\"JS;BEF]/Y=,OV-5K%-@BO>BBQ\n> MHL50S_5\"92I(CPRWC\"3MU<YIHL=0`]#`:#AKOVUQKR`?`N`[1D(CY2N:3_9=\n> M`Z&49#%EGK0%R]I<`^M!4UVS^GB2B5#YA,)R&T5F\\6\"5Q\"5J-]>\\D#R*K[?F\n> M\\QT4\\QK:*-BS\"YOQ85C;H%+QR9YOX%@\"@QA5L5W>[!5VBTICK`!K(I)0V55*\n> MV#!N82$;C6`C87JS>2\\NH2H/R0QSV@UD5D#CC9I#?R^>:[/-^K\";[F$1N+9$\n> MG<:S>J_WVY2]0\"_$SP3CJ2*^!7;:.T3:0VM`L\"UPUJ3=Z`]/HT&1*O:*S`8P\n> M`%:<=+::@BKOU*)H`\\\\<S@>8^(W+(.;!&V@[W)L84374'%0,8$AC(#S\\8$X0\n> MKS&'K#F,\\#`,7W!?&\"'AH=<`;IQ]M:44`]5#LNMZB!(\\-2[JE3%UP;$^7^\\/\n> MU73^4)*=9%FDF%7[&_.72_`VFYWS.J(I#_-[\"C=J77T8R/=ZV`8?&.RN.*7:\n> M](*0`3PT+H834L[!&Q(C*Z4CG;8*R:IES9T\\+G=P)6D!EA)\\=4EFZ`9@/79!\n> MY\"1Q'C!_W6K/-Q/X9#Q\\XB/,IMLL]GP..$J<NP7+AT+E`4\")U%H90ZJ^UNE[\n> MJN4M3%[JYS,CRS:T5\"XIAD1&>,8I8\"<KI-2>W#Z(\\P(^*C\"C&`\\+^H.?=!OA\n> M$@LR:CA4/]-+FS21@\\JKF2FE]>->%.6/W(;/;5L+LUE7=K-Z+$62P>[Z'/<<\n> M\\`QYRD.]4@:7MF=_:O,+4<A5>G&34[=LX.$U&9@%W%5V(LG':?=B1&.&FS!5\n> M7'O,1%)`180@]]^RL]R/^\\/EX4O9=&-%;.3M$8]8V/DM+>-.?+W-<V&.-'U+\n> M*/S)V4\\4]7H]H'J8;?QRF(W:><W4F:[!I<`5+:(`(A@>-1T=DZ[?@L&[)?4-\n> MWW`CO>&X\\Z5Z:;\".O]/OB#+JXK,?>KJ<N2[3%F!\"Q+>=H4,SH*\\O(B;,%ES!\n> M7E'L_<=LA\"35)IQXJ@D7]#MV`)3S7$?NX'%B$-;QVQ-8M4Z@]0I?:I<BVZ<V\n> MZ^$'-;@45.TNJ]GD<CJO!&1LTD[F!K5>9OG$;ZN#LZ(O6KJ>VM:\\#$FEN`*%\n> M<2%OG6.\"C3HXFAR+@;W@I4EJQJH/O%G08:-NFJS:9L<BJ9)!OD;M(Q/9(_/M\n> M9CW?K']\\;6O.LHEZDA%>>F@%::&\"14#L8B^OVV#!%@0+YF#.7C\\CJ*](_++8\n> M-*).\"4L2N]0390E8L'%L_:0<+%BMX5YF3>U'[+@DIAG^Z@$/\\$$;GY;R<Z<1\n> M_9L<,7H5IQB]4C%ZI4$X#O8%9ZK2,!*W^'#\\\"!@XY.[2.&Q:)^ZSZ7$<CQ@$\n> MERA/]!/%#ET/W%HQ8)WP6:*.N19_WK4ZZ/N\\DUC8TO*88V@Z\\>!/.WN#\"F=(\n> M@VJ=/,\\#+%1V]F17S[P($4V#A-?+3L+\"2GL(R+)\"O1T$/RUJXZ>-$P]0Q#F$\n> M$VX9=E47/@VV-FO%81F0Y-;0OT382(F9$CV;32D<.9;!^OZ)FYEPJLMF/J^$\n> M81)N)TA.QZMA3S\\B(V#(<4\"!E0VTE;P<(60)W0?6(1H(1DIYP=\\3JSI<N!!S\n> M)><^*()$@(71:TO'TSJR*?DBP/[[5&[I\"&UD(W+\\G::@]N+[6.C592$K7)(=\n> MN#\"U$(2)K_0[5SL[GN2-R\"--(S)K$7V?`:Q<9KLZS+RQU&P:D8[9@6=:.\",!\n> M(5/[`'!5*%Y0&\"KZZ\"$3\"/<WJ-:(W.FKI17Z8$E,)J6_!TF-NS_L-DB-;64<\n> M\\-4P8O+9\"`-NHS:XC:)1P01H]^U7\"/AZZ?K;;^03,RS.M;-,Q1D[9NU,-2]>\n> M'C?Y2;Q'7A9MFQMXY3COL[FY(8MEUJI,K,SJ=\"71;<H2=+70@))N5[7FWPCF\n> MWW@2^&/0RLDLU:)>6?;IN.001[;GBU\"+)AU!^'G8$='91(&J2R9MHELA\\),X\n> M^)NBAN\"@1R'BL^\"8&E^CI-UV\"[_D6UX]*Y)8`?Y=]_#!TN><%>TIMSDQ4^3$\n> MC*P9WQ3[;KINMHK-)5>?H+8]A[;'+.WS-2/A4]9'N0I>K!'L,/76V><D2;LO\n> MF9M]=/+T/\"&+H:%7C>:JE,1C1#)6G'!0<0TU)CW,BZ9S@^[D<,EX_OWK9TQ2\n> M*[*FF0E]H3EH%IT<-%%2*!.%<F]T<=^\\:)T'<'L=PS<(5E)H]I)%U\\J]X(0L\n> M9,9F@R??<#\\\\^?:/SY!]QM3ZTW1E:\\R3,3'WKD+[1ZELN&R9Q2>CLM\\L7K2/\n> M*-(]\\3+J!&J@0>TY@J\\]/;15X57^888.IVTV#[*^/I+(80#G\"9M1\"=X$\\[:U\n> MNW^1G;H#UQ40AWDGZ\\%H'$?'%JZE-GV!,#KF<FM8R27P+H42VRX&V==HSA^`\n> MRWB`1^$S-=`E-0<BNFF1/9E3O7!<`E-U!O/4H]Z41LH1-%:,D+&EZ?CHIE1N\n> M6P9F%LYAE:7,Y[7'$U/O6WRR$7VSB/Y-NR;9[)YY'-:#/V[G\"FDC&+DA+P9=\n> MK?5KXUR%DERSI8!W.%F2!MS#KA#UIV=ZDAG_X\\NO/&/C,PGW]>]\"]`R>#6S:\n> M6\"@H8UL3<99-A\"\"82JTB(J`$I*\\`JHM9BS>=,P]WF-^9WM/)O4*^J&P\\.>.,\n> MWCC21_,SO2>G&*\\HY:-\":\\4X\\22JWUJ$T2_9.C*>'*$`SAPKBY&[Q=#0'NM%\n> M'B9Z;V7'B7R'>>A8@X@4+[S?V^\"DVBJ17*0-SZ<BE`V:F!1):];--G.8^\"-Q\n> M3'[C>@0QT+`A_/?2=P7T9)?(\"2_LP5YF+3[!`J>'\\)RMC1FYL+FVHSGIMR6H\n> M>,U861`VO-EQ6'10KE*WQ9:0)N+2]!5E8!3C\"<!4W*(M.3TCPP,I38FL.[]F\n> M,VA(VMEYJNTC/(21#=]0I\"HA-$>#%]_17NRY'-D2I0`M4@IG@:7(J',FB%GT\n> M_;T[VS0>C\\I^(U[9STQBU\"[&0D)):B\"(KB0H\"I1:%?ZA;K]E<VK=*&Q$&R,7\n> MLZ^_LPGN9`Y*#>>Q<Q!==+0B[.Z>!OZH@0\"-;FK6AO'$L6`2EB@J9QQ0JW5F\n> MP\\,XYT`*\"M$9M6+-5%OKKE*59&M/#8>8Q$`/YSP!%(&SJDC(.>RF,VNDX4@^\n> M5=4_I&JLAEKRJ^>(_9];@512+5:'F006PT..9)1.+K-0!I?/)5@$OCD=%:*$\n> M5HKW\"4=L`-$$MX.//L>]@P..Z$W6F]TERZ0^J'?J*?/SLR`SR5),4>+\\Q>Z<\n> MTI6.MZK$A(7.H+\"KZ(2=\\]*S,\\T;C;+FVFHUTX1R?\"M@QPH'H=>IC.)/H?*4\n> M@5J\\JIPH/54U`)$B`6^RAC92HSZ4?E2Z!(WE:\"J*]K>9-EC;A5X]\\KOT2)2V\n> M,\"Y=!1DAW*`7JNGT+#VRK;*QXLU4GDAT4=7,,&QWFUE5\":X/BP?*!XB)4@>M\n> ML,`J=-1B/%D>-9XDMQA/)%,(7>.^%>4]#9.W@<N9'J1,1_#85N&($X+&T$V7\n> MP>%XQSPB00-JF\"C;V1KO$\"^\\,TSD@,U#DB0$\\;CDM?/G@!&:%)1Y79-O9,@9\n> MRV0G9[+S[/*BFI,1&MO?LG5E.K9A@4L%!;`FC\"B(VMV_-Z>=?*J6BI3,),E+\n> M8%]I*U:X@,E^R9[6ZK<]M&&;S>\"IIJ1*2G\"T[$,.W0.C`'/@UN54$N5.]PRW\n> M?KVD=>^@.$OD()2].%01GX,E@S=K#Q[#%L8P6QU'KU87E`!9P%\";,D4&[27/\n> MKFSR[\"W+2]23:*$V5G\"KNVIVM?.`#+#YI`D8'LD5;Y237RW?D4(&!;P7P]Q6\n> M.VH5<9=F2=YY7<2*K2NFEWV3;R7M$,>@#V.!O!O3&#G3+-B\"(!TL&6X!TRCX\n> MQ-XTTK;]D^F\\N*M(FH/:PUI8.N7J-)`0>=GF;/>-&M;YS6P:7?8/P3:]WI$;\n> MHZ8@$*J#T:+?;)SPABU[CEWS#1=EMC,GGZMVX`A%76$5Q)BFB/FNQ6ZZ7<)'\n> MH%;:@]C%>HJ0Q*`_8)'1!;9Y/!TNO1#[_O.O=_!2.F811TE3R,=O5=6\"$+EH\n> M:0^G(5\\@<T3'$KU@HPH'16/.*\\K9S!NU<D<6F[7'I3<W_XOHWT^1OCS)/3O+\n> M$_''W6W,]%X.'9Y9V\"=5;+O=O/1)\"8\"F]@>!6ZY\\O_W&EV50DO4HF;\\Q9K(X\n> MX(1:+=.F>89I&7JHXGOQ5W8X3D13GW%7?0Y3@1OV>NC@KJZQYV:3A^#&S_[O\n> MLZ=ZL%2/]8R[T0E&(#7.;KK>TU9RS)WWV;)W[`Q?9L?^CH-CQ?2QA1V+$N::\n> M\"VW''AMO$28#QQ7BQ<48$A$JV,5N\"HOM;D/&\"7$^`7U4\\F2-$4.;17>IKO@,\n> M^FOC@O@:2,]R!#0!E:/Q(#E(6K7Z(C_5+#@\"FZ*J;JD=8]C3Q*](/4R\\F@)D\n> M0-B\\)NF9G\\R@J?9ZYAPGKEFVZ=S9$*>%@F\\*@_L=IIPV1HL`.X![QH@S7`-[\n> M2K,[KH)_LYVDR!D1!+<R%A_&H1:1'UQ6).LW^\\NA8AI;ZMH*T^><XTL?-<-F\n> M>7'16M;SUD4W`K\\-_HH66KN;.%F`QCA9&=FY_(Y9/F'I)?J0>`D2?6>KZ4ZS\n> M!C?$<M(]:7BG`#:0K5'4ES%[3TP/%DFG;:@%^?SZV9NG?Q@N/,%8(\"':J88]\n> M*9^UESP?4HWT9>'E^;1)*$149VM3B1!C8?MXGBR+N(QZ/(@ZQ([YO8IO2<OF\n> M18[->\\@\\0XL#S.A/_!=T],]__/;;LC0\\89AV*Z'0GHC^]?.Z3HB@OQL\\79);\n> MYHYMYPF[ES/^[%\\'+TUK9B<O0\"20KR4R=S6]5'2&Q>\"/YFI\\P%SUGO/6H50C\n> M3'4\\(KT`;[$?^#-QI208`8LKU(ASTTQZ`QI/IKX].^[YUZ<<9X&]N:APLU7U\n> M%?UBK(2M&?W*WHN\"1DBLK9>6@MG&EV9XCU1=0\\RHX8Z$&#^GD^K'=PIGNV+_\n> M/;MG/!%@&\"`MJ(.@V0BBM*@-YW6U8PP_9RE<\"#+T08O;I./<KW)\\ICG'WXL_\n> ML-D9Q)((N'ZU8_2=VA>^\\7X]7=F`Y)=8R8IOZ_T2:3$CVGK(6*H33_\"PN!E>\n> M$B/JKSVKKRA0U.Z!I2)TO,0+Y+687E[2AE@/D!F9*FG6,K*E)UNI@QI[%\"=@\n> MRIL!Y;0=`WZ'&>?7RPUT[W]HB/UA0DRK1D<-?2/KFNW<.!4[H3?4F8\"L(8YY\n> M;0C33&]JI^HS_7_Q^LTWKYZ]]MG2AA#AS?F1M-/,(SY$K%BE=.`1^P3#0&G$\n> MY[5`8'AZ;#B+?/7[[U[CE.^KZ0YJ%T,-#QL**Z-/KQCF*5H(-NCJDCD>S9)\"\n> M*`[D0GXA#LMT6YB_2.-P61XH(!MW,\\CX!>K0+,A<:7INATK9)DDAL(A4%Z$Z\n> M(6_Z-%>##[%!:NYK]J20J(EX@ENB@E\\DKQA4=C;/\"S'3\\RMVV]\\?KN;$KI`J\n> MXFI-?J$$):5_K[R_-Y)JT=PUJ)GGC1C'2S8Y$6HI#9O[@+2D7*,BP\\R;G54/\n> M\\,[9;6IS28)*?&=&.ZT0\"_;ZL%E37N;W@DIKU8U_7+LDX_OF<*,\\UE/#6YB1\n> MKILI#>/WU>Y]]4@<%$?0%*Y8645^D=#7O)`5((,Q;7!O&:R[*-V,E/H67\"P\\\n> MB3TECR?5B(9T@4K^_8_/6\"B+$@1%0I5(9R>>3\"B\"[\\EZOE,URW_PS7'%3#A/\n> M[O_>K(\"=^11.@Y&ZZ=L;;T5@/I@]SK@V:]C_J2*8'VM8L,.!)BS&EGQKYVU#\n> M6#T9683%)F^8!,5\\\\?RF68/WURNS:);OQ?5-FJNA^'/3KW?0R&B0P#O6CC#9\n> MHSVMF5\\3)+!]`_V=1'Z;N3;[<XH3+;82@=11\\P9`%B\\E8H`)I\\SR4$\"0'=0'\n> M819BJH#K>&#A`S.08#.:$S1)5`<_L]G\\H-/DUB7H;JEG;F_6K+:X.`VCWZBK\n> M2(,\\?RY/A7J+V>O.\\TEGC22V(GSVU=@A7A'*#2-\":;V@(WVH]OQ(QQU?+Q:\"\n> MS\\NKX)//`.2\"YSS`FA3?++,'\"^D*$A&S\"\"26%*8A.@_BG:1:':YYK3N7ME_D\n> M[S^92T>J1*64LQ<:726)7\"5K=N9^(>I38D(K1`PXS-^OZ9*CNYEO>K$X>-<,\n> MWS*\\K2B2FB-`\"`>?IL-L<*RDH>ZP8@0D5,-??/TTE`J>_.[=-+731KP=O**;\n> M@^R_-(5C9%20,`X1D#4>0\\\\5[V+9[.8OUF&KYKKV=.4W5JE@!1%A<AV3[K2=\n> M%+O)4287S8I(('<&21M,H7=\"*:$B02@G\\7$OUCX\"L[KJ,G<F=ZXYBCN+MC7O\n> M:RC(_\"S1UD,%UC$L\"IKT_;%9*RV*TY+!7+>#Z<7^`,YD)@F-PZAN%GF??/5F\n> MST@,3\\0+T]ICK-'`T8G(ZPX.@BB(7%>L1M(#@J9YH\\QJ1!I95?9.%:TB(_!H\n> M`;)M><4%#A.KGZ`<LM[FR7DN,0V;2XORU>';U@/FQDR/*+45F(\"OA]%*-OQZ\n> MH*KYR9@M/L2P0'S1R`CQ;IZ0^]X?V.32J!:/\\U_[B@>OXP$C@=6..-6C*&/C\n> MR;A'&4L'-M&Y%Q*[4!*K&0)!6H=+M@^I([E#\"6$E)2.@MX*7GT@8(0N'JV9M\n> MN65H]FIK@\"2[X';_5S(LF6(6>/_[/[)R`46EI?X`_\"<*#]R8@_V%Z!PYSLIP\n> M@<W!+.05CMC4,UV2W]/:B)X[B3A@H?Y=Y\"XYYSA.Z:IB:C_'A9L4;44>N1M%\n> M]&\\WWW)B%7ELU)38DLV`_%\"^86.\\7.IF^29Y(=J\\,70V8;R^11R00P$.A\\%4\n> M+@Z[JGK(N)LPYN#JWAV0/RHT+[(V*.70T^Y4DG(>^J^I!1FWFH:E\\,82C\\P;\n> MD;-?;9?,S--Z-Y1UEGGOV6;%V[AA:Y]<.]5Z3TM\"IGN^L816*#;\"CNA2'2B@\n> M6@O&Z#&R;CPD0:MI[<,_0%M$&@[TG>9H.]WO23-(GDRSW<WV@-C`.0L@YC;X\n> MFG2%<1:GA$P9L;;A_V!/752[#5&G'[(O?LBI^2L*N3ZH%IW&DT-T(W4)=N6%\n> MN!%8#D2&7J.L$1C^9D/2A@M_.%&1]:V]'*OE%5C4Z25%[\"F;L-U5OHF8ZXR8\n> M`/,4C;.^5>_93UV:'VD?O>K*WG#^H]7YN#&U:($#PIV>3=(VHD?47Z,1.%R*\n> M0*C-:J;>M\"MFYJY>\\BS&H_08S8\"\"W\"EE:\\WSUU94<R:WI]S!.$XFMQ#3%U_]\n> M_JEJ\"XC.22>2I`^H1#KQO_O*D*($ML;^,F]FJR_>O.\\IE<63XZ5>5KN53X7/\n> M/0?E\",Q<RLS<GY;@X,2-`/OVFXJ.78=7X]R^DW%'(P!%E]Q?H8)&XKOVFZO=\n> MK`(*Y1Z,<T0<ER(Y_50`(FOA:Z>%,V?'\"D,B#!T'=BSD:G1DG(O37WW8/O[B\n> M\"_/O^58X_K^NSC>[Q1?;JXLO5.]92!(+HD041B&8G_O-4``80FF,S+.7U99S\n> MUM)>09,E)?;R!P[-]!2H\"KO!\\H\".7%]?VXXLN\"-6,Q/;3G!)BSF_VMB338IN\n> MQ*`WN]UF)UDL3-M($<;</K&H7?-$B+'M,]%`,Q,T]IYTP4N2^E=SWB')2/UC\n> MGVZV-PRW4ED=4=39$6=)/(H[.P)[\"3N\"U?-F*5;-C`C5!D;HB%3P]>':S,`Y\n> M0ZX]/WB%9MJTTVQS1ZWR(%+M@?2-M`?B1!LCF)Y5\"51:N?O+J>2O45]VZ/?-\n> MW2\"X^TY&,ETUI%L<IZO=90N!4$4TG9[(0UIK%/4[21!''R+8ZO;W=[_`>$<^\n> MG`4<NBN.@GKZXN5_O'K^S1_><#2-;6G136S`FO&@A4HH'20/[A=\\_SS]KR1]\n> MCI(L3C2)@T^#50G?2P^27''[WBPM'XN458O6D:HL@]^PU!(%HB:9ZG#^+ZOI\n> M6KAV]2=:2GUF#RVFP;!(<(\\&-Q>;N:1UR)$1FK/Y.LE+$!VJ-4G;)(Y+5X@K\n> MN((O[OXAB7&>/(XUIYOB<C,/8G*\"('^6TY'XYXT$Y20%70\"2$7S68=B1,%ZX\n> M?0]FHK(,T'`IF6$H/I^=\"JP24/3N4($A_1#[$WB.3&#I-)F%+CFG9%8M!^D[\n> M&DG@LA*;KZ+0NZ\"\"\".*!P#B+*(P--_-T5&8\\ALI?&N$#[90CBJE_;>6X[Z2=\n> MY>!;:H>1O))2LW62_X=A\":I_V[I]8,@E(QOH^O(G:C\"F\\I-$RE,K^W;AVA5^\n> M\\?J133T\\HV%!`V.K*:6:\"_/N:!<:JQ*Z6N\\E-]%<H1^TKG3$_-L.*-1[[)-V\n> MC3/U>WHG6FU/1H5?&F=40&C$=W9],&]R8.9&AM257I'\"FB?49K8V-Q!A<NVZ\n> M4W)FI\\1)3(UGSJBCH\">,MZ9(R_/-K#L]9W9ZQ+)GOKH\"_*,($K+350!1,-/M\n> M]#`STO@MBR9?6$UUIV,B&)JYYAOR>.>6\\@6)ND8T1)7!=FK9]&@'1_1O-YUS\n> M4II;:MSCXI^(B[\\GV$'L^QJQ1Y08A)PZFL$?7S][]>`UC`449;,8O/C:_(.P\n> M(U;FJ9\\UG)4:=_A\\-=3\"GES5$E!@VM$M-W>K%\"0<1%0Z&?G-!+,&!9!E:_BQ\n> MD)*\"FSSSZWY$6F3S'+1A*M6S`L>,QMZ?Q'`@.F`DWL!N/'A=Q(902/MG>2$L\n> MI0=2\"HI.'F[,/[#'J*D$Y+-&\"P+6-S*23J$*RHQYVC](['&I<62<9:US@\\'C\n> MOA#VY??L,>DRROAV\\(\",PTKN:887$:_1HF(;$U_.^P/S6\\U!#!:LH)H9;FO]\n> MP)0#T@6!O#3JOA1I*;'`G`'T*VP;M'UI,P@XVR0*2J1,S5Z?\\RNG63)LP52P\n> M\\\\PP#>,UEP\"C)H#K5U:\"D;L8\"U[M[GX6/>%0%A)U\\^;)M]\\R?R)&9LN@:\"JZ\n> M%F-\":D'Q;4-T@^?21K)_13I83KAHZV#;-P\"QC7@/;C)(@!1,$XN([`1![GTD\n> M:Y,]&[XF2':YA)[email protected]/T]9+VT[EDU'I]F$*MSZ?^#88DV^ER\n> MNC^(:ZD1&.-(%=GO/!Y\\8179K7XBRQ6#90:ZY:BKXUJV=5QM+H<!_`&Z?DDA\n> MO&:K*0B1ZAJT\\Q`_?/49YV'858+MI$#<+&.K;UFEBKR&KU\\[=B8!X\"$:#1^*\n> M&61U3Q-'8#::48Y6=RB>2((X1\"+/.2GDX+50#;36@3]X[.C]7IB6:BW),00C\n> M6>]YANMX\\_3E%\\\\IEA6,\"/<&^:_Y`C8\\].;2:@;7U4PY.QVQSK[,6-OGW3DQ\n> MV9IJKZ:A=?@.)HG36VNXUWX[A84.<7YJ_KX.KNA%Y`V:T?FX0A(MK299PP3I\n> MX#7K*WA5\\`R)J8F5*)HA=.$/638\\9U)QCGATH;ENU%XOM-=*#FKU=KZ>$KFF\n> M+(@<?0:'^9^GLP-PJ#B.S9LA$;<XG8F=:S@N9/#978KT8X>+6]UG7Z13:(TT\n> MD5=K4LTAO`C(HA'O+H^!$XE.3/][<]=MEP@DD&B;\"6[!N:%,^^548+(NJ\\L-\n> M?+X.=\"M;A9ZH)`,=9N6K,!>^\"G,9M95^*=U,;^W&@NU8NH'L?!#7YIY;?-1V\n> MBEE;PT9`%P1^>TUX;NR%CIE7'Z#HJ\"_*TKJ@,&I;,E%8'5)M]GNC-,X;I6(!\n> MN>.-(J:.F)&'-4(+1,%C:G`^EKU[;AEYGFB=$'@Q.3P>4AAE(FDN=X-_9B7/\n> M>K,$4J=W\"`ES?H&&JMTY!U8D7_X+>3C^BQC:E1U@<L;<@+CNH]>+%NU;!)R+\n> M6:@+PX535E9.'\\QVWJ&/+T&7%=+<#K@#O%^<@L&#U]![3[0)=E(H&1-<N9\":\n> MRCI4P-AF^CHC]<B.LQ2+O=\\4@]MDPZ.I6!%B.K!PCJ^A\\B$\"B)*=*`]+U7\"K\n> MDNZ1$)G,Y'MY7A)@P82W`_U-N]XHFL\"58-3I!5%[U'<O9C5L58<[0F)[M<.=\n> MQI$OJGF2:VD6^12C=JOH#,8VLHG)@!A9EZHE$K;`1MM\"P;-GZCXSTA)?&L[%\n> M+<D(WL6ZN-5ZHMFH>+5CFZL&[email protected]$^WVW6B]6-Z+?-H1+'!V)X%=9M=V7W\n> MU5GW,DGR1\"X3G5SP4E,+8!>`NV'NV$\"Q8]JZVVP.P;(68X_QD9P%BMIH>1^B\n> M>]]7Y$'PE2X,FZ_'L8=N,.OM-99D/C4T=3T44Y&Z:OG9`!<>(NG0)0#P'<DT\n> MXW0K4I$=N(SL.H*.BV]XGT1,%/3GGUGM*)`+%U25$6_,`V=5T6P,O'^`KRE4\n> M(.B:YW`@>Q#'1JJ2N<URX;H0>L;S/S@&2F-1-R)N$(9B0%<1ST]6:4;'6%;#\n> MP-]A.C=\\\"/'8DO<=QS[23D,E)`%A%R[.57)ZRQEXRYT\"%*BW@^;>,BLFECN(\n> M[WJ9DCGIOT2$6D8MH)/&/])P4>']ISH'1U;\\8.S$0ZI==L=/!)50LD6TLH[=\n> M?)`T39ZFOJ-D9W`3E;@14>PJ%AEI97DK2?2)Q$\":D?D8/LY?=JD!CI)'K5*G\n> M!B]@<Z*\\-Z12;JH.ZO`M>NRA4$?62JY7@<)[P1/3\"U6`!RH[?0EKYD6@T8$6\n> MM$9-/>+YV^(69QP3O/72?NR/),N9F/]%]&_98P;/9+1_($SS/VF8(-G!EZ0(\n> M&09V\\\"@>T8Q`RS)Q<2UG0H8DV\\I3X1V)\"FF^CNA\"43J()EEJQ`DE%5)63(5^\n> MZEF%6>'TH)%FUQ-NU-N;GC?\\Y13IVRG?7^TA5C#KRGBJ(7/+\"X:^3`+8V(6+\n> M4#RTPM(B!;VQ@)`:2VS875Z@2J#P'X(':'G?4QL/&9]\\SJRPL)T(,$:^>_+@\n> M81S%Q.(H+MB[R:>O%&?J0=N(%G[.,\\3@;<+ROV5UDH#7>+O*GGJ-1D!D\"&)G\n> M;3O@KF\"*DZK!`BBV*T^K9+QQOK13B?\\DNC^%-N@=IT;?+RN<)>:3<>'#P*=G\n> MJ+;N)TJFL&W33%P:W_5(*$P8V-LXXIO;C,JL2F-:FP\\%8:PG7M[C39.\\C,;9\n> M)+AWT`U[YR#DH<5XEGKEU+KYIJU])AMHZKFO#)TKZYEW#PER:7)V8:<@'EOD\n> M2XZO<'!/<.`A=?0DX6M4'7GF@<O.PG/9<=\"7Y)V#$\\.RH-R3W5Y;;4!PYI8J\n> MT@OH.^]:5D0AG$G=D(.5ZBP+]*(\\Q<P>0:G!#-B^6M5#`9&V*[;G58I'G&5D\n> M-_A3M<)EFTS4*,XWCD>XQ+A._/=/.-;P:S)G:=V8*]A<7S$GNB<-'GOV+C0T\n> M@,=0-[\"0>P8^5EM'GFVQ]@V0J#&!NYO7CS0;(WT]P1F3^[NYOM[^KZ&Z!IHS\n> ML1V*$\\]^11Y#2W>H3%%-:(^B\"U/TK[I\"?[TRW\"HJSW+[!54.F64G76O6X(8B\n> MEKI)AKML9IL5B`;+S#\\#LG2H?M=`\"&,Z\\!\\0@B3WBPC+#%;.J]N63'2;//:X\n> M-5JU`N%N_.9W7[*CPUA!Q16'DNYMEKIX?U:-Z$-(W06.W;K6>Z\"UK<!!V@<Z\n> M:&N:I@W^X#</2-6P(XP<SNS*$`A6JD.GRI$X4/)*S-U*R!%]-WA\"*2\\\\#T!.\n> MXM#VJUBX2KSEA!G%PP5C:#\\E4TM:0;#:M,$?$!'<4(X,.BO_ZX$RJXS(!O3N\n> M/[\"J^1;'1/:*C3P?\\/Z.H6XQ_XU:;C\\DI[-JZW_Q:66;)(LR('^G#3Y:*&1N\n> M-\\8(O!'LIYX6A?UUS'%J^TEI?_[*_:%SL)<4,'I_:*0-`:=R\\M.>\\O,CX^D#\n> M\"!%\\AQ8ZB#C%2WB4%Z*]T.,\"S9#C,>%'6!X;T/P0]LA+PZA@E51!UO$AM16\\\n> MX3UV]Q*IALQYD$5)$;?=L[3>AF`A+'(ST44XHPM(,^5Z1VX00SSWFD1Q/EP&\n> M,I2#;;<AITO/&,'&ZZA5NUW$<7)LR*M[C1@NI$.;L!SWU49#7R(72@0[>\"<Y\n> MC[8IY,].\"G7;7I$R*:G<WY8<[5F-V-H=P<VK80>;;>6EF0=Y)'?&9`+)PBMA\n> M@1C;/=\")2T=QVRU0![$9OO?R2AU9VBVGF4I2'@R9)/><%F$*)96&ZA@I;7LE\n> M^!2-5N23+',U\"EP'BSJ<;]&).NQO)4\"RFXZT$ZFTPZ!UI>14]2M(G=36V,-*\n> M0':B9M=H7PD#\\/A97I)N\"+J'10UU)8-UQC(7@M/H&<\\[Z'9\"AF+/1E&W8F4#\n> MM7%%;(#&8?F)$1G\\=5\\%ETD'49_]JT3^%-:+LULN24_P%?OU!JDM985\"X3$E\n> MU_<(VIB.H3T=&2YG=,30GG\\R0[N?JIHB)&]/BD.[QLI?%CWG(.?>P\\T2B\"C(\n> M)K?B/2J2-(($,K$[+'I@S`7JD=:%HDM]R,<N'FP%/PEHNJ),@3I[!0[2A0&M\n> MS#*Z7M:F.KJ3S?6XW(FI\"4!F+28WNH/+/<KD3D91P3#GKA.9H6V<;ODT%M<3\n> M'*EHF<3'65Q\\4,9]'&[DL[@G<[@$03A!SBF/P;TG?QOY*Q6/H&6V#*ZGA-1-\n> M:.'>`..TWG#\"H4/VD\"N($U?!(T)<W$NTUT]D.YKN-!TE[9PX+H./36NLQ@Z^\n> M)8Q,5)VD9\"E2.D;E$23B]]:\"J`JP9[NS\\,BC#L%+WS1F1/,(E:?Y).S*3PQV\n> MR5B=\"CSJHXZB!QG0=H1>+@:&\\RC'=`ZYQX0O>5O_HI,ZB*IR4B?W]8^`:I.'\n> MDC[^X0,D$7Y@YRPG!,JP?Z7VSX@S>?FIYH_R`1R;O_2A))5_^(`3%[OY&\\/9\n> M..C?Q,T?X*X^S?PQ7-61^<L>/OC]:OJ^,AU\\94B4F[Y)WN[>9&2G+QG%0?<D\n> M'9'<Z)23+[&2H%/D;:$&GJJR',[!GC^UEWIYR:/DBU;=8%\"?)N(S9+\"!<H.-\n> MX,J@6OXTP)>Y$/B]N%3,E+I'O!E&?$-MCG/,'DOE['1(\"\"[:=/7]@\"+=\"_!@\n> MO1@1\"TDOI-FQ./TZ4V6;:EVCQ,P@UO/I;GZ.'\"JD^TEB%P6P8(-:LSH,)=L0\n> MV\\PDRQ'2P*FK.!@K)LWSRK2K``Y$WJ%NO;@Y<(X_CJN,_+2JUA563>D,,.R#\n> M6W\"V45$>!H$R0IO9D(?;L,O5NX7U(5'JR\"V%$8J6[/FD\"C0U.HKZSDFC0P&F\n> MX+2X6%<537M1+Z9B`$HRW+#4>>VXNTZJR(D;;#CLQO`6Z=BIU`7,[I55'S-_\n> MJ\\CFS>!K0/83?@$S++ZASW!KZJNH,#*FH*]>]R,V>3Z4=YI;SL5:6GS#N5-B\n> M>KI+MN*REP/\\AACWPPR3.S,AO5=+$XXX4Q%`P8$KN!P;&E4^5\"OB(@KTPF1$\n> MM[XP<[G3*TG6B>F%RM[3`\\%T313#?/#@-P\\TL0[+E\"&KZ7)OL_=;+:EBAXHX\n> M%GE$R%JD>`1B>?\"8RD:3XE0>4]GHQF)NDOL\"Q!.G=Y]P'VRJNR#KA9?HHV/+\n> M1$(O(J(E>TDV[+HG.#N>F0V#X!Z`%82=@57W^\\U*N&2.@D44G21[`7&^5CS>\n> MEM='U,?I&M$O%Z6MH_D3N+=QOJG!]GP+;H=NAY35I]Z-@XBBX$X%CH8IO8)1\n> M#?$\"6_97V4H=XU8=D@]OL#\\WE]WO]A7]1PXX>IBF%.AK\"R0CFCWJ%CXV_^$>\n> MN^\\G00/F>_+(.J>;\\W?'4OFDL?E?1/^.>DQRN9CD7K_X[IF$KWU-5-FQV!\"\"\n> MC>2$/[Y^^>SI\\R?D%?2US6CV1P*CD\"RH9L&_>B9FPB?/20<MT'0IR5+6A&>.\n> M%68>KR#G<((:^LT(;IR4AK2X#*KVO;F'Z:^,:_:6X).HV[O--1E?Z2\\\"+N8I\n> M>Z,QQ^FU@)(IPBT\\;-BT+$!99`0F/?:<A0FK0D&.I-U8HO%5JTP+P_\">'A!2\n> M%Q:+4O\\0W!8@U=U1A6\\5[32VJJPW0SVU$@S'NJ:W6NWEQKM)R`RH^[>.K(NT\n> MUEU;V8`=`)T+./)YFYF<C$;1;>?FU\\.EXMCXIP.K-2E&_4>`W5>JGZ$OHU;B\n> M42A[Y\".*EGFKS7\"3)+#8MG#R2.,@Y4-Q)$^I/#<G_97C1M#1.(%(@TM%8\\@F\n> MYLTOX#Y_@2`#DT$,&<4F3?I7NRWC!\"+'V\\&(-\"!NV\\7IN(S:1H:L''OPZ'08\n> M!`LOR(3+\"=PNMQ3G+!&FS-B$^,FR#4C$M[=!R]W4+4/DZ38]XLTQGN0/[ZAX\n> M96D9:#!;N]GR)IG3Z,:#37BFA%K1YN?\"L:P5#@N8`5@X305C@5NM=P??F)=J\n> M3!BG%HHJ9_=\\@\"O@Z+#C=MLM_Q&=TP>F!L''5I2%IHLJ8D;>1`JNP#:63-?$\n> M?.NY5,-$&6!(>0DU?+@0-=\"SE/3JJ^]>[X%<Z7E:6\\\\9%]I7\"T0TVT]&XX!=\n> MPRC\\1$9>]O:E)&]?2(YLCD1S+.Z>(M9=[O-:4I_#*>!RLU90I(=B=X<5'S`&\n> M%#4Z*[email protected]($[NI&E]O:6MQA)4\\=Q+AV^5.1CQH;(C9RM$U%5M-?9`?D=M@XO\n> M6LW`PI$P_1Q6FA0>SP]\\/<GYN0?W<D$\"Q8%7V<N*13O4@>9>\\`F:;7;;S4[U\n> MM[SI6KJP,+$6QPMP/R@`VGG(-N(J<P6&G?27UU/6V>(<*\"0<^/]();O:`6E8\n> MCI;WC-BX$$'0@2=HUH9H-0>*,!1+S<@WA<%^!9\\J3(_G#!2YT'FQT,1MF]/1\n> MLB+IP`\"3=ZP<,*M<J>+;B`DT.7P\\Z:!NQ$2J`\"0/!67&@6NQ<\"?H:I>;>;4C\n> MM9@`X[SSSTQU?%E5U9T4X'O:J#4:-SC*>L+=\";B60EM(FL;8W_;=X=8?7Q1G\n> M;-;'_1@G6A'$-S@LDRTD;ILI86W9-8L%:V[)^)$4[?7SX31PNP<NXF*IL.Y-\n> MN8:7//>V!@(!K0DB^$OF2]BR*+1E*W#;5@I4IM;<,QPP13ED&@WK4,`WID0.\n> M4/6=/05[[@.[OKV^-$P(XI1F#(GY]%__561N/^\";J6LMG`=4R<Q319QQ@\"25\n> MV6H*A_VEL]2*Q!&F#@4ED@3?*,-S2P'<G@5\"CI.SB&I4O\"8D$=0AKQJQ0[<8\n> MY\\3\\+Z)_XZXY(OF'Q/V%,J++O?IU0]HIE8D\"SS`O>1CIRC`^`H.#_9#VFHP5\n> M\\D0RZ@A(T457B3QC.+ZW`RY5MG3&K#*.>G3&9TAMYK.#(O(+`Z2*Z(ASCGG?\n> M35>'AM`)Y2M5\\2%KIJ]4+N)`9XNP/XT\\5UY\\W)DF3ALG)LXJ\\K(*ZGQ!U3\";\n> M;LT)$V<A\\>V4W;47[GUI\\]M$`S$V,[MJY]NZ'Z!O\"$,W!(_PS#PS:ZL:=L5`\n> M*G9B(F`\"/3#H@XLPYS_6+K[*^I#3%]'2==_L^W-,'P`LNYQ]=\\&E)-2M*,FP\n> MD[[N7]IOJ?^Q*(P/Z7T,-$)OYMZ*LE?TMHSCZ)L*1I.XO:QQ/!X[+V#K.I*J\n> M]^P8^$]\"%%IK9YM6@%J%QO!R$RE]B(*%>RNIGT7MHK9<S=RLRHHX,=+'37C&\n> M*`ZU%BV,=55W,RS>.`M%!K*):B-&C9.,*[[6C`Q-Y)?L$B%XK`.;]S054GN)\n> MVZ8\"G1[:\\Z*`C[S#*>:AM$S]A2QRY$.1H_E0Q\"S26LO\"QUF6GAVS*Z@U`:G4\n> MS($W_T>/\"?X3#1X\\N2(.=NI90^#='JCSXZ&U-N1%$K6W2,'>E$A.+7)\"B<@'\n> MJ-)L@H/0J]LNN,NM[6\\@;`,OJ)=6#6DFG?8RC&@-.$XOA,PPD$QVX[%Z$Z)9\n> M\\Y-\\G8B(QAG-Z\"\\\\L0&I_`5\"O]#$POP]+G/2'3UZU$KRZ_\\PG:!_\\?D$GG7?\n> M-NN_42N2,#C!*G!SOU!'Z`8;CZ0$\\L=\\LYO^S?;+E,BE1.1*I#F72$9PF?VA\n> M6:THPB[W:^TI9EY),6B`7D]7?[NXVBUL;]!FW%,P2T8<M)$PR+TL5$E_:TUC\n> M=-LTUN9G@6`/H67H38K4-\"?,JR@'-J*[*LCQ/842BC<Z=\\AT.G8#:<^U>2'9\n> MF]$1.0*:OT<V\\VLQJ8@*#H:6A0>@KTF)(IL].$#\\L2$%%U6PC5G+()&84Q><\n> MN84%6N)B@)MRFBX(FQF6-D8ZZ5$\">;?ZEZ9@-AKQ42<D$C/IP>*Q=\\+&\\S7_\n> MA41`K0'%QMF=9X`7B#\\WDD*P/Z/N_N3%*4>R/[E8T=J?4<_^]`K:_<GNDFUE\n> M)53%\"\"3R;:6],\\H[^S>BSAZE]YPCWN\"1V^!<3^'E!K]MTN27['\"V'$LZ^U-V\n> M..8B<8<WV.%<5\\G*W-C.C]A_U9=(IR<$Y$@).22B?[O)Z?^!&FT_J3U9/FR>\n> M!%_EZRP??.G0\\BD^EYITS,%I)+S(UP7)ZN\\9RMDWE*PV,Q>H`M!@GZQQQECJ\n> M58#H%9N5(!2E@XABZ([ZT5&JP@C:]W%QG_/^:U;10SUP]WF/Z;Q#BS^Y[W&'\n> MAO_N&T]..WU=P*!S^X7'6]1<>\"@`Z\\WM]QT*T'U'!<:G7G>6G*#4Z;>=I294\n> MKLRS.P]>U#IXI,R/F0;E'8.)H2<=BTD<Q^(>#R=OG`'F:A[\\FA)P\"L)B*-PT\n> MZSE\"[)UZ.@JVUU)S($@X&T6OAG!SMCZD,&8-`$M7JAJ`^A0Y'$T=S%ZQPR=W\n> M(PIBK0.-6CU8-M6.9!36/6,:OA-UIBJ;547M>3GZ&F\\?T\\5TC!S9;B1=/7(W\n> M\"A@1X\"S>:N:6U8!4L:V<N@_9BY638)(%L3K,$!>I(&@VC(HF'&:3Z<'B0,7I\n> M.'4Z(HG$_2/4>T2?OO+4>U];S2`*9D47GK&Q:+A[.W,MC0O[4S.<\"B>OK<YZ\n> MX:`]3:1:A&U:!)+,++RX*\\,V\":0\"<879Q=+ELF%;0^3B)WW#AV]2:/R1S3T%\n> M(\\9VO?&=.Z`&LFVR&C$N1GU`IO^.!H!E_:VF!6AHC9&VJG;3_'CXSAO&W.8K\n> M$S<7<;E(\"2BS#SSTI:=\\PZG[UF61\\MI(2'7HS19MJE9+4P[G%$Q3C35P*0V@\n> MS(M(3#3'\"=`6>[-KGXI1IBP%Y?'?Y6;XUD_Q)J..PMU%%W!L[5W[:O@^F`FZ\n> ML\\*](M%BD2;C@D\"O)XXZKW3#=KOR:@2-F&^&BB'I=.ZLR,\\4_OX]5269]P\"^\n> MO2=/5;*K7+/Z[^\"._M+RU<Z!\"93\"H9]`W;?M9HF(W.87J\",GVU.XL(!ADN(*\n> M:$)S]D-A5(.U6%1VFZUABS`-8E-;6JQMAGC9>C8VC5-T*@$\"!B)X?S]DG)!,\n> M`$R\\&C)\\HH\"&$A0F]<8,_IT<>T8E7$3>@7K?11%<&XHVEUP)+;RB^::RJ)Z1\n> MI2OBLZ;]WE-4.`AUI3``6V$MX/H$X,A$P0NF-MI\\<+D1\"YP%[41[\"Z\\A&8A/\n> MJA:]RU6U2=5#&X?1MUT:V2Z\"D2E]P[[Q)H!92M8X(+F2Q%E4`R:T+:.:9,/U\n> M8-<E%\":%OI2H)KSG>O3?=XZIB=173V(-%-]&]8;LC<:\\ZT2B3ZWYPR,O-'_.\n> MDU!?;L\\FN90T=TYR]I,A(]'UP`_>V/`%,/?.#M$I6X7\"21+&EOGQMVH'S)\\=\n> M\";\"2\"6'*67?0R0Q4:G&E6(]6XU>H@V#-]F+9J9&B05,=7&PH22U].*BEVX?6\n> M\\X[M`YPFYTK#YQN7O,(.@3%%H7MN91$N%?)S'/3.NN9+#;Q7;(X\\R7C@6ZBB\n> MI04KX\\9=.D-FF#AYW+X%-,/@J_!DMKV=MSMIK^9<.KD'50;CP%XA4\"$O_#G<\n> M6W:`N\\/-#S4J\\BH8J#=5D5@0RS.&1&\\TPI)#W&D_:VI-OIY9[8ILHHCL8\"L!\n> MYA--0=L,6E2;74:OZ0XY9S4S\\M@!DRNZT&MP`Y1'M7BJ'VQQYB,M`<*+4<=8\n> MB4R7_&'Z2+L/Q^P+\\I%QSA`,2<R@`0^'&M!L#OY^2RG;$?[3TP7/F]9>BN?B\n> M>K=SKAQD)52%)..JP=<,KQ3`=0R/\\\"!C(^=0#MMM_*GE3:4,KRJ'@J87D:\\(\n> MW0JZ49\\6U((FG'/4?*0HKLYRQ<DS7/I?S3_'R#OKT-U%J56TE`C$SO;P8KL8\n> MJE7Q&4&[0CA&56IYN2U@@51>N]*[%A\"2-LMOI,`G/L_\"-!=F1M;=6.(+P=]Q\n> MA(X21QJ=XZ-+:]A4(\\XQ[*]%U^U32K5**;5F@0-0!(>UVLK4UX(>P=@OWI*\\\n> MA^+!99A1U8,-WF'=0\\.7%XM7@5L8&[CFFZ&Z>9-_$-Q&AXP/C)T0U=8%38CQ\n> M]X!.6JD9&^JE:P5P%,'/>0-9/WIE;X`$CSRB\\T9TZ<&>:#B!QCL626>^CD6`\n> M,03@1(Z@V?3[!A9\\Q.Q!7T7.&TOA^SV?WSKLR\\*R4^[6!(H@;PZ1'QFLG;9Q\n> M'EUX-ZW>%-<;I%:KMGI9::8F)<^J]Y(@@W?X>\"BY5*V-LGUC,(+0)1F7)JH\\\n> M^\\$P[&?7ZE7G\"U[4YM@*[_MJMD$@&5J*:O:D]3VTIE:8<K&%,@<65VAIS5C.\n> M\\:03PY>9_T7T;]HUFF?_(+#<YRYCM4YR[4URB*2RK#PS;.1-]YE.=VRD(NCY\n> ME.%Q<_Y6K-,2D\">6V(,X5$T9<L<&)4#-[L7VN<N5C7?I.+O5/.LX,].Q']L]\n> ML`P!NXEE;(!/LU`U!YKXP/KJ03/\\=K`EYTAB(TFUUXAC(!2^V22H(,[@X\"T)\n> M$\\[81;.&B^4_Q[\\A%9$G:C[8_W7UX#=<3P[UE;/QGA5QH'W\"1IEDOO(IB2Y<\n> M6JI:-_&25W)AL<DE'[(J-1G`FD'3HI;++1LUQOW3;'6<Y\"O*47\"8CJ47MU13\n> MYE4RI+>G5:FSMR[B<HK9:08CF8<2\\[E7QW/#35MMO!C#X[3M9\\K1=VEGON(D\n> M3D-7!SJJ*[5]!EGJ:H:)<?F>J<]3RWDNC0S*R*`,):P)4.=NTU66_L04V(G&\n> M?^@]#.SQA$,`[%FVTQ&-?-?BRC4BYO(*_#>N2<:1X)!+/Z(G&@CG^B=#%5@=\n> MQ#%\";'5UTD4@-8D^;1FP]9+N-4V$)T6U%YS*NJYV[IKV0;9JL\\-I\"E4E!\\F\\\n> M$=`+A')>\\W_JJ]V2>:R`?>9:D$\\PLF&[%X)?G\"*A)%JGYGE@%$0*U\\9_3DA5\n> M^,\\I_6M8WLAG><^%MY*U*<L)UV:69MQ=FMAB)HCPH'8)^)BJ\"&3]''``*O%:\n> MP%IFZD+\"-$(C1GC_*]7U_2H$,)(\\N2VN4/77*_BE>(GT0N<\\?Y<L:-[AS,2B\n> M*%4%TO,[7A+K;($8%$2R`;P6'BN>T,0,]<X1*)QZT\"@;7'^H.`M[D,'72VK.\n> M-T4$KLH=#);FI\\-N>C3I\"K)INRN34H@:-DE39E:VF`=T'E[9UDU?4L$'HJYD\n> M%&\\O5Q/P],ZY83!;5K/W<''E%)IQ`6\\;^`_MS=:&@.RV1K@RM')>W@OBRYX?\n> M\"-U&-#+(H,[P!M/W3O'@A0J84RV2!>X.4<.%N=3=D-1S*AX#P)H[(Z@G!.XU\n> MU8LV\"C9-Y>%K]-_I1E*D.J-^$B:\\U#OB)2KK[TK[R$V,K9:3`\\<\":8_Z4T.M\n> M\\:J_^@!E2[/@18J![0(SG'.UEU+>^FD%\"ZQ@7E8],O%U+@S>\"ZQK/?KQR/%U\n> M?/IKC>B\\$EG2A>_U'=/.FNI!K\\4!MQ6'P$YVCG;8J+HGDE8XY>Y!(S+C>P#*\n> M(3_Y\"QQXUAM59+N)$217RGLX]`,U5?$9>/\\L6X*Z%T+!70'R-;9T(QY@^TH0\n> METAX6R]0MP9)*((_Z[4(6[FVL6UBSK%'''FOK&:6XC2)'.JT3E>50[P7)0Q=\n> MG)SO.!_9(Z!+8)/]>(ZBZ;DF!\\Z[LGC1(XM[Y2U<)/@'#JX;UGXB]ZDX[5KU\n> M0NVK-6`U\"GC5J*U]`FXQG(N>7TX7!\"H'F<K.%W3TNV9F<;+X)$1Z%\"3Z^'K*\n> MV(-7%RO.9N44U69'$X#$^3`C'O)YS0UPR*Z+2VP\\;HCCCYFK`]LCD7VR:07&\n> MK`[E(*?[(B(D$<$B\\?M)&,(P1:>L';KX8\\'[/\\CME:3,\"OO3#'Y89]I7!S':\n> M!2>S;LL4Y$:4]45S]0D5Z^KZ1TS\\6]0>\"!4D&JTV9J;8+>@.L>*?XW-=P5_3\n> MKC.2`3&?>BO6*AIP99.6:)`4^1'OF(@D%=*U;=EI<1GT&?\\!\\OW>_PL[)8WC\n> MEFNB!'C22*4C8_\"I6U`7\\9SA.,4:CF-!A6=&_E;._5][?1*`]Y&4\\`B+I0H.\n> M:Z3GI&`',.0]22:<552=\"):(=XQC-NCSIW&!3],1^)K4??K[C2&7XA#`GZ:H\n> M-3+\"-GV:N4^?3B_%9T`_S4?B%)1R'O&,]6A.\\#A+D[SCBYNFG%'QP*8[T1`(\n> ME\\0I(Y3![=,9BKK$YI$0[&V;@9UXO7Y-3\"?Y+U_NBLG+7L/(7:9[$C$0%L>;\n> MCHFJWCFY^'1_(PJSB#/:P_X7TE1?*6Y-3'P_>;DK6TY*E\"LVHG^S_TXGI=Q'\n> MSET37O]FX%F:4]_27+><!M*D[/@,+!D[U2:)[email protected]$L4F0!UMFZJ><VJW9\n> MJP0Z'?JQ9%NFET:TOZ*-45=309E\"P#KB6^#:J_ENEAZ&(Z*\\ALN!\"UGS6'5L\n> MTT:^NPC48!*VU'#?+'\"6I*R0V]6%KF6(3Y=YF1GB1DQUO]9N_U=#30<J&;N\\\n> MXH?APL=EE#\"R!?6_8L<+!<GEG!]VX2!A<#=RQ!2V[8\"L(2^2(NJ$D8S[HI^=\n> M\"M4^2&RQ8??_YJ%;=5>ZQ5YO%]`0&)I.K?LJ,V:\\\"T.4S%^4,>:I\";ENVGP%\n> MXO&(D9UUK:=L1[I@G;'X>;#J6%\"3*.<QX<V0%I;].\\9E7_I57QEL^>%W/7I.\n> MNVTYZ6UO9>QR@:K(7X2-9>;W>RNNR*1\"I\\RI1`@]\\V(W16:E-)+<<ZUZ+UF0\n> MBF/I7^/\"\";SN.:R/J`WVP4X:HXF@J//.MN-6NL;@_5^[C2LC?MS>+X7//DP=\n> M.U0I[(T5:1:Z^<]G+@X&VUM:A1M*#!58X#8?T]V+=&&&BX&R:2QJA_^/)+ID\n> M^OE;19@_7WY)'T:BU+%?+0:_O3HTJ_T76W(5F,EGJF_Q/[N<'I:VECSF9)B$\n> M?6-V](^7-S^ZC0K%SB*F^Y&JRA'M\\O=A9IA-D>K%-HJ`HJI9O95=SJ4$(XN*\n> MLN#]GS0-)$\".O*#Q\\5EHB9>#5139&;[\\P9VI:)PE]N78GJE0$2>!W2W-\"DN3\n> M`-F)%.QGX<\"F%RZJD.D>UHO%2.M$)L6#2)5F':;I.5\\Z!41C>4IEISTM7QTY\n> M>[)D>/A98\\K=CID@\"\\W^KU=36(20L\"+,E.(-\\YR-PD_V$@D&1K[7RVCNW5EP\n> MHW@K>7*V.\\;#V5PQ],=,9F(\"W`9A.&:J_/*5?HO(]V93_SD&4X<CB4C=]5F/\n> MQ59E^O<VTIV=2]@+V.E,EY*N#&\"=(R1*NF90;.&=^HWL(@\\S#-V*NRVFX2G\"\n> MYX&S5`%GR49$&_ET80Z*Z9-'7V:.':\"MG,0CSA+$J><#MSM*O,XZ5;WPQ7?`\n> M@U5G\\VO4ZU_(]\"!)>KWUJ')&++JX\\6HG3P&J\"9!;(,H8A-34FS)>:E(_+58+\n> M(TEL4-7\"JRI*TJ(OCOJVJMBAH=%X3:V,UA$>FX,79`5_'Z#FJB9#O<%C$H,2\n> MVD:TES*:.%XQB7&62A=N(V(_4+<4`EKBM!Q^\\=F%3^R!HX].L;NF!UPU'TQ;\n> M\"HA%=)1;MMM0?3&6WC;T^\"!-PPQ7!NTKZ\\*UIZ`F;*C@CH%=$)[M78#T!0V*\n> M*#,<8W/EG0;K[$O)+)N=-T%#P1;8#=?1`+\"GZP&MG3V\\54^UZ%#!R2=[!7@9\n> M...H\"D32G[]]\\?3_3,J_B&N#Y*ZD;-O*5`@4=$0^[>_Y\"#T/6(K72YQA\\+'?\n> M,E?A[FZ?&R`/Y(N&W)Q50VH5&AY#8>K%2`!.+M_MKM;(P.YR48F%P@EI\"V5M\n> M%I:UJ549Z-B-I6]\"5G\\D247#C@,RCR4B9CTILN$!*P,2'6O29L]2CR2.>6<6\n> MQ1!V=F3M=LSV\"]^X%`:,A(S(IGFU94!B>(\\[8[--X,%L#D^9..+N'&LM0K*W\n> MY9$H=UUQ*@9SYB5L<4B)?M=7/YM/GWW[M?0`.9NE)>PHL5'1=;N8S23@M259\n> M.-AA)U3TZ:8<GB'3V30>90%<++6P&#QZ_L\\OOWGUXL6;+Y2!J@:/ZFTS(]OS\n> MHQDK'(3O0[%'='WCQ4;J14S5/Y-&F6M4?CC2SQ94:,F_]KP5'?H!FYM8BV[Z\n> MQIV1,(\"]7<GM%#J[CE/04M.BXF8Y;,R\\QY&7A,?!UE>#S9;Y`;3/;A&^`A:<\n> M/FU(G?Z%$B:ES_0'GIM61CB',AEQ:HV%Y;K8#!76B5UO9,#YU4P)%)$5C:1_\n> MQ#[=$CG%.Y0-(92+W*.T_CTP6&VF<^<Q)%G3Z6Z>SW><4,[SD';I'IM+-O!;\n> M'S%=OI8.I##_B^C?O.L,4OS7.8.DD]PY@W@3;CT(9+HMM)*=Z*@]T9HE@/,^\n> M6O+\"$0?/:W/_L=YAL>$4OY)4TI)'0FQC;P_>LN\\U\\CP(T?0S[4)FA6(;/EO@\n> M;*\\WN_?#A30EJ;N\"F!1<(<%]PLV8HTR6UHBZP<!J6\\GV!V;LAA@\\LR$>62)%\n> M-NU-37H:\"`0$^\\/YT&8N.0YH$RC<#>-5D8PKF<(8E)Z*03Y;S5VQL9JT81]1\n> M8D@]RV**;)1)9A\\&S#.'MWK\\LX*'LA$AGTQ8N_$=\"\\BU4%P29!LGR%:#_[/>\n> M7*]UEMJH4H\\MS_>]NI)W8WZ8.>4T0AXYH8U6<!`%$H9Q\"J^E%5U\\,_$Y(ZVJ\n> MSL7L<,0PJ+VT#GP-%H%3MB\\*4DD?593F;PQ'9`F1(36!34C!B\"I(Q`J_IT9\\\n> M\\=@SE06@YA;?^,=#RO&\".-:C.\"&![T5;5@YM\"\"Q@FBI9T]$R(\"P'#[[X+8[)\n> MC]5/U>Y'L[M^W%X=?FP.7WZ!:\\#,Z'[S`.5C/X`6*\"2^W]'LP6_0\\9B=),6T\n> M@'+LGF@'0M,7IX5$6G]-)@%0&^L[N7'D(]3I01R?D&*%BG?U7*-QP7\\*A'*9\n> MDZ5F9/%=0Y?.]=$I0CW'0]*)P#,*G<X@`S>:^&+HXD4T'MS:B:/`BS?T:>EP\n> M?KSD9IIZS4;6PK-5V&981&\"2\\99>33&8\\F+26FMVJ^(NH;%QWEI,)I:&2>RK\n> MM!K\\=DC!I++&9<Q6&QM-&@U^P:>_N,+RY3B(?/8Q_L1D@_Y.8-AA4\\U\"`Y?C\n> M$1/\"7X8$;'XQR+C*R<2B4X[86C.AK\\KPXQ0?)R,HNF\"OX3!1,3+%>5_=20Q+\n> M#]EL3-UG-H0ZR14S@+_.Y>NQ6G!0EUAP\\*=D$D13PS:<LQ\\/1Q*FMT424B2C\n> M8(51;+]\"&SZ9SZW6U,O&&61X4;'>Y\\`BR;ZFPM;>97_T_;8J07M:,Z_3K('$\n> M;1W38;VA/T@&$`;]!<\"A%ZHBVF3F[?:D*G+TK=*$<1YL6=OAZU#-ENOFKTB>\n> M'0U<#G0OXTZH'$(W2@(@LED606G#/,E\\6?7:7;3/A?\"CODJ*/-L87=B[/Y?2\n> M\"YEN<N.;[F2M2DI!1%V-&:@%^(#$>*K)&).Z]+.1:&)XSA(JD[MH*V?8O(9U\n> M6(;Q[0+8#;7AACWNINOF;]5<L8^E1PO-H'TY\\!VW:#/)4OOD5G28HFV^6JT>\n> MS:M5<]FPS<:!(\"YED,,E&_\\8\\H<KK\"/K2V2$J,)WA,GLK8V4%HP\"^9;7F+M)\n> M`R9G4MD\"4['>.XTC!G[.&.HVLD:FSS?XNZ&HJ3\\=448RU_*\\W<RBW0Q\\G1`C\n> M[;SX=&*P271J?'S(A4P-:>Y^7^T1&%M%KA?Q*&\\%,[$+CG?Z'#Q;Y`[>TM^@\n> M(GI*KAOF(^F*(;%A\"E;C0,[QU7J)T.PFHI1#R$Q+YU&\"?DCEA$QD__K`,H^2\n> M8A;@612\\1O1ONEC@Y\"U\"/X:]<#C+D!%;^-#VY^R[SP&.0TZ/I);(I?7FYP@=\n> MY<Q%FV:3IX)P9HQ<GR:))'![AE`3!F=Z08F5S8''\\.SDN2IJ5)%SEAQ7<#%X\n> MLN#\\Z^;_%XBK[T^7,S;_B^C?XK_3PCP.+<PMJ.72*J!>DX:8,,;32)\"\\X8E3\n> MMW*;VJ235`FE\\:J5IHBOL'<,]Z)>D2_AQ_;(P[\\@U%22TYQJB:-'(L>Z3GT6\n> MR&9M4(\\\\1F&C!GX^GY$#]$&3O?1`/546:9!AE\"7?6F\"RBEHF*X!A(&#Y-HL5\n> M86PP0EZ?P0K8%<!2_H)\\;=ZPY_2\\JAL_<-0,^BF/Q'SSZR^X%\"/YF0+F:Z80\n> M5_\"`=U_^?9B.Q]$8>,GL:B6LR\\^_H;^<C2?>7XQ`^!O!JIBP78O.J53&?YD@\n> M,M%TM!X\\7X?T`]D;/3J/,#^2I`\\\"?.='%Q('B8R1(P064GT/?GYX\\X`I/QDU\n> MK(V+><EH.?AYJ-GT2,4!\\XFX`$@37.%$*PSIL%@PHH6.5MTF]P3S]IS5QXX<\n> MHZXXY[H6C.6+&XI=320T9*\\8%1@;SS_Y-E1V\"1K6VICJHCB)M6L<S6J:VQ*S\n> M[+8DA9#^5%'=X6`9ANV<6I:UCQ/D<O'VA,!9(XF+;/H?F_5;68?!KTWO&#F$\n> M4H..8?)$XM$,65?LI/S\\D/<!_I;'K4;8NQ-_)6-IH;MV+R'!R\\KC2V19:DLW\n> M#YV!F=F,Q)1,&4F+22$3W@K$=%M4%HU/`;HQSL&4U+CT]]23^BVU30/Y%6LU\n> MR.B3Y2M:I(<@%/AAW@Y_1=:7?S&#CNK!O]P@)O]__(YN!TB^.#XDHM#-5:TV\n> MB[>#9Z]>O7B%JMTD/S8_J]UNL^,L)BP;$2?Z*YB*406P3VS0[?=__/9;F>,)\n> MD@C\\I_P`4NX7O'T-I_93Y5BMI;!]3+MT46J;^\\1M/YI+(ZWDF,NE)(I9<(`:\n> M&TZ8.!$;6DGR>/'#T7PPTB9JBH%!0S6AE.3&4).!5\"5N0/6N8N//].JP(0L.\n> M.\\ER/>-,5A<<:^\"D#X/GPG(?-0/.1Q9=<8TSS]'EO/!&+H(7HS@<,T[\\6S<O\n> MYJB\\'7+GS'8P/.RFMG\\5&SXJF;A*'GU)\\TGA/C_S%D]2B'SZ5W+VI;_*\\4@X\n> MMX$L:HWT\\O2AX-1'\"6<[^$_.D\\`)#G!:7BCOY;PU]%PG14ZXB'P5B.5_ZA$@\n> MN6_OH#^UNRV(D\"5CB**_5CV-I:X:$NXL^DJYF>NL(MMPHUFIN<*Q5F@Y]A;;\n> M\"P'09LA4LN:UK`,N.8R'J)12L:0DQXK(DC'#JWH+._CU3%=1,E#DL25GZ2B.\n> M(ZEN&=`K<I*,':&86:+V.]HX=\"#9I,1S\\5;>:-')J+O5E+*\":NA.*T94*$*I\n> M!`[?^ZV9H4.MN^.AF9I?O04-6CS$OT*%I$=F$SYDTY;\\O+&]H!2*-[:#P68S\n> M?S0B(VVV%I=)C$Q$_XZ[.OSR'Y*44;Q,+0A:J(T+XNB'+<VZ\"A@,S3]<M\"WH\n> M78.O&,G`?\\/>&['!5V$#%#K!A7&ZCLP8TJ+GP\"W5?4Z$D*E&I+EX%?\\@#1F@\n> M6*OP5/K$?E0[:X\":*_F&%`(M&AUJ2K2X#/TMK$S/+3##*GH7[A.G0!>QX8T%\n> M[>>,/DLB7>#:ES;&PBK&^::1-T+XJ4-,WL3>R#[K?&LH(^W%.WF.M`+-0)AZ\n> MV9EF$Q)W8K@B>*XA0G'H5B&A/?\"G\"*>4-A33T\\-24.?)Q&71^.@UKC:O#@>4\n> M)^8<\"@1`3XQ,/VU6#^&%R#??(KSXQ(M!S./</^I\"#N5BT$;%[!S??!7?9)W0\n> M)0?-)6EET=NFW5L^])3H@=S3LES2XG\"8F;I3:5!9),;2C4A<G)%3V(-'OD#-\n> M,RE2DS5WB]$]<E9W7V#:;_R<<T7._5`QVP^:MRX!%*9?MV5NY0I]/9LGA_>K\n> MUY9.O59YZC5@<--=IHK$=]:8O_!5_;K<,+G5HBG$1B;]R0O2C8!E8O/\\`N9Y\n> MS_[O;7G?)+^TVQB!)[0=^!Y<->_AV+`D'$9$\"1=EQ^Q>]YC=Z\\#L+F\"V*J_B\n> M(,.(KN\\V''),?MVFYH56ZQ8R^'J_L1N*BB*$;)2T(HYCE>8Y#1M#XRTZ0GW+\n> MV*`I4(86THFR<;\",;HFS)3<^M:F1KD07*G*J81]F/1ZQHT\"/EHH_M,ZRG#DZ\n> M\\GA)P6B.QY,38_&C0%K:;*=_O:HTCV72S@/J1$3E'R0U:NL[CJ<Y:@[SEF@S\n> M>\"#(WWFKCK9%C).<0H;\\_L6;YT^?/68,W)=DUY\\;MN\\I!O?8<&AG&FED2G)3\n> M!$;8*,0765[C#`+D377`D:T1ZB?)3\\=A/$^<3UJ909WWCD?@`FXMG,<B'Q^9\n> MQTB^%&3M>-R=Q^A>$VDJ&8=PIE'_3):07OUAEI.B?Y@\\F9%;=)M,(9[D06.U\n> M502+VQZ+$W'Q4).8YNW/65O^.W=PS3YD_/9D-&Y]+<<A\"K\\GTS6^C]/6]XK%\n> MGL1`;/5CM)(\\/VN%!1EI9^0Q;.E8PX)L7+@?@V^#_#L!C4HN6H'\\\"$YU%PP9\n> M6<=0/>8CAT5V=CW`_B%>W\",)1I:*)=F%*YZ5R)EC_:;I_)),4++&\\A%;3F5_\n> MU?Q*=,1K<U=KAL?0ON6E@I0(5-91>\"EO('X)QC<QY%?K!GD<P>Y<5@3DU^PO\n> MD;S(RNM1CWE,F(0^#;U'^WQ6&79GZ;NX'7@._.A2CBA<\\(3O15E.NEYD<H5Y\n> MQN+\\>8I]<=`&OEI]8Q=0(<:M?2IRMHR6HW0(?N99G-A:6?J`9OMSE[66`0AT\n> M`GR&*=+XZK4'85C;T`3N'*6R:@0\"'3R.TR.PEBW2`C4LA?(U)T_-.\",<0Y5E\n> M[.`+PP]<$C6;+R(?Z8*O.)\\N$C%(4H;#QOG'4$%MMAG,5M64%Y]=?7'/=*PY\n> M4!HTYQ6#0)BO4(M+SXO)&XLW=L464^D!]@I<!F!2%Z^B\"R`>JFLP%K4A=QNU\n> M1RH*9H]IRZ*:28H-.]&&U\"TLW&7D&=G8:$O.%]ZN%[G%'2D8&V591%1A/2RS\n> M:NU`[IX`)>$T;*96A=V(DA*LT+S9;U?3&X)A@P/F\\[5Z\\C3L@`2YH#DHIRX&\n> M;%ZB*O(!(JRUD5?1\\YZ33G@H>_9DAQ,\"\"-:V-D126&&KO0^H6\"3H%Q76QY)X\n> MT1$[,)5^8R#B)D?D3(_B2R]3B12\"IFBMKE&^MPJ;6&ML$6#TBM>CYE?R0PXT\n> M)>NHFYVD-^.0W$STF8]:[^>4(:6VQWWI526(7]Y5)7A;MR4F[[;'KI[M]K`7\n> M_9PFE&X^>3M\\P#H;!UTO64MRQN?.R['D`(C;ZI6)^5]$_Y;_G4:\\29B=]8[<\n> M,?YD04_9=O`Y-E?IPTSFBHJQOJ(S4ZG.%(QC\\;V6SJ6(/:$W^<-\"5P[P\\FG:\n> MUYW,[PY+L4?2-D06&=_K$FO>6']*&Q80\\!#ZV..()J*4X&ML)3:@3?@-)HP-\n> M9]!12J?9\"C.*^55*_+\"'Z\\YJRC`CJ<3/+<5J#;[?F9QQ:[RPIOB8O=H$\"?XU\n> M^6S6T*Y>JU?S<J`A*]!,M;0Q`0TA'8E3'#BM\"D@F%\"=16SH4OA#A'PA;(S#.\n> MS4YRJ\\(%GS3Y=&F2,=E'&V+'?G5.T10]76<`F]B7P\\$XR;B`^XF+)IV9J:0Q\n> M@)>F8^=Z4OGB9HHD^ROU$'MA;MK=[+TL+$(U5QOHWSSH0T5=7$0:FM*5F(<V\n> M;L;AH?/ETD+E$4:+*Y'<KW36/&=\"1#4X#9MOC?=S;$?]#A0VN,@ZJPH`%%Q3\n> M5U\"(R<AZ](73^70++1022E+WBA%G*3\"+/\"<C$8QG=@1+ST]$D\\?&91?:9-*.\n> M5\"6^=6\\=I98A5+1624+=[QX,`T=C]+!29UDO^%F\"F:SWEW\\5<\\?&$C2/V0+V\n> MO=RD/D?\"R45%A=6(#QZ8]C\\R].A77NC;F^$J&OC@IZFW2TG&(H:>><1;P[.1\n> M\"*!,RXX89.@(WOM\"U.QR:VT<#YV]))8;`^?WG;/[X-;87542Y.`<ECS_#,_V\n> M&H5!%$\">APH$IR.,H)1#J)AA8--)+JO-L93VP'!<-_N*$)N@8YS[3LGLU>/;\n> M\"/8<L-KU[7DD@/:L;V`O+\\ZUT-C5J7WWR*CKO)<H`&<*'[\"=8R$/7F!(;1-)\n> MT/:B!`6!24.X/R&#C&S/RHTE6V$]$`IQ\"O,6M8*PZK'\\@8K,=YP,H0H.WC0U\n> M'216.F.EC,_<\"AJG!=S?$;1!=(3>([Z[!973!N%R[.LDHZ!54UHBFH]L/%\"6\n> M\"W(AVY,<!FF\"(`OG`FH?)7&JTS-K9_U%)+E:(%WPC?I&.Y4IIJ@G3VP+8*E\\\n> MS'Q/DHB2\\(M?,Y=`-L;IKME#7_&U4VVIK3(I)VJ<=4[Q.#<U'1K8-V.K;S<_\n> MDJ$&Z.K^YZK(-@9(ZLUFY;%DF#S/T$G5/?2,_0N\\$BXC2C(P_G\\78S3IG$TO\n> MW@XX4EV,V;\\#JM`B,3]!E?_E7X;9:$Q)@TN0;/Z4U@-V4/[TA@SD7&U>2AMC\n> M4Z@`9H;U8J\"1BRE\\#!B+_Y0?B'VMZ'S#BEXF8ZV#Y*/\"KP-D0\"J9P`'X/YF[\n> M'(WX!V,V3\\HN4B;NO<\"H82,)E\\ZH4;%#O3AUP9#[7Z&91^UL\\+VW:MYA]CI,\n> MGI%3OC'6LUC2/650$&#=K#O^;HS$SB$]GM(M2Z2U0&TFZ2#=P8Y+P`#S=Z6G\n> MEW.Y>I4@<:YEGRID%+!$W_A401%\"L`_Y5-CT6@YNSC<:1=W(T$6_C2H4R[*1\n> M^5]D_B4/^Y;5.QO]@V\",9:N^YWWJF3WJ;G31L/;3/+/%.?\"/$8IB,PJ`_6YG\n> MQ0U(-OK\"><244P($\"3%+2S_&VU\\Y#E54@3\\2&+-[65&.7```(\\Y;@F;7G,)@\n> MPVVLK_N;4_*V%-EG`RC24-40%9.6?&A'ZG&;OW/J_W$>M`((@/I@:.@PT-\"S\n> M.K\\,<_#6@QV!W/5\\S$##X];76S7SX//@SF#CPB3MY/B]O+PZV%[3_\\-^.&FK\n> M1ECU8B32/$SD:V31N)O(-XEC/]R*8IC]<\"L!0O?%3(G)9*!GH7Q\\S0(6H#I<\n> M$[!85!_A-C4QD@\",L5\\#=V5,SCX=Z<DL#[/7BT`8\\V`W/3)'J7VHHH!]AC9B\n> M8:/OVH369RAJ0O@OWP[%\\)B.XY,S\\/FJCXC->VDK];(\"4_^$+$`$F6MU5F;1\n> M)H3H\"Q`:^CO;ZG`%>8#4!9R&5#O!K^`J1(D;%U[6QGA<3LZ.:D=NZ72934[H\n> M]/AAZ3H]@4[;=GH\");+K=#)*6#<_<KH1V,_BK&SO1\\/$Q4XWDK/JQ]>-F/T)\n> MCWS#$+([/MNZ2M&//*_9-:;QH\"67BI7IQRA$=1\"D@*W>EX'ZF\"#'VA1RDVC%\n> MIZI/WUS\"6S)$E`1MQWYF'PXEVC<+!NM!88082Z<\"74DM1J\"TD'G0.AG\\F?GC\n> M\\*K9B7M4/F+72$\"(S0.T`@&'\\6;D,=Q5DF+4AL`\"AA1]%EN8YK=L72&]E(OR\n> MW:PK.D3<J3W06,4`P2E1@L#*1G!C:S^X1K,V&>YKM[G:6C42W(LIIKP`:H&?\n> MKFL>K*#BKEM8&^=:1/19JH]J'[1VJ>:6)0?'A.Y0+-!2.[Y4@LZ,DU82`95T\n> MK[S:D!2LLNY$UM)0RPPN(W\\\"VFAG_):]<OSVO9`Q[@JB>2\\V\\YMA;Q(@<IH1\n> M;)8@N<G\"6UL6Y%U?@NU;!W8F0CV^]-`>9*$1:U]F#\"8FR'UVP9<M!&56*[D\\\n> M75AQ3L/#AB[BZ#QX(K]G/KJYU0U1E\\2S2</4L2E)CMX+#@\"BD1FK::Y1L.S'\n> MXNT][:?%2G.):UJ+R(TN@J7D3#X\\I:HN,N?*2$-I!YM(/DQLKB%GW5J$IR=$\n> MHW;HW-[NYB,3+7RHS:%-U,8X!D(4-9^,OW=I'Z4C&'D$F-6(\"Q2FI@3J<KIN\n> MMII\"I@&>/L9]SDD1GQ^8V8:#8#-X7Q&*@NG;;CI[/PQ\"Z\"R7[,\\CUQ0A?N\\)\n> M5BN-84<RQ`\\9]#C.E3\"B9QJ6AXGQ9G%I-R2',%VM`V:[84[$<(02;=H']*WH\n> MV\\.%FUON#6Q8/,E+K9(MFH*J*O`<L&)F>:1@*:T5KQLS8NHO=K@L>N.,W\"PA\n> M+#@UL0<%?@N9K$-X'G2G]KL3P:/S^8%'(I8U^*9;S$5O5\\NZ2IL>A;#^'$I>\n> M9-\\+K)I;!V&[VJG<]$CY#K1\\,C7?-#:R3<GF-G=T+-B0UI#9&G:E)3U>FU0&\n> MG2!Z+I1P'C\"!;(AH2:&Q^5]$_X[^&XV#6<S(0Q8<FQH-YF/94NY-C7Q%DA)L\n> MTP^%8GMK:*@6[JFME6=#DNK!:FNF;.1F4QK.0$CL[&Q36&(O/@P.)!2:.(<.\n> MZ<IN:`\\71FE_V3DB0A<MR*@\"O'+F2D-\\@;F3B\\`.?QKKEXB&/>P>?]OI#32P\n> M!X<FC)%/LP(&9.$J@V1%,%\"TMJ(R5)POK7U<P)<N*H[4Z!961\"##DU\\=)%:$\n> M0ABNUC99:)[XR2MX099]-V'M+YNUA6+AA-LZI_R??'G/W'7343*K;P2[W[QO\n> M<&B9,:;>\"^QM\"2@FD&TX:[SPTL$L;,8&P_[,P3,WM3JPN]A/^-D?/=C!Q#HZ\n> ME`A`)C9.=>!5P<D$70OJ:P8N52`BNQPNT^(H1Q',HUZ&G)LPTGE,,E68T1Z+\n> M<TO)^>ZQEY;IUGA$\\4#>!,^U94P_IK?VIE?ZSGNQS$:AF8V$I'AD[6Q_T,0W\n> MPS#MO84-5O;F:+!V*$V31U]Q1JT:L<L*T^!RK*YP0N;Q29\\1R[R.1\"<Y_6DA\n> M6L5#\"XZPD]G/P9\\&2BS&_NRJ_LG7AO%T8]+E?]T0A%>>JCJYZZ_LG%MPXJ_8\n> MA.2AI?HZV;H]'80M8UI272N@<[VIB.+1N'<N\\%ZA9\\Q>,+?0]HQR7-,?=);F\n> MS4^.270V1:CHF%S`Q#)TP1&XX`)L;^>IMU2C2AWJD3EW>UI&()+8T%`DMG+\\\n> ME4.!KH%Z#%&\\F%$&+<+UN&AIHHTXCYIUH+/+K6J('TN4*VDS(L]\"\\P30X,Y$\n> M$H2#<@0L]!VMP%A*O4ZO57MAUN$>[email protected]!LQ\"3=(?%$NF\\$DK4BP:!$V=\n> M$)08%P7MS790HK/J_*NUU/PL)<:):]8&*D;6NN,7T>#>L3BJ]42317$)10N'\n> M+L83F)80E?IUPSR>ST'A?+IX.F4?EC98.1G!`(0*WNRFZSWT@`NVEHE+WW:S\n> M0LC>;+/9S4TC!\\FD<VG83XHX'`$YGBQOK\\Q7]-_J;1*/W@[>_7J[;-BDP.QA\n> M\\Q.+':^&-0>#T;#/I9HXXVK\"CJCBG-0EEG^\\(-%#/\"IFTYVY7QMR@$0UR4BK\n> M<?WU-F&2%#W1V5&2`C]%-R&?ZQ]-A]T&J0:_-N3R(:?!&VJ\"/['WI3\"=J;T/\n> MQC)O8_G;T)P;^JN-FD;$LIDGB4Q,BB0_.QXZ>\\HFC9)Q)K'*9`CTYI+FBU>S\n> M9W;8'@C%TBMI>'FSW1S>TBU!H8_XCPUYY'RTU'.<`>+D#].U?$S;^@N6HF'=\n> M5\"LBM!GHE^P%\"0J4NTS(::T<GL832V)8[1;]]PL5!'G(XL6(?>P/&3M%]K';\n> M*8O^T8N#8_MPO\\(NH3+[MP/L:`GZ9._'SLG6[_=DOO._9_FL_UB+H-0.$C4E\n> MLB2B?^.NN2SY1YG+)+HML7=Q?[R:@]ROH]`4R!<K74$/&0.9(4=FZAPM-D:5\n> MCKW4(^(!R:DS2+1J.?OTQ[\"%B9NGR-QY+6FF%\"*H!Z&MS3098CMFR!/?`A%P\n> M\"C8\\M&5E+5EZZ?`)Y<A_;]D$-NKEZ?V\"T\\0\"'IT2FY9/CL2FX>/(7ZUV;-JX\n> MB$,3O^R#19B#AO$W_\"\"R@)OQ@R,Z3GC30/UM2$%H<_7UX!SU6\"9D6WGB6UTE\n> MKLU)=>K^R#Y`UK)[1/_977Q\"<J-V?G#,LA\\'$C!,DZ*@;WTN&=TL9>JDAS([\n> MK*ZS:I!@HEJ<V,CL/JJECTU=MCMLN%#O6[OKXO$H]__0VG;QJ\"CN;4P&G]9O\n> M3([C)(AV\"H+*O.@\\RO_;^O!#HLJ2+*SD2%19\"APNWX0*1]R3@^=\\GD#'W8`#\n> M8/M@DD5W!R02C]J3X.N^0\\Y#C,YC0V;G3W_(17'$C\"Z\"I$B1=LS0@FM$W3AI\n> M6>[5P,#WO[<S'JK9,C3\"(VY!G#;;1GAB;%M?[^G3N.=;<]Q&;?,^83(1\"#%_\n> MOR\"[YNCAZ.WP@=HQCYC6#0><A)-D&-JD?Y)ZYXB..A)K23Q?.[S!GZ3.'$6&\n> M7XU/GB/B7N/>.8KZ/DZ3]L>B9L&W!)AJ+A3I!C\"U.A7[WTJE#*KE?>B4>F86\n> M?M=[8*0H(VUY4]-:L\\@MFI8HCI1(=)5'[%22,$Y7=WT-0PTL+EE?C@/*4I</\n> M)ND@WGFA5RT=RF1<<&FBHY%W(S1MK`R%X&$5EV\\YDF2B8C2PBQV75M/$`!@C\n> M!Z@@N\\$WKXE\\4!^UZ_DF0G!<'&/FV:I9L^]K1/*8V_6N&I=96,*KIH*`.2DD\n> MWDC&-0_N)-F5\\'/3:5%-\\4(C%EL62F$%%5:$YHZ]Z3FX3W<*!2:$OJ1^>L>4\n> M3S,`V@#Y:`V-81:;MH5*_MPU\\0F$8\\J*8VP6<!E^NF]F@H]&D;86)CJZO>(\"\n> M@#;>\"ORTX!6`WX[+@;J0R#@!;]2=\\EYS:\\XHW2U#D00F@DB-4@UI2J'(#$#-\n> MV]!>J\"XTUP2C%(1&Y\"\"OV_LISTK^XP^199]D4C.U!2.V$F>D,DQ);>9(4K@,\n> M)<Q/?`W-5D2+6+!N\\+)OFO`Y<VNJ\\TT/OEI.XUR=85E@,_Q@UYV22G'!#T]#\n> M8!9Q=C1KCE2D$X:S9!'0[KZ9=;=H6=H6JEYM96^.K('8LP(B0I9988ZM8:_4\n> M,->P*$YA0P3DKA3G112GT78,MSJ*^MQD!!G&^P-QDW.9F=H/80SGO9(4`$W;\n> M,;AEJ6D)P*GY7T3_)O^=EKK487&*<QWI67N<Z_B0^K`UFM?\\*(%65,N>&,'H\n> M;O<TT2U`3)8PN(B].6T8W%D>9V=A&!P[8?IA<$6:1D$8W!F[7':BX,IVX*!+\n> M(>NQC1H:*&;<IMWG;\"Q^E-IG]I%\\Y#H]D4B]R<-8>QV/<HGG(VT@W..@$HW'\n> M<7^7?\";MCAY!^Q8G:>SZ9'[\"@\\_U*4XE6-#.6Q1GZL7G]XFP.D9^GS@8:YPZ\n> M?[B\"8P4)!'S7+!:5A@>.BW$G.2CVI\\T$IM(SN1:`S1!7HZ?#`#.8'20/7#G]\n> MM#!0P%(>KEV[=$@X&(CQ!RC6T&8&C3E(`/!U\\\\#G@BVC+J(FXL`67U9?4KR.\n> MF?(IEN-R0R!3YC_S\"EA<K[[Z[C7\\37AHN-BD4V3/5ZL)]0%*9@TLXL-%LDD+\n> M)9%OC\"KJ)I;C26ETF!9@06Q)089:91)H?VC<$G2Z+E7F;#\"C^$-P5Q1F`Z7N\n> M$\\<-7&[8<2SBC$'LCK;9[QL;ZE@'*#ZR2I6NDEHXQ5V7?>R0&VUV!01JI/PB\n> M7!_@&;L<D/2>74TX>HA%^X,P\")QI=B:IB%B+P8&_$F5PM9W#JW',A$\\=<>85\n> M85^Y:TFLH6PS1=1<L)](^^@E@L6=``(:0\\TN6-5[=IE6E#3)5;AK#A))27Z1\n> MXM*!.-K`8,AZ2+3AV_&I5ZT=SQAQ4Q]'MI8<Q4`&Y8Y!@4]@\"!ZF!3\"8*)82\n> M#GALB-VKX<'&2[_E_'`\\<T`<M',S=&&P-GS74'_?6X&V_JJ!2;V1G#][<Z\\P\n> MLA5%8F,91<<!-^F&3?U:S.>Q;<`=_5SP],;G,3:ID?&@`2&>![CP%E1K?]CL\n> M*DXBXI*[S27SB@!W+!DHHSNW<)7RL5D8FUO0HM.\"T&VH5F)4QK2*?PKG.$T%\n> MUD,V)13X:$01:F$M:T4<'LR!KDB$UKPPJ0@*VD',FD0T3]?3A<X4REN?+`N5\n> MYTC6PZ'FCA>E0@.#NV(G+U:;\"WAY2C(&A^R)\\69ESLD]WG`_O@*$(!VL7S]E\n> MYR[_#PSQJ,D[J5>6N$>!,ZR_J<[email protected]*F)D&=NC>;K.J+$'N$5#4\n> MF]32'2]!YX`P._:.#H9CYE9A_'&#+\\5M(AQVU#_LTCH;,S*&U+^TL(Q8O+'F\n> MQKEEXNC\\OQ-(H)GPLX)A746ZNB$+[:;2<6!JU$[*1'.S\"'1W(OD=&+T[&!V^\n> MAS+O[VP[FT`I)]\\\\^XG(?\\QH)PO2ZJV558/#Z.!5M>*#EJ1$+<TW.WFCGT&N\n> M^T,UW;X!*:'$G/B.^@_JHJ:O)/QP01^MJVOWC3HY2M\\,9:4F?RUUF3<\"5)ID\n> M;(.\"\\<$;[&]L'%S4%^JW\\(3QH68JQ'4N%_8VV'(2/J9[1UV[%>,=9'JE2\\:Q\n> M\\\\[1)`CP`+0=AZ/BQG#Q;WIO<&\\,O9@VP)T/.L294\")UV&'BH_Y-MG_JWNCS\n> M5I6D!&,/VLCFF'S<EEXR\\[^(_DV[YKOLOS!/6V;SM+6\\5`^47\"D32!+?JH4=\n> MRNJIQ^9,?26<EL0B.-0I]KY45)X`1].GK1%[FUVJI/V4+7PVM,%=-\"Z1FZ3Z\n> M`O`AHQ*UG.<YR:#+(H`8RI`1@HZ#N1;E@Q(A-X!LB#F2\\I(9/9^IXES#2\\T%\n> M2A6ZY&Z2(L$K\"K:NISN1SU&5'D<E/E,MCDJ<:MBS.&.4MV-+1G\\6F5MIA2Y8\n> M22J^YWO6K/@@N16+W%-_1P\\M$JB7I3RR=9YCW+FBNMGF\\[C3_\"-J__W@2QK&\n> M;K\"CG'2EV>W>E]/#P0Z]L>[.>\\FZS!FJ=PA8TE-^Q9[7BOW::`I;`9`(,>VF\n> MXNA(3NXY_(A\"[%^KW%JV])ZO7SY'EK$1BDDL[*(ZF\"DA,='F^'90!I3)F6:4\n> MMW0>%3TYB;W9HC_+'%B\"K:L%0_9S9$J\"?PTO59`,2-E'GA.;OI)7RC$#4)(A\n> MW27Q5^\\!/)*<:39[LUL?<1X[PTN#$9HSTM;\"`TQ;'./S!:2K\\7@2[16F]8)A\n> M28FC5L`^ZPO<PZW0G)5)G^>\\G3/ZL\\R9WE]VRDJ&59DY]`L]B#(<.6>MKH-K\n> MB_H`GSW\\%>8-]M8+EZ:L+$D?PGV8BP,3`8_3%PX4HV%0#)Y4LTR0V$(X-(L,\n> M[7)\"RQ)W)HG.`[<^H:OVKJ%Z*NIXE)<]_J2Q3<'H)9T.Q*4^%M\"1M\"82EKK&\n> MB:BL)5L3^FBLF\\A$(G7+%4$:V#@&M@;O-7@^,]X:AF!^0R$CJLJE^\"+VR$H1\n> M1^,=$9865EA2H+-E(\"/9[-L`CJ7^L',IHB?Z'1G\\61+5=TA`W*Q`0Z\\&GZ>6\n> MD3C8W#APMVPBI_><2E<`WBJ[$%>_8F=XDLSYS'9?$=(A#'OQ>0%C[V-`.XR-\n> M\"&B1*^=YL:>D?AR!)J#I#`JEAI0XS8ED]&6.Z_'T]?T\\U+U707B6$>.\\(.4S\n> MU9P!!18.SAJ,+/HLQP%J@(HW\">RO@!I`Z5L)@,R4_+;ZF2[RS>Z+_;9!`A_:\n> M@`7\\+_W\\/JK^^$)K7TK\\]!CALA[+3>5+!GFTRXTO`2-M&/_E`%\\2MZ0,NMG9\n> MYI?$;3,BM,^;&UJ@C#GLTJ4%&R&880JPI&QC:IN&KR?Y&Z8E_=%<30H;\"S?&\n> M1B24MX/_T962R#-A0A;FA#$_O!PI]>!7&-!C=_X5#T@A2)WH._^5Q>!+TFQL\n> M/0O%$('3(J`A\\NN9K,.07>MH)-G(2P7SYM7S;[YY]NK'KY^_>O;5C[__CQ__\n> M^/*K)V^>O?5%/3K]/(Y'7]+^52'*O&<(E`D9HPM!+-&+\"5;D[E0\\^C(4B6!;\n> MABJ*H5`F0$+MUG6L*E\\$2P$$\"O4,+_V0;>9'BC+K1-+)^M&7N_F/4QK10>N!\n> M1TJ/P%L--!6&]+XLK8OGHK(8F1$#\\:FHQ=$/5[!IQYJ:6,Z:A)HM;;BNKA2+\n> ML8\"`1*.&4_K18XQT@AZ2+I@'3#ZQ\\-'^%]Z\\#K)Q5&3JE=FXX!;M(ZX0W*GU\n> M`*:NWYH_CKS=%$D,\"]1T2^DAQ0]FO)>`T8HFS5]_^<7\\YKAVU\"-I@K)X0JK]\n> MK'4$?L4WTA=\\E3XV$T07R;L@>=74Z^N7U)O1KQP69L8)9!C2<M3)-L.3!.].\n> MLD.F1=KGW9F?@>LG.?&_T;B5>\\8M=HDX:[L),+M,5B/S6>3GY+5;P^K7&K8;\n> M5UX\\$PECKREG$F6KL3D%(IM4H!H\\)_:LGE(&=C-_ICE%III7?N(Y9<8!3]B3\n> M-G`L@<8:M.KGA'3QE<)Q//2RAWAJ@2\"`2\"6=A_:-IL(3E,T+,Z;W`?AR&'+D\n> M+`HN^;N$H!*;`];&)M8`6\\X;VD%<M-\"<VWF!(-92=!OL<F8MR*HQ.Q(IWY*>\n> MPN!`Z,TB[6QKLA1]K?:#P8Z$Z$>#S@FG)<S&8HR7(596*.0$9'`@#>#N+*<M\n> M?SVSZ'26]TT)7E24=NZ.5@TC249(V#Q;5K/W>]9F0\\,C4=37\"(=1\"3X2Y&X)\n> MAA<^F[$7+1>.>XW#%QCZ#YE`/(F*D;-=4LK#GJNS=PL&/IW-X(.^6+&Y(])D\n> MK9)5%@J&1>79^WI]6=P8HW`]04QR]<WH1!CK+4?$ZK*B4!UQR3%DL+'6?!=]\n> MM`-=;&WNQN53L1H)#NBSK+?5;^CNF2AH*S9N``;`\"<:BP0*,,JO\\(/G_K=JY\n> M*/WN&+#-K)!%^4YIZB.UT4!X]VPT@1R&_L`2[2N#&\"`5`;(MO&Q_2Z)O?#O5\n> MK4[45AD4,%YF.MX;+G8^H*QOCX=KVA:FVO<T2^8[UC7:2A$H\"+NX0!9<B/U7\n> MO#=4?O*LEBK_!(N`_7:]63\\@5F,H6.13=K<A.PIK3B<Y0?:0/1HH!4Y8Z7/P\n> M)Q`!YBCJD$_7X-BE!_!C95AQ=8A'J9A)[G9TIW>O7SW55Q3WH<[O;]\\ROG`\\\n> M0FX^<T63/WSM=8:=V^V+C<#XYF24.N8*;[^&)[PKBIC`)+5H=V9RDLQ!P?FQ\n> MD#.D_`F3#*O1H17BQ_$3FZO=C\"6@N1#L%O)+LY?VX<7OK-?0>DM:7YXH#@\"&\n> MOF.OFAH)D5CA>HTT![Q*@FB<+QH^T_L#R*%$8TI>$XT!L1YP1_+':,#^46F9\n> MA-DP2\\PR3/-@CR:[M_,^Q.@Y,8J#714JKZ1YZ5N[(.`N(>!6+MQQ$06I&OVE\n> M.))X/`PF03<X#8LF59Z$EFE1X2.:];UZ91&WXL<H=.S6W*Y5Q6#*)&N/2/]0\n> M@!%W-FL;\\MB0+MXCDA\"&VJI:7FQN#7K`.*?J;5=Y]K&I5;LW#%IQ80TP@2?\"\n> M3C,,U:IG#[^R&+F<;8;2)[C,X-89H:.4Y+QJ+=6>)5ML'I:)<_AA;:5/3AK_\n> MUFD5H.-Q>&N/W:V]C/IO_WW/\\#FN\"4R&Q;X6(U+$]U?=-=A;8BS0Q@@^A,9[\n> MYME:?B962B=0^+@0$,*MJ3)$1.#*8A2=%L_A;PHO8D$RL\\\"_:=+VK+X5\\N\\+\n> M1ST?2/FR5;X=JL!QJ$B$K%[:%%(YFK0PSURXM]V@>D90A=G[802&37ME734@\n> M1NJZLM>>71?&G8T#P#0+),,@&;L-A:&RE@FN#A[PGYU(`;U/$L*K]D9D)*<T\n> M@$ACAX$D\"6.KLDA4DFSS`-\\Z7/C`SX'V?PE\\,3X+<-^%E@_\\9(=?7_H<),A@\n> MJ$`E$J\"L5\\5;E]TH$`\\!'LP#YY9)K:P#@\\<Q5Y%_9AI[1.K>D\\KGVKJ?+7#&\n> M(\\N;<(Y6!LV_*X-\"N)I%.P[I+8V61+G]VEP/]#!K#C>:K;0HVX%$G-:`G\")I\n> MLW[7F-/]`(^_KW:K9JT)%\\XDXE92'\"P&<5Z,BHP>N\",<*WL\\V4+0;R/J)NWP\n> MGML[/FG'%&F_'R7H[1^N#'MO.A[1(*[6S6QI>RZAM6!/'S-\"7TLS`ELTI(+;\n> ME\".#D:#XA@%&M:7=43A&<1OWXV0`^4$-+:%Z>Y1P8&S2#B)2#$-,A2BI'CR9\n> M3R_W#S3A*JPGMX[I[+0QX2@KB)$.JJ7$*<S_(OHW[]KXBW](B\"X[E(YYJ*_A\n> M?Z4>;-382WJQ6>Q$V5+[RI;(:EO8F;_T(V,X(Y$H;A8]R2!-<5=7;:MJ\\1>.\n> MT16(#M&&7#0K\\O6K(K91[ZZ(FV$>D.!$&P&$3.'ZR>@6(UY+\"AUOI+;(1_SP\n> M,UF]@]8JEM1:#0.8B3Q3'50)LPY,3()=)HJH#A@'$-(I#Q>2L&8*BPY/)W9`\n> M4A6XJ+UVDAEB@;SC%3*?&SY\\O>;'S?;07#9_8]G;BY^IK`)=L/NHO1S(!KY7\n> M,,&SO157%(5#6%)^!5JME-'H7Y(\\1T7(/8(GM\\8Z*K:66<-P80VI,#]IQE\"V\n> M68LNS^P[\\[MA<Y7-CH3M#E4Z.W-Y4^M/'U[0>BQ]8_Z\\H6DG)`#&;_4,X\"V@\n> M>5(I\"I+EF)VT`B/RP=N'\\/'XVCG7QQ+^M*^\\Y`5[VG9L[W\"HCN9NW],%U>R7\n> M0X:UG6W,4EE>\"5B@Q\"@)BA-Y'0A\"XW3VOEK/R2#<\\F^<(=/IT@/QJ4,6$MX9\n> M9A>[C0]74_9=X^UZ6^X)ZD,)&D[M**O>D<Z6.I2A8IH.E]IKMMV#TV8];YP:\n> MND&5^HI>*>\\SFUF>(<L`#,]ODZ2P)GW1%QMA9,+A<NT:R,L'>WM;L5MVSV2K\n> MA!I*ASK3U'@\\8O;I:.OX0)JOF[59V*#UV6JSK_;B'^,U7_O8.X$@'C8?C^+;\n> MFZ</I'DZT]RXF.15-QTF=J]=WG9!11`_6/,E>V_S%B$/448>'RH@\\Y!@8>.X\n> M4,<=P;\"*1:XYHDN.0C<HA\\3F);5I),^Q#3OD,)<+3CQX8?'A0:31BX2EAJ4D\n> M4A$73^OL[?DB>=LN)A0E*BE3=KC:6G[Q8'VP/(WQQ4U0OC2S[9>W4Z[\"T80Y\n> M8M/#*$[*UCPY+Z6&F5TH3=>MZ1\"?FT`\\LY\"3=3OV*G2/\"6M\"+](,B>C-?$J<\n> MYA_A2*`Z'>E;?08O*W@:`'BYHDTR.^C-&BG<DM7VJT<%]TJGKML\\A\"S0C'F?\n> M=\"^)F^#D`E[\"0Q?:+Y%SF:)/UO`LN&\"Q)<2[=`<B'I?%&5JDY8E\">H!-D^43\n> M9C_8I[R-=RFN%0NFEP&U\"/`Z+1&O^)Q2,$X9G-.H?4YSAKJB3;/=5>:RJ=\"K\n> M4BX7%CWWM@,J*4<:2-!2=O`ESX-BX(VWP@#\\J]S^_VJO?M-,!&05@7=]!V17\n> MJZL-[XI%YVCF!9@`,&R%OT!HO$#J8L%WW[/G\".OGH!^KD:]<DF=%_EH1'X_2\n> M[:,$=0@Z-:NV'._\"0'QP]R5,VDA=\",*N\\^Y#KSB9,,W$.7$%O(_@]^EVTK('\n> M.=7=8KP#O6U\"$3*=92?(NUO)L_D@D@'NIS]5U\",FT>)?;B,CIKJD\"]T?>J0]\n> M\\I,#N\\NTV;.3))X#BLF:5O\"RNGS$9[!$HN(-H)3-7N?0!B6F-;+7\\WUP<>,=\n> M)W/YF).$PN(RY]UO[GX^BR=I?OO>IP^\\5>9T($H-.U>#G8FH=E.Q\"/MV!NA[\n> MOV)_*A;JHMNBCM@:DXG(^7X'S;L?HNXJ>2C,R6C2]7;VF4(2MEY+E%GC<X?O\n> MA#6L`_=NCT>].A)4X+$,:F\"\"*W,,FYN-+$%%IF;W9=2#5!C`DC]9AV#>2T]G\n> M+=@Q%@74;P?-)S`#^PV^;SM0.S-P9#,QU6&#\\!VTZ0*\\A2W2+$(;LNE\"EUDV\n> M_DP%E3M!D*ECKXF#7;F6G;VD;=@,[=*!DLH#!8:/8,0>G&^]`\"0U^D&GJ;9K\n> MNEJ?R!2E\"'7U=?;D/^';!=CUN#/%COYPYB#O,L05*E=@)\\U5=W(39B\"3++V=\n> M?\\4'EGW>WK!7B<_\"7H*+7$KGMH(3;V&3Q;FGH^P/T9.95\"3B\\WV\\-W#LYI-X\n> MN9D;0=CUYQTZ8U[2G@9/S1Z(F%5**MERC<#ER5Y*G2D[7($TFO;@THT5'*=B\n> M_TB1UL1WW65$U]`$UT[_H-MC*<#]6#\\V>/E9`M!JD8[:-`COA&UIC]QCL)N`\n> MP999'8\\FT:VS2FFD549A-BY8X8.L\\%%/%3_<T',W(:ADJ,AJ42(HS+%_NC!@\n> MSA1]5$!PWB;L1H\"$?<)ONKJ[WB^RQP%8>=OXZ0,[?M-A'GT:2DE'1L]BNC]Z\n> M=59W?5Q*'X59ZXY_`LT*/$'\"\\4<M;QM__.',=G>=C'Y2WLZ#X`,7?P#UX]$)\n> M6'8G@#4$ZL^D)CSQS60+FG-<4B$Q'2&<V11_W[,GZNY6AK8$48IN*J:L8++B\n> M*C1,6_SKTFOS9>^2=?L''3V)X1C:=22\"<DM/-Y^D-\"[S6^>2/HAZ8SF4H;OC\n> M+-%D-I'.IK4NLM..3`#K)/S)3*!^X<WH;;\"(9[/JF<VZ)4?HZHKW5N6YQ$3^\n> MZ3KNOB4SE\";IK<P>/G`S1&&.`0USN1/M?0+=7Z11-(S=,Q#5NT]Q$=,#XC_O\n> MI0.*,7)\\];RL=-*W9AY00B^S(UO\"&!CIQ?.O3NI+Q)UIJ>_'YG\\1_5L<\\<$L\n> M_BM],.609^-A6B::)[Q_BI+<!NU%SOER5ZT<S>RLHIQ\\7BT?0,J?(>>\\=8Y>\n> MI/'18Q91+^CORM<+%JT&A^`G1.6&))R-\"JGD-?)\\S0BBI-;:3??B\\0?V[M)5\n> M(Q#A:XH/7!EAT#!`43K.A9&$^DR=XUQC2VVL$AGK:J8:7(YP@)`:B91JF?L%\n> MA\"MH%*X.&XI'G*%-B\"/$>66IQ`.[)3\"O>A2**HJ8V<MI]J+C:YB[V2.AR`Y<\n> M4R(0PGMG$FVN$Q9Y-[MF0<`ZW%?6>F@I7-A7^T<80(Z9@XNMZT4QBO&'[AHJ\n> M.ID91I'?<F^9GT7NY%R:Q>\"@T@NO]Y77.8BQ2T^B2\"?CJ'!2K-\\;GE3VT.O)\n> M2ATG[?`[email protected]\\JU<*]08L`W:QG%5E%@O.HHY`DB^PK225`)3\n> M=PLW@EDR\"SBJ3+CDJ1&71&O[%15GUVV2'(C>V\"BB63=I!?-X'N3!&>=ZMG+)\n> M>^LC0JN6QOB[XG&M-^[`>=D47#BK^$\\(T]-1<M5R-071]6T?'N!;L8C!/H:E\n> MYZMK.D6`.I/<XLQKIV:J!50,<INA$'V)0O/0G-58C^$.'(]8B]A2!3C+#CB+\n> MCB[),SWUV1N&-A^(K_#)QMR<HN1[\"DUT)6;D7-L5@:ZK#BY*<R$N%Z\"'=OWK\n> M,+Z-W2F\"/*R+XZX4$HOF=70\\R;@S;6T>>ID424\\O6<QALW3`'BZU6\\N`^ZW]\n> MB\"\"0U\\C;[;4U0@G0LQ>\"^?/!SL+.L5:&4XGB%'!^/\"`_$\"$E3;(S[;1%,H([\n> M(436GGTP;_9J\"6.QW\\\\RY`QBW3F,37.HLTO],8N()^K,H@2%P\";)')25D>=6\n> M;&-P*(1I(2P.R#>]D7%M)[\"%?Y;B>)1&J*=UFMP17_JP/)R#PBV.!/E*$\")W\n> MJA0G.>=3JTR&9%EK93&/8\\[@S3D,J*073@??-,X325W^4?O'J0Q&AC00E@89\n> M[-NP%U[47&S3'FC47.1%S<4VIX&+FC.%!=WH<OJS?E..(_\\;L\\/W/QXV/UY4\n> M/XJ\"X,_Q7^1;SDSPU?2`M$0EU`_7/T)2<Q]QW@&RXTD@'W5H+W],$7]]9[0>\n> M)>'F3`*!TWB>C\"-)*J!A>XMVV!Z[3BT[87M`)H'SE\"1Y8*P+S[E?;R>-Z0X.\n> MK>8%T*P$]PV.:V4C>\"HG3R/([7$3H'_)/R!A@HBC@C<0F<=?/O_1HZ]0[FF<\n> M%V<O+]HQ7CQQ[$SU*[OC'K,C@E8V])SV*9@BR^>_>FC>Z.2)@Q2Z_PVZ8^EU\n> MS<2964X;_Q!$*6MF`480127/?O9-.=:<:W,=L`?5KK(Y2JBS3+!_);AV%6UD\n> MN(0!L4(R&6FSOR($W=_X&0UD/GDV?VO:HDG+S(85YZIPTO)Q)#B5W4FC?MPV\n> M8]LHD*U*\\[^(_AUW7:/*_T+XDXEUC7+Y::J?*66A0VA3!9:?+M+F&*&H08XX\n> M^_5PV6O%E93'A'#G%HC$H,'VRA!JA!WF7('BWEKKN>?HXB%*.9LX%>9PKE]+\n> M]=82SOMD,N)@*MH'_C;QE!Y^J4=?4GS>9K#_\\^@O#X<30]-SRJ\\=M3ZRX9P5\n> MA7/B,T!`>F&=-+,<O?,%=<V[3!>Z#?50+[6G'%U#3;DKDP)$S9_*2:*$:;K;\n> M36]\"IP=E/#634#NJE6=Y`DC&7WLZWH7J>!_;(\\JLI\"E)T&J#V'8NBD?(H]U#\n> M_D=_,7SE[X;O!C&&3>F08NTLY:-9BI>E@L4$'1/8E;`SIK>FEHGMKG!`/P,F\n> M4+;C0NH2RD*#-FO\\KZ['DGG)NX),1R7SR%O^5A;J3#(KH<?\"S2U.(/H13Z2R\n> M:N@UIUFB)>?>+%A;RMZ:.I6<?LE>$I0])<LFQ\"9/9`,$?%K/]4$Z&ALD;38L\n> M`B\\IAMXF;]+@8WX)O76,7Y$D6NJNI'S,297LM$DA9D?^!9>UG3=F+QS5M)>>\n> M!F$3^1Q/(N$T`O))<`.2$LDGGU5[_+#A`97J5]JN\\!!'PK_?:?@W8OH1_G\\D\n> MXEC`WPM$''/>:A+F(R<G@Z='1K!6%K6%AGHM;:A7Y<=W:2X7&_U5A<RAEU4B\n> MR=64^S%!7A%%>3&.>WYJD!>'T@/S]KY!7G\"JA3L=?/)%.HEOP0GQ@I[\\W%M1\n> MP)?GR)\\<^M,%S'F8[F09A3*H>+WXR4X$)![85)79&ZV((H?8H[$2*K1M#='-\n> MN*A(\"1HFT?*Q\"J)B%<ADNNZ*^)ZG'M]1W+=\"HG+;.HW4PI@`+J,/?48CF@0#\n> M-?+LO5B/VHL0==?X9N=$4Q]1_H*!/-)1HN;H<\\$1*)*3TYZ$`DM/K`RC`B2M\n> MY`]W)??PMJ$F]S`\\65A)?W(/8=$DOJ2ETYZ8_T7T;_G?B2LPD;\"7+JZU/\\LN\n> MFL?,[H\\NH$=3/]T:S].)X&&>ZP,\">%I+S)$$S(1Y,3S,=?DA/%'.1JG#X)DJ\n> MP5BWTUQJ]*PU5GN*P&5;-1(%1RC$+V)<>5>3C=OQ$F/O!8G)ZK(?2>Y4*(!5\n> MYRGXX[OYBGW5B8)'PF99_6?;7<#J:X*XG\\C7[G%<[S@;]RUU'X2YCDVBI4I8\n> M`<V0?M&HCE\\HP(4!QP'R\\NC1HW]E='#Y#P2<\"6ZY^)?7EXP<]<NWF_6<O*(-\n> M<8'E=9#\\\\K\\W:QA0?WDYW5'\"^)AR>T*:3G]!V`C][8>F6J^GK%V(@>:2_?+[\n> M%6=K^N75YE*\"X6)(TF\\'%-OC8:+'2=E*)71K8%*<MO+WW![?$V>C=CR0L)Y4\n> M,!F-'CYXNMF\\E\\@D<W%L7&12G$$L#B.3AAJ8%.=E*S#I]K5\"C45&[C/]:Q6/\n> M(>CV+U8\\AFC;OUHQL.&.+%<)[email protected]@NIAA,\"=_!>:'HZ.^87G1X(.\n> M\"X:3SWF?N@6-DKB,S]JQ>FDR=M$V)=?WZ@IAP;!)`4B80FT0;D$N3A!>.<PO\n> M[H\"ZUP[1'7;GP18\\0+6KKU;#.6GV=Z9R<[<+4+,Z,<<^#)BXDK%)JP6$C>*1\n> M(H8X0:T1!8P@%D/^6?N*?-;61QVL9B^2T,=Y,W+9;':U<_9&-&QD9:!F1TSN\n> M.F&S'-T@B3U4(\\1(\"FLSWNF<R1,B$'-..@3?!B\\4UTN@UNH[((H)SN\")=D?A\n> M27'M.`VOA3$-PTW@:1^UZ5LR5HY!;B_>-EP]_>=''`@.^0,KP-8ZN`@#O$I0\n> M@W'!$_>XN7C'.J@_LZN=6<D5A0'_16]`^G).!.3/,BM_D;`][.P_VPCM7X;P\n> MNEUR]W^ASO]9`?_/S__R%\\;&A1J*^3SHF4=LU>.^U>+]OE3/+<%V8)(0X,,_\n> MQ&8#5*2\"G2CW)81/+8D3;D'&J1];3$GN%<@\"I@[AK^+4!.^2@P+CH(7:O=H,\n> MSF>;U=7EVIX+,NOHN:`V948YSH,FE7\"J&`+31F=-9\\Y+A=6HL-S!^#SD@`MT\n> M,4EDX;G2RE4ZM(BQICLKZY\".,81H^XQ6(RJW%VN\"XY\\289UY0!&4V@D4I79S\n> MM.!$!\"[W`\",KT53\\Q-%8>P)JA'MQU4J^\\(2QE^0[&^=+PP2$$9A(_F)WN)H*\n> M)H]8EM0J\"*]NC<W=+F_VEIU@9,$&P>_635)\\:XS<2MFMB&]CQ_8%DB8%B:TE\n> M;3BYXH$EJGT(9R\"_SAN*00#V`8!)$$VB2=!K-8$&X[/1TW4K(WUUN3U`!<(L\n> MED#S:Q#<F@$C'$WQ)\\X/('E#02$APYU3=%YD_B7\\W9:B,Q_]@]*TMK)?NBM@\n> MH1@E/EHL96#P#+?$?S+'ZHE:UW:R%$MUICF'&&5+\"#7P+;#CC5BT:Q@\\PRKP\n> M6!PD/@/!/^_%E5HLN=[>J]W>L^I8[K_D7^B]2*+>1AN--)+&B*0UGC,J),*<\n> M,QC;#5VQP_]<A[WG\\$;!K6IAJZ9MYX/<.1\\PR[?<R!S:!P08ZUZE_,4</P-K\n> MG&QJ#H0<AAE/%XQ@L7\"R/T\"O:*1(RH*%A9>=J7/GL(DLRHI#KV(SK&9]]&G(\n> MLGVQF;]<N[`HNL0\"[H5U!321K\":V42;6<P*Q)>)89_H5^D$W-@V6!2R@7+-0\n> M+C,\\:RU'V1[BH<T8>SD@9T<^T+7O07%Y@Z2IAI>DFGQ]\"T;C]%-G9=R5_B-?\n> M+I6-?WGS(XK:/)B<M\\B/7&<VF+CR'1D`B2=_.^04AIS6R/MX*SQS\\\"U45I-)\n> MW!(*&#V0\\(WBED(J4X54J+V+E/428R\\T(&+MU7-?M[U5.+6'O>A%VA3$!NMJ\n> MYYP->(HG9((W'=.,MY@C]#;.I;=R`)`5;/M8DK7FHUMG70Z\\3/J/^,G)6N,^\n> M&!)6`$B_(][<NF!<;!P4JYF9<ARFEAWLSU7TB@;;<UXE3MB:MAINF=OHD3:C\n> MY-[;`E705+OBQ$.:]7425F)!#$RSF]_MJW-6^(O#+G4`;\\U_?J,I7SV`%,GN\n> M&@\"D`*R67,G[4C2HW=]A<M4^4AB6M,SSLUA\\T>V*UC9T%'B;C'K/\\%';S?9J\n> M!69:PF(TWF`O:UV.RA/5`9$NF@R6X76M>.FM!@/J^K*E`.U\"NV\\%R]=F0UU+\n> M-E=8[^U?OC><\"\"/HCL@#P$J;3Z>7/K*NE32]FAA(U_[E]YO507*X0M-@Y4_;\n> M1)H%[[7`&:'A1MX?;-N,9_MV4#KU`F/5MM4++N>:)Z?[&S\\I6EF1^V$ST(4'\n> MBN[+,+;\\0>U/NP#3=J:=$J;TCIP3H_2,?))E?2,W\\@RGJ$O#D4MR$P>]$=B7\n> M#:')XXC^'7EZSI+XKG>#IX*$`/]=SM]7YA#@_VJD]'4(BO`^LJ@(C15,8L[J\n> MI>@(`HT`1>=^F-+)0CW-VB9B_6$8X<^X.*F&I,@54(,B:$GUK0@-OI9MKRSJ\n> MQD5'.3[!<P\"WF2,KSUR\\]$,0G+\\:N6TOVO*`9@8C]RE:4C9H(.0X9\\[:P96*\n> M3!RQ4%SY6=0(:\\!%`(=F$0RDF5=3Y::%_8DX@HK]'2I!BG59++E3V.A?;<+P\n> MURW3BRDG()5<+G7+?5^V]]O!<PR@6;\\'D9XIH'R0BH;JF7+R/`LZO[+QZ8Q[\n> MAP/DI8%!@&H[$XQ->&D!&P++S;5,D`3,LC7%[\"M\".=Z:/;S_'XP3A^TQMWN#\n> MH=S8%V;/#@.>IY8Y]<MJM27IR6:M\\S)YB&/K2HSIN\\W*QBY@=#6Q`NE&LU*V\n> MY#:W?QAQ!3V!#G%7S:I&=PM#)O!:1LKP5MR/T)=B+TL+\\^S9U\\T@/8^'XSCR\n> MC$!_XD,UO>&?-JKEWU$]Y__+A*F`@QQ<.*_#B(BA0E5N<`]O+(/3<Y\"0]7=!\n> MQR[RQE&[7&9+1(4-14NUV<U=?@Z&.U#Y\"D-\\3+QK2I=3-R7(09A9SLPZ;'F=\n> M\"B\\Q9;!K3QN`L'>B.;ZXW&+`C>AL63B7A0WB!(5YQ!-8H8.PZ`4F?7ZD*;*$\n> M3M=[<U7ND71-G5S\"/KAF71B+[`RE90?$Z:617+ZR$7B+',3Q*W2F<95>#B1=\n> M79QUL^$<)(\\U(Q-Q2+VNIJAZ&.C7)H+D`2SMR`YM-QY_*I9V*@A$=F;8N_6!\n> MXO-I.I,$Z<O)GC13J#6:C9OU`4XXHF45I`\"KFK`3B+LF`\"Y.DDDW'8J,L(PP\n> M0DYJ9';OKH$()*K@VL4R642V;E/L-BP*SP5#&]N/^^9.]3.2X7%M&K[A=4QS\n> MT1.Q=DI]0-\\Z@=F1$`2%<3^)[3Q,`6PK@$&:@85F_6;H.9XN[`BT^XTBRU6>\n> M:!FI`PLMYR/L\\R2#S&@S2NV'[[H&+`7(5ZWZDG0*A#\"EZT''DY1RN`8[>CG-\n> M?]U9DB98$A!KOVWNWSB!KZE55K.HKZ>OH\\S3+J4]&GBF(@XGR(%/3D.-%4R#\n> M;P?->24)7Z<H[.OZ5*H3C&96L&\"Y<TBME].YC2FL;]F-\"W9TUC_HGEQPH+OB\n> M@NN]S\\L_=(Z/HF44W9([NI37BQ2(F,,<GF(\"L%F%^&*AC3+0![82F+`B`[*?\n> M',&\"K;<MK^V`R#!@QQ<>5E<WDI#!GL,9.E1K.[7U\\=U=6YK!<2CZ(4`6:NNY\n> MC5D8PVX(J1Z\\Y(5FO\"4687MU\"&S)V`Y@/`1,3.-I\"@'()T,2=+N$0,Q$,8`3\n> M1[^8L6#FR\"6P90IR6.YY\\5=F0@&(19N'?7C)OE38&`=!X^6+X%S=H/D+FZK/\n> M#Y*/6-X\\\"%?76`ZY,2P/-(#(NLUWON=[:^.\\(X#JK)$DE*9.(!A1WTQT_8?K\n> MC3\\4Q#-X4`N1XO*,E#;0Z>-)8.@ZT=(#T:@1'_A]999O34[8=*V1V!WY9@=T\n> MAGV2-=6@7Z/FV]H$PCH9T63<YQ1F1=,7Z((T3``:&Q_)H`,C*TZ4<(A8Z/ZV\n> MU(M[Q]DR]F:_BZW,BR.F`/IJNB7_YV[QA61%5J.(84O`MO*-S_=Z.AKW<4DX\n> M<D*@_$V[9.)\"]D5[X%<WG$&2H6L.R]WF:K'4$&/;*7<1+MU9<[IBYE46D0-K\n> M<C0$48,T%VF<<E88R7&MS,/\"72=HJ\\\\VVQ[)PL$UZ-HRPKR[->FSS8#NXY\"X\n> M7B\\;J/5H.^'2B]*8L&T&:^;%]0#L'YJ[SUZ6<]0)`[.C-M3E=QO*Q(@^/Y3`\n> M+G'^`!VJF%B9.XJ(P4PNU84B%#*-6(L):`E;+.8J`9`->#LOO,E\"41D>NJ&=\n> MB)D7@4^\\F)Y3&E?^/##$+IS,B8S/M46BGCG^?A@F>*T#.((+^YUY9[A@7D%S\n> M2UWN;7K<2_8J8LER:$WC8H;R)(P9H^95AP,#3X;B%P\"%&Y]%AU2CX>SP$!#I\n> M?6$ELSH*13,G;YH)&I&],$?.^SQNFYP2\\[^(_HV[)J=D6\"8]%J=43&F!Q2GJ\n> M6IQ>XO<WKYY\\]QU,3[68GNI^TQ.)=(D1Z2AN1]2.U,I3*^A88`6:YF=.'F$K\n> M#\"?S\"EPC9A(KJ7;/P&#O4]N(PE,ST2(0^\\[)IXGOH#^0001,_!>^K$3IYU?(\n> MEK@&W+9I\"[8'UO3BK^:/!S'(.XLX'0NNCH@VM$+<K(3G\"3!,[8$\\<M(UQD+!\n> ML=HSF.H[O[S:RNF:>;+GR<-FGZL>!+V;4._$4909BBV&LQ=.7]0LY)+*G+LO\n> M:R(]H;4&+(2W>T?MS*WZ)E*WUH4@X$RYG<OI_H#4/5ZG'6<\"\"Q\"TT_OM%`8G\n> MOA,K&P'K#7P9U$'=#L\\NC)`=[$UQ2-W1U;IG6+I)I+!TPJ=))X.5X4,A*G'?\n> MIQ7V#G/O'J240[%F(H'YFEJX6\"]8WDC03U]^\\?PEG=PM)SEF<:E9SS8,BNM/\n> MNP\"DPE&]Q<M7K-\"200TM@B`2=WHZ@B#EN>5-PO[CEM!CL/`6HO$7@E9*U#!V\n> M0;J+JE-T)DTL2/DE>_2=Z`_<<2(A8O]^+SQP4*;R.E611I=R(&XN+PF!&D9&\n> MLO=!;P\\5W'MA2H0OJEW`+77V<KI=<H[?O7>?8R`4<V?>$YLA`>?JEUR)*S$4\n> MAVP/)AI,/I`[YM^$>7B$SG`PP`8@?&8)--%,Q661AI[&<C[,Z*A^W2PX[SLH\n> M*T5HF/-U90CXU,E`HK&UZ>N1UU`#0=V&=;MUHC83;W&)48^^/JA5U5._6,+7\n> MG7?>S#A'3+)\"!1/#IDW@_/6&_:H<9=-MQ\"B?58/UJ(2MMKXSV_U?5Q:4\"PI*\n> M4BTJL/R61TYLA1G\\3]7P'0V!%5>\"[HS<-1PYPP2+C9WP]1*_MD!/%047PB+T\n> MX*(>K9J+[8)WAT8O7&+3PGH^%S>(8!Y@V?)J90LF-#U</6#MJ]GVKTQD1:+&\n> M]6!JM'-.WB47U7R.CCUEKY^M8=IV4'E,5_L-`U;S\\H,4^5VDQ8CC,7L&S0`:\n> M,UQTR$`3@!Q:=8*LG-L.C5C[]I7`93H-9TL\\-,R:F7DH?%4ZS(6%E6RIU\"4/\n> M^90*>YJTRLT2.&%6<HAD(@SF4I00G`0%ZMN,,;G)90V*\\)UD3T764C^-\"U7+\n> MAAG98+P&JIG3G*[P/WC3.AX+J^82YI5G2Z$SX63K).]U2UX6S2B8^\\ACE#TP\n> M1&8R*R=PZ+EKIV>#:18JM);\\,2?^@_GCQL,-MT>6`6[=C-N<)Z1:\"SA@%@#]\n> MW>'?_N?#/\",M'J[\"?[NH%H;_YQ3<9*4]^[?5ZF=R#J2'&\\,&Q6?_=K7[^8SR\n> M/IB'FPBX;_^VNV[._HVRF7%)PXG^GBKZ:C,#JM'CLQH$<?^%U^YV'_W/[Z:+\n> M=5/+>7M\\%I^/\\NB+?_XZV>Z_:F:',U*KSNEA7M61>XU.>K^_N#R8+AF!=-?\\\n> M?+:].D2FF=6C^.SOHS/3I\\5N>O.?9Q=&2$8U]+>1^=/YR%3N_[O?+2Y:W\\7A\n> M=_&Q[Q+]+KZ]OC3\\[FA]V=G?XU/ZEX??':VOT._NZ-\\X_.YH?65[_O*B][M)\n> M^[MBTC_/G04ID_X/L2)YBD_B6WH8)ZY&ZMSQ)4[=A]2[XQ]F[D/JW?$/\\W;3\n> M1V8G+MI-'YN><;OI8].#A7'C/=Y'K(P;[]$/$ZR,&^_Q#^-VTT=&G23MIH^,\n> M.DG;31\\9=8*5R?%)?-L2)EB9`KN'G(V.?XB5&>.3M+SMP[$[@;G]M^_#TGW(\n> M':!_^SZ<N`^Y`_1O'RT9N0_+4O_M^S#V/LR.#\"8R-VA$\"+I1/,[/<R-1EN>C\n> M,UPNY,<3Q6>&GNX-^U]%IM+MV=\\!CT^Z4;^UJC[[>[6I#:_KOUWLSOY.C!\"9\n> M@X+W>_.>VO1?[J=G?V^_VYD/>\\JOSOZ^:M;58>._O#S[^^7FI];+G7F[Z[Y>\n> MG_U]75VW!V$:,XSZYGW8K:5YN]Q<!^]6,_.R.E`G9M-M^*=W]D^D40O_=FW_\n> M=MW,6XV;9<%?S7_-VFUVP3`V!S..#04>!&6H&[0TP<LYJIE/]T']M5FAVORJ\n> M#=,4?%[C\\\\[K62UU=_YRC6DR7$=G#`>SWG;G!']8F_[/K[9G]/\\B0>DXH[.S\n> M[.Q1<K;;K%;X>WQ6_3Q;GNVO+LY2VGSXP^75ZFPZG__7?<SGPA3QEWY^M-=>\n> MS51!\\\"MHE_[:.G)G7WRUFUX_6ZV:[;XZ^WOT3U^8,SA=+U85]X^^^:<O#->V\n> M.W3>WNRF<__WSZW?-\\$?@_K,V1).\"5R3#$A>X9N?#2-HUA!UHB&SP0S9,//F\n> M=T;[.MW-HG^RU\"#Z)Z\\!LZ'X*?JG_V0JP[P=>,.SO[<8NC/^XS,2#:OY:]KE\n> M9U09E>3B^L%<R@:?\"H&@CDECKJTH!B=HV&3*S67^$ZW->,:3Y.P2(UN=%89A\n> MT/_2^]69(7*&U=Y&1\"^+9#*Q_]W/HC&3^=5U]#_/7FY6-W24395)F>9G2&IW\n> M>9:2DQV>5_R<T(VSXF_DV31A:\"#SJOLS0R:C+]XTE]7^T:O-Y71]9LYJG!.U\n> M/J-CN*\\)KX9`Q,V]>AF9@D*3SZ`=%7V`UK8TU;4Z-S978YYQWT;\\N.)'XJ16\n> M_`$_WKMC=/6,D]/[);O>=\"M-,O*&+',T3<Z,(^K4F7\\V@KY$0?$D3\\@O,\\XY\n> M8?K]\"D]&`%8HSLB%,9O<LW0:FQ*3[$,[GHWSLV+\\0?TF;]QQ<:]NVXU@_H)]\n> MC<-JA,_,],2(R)>$KS0QTYD7M!6*LJ!GWK\"T0)2I@IZ1;R^+S;,](&O\":<QY\n> M$2YE?Q.JU*JUA[@CT]UN<[VLIG-LR1.:+ZCN/(FQ0_5[;M[4S[N0A;PSPWZ8\n> M=K3-UO:G+@5]I.=V'\\,R\\'HN)W1DZ#%)<61N*Y&-;`E,\"IY7,D%<G#Z1Q^XQ\n> MZY*3A\"P/0D[PK.1D/'+DA)]OK2Y<=PHTHUHNR=\\XQ3XT==%;\\O,UC^3AZ#W2\n> MMZTE-ZM#Q2ZY5US#G0M^9[OXP+UU[=YOK1/;.2)J?9UKK;0YPEF.A8Y'H`NW\n> M?T_GC;_'1&0YKW*N9>GODR.+<@(IQ3\\A+>5,>V=_(#-^0$QOKX[(0S(:MVO[\n> MX][4=>]ZR&,YK.?)=GN?6D!PN_?7M\\]___+?[U,/4=NDB-OU/(5J[-YSE([B\n> MWNI>J@+W'G4Q\"Y!EGVKYZ)2E'[]ZIIHR^<C%PZW3N>/OOW89:=$[._+HVI'4\n> M1>[0D3)_T<X<]_]I'E2[&)W]FV'Z5&DY3N'V5YY-U0I3D0GF,3L4B1%ZUM:A\n> MVU#TT`)MZ`>G`B,[=LIV;.N:3/'\\[\"E9#U[#4Q*&[TE<N$29LZ.^J)OUOMD[\n> MV$[R-_)3&9I=F><G^K@\"N]1+`-@(R!!YJY[?!\":Q'?LMK?@E+.D(P6T`/A?&\n> MO?_Q^^?_%^CNF\\UJSX5OIK,9$J6D<2S1\"U3Q#!:6ZF<O[4Z4)D5QK/\\Q]_\\`\n> M$]!Z;]TS>VSG<$*C3/27C(BFV>W%S=,5G3.LFHLPAI.6#81`IQGXRD,HM+Z^\n> MOM^KV#-;/A6I^5]$_R8=W!RS-33^]P_/.%#WI8#>O'I-#A9DWGBM.#AF?-^8\n> MKPQKP#.2IP(A(?E:J;:8$23<#D-$MI^!5?VA_#U@';UKR0T=P!'*?+`UY:V$\n> M:!)\"Z\"[(QCK=6S\\IV%J>O'[ZG%)S6LN-..NJ39J]@YHY^SLSLDP*R^UA^K/X\n> MEU*FZ2RR#G_L%BW9^M1=FLQ:C1U1VU6YLU.WUJU:_,@X_R(=)S(D73=[Q:0:\n> M5(8#V<D)I]YE,.7J+J#0%W$SG/?'U&O,ZG6U@E..!\\B%([*0$[+04^`RU-LS\n> M8U$$EH(R11ZF.1ETV39%`U^Q@6GGA>*[4XW(7XL$)#B\\\"^]`>Q:FAM,+[K=$\n> M8Q@9LN89,#.]6:R;OV$3P!`[)W]$^\"_LK-L735)>:`9;&-DY9.4]]?#F&K;V\n> MN=C:J\\-,7*1E%F$E#(\"_J8*@J`#%[;I]:$..D\\?BU7K^$*;*B#$>-N^KM3='\n> MBM7.`.P,E4SS,'<>ALY\\%GG'W'G;`\\H8\"_&^?:;ZEL&:Z6490!J%V+JU=R1>\n> MXJ_#[,+4$TVK.R4S';O/V_*`MF?0I*U@45\"4F\"1^$;_HG>*;>%'ORW80KSC\"\n> M7EF@^4@#?CV3+L>?;>9=G`-M6VS<2C<DZ)^\\Q5<<B%,R1.$CJH4\\.GC]$`.E\n> M:))(M-&L8!S>JG^R._3+OE\"(LJ!@S#?^A/O[WD&3V`O!!D/),CU]M-]<[69`\n> M'>_4,.,:_*MOX4(]]$@3_940'EF>,2.)^,O>Z05[M-NK>!9@L%O:X(/?21#>\n> MF%TD.9U;>-$1@*@74P?D?&Q?\\CH7-'%#NV<MP!)#MVAEV/'`^F^X+'8R86',\n> M>/=2!!/UO1]U$`,[^[U06W+5:`36%-16\\;1,[>K&Z0[LPO2)%T1O'GQ\"E=0>\n> M1!_9LR-M4=WG-6N0`O\\MV3_,KMA224QD:;!=(W9CDN`SSC+T-H!1T'U`EX\"9\n> M<X$^;:V#C>O!K$Z=-R.ZI4E*:O%<B&R2B4?2+@)Z&5T\"+J`=GQ6:UKH!Y`L0\n> MU99&?*_(=<=')PF#\"F6C`W[7XO*3M5V1[Q.P0$_8<=T&3/;E#@OO+I?UXBPD\n> M7`N?<(5)B$`J6V>$HQ[A:<A1L(B/OL\"B;]9P>N!(LMEF:Y&&V\"V&`N^V&@4F\n> M=(L\\1FA+<KHV]6XD5!`*5[)9KN8B*M#ZM*,[O2-N1/C,9VNCUA%]/_@/9N(V\n> M5[SQ;90K=FD]O6Q6S908\"8YY#\\L_''IY,SCS-Q3J0PDKW!^NYC=V61%4/(4O\n> ML3^K''\\ZXM0M[('D3Z]C?A::]%L=12,_)I3^P'?($E#_<O1VE1R*'!'`<J,G\n> M8\\F*H>[-G>!2+WH2K/?2#X[5.-BH#5XR#//R&:(K1X-CVHG51_1QPA`RE1<:\n> MM&\"(\"W%%580RB\\$\"_S[G<N=\")90RL9?KT`4O[IW8$`3V$@4K$%K3:#9?R]>P\n> M/*<QK$711H(A/MZ%#6F0=Y`509\"4G4]@X.->J3/S,G(1?&WXCH6&&6-VP8[O\n> MFLV5]+W$0GJG6YR(G;<^''-0,=P=-QPJY'>F$^5-$IXO\"=N(L_W>'#/>2U.M\n> MAKO!*`4<HP4)J4LJPIF)X3;IYZHW%'1%0>@VL9&#-\"*&&HDB.*.>Q2N!*[\"#\n> M[Q@P;(8@_.%>.H+8X8%T,.8?+HD`F</\\EQ\"'$[D'LG1,NBT$AN//-6`4&+[C\n> M-XQ5)R+QP?J'IO`/389+!XBR]#S'P!0UEBG\"+6\"%H';49@\\S65MF-/(9N?X0\n> M.8^4\"*1A$=UHV%,[(X%??-F3QY2V/O&9((3>EJ*.;M@3?WK@>OU`'&JJ<IPZ\n> MS3>BYAI%5*LEY,6PH-Q)[\"OX3,H,KJVHI)O;O\"#E#9&OV90IACKP<\\I;24F]\n> M1J@0N70S]T120V2AFB@J%+=N'X%KUH*8\"&``NN]KVSRY(B(&/,E!MM\\L&5U]\n> M\"^9LKI><#5ZM'*26))G!7GY]N!3\\'2HDG+N*7*.6R+4DL,/=C3BT5E-)P]CG\n> M/,EQ_)('1&;YO61FU$XQ;Z%K/V6?XE=$27^8BH,\\1M*(2*]`BM%@N5EY^`-+\n> M+]V<+V&82]/SW,91X:PBY*\"^W;\"WL*;JXIWZJEH]^WF[<S-\"Q$(:TXC<3FM+\n> M&ZDJ08'V7G]BJ.DL&!4.&U7,J4\\K08[DFXGA_!P@%VT(\"@X8U@%,=D20\\]/9\n> MLB6$8FM4*]$/$`5ZRO/.5[B]2X\\O/K83)T_1+3!I2]VR!9;^%J@#P-ZMP]M!\n> MY\\/F!>E2ESVRZU[MWTQWE*C*S3WFIE10BLJ++'=K@HB&]F(*W7ER.(B\\+>.;\n> MZ/4R)O_V]@?#&O7:&-OVLDI:0VPD!'?*P!(7>L:3HRP\\^<)KUX0GF++@]@/#\n> M'MN6%ZTMW9=!D;:TGQ7+IEPG`8_B<VFV)HE@7[OYU!1Y_JY;>)M-='MVLRU\"\n> M83S(/VUWVS+<;0<T]2WVJLP*=Z?,HAO=&,&V\\Y;%`87J#0:40J*NC0_I91?%\n> M)_5.N$>-]AZK5#EE=Z4E^TLE2+HS*X\\@Z<Z$:I1Q0@-&.57<W<[%5/OSMG1G\n> M5--;A-,&4>RO\"(H/CV=KGBS!$`3$-\\M*`4*!4V0(?NT1?\"],])99:L0!'9UX\n> M8G;29`PL>2/)>?AUH(0+;X_:H%0Y>B(I//G^*^XFM:BQ'LF91DNALX`@Y4--\n> M/)[HE&&=V%\\AQF6S0[B!Q`*IP+W=M0P-._<.1TL`&*&DI^H0NW-U(</==^-D\n> M1</MO@EP20]NW?A<N'X(Z(!,@:4243MPE?K^?K!_3-P:5\"Y?4BTKP]+9,!D]\n> M0[C2V^1\\X8;I@D);;9MA:A,-U?T[:L*\\J1YSQ@4)G1(J/@M#WCJ41+2#?$9!\n> M$_W[+[A\\':&R*.Z@5\"VS1F;^%]&_:3=4-/N'AHKB:+UK(4H(0;<[BW5:`:GF\n> MQ,R65%?'&8(.J>Y.L`_YH3>K!M%;L%(J1<K=`R%Z\\E:Q-VP0*NXX)FR5IQO`\n> M<43NG+9PI]M7B\\+NFU-:($8C[L9H&-D4,1J)^2/':!1CCM'()R..T4A-R7O$\n> M:$#L(#[T<XC&YQ\"-SR$:GT,T/H=H_+\\L1\".-SW-S3:?FW\\\\A&I]#-#Z':/R7\n> MAFB\\WM(^>,VZ;P1JW*1!?$7PZR8)_A;\\NHF#OX6_IF?FA?G>_$OC-N2'_F]L\n> MARU53,],+::1FUL_^_GBS'2+:DMOK>WBS(R%:KOM,],CTQH?9//+],#TU91$\n> M&Z:DF?&?\\*?_OXCRB!-$&\\#Y-RO9)_)V;V%\\)B6H\\`DEZ+.X'-E&X,-[0BM:\n> MALJ?4L9^I^W<[>/N^J/MG%#&?J?M%/%I[4B9)#ZI#'S1>:816,*.V7>62,@W\n> M5D)1$O(5QR,O$S[@Q[O\\]4DWPXWCL7>9[W`?AN=ZQ^&74O3>QPF9:B$;<5B+\n> M4S0>CQ)*<^L)GV7Q*1.($O`+YA)W;VZ4X\"E'\"9YR/$X0+H$/^/&.*<_BS':7\n> M'KG0O:8\\)0MI=[*@7+F/$SD%4I%_>U@-*1?NY4\".>KH[8%>MH+PXOG)9D8I_\n> M]ED>CY7(X)%G!1^<-JN3D5:%Q]Y9/5;\":_&V$K:/I[>1%CD3`FD$SS+$>`S:\n> MP-^,CP2Q]50'3WVI(KL[O(=Z*F7X>=Q'DNYRQD?^Y4E[Q['ZZCY>_1G%)Q4=\n> MEWQ#+<[W[=W;)7HYPCG.TE2?3R\"4>9ZZ,O1\\2ID1!QVF^KR2YU+BB]SS'4N6\n> MC@K7;3QSL?N3V'P<MZ=_)P;%^]+9G&R^857=DWI7/69(>=)92&N+O(5>(Y2S\n> MP&6<Y?I\\9_\"2=EO*9`BSXN>)BPN5Y[L.$FD7M0MXGO0=I+O(W@1!E\"?1O5.J\n> M*CKA+J_Z%OA8A%R:D:M_P3LW)U]MGE=^/Y8@+[.)@N<\"`_?\"Y!`C*G/CENKN\n> M0+F3FN=OO/>N^=.CY8)19T25BE%<TE(2?2SBM`193>6]/L<ED]N)??9&C:(X\n> MF[::D^)!3VN^\\)XG]OFD4;_85NNS_;9GN?.R-+<``C?3\\:CDTTB<\"+WG:.V4\n> MM#S!<SQI+[=A&\\\\16DW_(9HWSMWO5?1/X0OB<[J_4ORG(WB:PL$'^LMTE'X5\n> M27*>ZZ^^PN,T=Y^;7H<APGU[\\)0YX6?OO9N3.U;C!*XL[P8<_FI?_>I>=V1L\n> MA(@B[9#H7\\'!ZU>W4-9X8C8=Q;Q=\\F,:<U3J*'BD#^ZDD\"A//HV77(@>[[@X\n> MZ3.-5L9S@8904_#,4<QW22:H(BWM8/!\\2A?\\,MSLW66*_/1VNF')&7A\"A`>S\n> M\\,%ORY'&(ON/2>L`0BP5$4Q%P1/#DF]M%X_>V^0C\"2UJ21+7I*)#\\'LWPO\"Y\n> M=[@H>FD%V9,'?%?S_.R__^A1$WAB/\"F(OG/71R5@!8@KQ'MY3F($NN-[>6Z/\n> MFD4'KJ4ODK]GT\">U/G;/]OO[1Z!C\\X,_5^HQ\\<B']XQO3B,@(&5\"#O!\\PED4\n> M9EFHSLB2C>\"9>?]32(BP_USL!/:_4Z:77^\\IP[S#R>W0UL>LBI@A,WRWIHBG\n> M%&5.F5*T4TRL.(-G$6=D);G>DU85(HQVFYZEV/W%&3!%GT:<R48=P?2#Q)FT\n> M[(1XGRK.9&EBQ1D\\GR#.<+>E##V+.`-A6\\09>3Y%G-$NX#G_`*42;0QTX-.(\n> M,UDROJ\\XL_9N!*?;/$T=.O+TH2.G$,4&D2TNSZ>H1/6\"PG/^@4I1QD'Y:*TH\n> MR,L'JD5E'*JS/$W-F3O-:/_8>\\J,/-WHR\"E'@64BVE%Y/D4]JMW&\\_@#%:0]\n> M\\_9!&E)`U7P*%6G/?CA11RI+`IV?+`D_CYV:],3I)16@KC\">>Z?W:!F_V3MU\n> MI?=MAZX$()M(.WB6H6(+B2PMSZ=H2T$$M+KD-&VIE.'G_`.UI5\"C?`IM:9I]\n> MB+8T\">Q]=YM$H)B>3.YEN$I4&WZB39$*2!-4X&[U.16(D\\*6P/,)1?+4%<G3\n> MDXIX=DL\\WUTD,$'>;;9$$1!)+@$:Z=L5\\4%QRL;&E[YA\\D1CYCW,N:&F*2U2\n> MP]/B?LV,5,6Z`+PM<$!3\"KWT'O.\\!;V5QB/5)N!`]6D3>K0Y=[:+#]Q;U^Z'\n> M*A,ST@>AQ8PR[W+=>,LM9@6A&[G'SDC!]?!(095/&^G=[>(#]_;3C+1,;(NL\n> M%.*WI1V>_YCTC;3,[4C+'D'@R$AO;1>/WMOD8T=*QSD>08;.QLB'`!DZ&V?R\n> M7I\\3D`K^/FE+\\\"@JPT4MR4D2_&FMC]VS_?ZC!IU#*4#92\"[9@!53)@YSE8YB\n> M>:_/$UP-_#T_>X/FHCE@Z:2:'LFV.^K3FL^\\Y[%]_E@553;RM&(C3P$V\\I1P\n> MP7._1F[DJ>1&]]#)W=$\\/_OO/XF**H]CJR3*D]0JB?!>GT5+AN^3M@6$1YTZ\n> M'56O@?6(CNK.YL?NV7[_\\6N=IV,[V622T$G-&742ZQL\\I^.^4:.HK+54<]):\n> MW]4\\/WOO7?,?.&H8S;)D8HULHFS@]VEN[7K!<](Q^A!S*FOM]!<GVOCN:IZ?\n> MO??)J?:56VU\\65%8(UO&I`)&-KS7YS*U-CYY;MOX1,/)S^5I-/RDY@OO>6*?\n> M/V:MB:5)4L?_B&6\"WV>.W0J>TU[>2XPB+(?W&3B.<%]W-<_?>._33\\*82+/@\n> M$*1Z?I\\YUBMX[HP:W(F,FL7CTT9]4O/\\C??^$XVZ<'R16*[DO6/#PN=>GDR,\n> M63SJ/@/8L5'?T3P_^^\\_\"6N6E(XY2B:..<)[>4Y'CC63YS9KIJ.F9]$;G,2;\n> MW=G\\V#W;[S^>-X/E69@C6*J%.<)[?1X[WDR>V[P9&>==-:?S9G<VGWG/8_O\\\n> M<1X+8/!R,3!9*0=O6<H!`^H]=J0K@BL^)QXZR>&L0-*]^[V*_BE\\0:YNWJ^B\n> M')]#3NEW-PC^'A<%?`]*_\"CA[<`_^DI.4O?Q*7X*=\\_$N`S>WD/>.SK]8+S2\n> ML3,FIJ4S)N*]/D^<*5.>>];`7'*\\!D7B_:8U\"%X0G+C_*Z/_),F87#_ZIC+X\n> M0\"8^C;E)GN8TN6,=Y`,:(_W*T!W]513C_I;U@TE\\ZBJ>-J%C]VR__^B3%)-B\n> MT/\"4-/]F=U/B8:HU)C,2WNMSRF#N^#YMLX&ZE+S7Z3B-:'+D-XZ3_Z+W/Z1\\\n> M.G:>_`_D%Y%FMQWD5U_AX`,YP:U?\\>188?F`/9=..I\"G3\"C>ZW-NGS_6C2M+\n> MS<:1)!=%RE9(XN;HO4\"SCV&\"]Y[+/J^]<Y94S]76ZW[#C<M_D69%WOF53HIC\n> MGEC!!_JKY`.3CVF/R*^^P@4H@'QPFAO7\"7/\"WWCOW9Q\\G!L7*<([%I^KM1G+\n> M=+5Z?(;_JZ>K_;U4^00AWO78]Z`X[F-,BO,>=WO3P[_>S[Q`]?3`FGMP*_>M\n> MK`LB[R$%W;>R<0?='EAF'UC;I).>8+';7&T_K#:V3P:U+:<_->O%!U:7=TP\\\n> ME.?^_I5!H&4?O);K(%D9[^>%.$[Z=MFOV@Z(M]=\"=@8H&,):7@#+9;,[>T/0\n> M8?>ICTQZXXYES<>.N5>6@0*>-Y/.$5A69P!>J6U>W?LL*0'PLY/,Q\\P=?$*R\n> M26>POU+@O_NO9X\\%^M[>J1-U!^MLL4VKHONF0<B+F$3LV*5!6!#2X./A.\\)6\n> M$2QZ),=6^D0PFQ9#G#[Z>G@Y8(BX;RTZ&`/T,:[)DH%4.O!*4<(B9.SP[!/&\n> MLW\\S7`T`H@MD%0'N9[Q\\`D\"RJ:XCQC:S6/?EL7P!M9\\OH$$BW^E[2G/M(*@\"\n> M?,$0*][#_Z\\]8'`&FV%8O^W504#\"\".VSVG(F;H$ZVU6SJ]W>8DZM;AC%\"C\":\n> M5/IP/DQC@;\\G>*%I!UNK`YVG:.>1*:WYW7MPK:6F?\\<Z<(\\*`--89!E%N'(H\n> MR@I*PT!4EQL`B'7PJ(\"G'`VN>Y!<!2M*L2M+0`%E-#$A=B5GP2!XLZ#KP#AN\n> MUXFTX`+_2E\"X2\\DJ((!-?OG%V7%$2`:\"[T!:MC`@<X'P*3-`#C,8Y/(V,$A*\n> MF0`T2('O4S!(LP)F6I><KT-1]3A!M0_[R'D\\@![[/1>DM.N-9&=@,'8`T5(&\n> MZ::6!`0=E,(%(QD*<J?%%UIX((YRO?'6!<\"PS5*P=*!F$25?V'-\">\\%Q.U27\n> M@KGEPQUU4*@\",\";V\"%Q:(/H*\"<@'<FWX((.`-S--'I\"5VI#_Z6JS4%Q2`7)\\\n> M\\RR`+@M`K&@K\";:;(.4*BM@MJ'\\6:]DA/`%?4I'Z7)[email protected]_'WQ-\n> M\\+\"`>EXK#.@FF'^_*'K+:&^-ID=@##]!Z,4>VE6U`T\"TD^35<Q2AM96EW.)6\n> M1NUI!&A<!SS05H'E3\\HTNN$1$%%=5)2X?LK9(@0YTD.%/`8T*\"\"I/8AY.ZYL\n> M>19`B3JJY<'E`>V1B\"20O+E[0'GC%9MS>4D/XD]X'>D6K'D+TN'MW_H+W88+\n> M3>EP\\.8E\"N>6$?,D<<B<:6A#-!1=`\\R;RR)383=Q+@/.4*'7H[N;!/#_X,$V\n> M+P?3\"\\.7$G3Q-@H@V<CG(Z)_LT^=:88(8$%W1=K&$8OB+&8<,0K58ARQ\\8AQ\n> MQ\"BE!N&(F3TSN0^.&\"2,'YE,?X82^PPE]AE*+&SZ,Y389RBQ_W8HL:0D*#%H\n> ME3]#B7V&$OL,)?892NPSE-AG*+'/4&*(L=&`\"79COC-@@K_CB`DN?W?$!)?A\n> MN>8R=X=,X!.9-Y0Y97W8&3JU6;'QO)+R/.]<[VG)R<G,HMU&9-W=2\\=E>$KY\n> MN7=Z[@CC&ZG#>\"L<[<?[QKLB.7N/90QP_/>NIVNNVVSO&^O)-76M:XPQ?]^*\n> M>@QKJVY%QU?HQ`WL'Y0/7U7R`.Q9\"\\I8<+R[G(4^YK@I&C*>5_(L[AKX)CTE\n> MG@]86EH=GM,[PZJ\",EZSMQ_%R?W;(9?WHD`\\'Y<7YR9Z'L<@9_A&GN\\:*JIC\n> MKP\"NKKR;\\J*K4H:>I:E[+72<CHR\\T;4ZWCN>#\\;V\\>@#X_D8:HK=4/!\\=\\\"9\n> M+EBLSW<7810J:84]/+DXKW?B'N]8+HJ5D2[38_]^N=MI@G$5/BZ*E_TE/CZ(\n> M-P:\"]7UC>-L^D(8#BN%]&8\\I/7T*[\\N8S+-CC@WB9_:FY^_3MO,GBA83.']*\n> M-2<Y?Y[6?.X]E_;Y(UQ>S=9+S_().SRF\"3VS;R.]+]A;-Q:4+/N,[X-1DZ$>\n> M)2^91$HM=P[ZE-;YF])[_EB?;JY>8CZI67.:7+-)Z08=/+=]NFG0*\"F#EEI.\n> M&O1=K?.S__YC7;KCG`Y)!I](L+`%>P['%':!]_I<X,;E[XN.DRF*RJCQ7)SD\n> MTGU:\\Z5[MM]_`L\\QOF#R4UPC[JB(&+,>2M-Q(.ER%GG,G'R:R?,=]R.5@4N'\n> ME,'S2LISZ!+7R\\]=FG_\"2`C]K>?&H_]OS#^/S\\Z2>\\T.Q2/FH[++37.RFEMN\n> MT6QL[R1XSI]RC5(9N4=1YI1[%.WP1<KM\\$7*Y>4BS<:GWJ0:/8TNT/.'W*5@\n> M>3_!90JDG$]QFW)%][].!3U'E@00(;(D_)PZ])P3IY><2'6%\\7PW:^N7\\9N]\n> MK8SKZCW:(9@F8:&YO,2#T[/PLOCF-!::JV-VF*N[FX7FKHHS+CU_\"`N=4@##\n> MIV\"A@8;Q`2QT-XI2&`*&*N6+'^^%(4#DI/_<94>`$R+\\B(#/GA99=TKS_$WI\n> M/7\\L/\\+5\"VX\\-2LWO[POW:B#YS8_@E'+U<RG[#2&Y*3F^=E__[$,24HNH\\(1\n> M($)*.`+XT<O-S\\_,A/#W'8:$B\\JH\\7P:0W):\\ZGW7-R'(3D:64>*,N'SS<UM\n> M^?R41`;AY_F910O^OB-FH*B(&5+-26+&*<WS>WU.[R-FG(`^=@KW=!>&U/A4\n> M;JYU.TTL&P8DH!/8,\"[#4>A<IB\\*O:>,L&XH(ZP;EQ>,Y<(]WXF6E+ANX[D7\n> ME?LNE^51/VMV7XF=/*CSO*-4O+_(CHJZ_.<)N%OD2B]+DJ=YKDO\"SSPW_,U)\n> MTXM)T>KP?#?H>5#&:_;6,K:KI[>3Y02-4`*-D,N70\"/$LS`I^.8T?HH_93Z'\n> MJ[B;8T97M0R>/X#%1:QN,>K`#MX?=XM\"IWMXW'LQ&0``R).4HY*!.5UPZ#;>\n> M,]YCB40&_G/:#NDFR54P47F_G0;+<5+S_.R_3S_RNN5J\\I%KEK%0Y7WA1AH\\\n> MCWI'S3\"L/.H^2-5CH[ZC>7[VWX\\^=M1D_\\C'&>X[1\"Y+Q/@DD??Z/&$4.WS/\n> MS_ZH.>B91\\W53$X:]4G-Y]YS:9\\_)I\"=C#7\"VP!077@;`*H+#\\//S$[Q]QW6\n> MBHLR:\\7/I[%6)S6/]_J<CSZ9K@>7;<_%<M]P&B9V7<'H3BXCHT@HCOZ'*@'/\n> M)Z@ETB)C1(]\"GD\\I(X\"_*',*3O'80C^CS$DXQ6.`U(]L.[&@R12YPRD>GXH^\n> M#8`;[3;`;GIQBGO*\\)3R<^_TW*5FR7(6LS_6U\"N0Q)V<(/<U]7(]`!OX2%,O\n> M:NH!6[ZWJ9<KZ@)`WVGJ_8QJ^1G5\\E.A6L9%GI[%!2.]C\\<I$`JH]_Q^PI9J\n> M<D0*GMO\\&;3C**H^-\\5)_-E)S?.S__YC^;,XF:0$Y\\33'*<*<,OO<W:I`>RL\n> M_YRUP>)@BT?12W8CZ<7)[;-*G=`\\/WOO7?.GCUIVAUZ1R6DW9&(O2\"Y]P@69\n> MV(N&BYQP/R;V>N0=?$IF!'L[<BM\\.W)QN1V34R_'26:[3(\\G7(THP3.)Q[NF\n> M9=%%N.`U9S0'7EM^SVO.J!;^<V?+<?8\"WG+0:YRXY4YJGI^]]_?8<D>A/6@<\n> MN9%\"3,,%&8'R\";&>129OY9$S8.';3BXN.F;GC,['`\"U)2E@;\\IL`6H(7`IK2\n> M^I4GQ[!V@@_TUYC!DD:I]ZNO</!!,>[\\(,CM_J+R]R(Y%9WE[HD<VT?[[4=C\n> M[)3$;8YSF))+)/%!K;@'\\%Z?)^+`-;;//4NHP\"QQ4M)L.V\"6\\`5=?>YS`<TA\n> MGNM6C!WY0,\"3R++C_3J^_O*!$65.A<DY:4XR[WELGS]N-<K$',I1%IN&QVEF\n> MZ%\\RIL4>)^;M:$);H!R;MQG>IO0XGH0+00!QYYQ-`0N1E*GWVRQ$^$*PAEJ_\n> MS&3VSV7X@?X:`[Z*O\"/Q*SN*6^5_,&$$K#%0<\\J4ZJ,\\%/U%QYQ*C0&R3EG\"\n> M.R=R;`Y7RBD<[+<?#3TVR>T)3B:)/<()Z3?DM/(SDT#^OIN:4#\",E!P:*AV2\n> MP_`%75/N<_UU]#B$'^@O)FD\"*7:<'(8?T!@[OXX31/W@=(IXTH3BO3[GGX@L\n> M9N3#*\"2`4QDS\"<C(*BQ'G9\\G8I0?CWK(8@A'=:DH4AY>5?\"\"Y&`/KVK\")#,N\n> MCBUE\\('00P+'=-21S\"3]JZ$?G$X63YN3S'L>CSX-64PI4L[\\DYY)LAISGE.L\n> M=R;O]9G36?+W_-REC8*?=VF!UOCWRI(W!=AC8+7PUW%,O_`#_55P2L@8+X52\n> M]A4./A!\"*,11?QVECO*7>Y#'TR9T;)_=]Q_/;Z3I&?/8\\20'2#ANT!3(RO0X\n> M`6BS?61QYC9T3.8E/'3,X(7B-,JO?$+G;5P<NZ/\\OY,+%A]'^C'6DWKK[9;=\n> M@\\NX:R;PZ-ZZF?C(Z1^GMM%Q8:L?3^R<^X]]F@#'=CMP4@>,&;Q0<%+YQ>\"C\n> MPBT>1R>5#X3OD]4H<6,=7SI>`/[[R0MPZUS@T7M[JEKB,_3>9^B]_]^$WD/6\n> MLZS3MS<?M**HK.PL`B'[W+<J.D\\]5H07+^_GD%SVVA!^]>4]L>,^D6V$5NX3\n> MF$90S:>PC%!%G\\([email protected]<5<^H=X+NM^KP2^M;]=]]+-#?!\")MZH#^`%>5\n> MGJ>/`:A$@$1_$M0T.8E5U$+RJPEA:=E%\\MN>!<!$A?E?1/_F`\"9*X\"%\"P$2'\n> M05X,RX0:NQP\\_<.3EV\\(BRA#-X8)`<6]?/'ZS3>OGKW^]V\\9H^CK5Z8_+[XS\n> M+1.043-XB=_?O'KRW7?/7CUX;=Z_?/'\\^S?FOR^^'BX&/SQ_]B?&E3-B-G`!\n> MOT=G#\\/W!)2W)$RGFF'5&-*I]C&ZR-.IN3#?7QTJ!Z8&9\"H`<S%85JT08R%8\n> M5AOERQ2,6NA<=0BM9=C;LQL/I2N=>.AJDRA$5UMV.VAQU1:,_P=,0(*9(H3#\n> MW>#-M\\\\80*T/,&TQ6&Y6<\\!-$<#:8+MI*.:=@;H6#*Y%2F2:Q!VEOIP3OEG*\n> M^%\"H<#EX*VA?->I25#^%3W-@7M&2X,6N+M=OAT#,(H0SOTV!.!.<MQ^>O'(`\n> M9697C,Q$O`'R'LT9F2]NL`RSP>&:NMI<;LW=,06>W/KJ\\J+:44^:M0-VZ]9)\n> M.PV`D82V&*\"1_33=K3>$M.:OG`ZM-EW>-X+^9J$0T2USKFXZ^'?4H@=')]!F\n> M`T&+8$BY8$FQ8YJU`T'S4-<6@KJV$-2UA8>ZYE9U>K$A6$9TV^RHD0R4^IBG\n> M-'4Z6CIU,F#3A9/&O'1+VCTNZ/IU<UC:[G=@[P\"X9V8_%10W.O?>CN'J(X)4\n> M0W<G!#[6;F:&X4[7!.`F:)6\\[0,HMBKR\\\"R;-@3;TG6/#]^BA<`F8U$$MH5#\n> M8$N1RF62M&'1C$2118!%HQQF`HM6,BP:(:PR+!JQBZ?#HEE,R\\^8:)\\QT3YC\n> MHH5-?\\9$^XR)]M^.B38:G\\.E\\SS_C(GV&1/M,R;:9TRTSYAH_SA,-.)MD#--\n> M?`#A&'X\"MI660?D3RM!WB'*Y],J?T(Z6L?T\\H1U*]J/M9,EIF&!:!OT\\H0S:\n> M*2:V'90_I1TI8_MY0COJ6PJ\\L!/PURRNF+1S2AGZ+L!?.Q5++7'S=DH9^D2<\n> ME*GXW3['^$Q*4.&3O)3AB@)G>'D\\H02CG'&1WA39?66RTI7)3O&%U+&@\"'O]\n> M<NFQND+:Y[N\\CHG_XF'BL;_]NX#`,K;C==2_]0>YV'=-8(24?Y]:T*&XH_[M\n> MC7H\\I:JBHY(V537SQV1Z')6W>>R7J3J(XS'KB6=JP[D56H(SUS)R!CUR80X>\n> MF)RPMEE9C*0J/-[=.&)GN01\"8;EQ/')A\";2]7^->^=M*V!9/[&[H834FTCZ&\n> MV\\+8+%\\Q@:<\"!7_26WD<\"WQB.NJ:KU$L%1B]T:C7F;['$>G.=NFM/J;Z^%&I\n> M<LD_5\\*KR9T7Y)_?<JAJP=GR[&,[X(S]-PH;;)WV7\"!].7+O:A>/WEO;[H>.\n> ME\"@8SRUGX\\+<9A0;P+.(1U[(C,&\\VF&48SBTC[2&T];T[G;IK3[F^0>MJ=!Q\n> MQ$L)(<?S\";0?Z7VE3&^JW[YV<G=AX%ENC)2!)[A>?C[EQI!NXUF*W>_*`)4;\n> M=P@]&4KN2YWAY]>FSC\"2W+<F1+V%-=T1D<4)IWDRR+_QE#5$&5E#E#EA#;D=\n> M7D-NA]>0R_,\"<+TGK2&B-Z7;G-#L`]:0+XN/7D/8>S_)&J*F^Z\\AW]$C=TDG\n> MR8FW],B[ID?NGI;R<I.??EF.W-UW0A=X^D?>93URM[64E_O\\OEWPJ[C[QKYO\n> MMT<:L@;)ZLX8QBP/>/V[`Q*IP&1R#^$@R^^-&HTB$I!XHJ\"#(A*0:/&C[RZB\n> M`8D6/OJ$(KZD=W<4(8H4\\7V$0Q0I2U>D+$\\IHA%<$&'CDU8^D(Z+DY;2%]Q/\n> M$(ZQ(1.WE\"?([6@E<WL2SR=M>W_?GU3$4W;@^>XB@1[F;ET'EC*QT9L2P>HI\n> M<5!G>A)`L:^>N9=*1Z8?:H:[MU*@SKB/:N)>\"H!`F7$?Q82>I7N-QC]_)Y2Q\n> MY_34=D*O>4K<;6XJQG3,2@YG77%\"[UA1]0@BW'\\>M=&QP+>@J,`\\2C5W\\=<G\n> M-<_?>.]=\\Q\\H3G`U:>*:9;PO>5^ZD0;/G<@-C#IU()I2S6FCOJ-Y?O;?)Q\\Y\n> MZCRAN&F>[#RE.&N>5'[/DYUG<,?UGCMK#4HB:PUXI1/7^J3F^1OO_4>O-5>3\n> M)JY9GE1Y7[J1!L_MM>91\\UKSJ$];ZY.:YV?__0>LM8@&$R>!W*VH1`E15*+(\n> M\"8I*+L.*2BYSMZ(29<9.9AD[D464DUSK28I*B\"D3)['<V7Y7/92D3D^39$Y1\n> M@_?ZG#L5D3RW=41)[)1$(H:=I\"6ZJWF\\3YVF2)X_5E64Q$YG@ZZ+>B:)G8HH\n> M?.[5%\\FH6:*(3]<8W=4\\/_OO/XG:2\"9;LGX[!8Y,*C_G3G7476LN&COET8EK\n> M?5+S>)\\Z#=+'KS6NBRP9V>M%^%AY7]@;+7P>]=UNPH_A=NMEAX_<;G<US\\_^\n> M^]&GH//2+`BM5\"_O\"TO;P^?VJ$'G9=2@\\R>.^J3F^=E__P&C5M5?F3C57YF<\n> MI/ICQWPN@^=35(S%R*H+\\2SJ0JA81%THSZ>H\"Z4+_)SVX,/T:<C*Q&G([AXJ\n> MRD@[*'/\"4+D='BJWPT/E\\MQ/KO>DH>)>DB[P\\]U#Y9ED%%.>848QO;,,^BAE\n> M\\'S*]'`[/+R[VT$9:0?_.:$='D]1VMV#9]D]>9;9W2//=TTINLW5<;>Y.NX.\n> M5\\'=/%[=\\:A\\VE6IJ0JW<CPQSZ@\\&9>$@(*Q)@20:-Y/\\#Z.S19)DR[)/&<1\n> M[%R'[7Y3\\'#P0B+E6[\\FV?A(\\'#P@?R2)O57?`P>(_Q`?W$`,JDJW:_>R&/_\n> M`XG1UU]EC%AFQAOHBSSF>&4Z4*>&])^R&N4H,\\^3`L_Z_4>']%-J`6DY&V>9\n> MMIR5HU);S@J`,F$?9..TZ-D'=(GX^T#P#MP^\"%^0<J3[Z^@^\"#^07[(/]-?1\n> M?1!^H+]2P:E)1[?N@_`#PM'T?I7`WSF^#Q0_XO1]<-)JE,E$]X'[_J-1.K*X\n> M%`NGN:=+L67R6\\Y903WQ'SMI6S@L69:%]3GN-X5`!R\\40\"#XE4Q&Q_!J@@_T\n> M%T,!R\"^)DCY>6,.H^5=>>@UGQ_`+@K]KGPOLQA%3I>)(T';P=^*E7<FT2&XM\n> MF1>NF9/(QYV+AT?O[>ERQNT[1B0<M\"J2C+QW.R5\\/K9MF#[3MBD\\8H]MX[\\@\n> M,X!W=<BO^-CRA1_HKY$LH.!)'(.?\"#_07VGF_TKR.PKS!_HK+KK_Z2L<?B=U\n> MR*]L1/-Q2V'>/O+!R?OGKJ7D9__])]A$AH\"-2?2>,,$C'J=(0-C,)0BY&==/\n> M1N_'?/T8(6(<I[W7CS`;)(\\`.DFQ/^@2\\%\\(V([^&C,=CX\\M9?@!)>/Q\"LNO\n> MY!@<3?B!_I+K1WZ-CP$_A!_(+X(,]'XQPM'QPO*!_LJ`]I*/07;E5^^8_0\\(\n> MCM_[E3#2TO'\"P0?99.)51;92^D]>W'YK$K-[^JUY]R8JS;;.\\E)N3?G^HVE@\n> M:1@A:3DA%9^T;\"ZF3%LV7/1$MZ\\A)67/]E665K<OV=3<;V!;^2\\$$$E_3>[8\n> MON$'\"J,4_CJZ?<,/]%>J$#\"W;]_P`_DEVU=_'=V^X0?ZBS<1(5C>NGW##[0#\n> M\\DMZ?[QP\\('@\".DO!HH[OGWEMKC']CUI$Y5FS\\CV==]_`M@5`DCI@5TY-*OF\n> M<'-?I!4`QH8U->O#YK[5(%W9QZ*LY,!LZN2#V''T\\GTK*CI3]->KZ>K>U:2=\n> MC#*,^K*Z)[`'*BL[J58^$%H%J5:3#A8*HN+W9E]_6(T]$#>[:G^U.KR20&Q@\n> M^HSN76T74:;9_[Y93W<W`A+TH3A!2=+Q=`;\\T(\\.?^A#$8BR#IS+<KI_LECL\n> M/[++T*MW*GY]=6%(TGNN_-[U,LAON[__C@CU>W:OBY8TVUR:K^=NV&=G\\;U3\n> MP/5XM_]J?[4U=+C:W2_-$VKK.N_?,UL4]&^]G:I6JP^IJJ]']TLM`>WKIPE+\n> M@-7KH\\,2N$.?)\"R!J[HM+\"&>W'L70)':2<-A>G;_Y>NK:GO/++\"<C[5;D:&@\n> MU*FS^]<UZJ\"@45V;Q_<_@YC_3]0WKNO6OMTK/RPH+1)_M7;KL]=?O?CN?K-&\n> ML%>?I\":RG<)%M)6:ZLFK>R]C3WXU0,5\\Z#+>7M_]ICX')O3'#A-W?/=::_8O\n> M\"5YGY=T<'W;)YYV]]F&8;[A^N_2,\\>C^_0.8PVR2]K(?'P!'AYV2=GAQQ=AY\n> M?,^%Q4:Y1W7WA00SM(36)7:08,T@/<\\>#]\\-W@S7`\\6?J>;#.6$;O1D^RL\\N\n> M!E/\"RE%\\O3AA2\"C@ZA\"BSM?#2X#^\"&C>H@TAMF3@I`Z$&`\"Q2(=PHWU9F/>Y\n> M^=>,BB\"-KAW`U,'K646-7@MNF>G_C%=-L88(/>J98F@Q?M5N\\,3\\.<[BF!LD\n> M0,*SG\\SU=78]`\";>PL.X(KRP*&A0$;(8I0C&(6!DH>A[!S*5QI&`3)72*N5C\n> MM(!<!,:UJ[:&WIIE81PN'IP/=[0G%+(IYL)\\NYK.T/X%T,`8U<JK;SF8[J5/\n> M11;=$-[9G+A2H\"!5!%DUWZP9,&GG(TP==$%7=L&`(J4@4@GA-#TER+3],*8-\n> M8$'3ZL%TNZVF.\\%,VK;[CHV!WE-54\\,E\"/X7HY\\E!'_V?DU-70.NBT\"H!\"-K\n> M?[,_5)<6<>UR\"MRN(D2.PH8%HIM9C4X'>&X`W+:T2&4`2F/P*-LRBB\\$.BKB\n> MOI74-T!`S1D\"RF)(+3\"\"A9L_A<8+,:2BYDQ!I.:,0F5!I!(@'(\\(C8OV!T\"Z\n> MKI<;PO:B,1P8\\@O[;2I@8ZYNGM5MM:,_8^@T<[O@^RIBG#Z>ROGT0&AK^\\/N\n> M:G:X$ARU-\"ZH_:49BQD3'8S!Q73VG@\"O;O2<46>V4R.)RI$09#BS&IL=<,PV\n> ME_5&Y/IJ`*2J85S(KI\\/9DM#IV8,\\,;X%S3WP9Y'3Y*$>F);#\"'TE+9TL-+D\n> M;PYM;+#9-0LCCJZPB795)9NG?70ZL&3FD,_IE#LJUCE:W,^2^FD6:L]G6J'4\n> M]MI01%R^V44TLHIFJY%30C`8#4ZH!5`[@NT6A>!N0L72A/;]FN'JIO.Y0M`!\n> M-D\\@Z*+4L-,TC\\%6>#B<F7.WW^@1W#6;*\\``\\G%>TEG6^>.3T@:QJW4%>A`2\n> M,4&,D3AXZ]$\"L_\\:)H_F0DQD<6L^*&^'=ETL4?!1%\"TZGID+PLX3^$8NL]2-\n> M3\\>H!3HY-O^+Z-\\\"H)/QI!30R4O\"N#2GCM`5\"402F_TEWV5/7KTF!$I\"FGPM\n> MU]L3,_AOS%=Q.>&K+1^;LC&9_HLV[%L49]D98-_B<2*P;\\688=^*>\"*P;X\"$\n> MNS_LVX^`__T,_O89_.TS^%O8]&?PM\\_@;__MX&\\QV:$*6+<_@[]]!G_[#/[V\n> M&?SM,_C;9_\"WS^!OG\\'?/@S\\#<`6G%)&GL=W8R*0VE_+(%8TRSCD:&3+\\S?C\n> M4X+I@RYX5=Q:QC9[CV['F2W#SW>72<NQ*T/6)ADJ/W-Y_N:$H;9B]XQT$^<<\n> MYDF9E2C3%GP>Z7TQ8N=O\"AKUG_.L';N'N-)\\;(%)I)J[8_=.:)Z_\\=Z[YN\\5\n> MMTN3G4_<Q!?):1.O96BR\\2P3+^7QC3R?$K\"DU='S\"5W(*`6WE.$H'.X\"GJ4\\\n> MOKE_%_PJ;C\\:XWMW&[,\">)I+[MX)4#5<)B]<F?S.>\"XNPY@M7(8Q6_BY1)`I\n> M?\\//W>FYPQ!%X64].:!>W#?U#UT<2=8Q8&^V9,S*D_MYI*`RP$FU*FO$S:7E\n> MXM0]\"(*'Q<^]8%3=:=8RL'<SAA8_<WG^YG0\\+*WN.!Y6]R#TP'@%D%SXYMY=\n> M\\*NX\\R#<M]N?\\9<^XR^-/N,O?<9?^HR_]`_$7UH+0#,<@GG#\"M-\\:QEV]1S;\n> M:.H[RP3<=#I.+#N;EIEE6_D]L[-PH0Z>.]PT;A7AIIFM/XV;/JEY_L9[_V'<\n> M=)*EZL#-LX;G.V+542:;N#+9W;#@*#,N7)GQG10.92@\"2,O@><7/Q2CE<++4\n> M/M]QS+$1\\*F&_]_5[5#$RDHRQ[*,4XSHF649O(]9QBFH\"?]YU\"MBH:@PNE+-\n> MG2+6*<WS-][[T0=M\"A:7>%.PW'+WIN`RO\"FXS-V;`F5D@5%&%IC+\\P;A>OGY\n> M+L@,PMLIQHPP5NKS'7N,RHQ'B2V#YQ/P':0=EKWN;H<QG48>J-/=[?!X\\L32\n> M#SS+E(P3#JV<N.=3,\"&D.NYVGEC&6ZH0Z*E3JL/)$\\D0)_(4^%>48<F0R]PM\n> M&7*9,G5E6`*\\N\\S(W1G\"@>(YY<.\";^3Y%,J!3RT2]!W=[E*.-);C1-'XS$/*\n> M>X[2+]\"$_]P&56/4%69%64#J8T6/4(Z[FN=G__V'@ZI]\"`+-43091J-Q*#6G\n> M40'3;*:G!L\\GG#13O2V#YQ/**-Y/J<]R.A7O9^*>3SE.DY'=L9,[^6R4\"!BC\n> M^$Y&F\\NP!O)49HJOV+$]?\\787L*BI^1:3U+/2KJ-V'%F1SI]/&YV0LCQ24D1\n> MCX1^760E=:&<\\%MY+(0ZYOK8`QN0YPH;`\"@(^0W8`/]%$L=Y]U<Z.1;$&7R@\n> MOQ*._.=?`N]PO+!\\(+]*OV&&GCA>E/^NS7)MJ4:X'RGI_UT!)/A',9;PWR,E\n> M_;\\+HH;\\*//;$2[\\OPM&3C;R`'/R[%A)_GN!.-N38F3OW##T5A]S??RXZ.Z4\n> MH#CC20E4GDF6F25E=-*1(<`Q8TV:+J3T'AA)D[0PJS9J8>-0>,LY\"P[8JOJ;\n> MT4A6%.GLO]!?@I;\"P=MQF1V-=/8_(,*%PKR,#$TCO_H*!Q\\(Z)'^4F268]'=\n> MLH@\";7'**IXRH>DHF9CGHL\"S?O_12TGB59XF5AQ#/(.(77E:6@DP?$YZI,%S\n> MBQ]A`_7EMPO45X`))C>M7]G1U0@^T%^IA+`#=$I^]17FE9</3EJ-4^:$G_WW\n> MR<G2R.VK(7(06A9YA]_'3AX/GCMBF*Y&.9'58'(MOVDU@A=I#N@G^PL^4_GD\n> M&\"T./Y#5H(ZZU9!?QU=#/CAY->Z:$_[&>W\\/V?`X!@N!C!5IR:!3Q\"=D$R*>\n> MV=CP7$7J(,\"*/,WP/H_-<U:TR-S8'_*E_1U;,A>\\T%\\IPR6,<5D667P,AX)1\n> M04C<.7%\"3QK6.*?G,L:S?O_1VSNG6K-\\#`J1T62-J&6S`0L:(D@>`3,5>1PS\n> M&S8VSWEYUG=ON`F5WW'KWM`)E5\\\\H8(14J3)D0D-/TB+C']EWMDHLO38]F8R\n> M=X_5.&E.\"(?6-)KB6;__:(P68@>R,K.\\0S8I+/.`]\\I(C&++;\\JSOQJ,8N-N\n> M<?D]5H8S?*&_!&ML5#!^V%&@G^`#09VC[\"(@Z2F#5R7'X':\"#_*<L='\"7\\GH\n> M6.'@`_V5EMZO6^#Q@@_D%\\E*WJ_\\&.<2?J\"_QC+F>.1^'4?UT\\]'I5<XBQFL\n> MYFCA\\`/AZ_77J+R]</B!LKWAK_(HRRP?\\.XXF?.]:_OBO3[G]OD3W-)))L(Q\n> MP5-G#%_,[T6S5^:MYU'KX\"A4&^/B6)Y)?CN>25$S140+?[%`<YQG4HE'?F6,\n> MVLBWM/PZ?DO+!R??TG?-\"3_[[^V<?.1JB`8(+8NF1]Z7=@7\"YTX&!X54'8]\\\n> MGDE^.YY)7@C/9'\\QBMKHV/8./Y#54)PJ7HU;<*I*P>K+[[4:=\\T)/_OO3]>*\n> MW<XSI5F1*'.1VNQX!-+*.CC`IC).$WBFM$CB7I[)`G_I;T;;<CR3PF_)+\\&G\n> M8I[ID^*/GC(LXIG,<Z$\\$[[_)#P3;1_E#]*\\3)4_2#G?`7BFE$&PP#.E1=I\"\n> M4A,FR$VH_'83&KS07X(*++AJ1]5\"X0?\",RDVFVQ]/C#'>:9[K,9)<T(\\4ZJ\\\n> ME'[__S9@L$^*NO`9'NPS/-AG>+#/\\&#'H';@\"=YVQGQSLZU,M2]>WK^Z;GI.\n> M<RCW]ZJ'*!8G$PCJ>=9)SGF\"`VQ//1_D`-L[34#^F13WKZSHG.:C#K!W5)?0\n> MSAAW[HM[P@@1\"TK)B]N]^BBTI-OKNS]:TB<:YCCIK.3'0/U\\NNIP?GH=K3_\\\n> M.*:C3N\\^Z#CV^)'?^SC2&O9`A#W=K.]W7V//]U3T(9O!G.KL$^UYKJ\\+]/;!\n> MFROIQ:&;T73]-%U=5?>L\\-.B9GUR0*]/A(;[7P2&QG=;T6$2-_>\\2_B,CSL<\n> MP0?5PTD6>VG%D^^_NG]U16>S?1\"M*#L+>5]2P=U).QQD=Y;N#9Y69&1XSAUX\n> M&F&A$6;9,?RT+DH9P'5N1T=;`!UM&P58-Z7Y7T3_CH%U0T@[C'5S&.3EL`0F\n> MV^7@Z1^>O'Q#\\#89^C5,\"+3HY8O7;[YY]>SUOW_+L#=?OS)=>O'=L`8V3C-X\n> MB=_?O'KRW7?/7CUX;=Z_?/'\\^S?FOR^^-IWYX?FS/P%N*$F!>P;`N.$X/G-8\n> M9X.7IO>;O1%]\"3^,&JH&KZX`]/::4;T`5Y1-J/QA\\!*?XKO9@-!DS0D#I!*A\n> M_0\"AJ]K55RL&>MMQ/8(.UBBXW%Q1PVAM\")9IIM!.!+A$!7]J:$ZO?42IP?3R\n> MPBP<`2(!],LP>=0C^RU!(FWG4X*:$ARF=X,7@C&UNAG&A._&S7I#6'1[\"9@D\n> MHK5&F`0(%8-`70/MK*%%ILW%@%?GPY3`',U\\WKCYI`'M\")IM0/AM]>!ZLWN/\n> MBJ[V%GGM<$6;I1ZLJ.L_5:LAX,>V.XM'5LN@KZ>`?=)F*WHWKZJMJ<:A,-%T\n> MY&G,TX'-6,VN#F:B%9`JU_6>1!=VQ(M@Q-=`GIN9N4(;9O.O*P65VQ%8'2%W\n> M->MY0^!^/S5S(O0+&<?\"4']\";0-NG>GYC$:AT&GYA&>GV?,:A%,HF'?OM.RN\n> MNK18=G.T:.J-)Q-\"%J0^R8BCQ6`UY3FNS:?80^:4HX)5-=U7`7B7KCB0P2+R\n> M.+RAN7YGA')SA'CD]=6.(.H`_[6<K@E_K\"'$KMH6GN18[F@<)[+<\"5$0(]IO\n> M\"(>L.S#3`8846W3G?$EG@;'P:$C5;,FJ&;L$\"X=5MV\"$+]KDN^;P\"(,8CZD3\n> MM%-\\T#&'7[@-ROC;.VCX<C._PD8\\+*<'\\Q$VC]GXC$)XN*[LC`?P<H/]P5`V\n> M!S:V=,/<KJ9K=/&,X/=O!NMJ]\\5F>V@NF[^9DML(G:UX/\\R]@0O6W_Y`$'RU\n> MO^-YULL)0?Q]3;2*MF.S=OAYM3EH%KQM?V,6X&=L6#ISLUUE/ZH%L<Z,>,\\'\n> MSNRNJDL4:KM(C!!'T[/!K$]RPI3;LZK2U)Z<Y^<EC_L5:ETPPMW+X4I(ZII)\n> MJHR;IB6.1V,ZJ82CF9F[(I_HPM$I?57Q>E4^\\8T3<S^#^K[1:>[;&1532^E\\\n> MQ1/JK7)4V365V3IS:[ILKZE`)D:RIJUU/'/C20I:F><'TZ(0L,K'_JPJ@N04\n> MY,(%PQ8N'&SAT3Z\\51#(R($0[@F7T<SO%7^M\"'45.MP0.AYZ9.2$Z$80)X$)\n> MBNO)'L*::2UJ1\"_,UJ\\%>A.@B1<T;@+(6S4@0GC)X]%Y;PX>,J'._YZO./JR\n> M4DQ#11\\U_ZP.U6YM+BCI)#`K&881BTRN>&:1UPK<JD=CSYR*O-TP(-YS/1]T\n> MR9L^--4U4[>X2&2O?(_;F&@Y'2DSM^^K`T`U><-/5XN-Z?3R4HZ&6^XC>TLW\n> MEJS]>SV,<6*HNME8!U`&T];5_K!3),QK%%QR5S!N8J5O!DOMW\"&\\5WE6P^N?\n> M[\\V%'-QZ@\"L1]Y%CO&0.2SHH@V^K@^XN,\\&K%=K\"K6KW`([_!7VP:/2^63\\>\n> MIJFI9&)F&!BQ(!X50Q$*^38]N_X1SY0;+)Z`#&/G@'_9F]O'7\"I433(\"_<,5\n> M8K;&X4<JBE+):$)_F6]`+LR<FLLSBR=GB=D\"T8W682H[!U#X0P(K/=\\\"_AV?\n> M)3'=J/5N0Q1*`?V):CWD'JP`@KLW!1=TLJBJK90$IP*@0B+BIG[JW>]03'[0\n> M2<HG$\\,P$D.!AJF?\\HWY\\1N&@\\W2L2-)N`B]B21X3N6K+*YJ#6166@%P2+2Y\n> MYO+GD-D`2&IDYE]FH@))6DQIKGS*Y>->+FB]:)H-J8W+,N8.^EC!F'ZP.HW@\n> M[F:3,2U#92XA7#,`0WTN\"S)C_H&[`+Y-*1JQ/.`8:`%JO]UD4G*M/YS%8VT2\n> M(Y0^VR$QH29X'CG#.GD19D_Q.),\\+P2/$^1KJA</+RM58KF<13CEN.)Q/FA2\n> M#3NH_%A2C(M;STC$Y'4S>*1\"CLZ]-\\F4HZ#DNOS!@KL@\\MDY5I2C4Z\"790YP\n> MK+##^;R46:;;>M$^,9-1ZF]<%%H,?ONE:>C!ZTM#Q1[\\AL64T4CDE#_0IT!@\n> M%?HO&*T+GR\\T2[VEOUJ84@;455AC=]$N6\\30\\G%US_Y%:^U9!+SJJ)@0N9#I\n> M?.]-9Y[#'6`2!;.)C0^<5B-J9\"2UO>7S1H_!JE48'>]3NO<CC%B1M!?*+QFY\n> MS&SX_6S7;+&3&IH-!AA.)K+3*O]JB`;O##D'PW6H<(77$.'H(M\\W,U/!E7EK\n> MQ!U#-TSK(FOHS5\":`T6\"H3ECTP9<;=!Z+?S50O@K\\D6@7OSY]1M3P8OOR\\E?\n> MW@Z9_4N3\"78/\\?KQ>&CN.(\"]OC&,UO2]R/+A#>V3([!ER\\B1CO`@+>0@>2C1\n> M.$C*6E/[0,;MM/]D/MT>AA-M_`@^-X!_+ZOJ8#?>^NKRHMI9QF6SFX/NH0\\M\n> M5&X,)`I&(OUL[;#+00=D=V+^%]&_90_(;BX@NR^_??+]]\\]>??'BY9OGWSW_\n> M_Y(&(DXA*!X&^<2L2WHFVH/.X.?AQ$]7OC1?ASUO//#G``1^J>J514C/9'B1\n> MCF\\1\\F%A8RGY#T%$L!4T%NK[V`PN3J>ILA5H!E,T\\XVWO:J.E#>OJ&]K07YN\n> M8;0_M(0W7$#+J0F\"N5\";`VX$VD.1##ZDU^`AS=NWAN,[5.ABEF<XS(:QC3/5\n> M>[3953!5<E4(V'S-7`QW#H9K`>BV'+CR=?:T+%J#6\"BKO_00RDV?HCQ6W/#J\n> M&,O/F]Y6;,Z$G60B'41C(RAZ@)&^]##29>H\\C'Y;2_MR$Z'AL=G&%$S<=R<Q\n> MW\\4?C%-W*5E>R\\R(SVR!UUKR!C*\\&@I2`LXVJ[4(6*T(AY_8E/$D[66UA!VC\n> M3TJ6/,_]JR^2N^\\W6/-XE\"2JXLM9Q?=2Y+875FZCA3\"G.W6L6TQP]H?I_CUK\n> M3\"JW5[I2G^H.%LS5\"GN\\9/$&GY$*)E!\"L5J/JE(9&&+N<R::3E4VVUQ>-&M.\n> MBV`6$@.*>5HV^WU#SB;O!]?3&[\"VM/FHE?W,]-!N8XMZ3M\"D]JT[F0%0/<MI\n> M$/SX@E6Z,+57\"-%3*/2>4(=22^=XV'.>RP3K0JY9U/45<8X!G=KS<K'+QE\"3\n> M$32'!Z;ZO5X'1&7L+#\\P_92T`;(LIKIJ8-@$,[V8;NG&1KF9A?)Y$1@]&CKF\n> M+\\6)JZ80^JB+LI\\6F'<\\;'#<6-!>LKXR8KT!TX#9LIIN3=-,^6+R9A/51:ZJ\n> MBV\\,[T/B'K5*&FI6)_.2+;$%]UPX)[V;;KMW_3L,L/[S:M9PA@Q-`[%%+<@M\n> M<K6:LX1N+D9N6=0*G#CB:DLWCQ+*P\\U6<?S1!7(RO($6<X:K5<@TJWF=IHN4\n> MB+)1%AY.?N\\.241]0]PFL3NL6EMAJ^HFX\"W,'F$+UQDP@\\)Z8OI&8)<P-?O*\n> M-+$^L$Z3]KG<CMR!I>TA\"8?[C;\"<,J7\\J4K-F_5*LBAX=2Y0Y]Z*\"@OM,C;U\n> M&+R@['/2HS_9[Z_,/IXA<0?-'V^X6JZ.]BR2^&9[:/;N6V$9@RPR-(K?/X*6\n> MI+)5OAU:S1TJX7G&=(W!&E*.`>*>A\\E(#B0QD039.U,Y$$<]TK-^'N1``+OR\n> MXN4KLNF0.75J+HJ-<'FE9\"M1';FC%^?=/`I+VA;K/9].]*_,2B.\\FR-J=JPP\n> M!(/W-*@;9NSTW,J8:S>-,F):/NJ:$]_,;)OBAR5?\\W([&HF'N)'-H*#_K,QU\n> M`<+Z`-]OH,->\\#80U9='KPP'.\"K-+(HE0N9/>@2+@EU9\\*R\\\\:3CNJ#$$,M4\n> ME3:QBR1\\B*P>O;+'K+&ZMMI3%G-O\"NJ-MWY[/@3-VDZ7XRIHPK5C/,U\\%F62\n> M1;!S.S2B@?B*?:8B2V45+IB*KO>&V.Q8.J80U826\\4G-VJPI^+&ZFBH]XSI(\n> M6ZB<B=H]B-,PLA\"G5J&9<+?3@*9[Y;-@2E[`5CZ4!=M;V4IO,#J-R:@LV;(3\n> ME&PT6T_M#BK.!\"[UC/=PSQ5NK5LDM8'A6Y%X2RWNH:)A!3&;`GZB2636%`8P\n> MKQ>/L())G)$.A;OT3LVC>J,L5`<;!8:%*>9]9Y;=$.LYI%>]MITPL&0=<'-.\n> MJCTBK%1Q7\"#W3B-KOPR.?FU$X/>\\L9*$=.S1U^\\'CX>T1Q^;SQ];RXB5/$3%\n> MN8IQP./?\\:\\$OQ+F$6R_#66JG02S;-73%D>$UT&N(+`/U`O+R;BM2`O('2YH\n> M'MV-AJO444T5Y'FA<)TWNV#[TC:\\\\1F*XTI[MTMP?)D:;=V=C2:P]=*$>@5U\n> M+O5BT;#^=5<]YL-\"SN5MN7!K^/2]'0(!2U/>&!$;4#7Z`:L[M`V[9K$\\B+W8\n> MS_,CS\"93(;U=9I4<DKHU+VJ)Y$-8.Q(2033,2AK)JJI[6E+9KL\"\\P\"[EA@SE\n> M\";A.0P1NF-[7OK[:22TST!XHE9WBQ4Q^)<EQS!T*%0FF+B_[INZR8N4!<\\GD\n> M^VTG+<*D8?&>$4LW\\V]8GB<DK`I9&3U7';4\"Z[_KS>[_8>_?N]RXCFQQ\\._.\n> M3P%[=5L`FRSF^T&-_#,M2[9F9$E-T?9=<VN:\"P5D(I-$`24`):J:K>\\^)W;$\n> M>64F\"BB15M]U%[OE8F;BO)]QXD3L71NEA`PMM2ON#G2U'2=!G&-^OVS-G5&+\n> ME:!V1*2.ER&W^\"P$KVK>C2!KS]]H$8IO1!8+U5%\\2L/-X]66Y-/!,H=\"%&'.\n> MDB-EM&1M\"311YK1JY/Y!1?6IH]2=NYBJDBRULD7B8*,$3Q;')Z6`6>!5WU_5\n> M\"ZQ*G3?J,+**/!\\LSZRR;N=[/5ZN,`;T0&YYU6TPGF4\\E)Q,;SRH)%J^-O=G\n> M#M5Y*3-GH`NW0ITY7P5(R0Z.[K\"_K\\WT#<\\W0N?$!ANJ'3\"!!AFR\\B663E(Y\n> MOS%5[;#ERZD2=VJ[[<VNF^.*DQFE;B$C8/-V&U<;%[1F7Z7>[P(FOEKATL4T\n> MD98\"]JQ(4&W.92I)U;W>+C@_D>Y%@]^Z;&3<E`.R,:OV\"5A#J@\\_,1]^_D2*\n> M\",S5[T47P2/&T2;B\",0I@#K-**=KNAQ;:6G$JBME:U.GA0/V\"D]C&KBBI\"$2\n> MM%1>5\"?W<M<]>-,4\\&V?<A56+N?XD\":L:=#2^,<)U6_+6V9EXVVI=Q^ZZ<OW\n> MR46$H:V9SIAE[K\"]D8+/7\"62*6LC(PCWO'0']-=ZMZK_WUB9')Y'+0/(I2W)\n> M)M/]+>Q.MKO]8QI/&ZH\"#6RU6(KTJ^G-MD9R5<TKM'!,)$?C6ZZL/=.&1IMY\n> M=#;%VA*ZK4PJ*#,&$Y+1&N1(K]H1C1JC9_'5M62)'09D)5X-[,3R\\%>U$SNH\n> MUKSBX=6XU^\\-VG7(R1?XFEM:5433\\WS/*B;:+%50GKI6%XJ#(&X+S=;7>OL>\n> M>GVG3JZ=V.^(FA=JC7'Z.WV'OY(=44KR4FTTK'99J/$#=2O6)%&&J/6%A.K=\n> M]'O*W0XV/;:1'S4+C5-:'&E`+P)#\\*=.U;C\"@.JU+P^BVP<W$OK.W2BJ+Z<T\n> M8M3J^8PTA#'=7OV>BK*>QCR4E!0*\"0#'PYM;+.DMW2238G2C[\\:;.:X6</32\n> M8IU1/7$CIM`-V=E!=_AF...F\\H<;^&9A\\NO1[>E]U_4<%C6JQ>H?OC<JBL'<\n> M[LGZ*]N&=+&2HC>I+=_P!5UC2^5HTEH[PZAPU[0J\\$E!N\"!-?NA#NVK4SJIA\n> ME`>T]+,=599I*PPM2JG!>7NP8L6.AR$D>=XE7;$B3K0L3<*CW<]6U&=Y%H:A\n> MQ\\C(WR)+L]O8(L/Z''V31[)-&,9*]^01\"-^E:6Q?2>5(OJM!2]64+UJY(P9(\n> M-JLDOLFV=RGB7V*ALC3T@#-)`^>ZGF\\@%<P/6,ZMNMU<C[L#0#@BU>Z%GNBT\n> M(J_I'\\)=<?IH>@%N`D@NH0*QHOTRCE.RBZQW].3LC7*$9)%IX>P_^UG0NFN6\n> MC,H]5%!N^5J6.!IMJR'V:)A`N''I=#MX)Q)_*1$-S]A<:?N]Q[N)(57.99]K\n> MIRNT\\WP6-,,^@L5=(0/'M*&66UNCIB%KQX9[C&P-MK<K+&LW\\P5F+<NEW\"*T\n> MU+=B[SPVLP)=`U*FF>:B@E3HD^6.E\"1O>:)/HA`M]EQT5`O6;BMYSEZ^+^D(\n> MV_D2P8CZ`H<=:J*?A+K5ZM#EL,DF@DN1!MVA+<3/L$0*E7!RISN')C#]#!^Q\n> MULPMV5YI%RN\\7:Q6.\\(.ET&;Z>-9\\-HU(*S[AM%\"YEH?[M1O$.\"6'=FM-C7/\n> M\"[[0@@43+F=$<:V-@&MWWQ#;-ZMT;22T>P!E*5F?]<D:44B#,=Q8!X&U]5JM\n> MA[0FFOG6WF(+GE_C>!8TVD!`;F.:VXTHZK2>EAJ=6AS&CP?>85HU0TD6F^MK\n> MVX\".E4Q#K+*YUN-<K/(,>S1[%6L]!>U)+#31]JUSEDMS-CW$[8LN@?I%S0:6\n> M'5?:U[D9&QC60ORY.H?NL\"]%)*.9+;`U70\"S46TSYQV3>4>=N1CG9FC;&C\n> M!#;\"/2E@3V;+]6%QH2^`5_YY:-(K:8<AC+W\"43>*!O1*[H_%JB?**F/@G\\\\*\n> M.C%^H:W`\\3NXVLVE7\\S'2]=6G$6G^9O:W%OX!X`1\"\\[5O5-52]`KT2B0EM,:\n> MN#N-!4YVK,RPP&(^=@@_1**<3;3=_L*QZ][>:A60'#AH0^-CBS'9&9&W^IX:\n> MH#>6C<8]F73HJ[F4+Y=+8KOVT6!QN.-I>2.'W[OI'UFETHGNGH2[:_3H\"II:\n> MJ-,Q\">6^G8\\NYD3J]//*:`Y(JTZFFYCKEU-AA8?U&8NEG=Z\\R%H&(G(8D@7L\n> M]UMW-74[NV6I6FW,>]-(6/G)^IQG\\KY>\\\\C?VHZSIR'9+'F#)=-?Z>,.DZ&_\n> MJ0I->PA+82MTACSC5M[)2^3<\"Z,?W+![!UWJJ2YJ]*9YZ?2;^;BU7/=05;2V\n> M9?BVJB/#*(B0>B-&R:)8CCU(!]*Z2$=DJS5C,VKMM('-.'\".&;9W1*.F95:D\n> MUNC&;-'_NDNUK33)$[E(D5P6MI/S.XQ6;ZCN_?Z!<?2P%_5%4\"OJ1^_(KOMD\n> M97N\":]/X)V\"QA#&4Y1,E6,E$,$),X,V!.*%FU(;'.K^-=L^AGK)MLM='/Q&3\n> M5G8^O:']=HDVQU&PZ:O2>/:0<8ZA#3^\\@5B[YA0?G9Y!JW3^5P`<0A5H9Q\n> MPWW*(IF9G3+GY`)`EUNW`9\\MO?;EU0^\"@&Q\\-2M`6UY;`M6B*%)]K<0%VBVQ\n> M2^\\LU_TE2_[7]?66_<W8/+:#_+.NR?@]8,<8H_7IM^=,U)L_=OON@/.B%>E:\n> M5T5(8T76CPSVEI?3+\\UMGJR<W<9H?1V!?'AJ-K=L<J],6H3#[0YW`&Q0R,N8\n> M4=M3TP8B9VI--A<E3[P66@YZ&$<QU1(8@CD:]`LMAK_1,T!NXMT-R!J\"#@<A\n> MQFV]U`*<C.':Z\"3-!CE853NN[`*J0/:$4?^#]8X[23OCZ!,X)\\'5Y-@1P9Z)\n> MY\\X@(_N#0$Z@Q:1_;RV'=V/)M;Z#XKZI3='4NHSFK<7AR6SJC3G,]C92YWR)\n> MN4N*7C5>NZ6G*<\"^VKN\\L;5LS0%.:V-%S*8[Y5SV>S1<&:/SI='WJE9[Y&!V\n> MHF\"@:NE;?EHY1;MDV&4/*TK04W9H(;9GR29K2`D%*9;7M]V^%LV)5ZG&%!BI\n> MV-U!2=3=VI2F7_!`MYBIY9*60FJ32-JDTUYO>BLR<Q!]6:7Q\\\"IB9:\\BY)[:\n> M[#P8W*C_-W_[^FO1\\VXQ6G03\\*P5)RP\"'<89#K?C<0)SE<[?%[4CWX@5X,I.\n> M-K/+U-XNTX[H;G3MDK`4MPRK7N5#FYP(W6HY!QJQP%*K$.QU`C/02,#D6R>[\n> MPL]WO$IH>P7N6->,?@OK/Y77$[%8AG);O)Z6VKYQ5&\\M'@;_\\*W%%^X\"T_@+\n> M8B-G<%?WX8A>6&IHF4)#L[2:Q-!Q]Z2\"&?L5SCP[TE&AD#6+YD@.);=:ZZ]P\n> M2\"11YH+MG*@%Z1)!CG\"D)*N7<HB0T;]PC;6T%0HEP?T9YW!A9+%73,OE;,&*\n> M=Y:0%UL2UP^]U9N/`4Y%9,L]8.B+^0#?#K`E+JE*91`G*;G9B)3RVAB;TBW+\n> M5<WG%7(+:,6\\PG4.P`5T:R_F@_YU%1WM>)K#P7)R?&ZQ5[5T%D\\RGN@P'QGN\n> M5ZZRRK\\8,?O$8XS\\9;=@<[R5UDO!\"8LKXS:&[Z6>1^J_@/Z&(\\;BN5BZ??&_\n> MOOC\\;[#.?S&#\"QOF7Q[-XC114Y0\\DR9_`&NU=KR/DFSRA_7Z)_UP1_!\\DS_<\n> M[GZ:Y%E.#W<!\"!;^H!;4R1_VZ@3/,57)P%&M_?F?31I<7NV?8N>[V0?_]M?Y\n> M:M,UW4*`L**+,!,N;+!G$T#+DAZ(];I'JNV\\/[T^J,)<S]79_J?)S>TA>,I`\n> M`^\\,))Q+]ZY^\"]5/%X2OXO[M\\:!3N,@/%QT+%^MPT?WI)7ZXH^FEDW?1.>7+\n> M_'!'T\\MUN!/E*_QP1],K^^V7Y:/AJGXXHK<8:^=!AY3Q>$#T\")&QJ2#1/26,\n> M8ILB%>YX%R<V()7N>,#4!J32'0^8];,^TCIJ\\^ME?:QYBG[6QYH''6/K>[R,\n> MZ!E;WZ,!8_2,K>_Q@%$_ZR.UCN-^UD=J'2?]K(_4.D;/9`@2W=>%,7HFQ^@A\n> M5*3C`=$S!8(DY7T!\"SL#,_-W+&!I`W(!Z.]8P,H&Y`+0W[&U)+0!RU+_'0L8\n> M.0'3(Y4)ZLTRV,]_K(,HRH%XGEUDDP-!JB@Q5'UE()?YN@Y4HC>3=XOU=E^3\n> M3;J;6]U,WM7;1AT<W:^KW>0=X2^0_MC[OE??*4_WXWX^>=?_ME,!1^*O)^\\(\n> M>/2P=3]>3]Y=;W_L?=RIK[OAY\\WDW:9^VZ^$RFQ_V&W?^,5JU==V^];[MEZH\n> MCW1]LZD7\\QO_I]?F)Y*L_-_>FM_>=LM>YJI;\\*OZ5_7==N=58WM0]=B2UZD7\n> MAXI!7>-]7\"*9Y7SOI=^H'FK46[/='+S@#8(//B\\:27OPRULTDQ)3!G4XJ/XV\n> M(\\?[8:/*O[R]F=#_Y$Y1US.83-+)DWBRVZ[7^#V:U#\\MVHF2EB<)#3[\\<'V[\n> MGLR7RW]>8)X7*HK;]<NCI792I@2\\-R]?^K4WY29/!^\"YDW?!OSR]2[A\\%.9?\n> MGO[DO=W%WF_>VUWD_>:_S2?J@PJO_E*]U?)#_U>8:DL2\\XE*165R=V^PGZXF\n> MJEB46G)O:E<351=*[;Y@JD0J-Y[(ZDV50)55Q40>*J9J\\1_QT\\^\\4+%X\",%R\n> M\\JXG$TZ>\"IC4@>ZBOZ>),J'%A&)R=!U@*7&]H++&3-1:*)G9O(((PN0UV<.N\n> M._5/L)F$DZ(BS&*UID[62OB.S;_T?6W@FFG)S>.J,O_N%T'!.\\7Z[8!R#(P&\n> MQ/05Z^>U/!<@!TOLXVG^,H'_DP3BD]2/B)+%-DIVDOF1[DTD%SP68U2:)Q#S\n> MP'<_P##3>K0'(8=F@-3OI634%0](BJCCAIB.$W`6T4H\\O9C0_]]#>I\\GEHX[\n> MT\\]K>:Y`))_;QQ.=B9`)F.TX@62$BG4D2I[:*'EZ*@JXF'4V>.;\"/:@[B0*:\n> M>17?NQ.(0GJ(.GVB$\\X`]DT&`*=0'Q_K2P_\"/8HBP',FU$8$Y)=$,<&51\\3B\n> MA>_Z.0*9(H?G9Y?9CZ+&>9;99,ZBR#PO^](^F_`/I<CT:YT2]T8$:KHH(4Z.\n> M&-P;`/O#=_W,C*P</A[P&5)4A@#5R9Q%*7Q>]I5]-N$?3@Q*(\\3PR.OGTXN@\n> MX9'7SZ<XR8N,V=NOY3D[R2^.;&*>G65J&@_/$KVRCR=6%(341/;Z^73^NLBE\n> M?CY%HYX^N)I>TSAU.QE']QKBCU;G_K4A(ES4=(AG^@MVHB@)F3SG`RY=4:BY\n> MJM]_945::N,=`/K?/!32GTLU!,XW-TO/!GB\\/J=N5FF4;GG.3TX>8.SJ.*5&\n> M^5[+,\\?G,/Q\\8BJ`^4^2X^?312`N)A.'0+ND\"/S,\\3G,PXO@)'%O')/M^<4&\n> M80^`7C4I[TG28S<*R5;G1*%P#\\PERKU<2I`QNADBA#R?:$VL`\")(0X@309J?\n> M643E,&=)TAR4A5Q.XK0HS7%8EN8XIV5I\",$Z'SS_`FF:F-D^D#A-[,,?2IXF\n> MBKCW%*B)L55+U&@9$9GXF255#G.62,U!6=CE)$[+U!R'A6J.<UJH9CQWR0?/\n> MOT\"H!N''!Y*J,<(^L%B-`?=0N7K#3/4Q<[/F?)*D,RX]IF!01X#T'#)UM+`D\n> M18\\<Z=X85:1C8-/GS/$HF5.`LS)/H](D18^G,S?97-O(9\\1X4!X`W*]2CB&/\n> M:WX4\"G2$.),./39T-I+$:0YWCL,<[ASG-(<[2.DU5SR>BS$I\\P2F/P&R14.8\n> M^Y=??_$@:@\"@1.8#V:9^(\"XZIY,,I*U=O5]NKT\\)1]R#$/RX!_$HS8(0Y_4@\n> M!^66Y21.]R#'X1[D.*=[$,*@S@?/OZ`'55X?H@?3O/P@/<CI_((>Q!$_!TL5\n> M/Q=1=G+'(%9J'0<\\\\\"58T/`L\\1%&GD_-7+J>E^3P?$81L+7J.'1%)T7`L\\1'\n> MF%]0!\">)>^.8;,\\MMJ>Q2*ND5\"?'$',G)/A^XI!;3^0[]#%9%(:]Y[\"GI\\$9\n> M#E&O19[G9$YI+,[)7I[=[R;[AV@LN(Q5:@\\/56K.(F#?67,8>;ZWL[Q&C,GK\n> MAYB/U$`DUY8*\\BB^1B'DV9AXFMSG*A_3^5#$:Q;AJQ&9=H00]G3.\"&$_VXP?\n> MTG:D\\N9C!Y2DY3DS0]8ECG'&V091DLI&24YOW1$1NIL867E.A-B6ZK0X$9>)\n> M45I!RCI#`<32&,=!_)-Q_-&4%$9W&2>IT5W&B=51\\C.K2SG\\0'6*J*(ZE63.\n> M4IV>EWWJ/!<?0G4:$V2VZ\"Y1:M%=QKG54?(SJTLY_$!UBJBB.I5DSE*=GI=]\n> MYCR7OUAUBE.`:.%XK)S6G7(<9QR>H3SEL<<:0GX^K2'D?%A[ROEP^_'8Y_@(\n> M<Y[ZE(.R_O3<Z<-QN-@<Y[0\"U89[0%6=YG&K=WHYX*[[!5,[*?)R@N,?47*6\n> MX$XD)F+ZRO>6%>ZZ[&/2IVS.HI`I%_G<Q2F<&MVG\\\\6C\\S7Y);L$&D8T,VC,\n> M\\RY339PT3AX0AQ57'(<55_S,&B0.P\\\\G)?[0),?%YN3XF9/@,&<EA^,<!H<<\n> MY_`L)THH%.1$*<_G)`?V-4D.SV><*%G/('&X0T^>8:38.,-(L?$L1468\\XK-\n> M0;G8G,3I8G,<+C;'.:/8D-AH]Q.I+N>=\\'2<-+%QTA%IM2]GAZ'.!X/AC'PX\n> M#N?#<<[,1Z14Q!$IE>.S9,KIGB&E\\F64E!JZQ3-*C2A<:(YRNM\"5X:^4*%RV\n> MD[=D7$U$D6IR]`I[#E+EQU-*RCR7H-?R?*K,_K),TDO.PC>DG8)E9WPO0J:+\n> MIX9SG_-J('N0JA-1K^74%XXHLT=6YW.RYV?[W<G^0;('#LT\\(/CT>GI`<!P>\n> M$1SG](A`'.E?Q)'^Y?@\\/#C=LX9Q1`3G!<@/Y4R?GSP<<)PJMW&JTU5%%W(^\n> M?`(_G0_'25DTR_6S5@)PGAR&G\\^J*B?'Q>;D\\%S&:'F$D>=S%+XBT*5L>G!2\n> M?\\]QLM3&R4[J[SE.&=DXY9C^?B1.R'L_XO!@QS/4\"&L.(\\^GMAPP(H<X#^+Y\n> MC&*CE-(\\V._/:!Z.P\\W#<<[-)[2BD525G[FI.$QYSJT,G6.-R92QE\\+2Q\\9O\n> M\\GAJK)'UDIR'Z5%B#:)\\>U-O)ON;X<I)=M=16=\"IB0R:HJJB`Q$-?'SEQYAE\n> MY\"(SCSV]QP4(MNA>194CS>.+3+^O@W_Q/U`];?`HJPK[V\\!45$7V`E#UG,CR\n> MEL049\"RR%\\\"\\<7GEC8MS3V0IK[SY_Q1)?'_.'$\"_E:%;9WZ[I\\XZ.!=`WJ(J\n> MOC^R%Z#,G(R+\\OY\\O=_5:'!B\\DL5':DN_TY2]P5&AS?^@I$=\\^2PPU=YS,SC\n> M69OET;$>96DQR<HDQ2C-4B)]++`85_0]C/`<JN<JR?&<T7->]09\\$5%M27//\n> M`Y[;2-YIP'L?Z\"[5>:LR:F1Y&^L_+X`T:?^M.A+9#R!OM*`Z;TEV?V0)H-_R\n> MV'NKCHP`/X!^*WB><I7D[7B==7!N,/U69$Y28Y&]`'0'XKQE!?]6'ING7@\"9\n> M:FHHH*M2M%M6%L<BIRF\"5_CGG&%_U@C,XE@]1[#O,^'?=^S'(=T4\\HS#Z5BF\n> M')&6ZMF%9UG@$?[X8L\\-1;)ZZC0<C7WO@W2E?N-NOF?)]0+((.J]<?\\>'X$2\n> M0+_Q7-5O97YDIY``#UC!SFE0_JZ?L^S#+&-J?!!7$MM#106Q)T6DN\\WH*C)C\n> M4YHL2A/UG244\\K7.!O<5=&5)HU?-]PM]A8D9'%('JJ[T/YBW`O]$V*)DIHPT\n> MJ\"I,]K\"Y<5ZUJHJ>H=8PX=^W05.ZS\\E*M=B3V)E3SJPRH@N93$2[4$U(];W\"\n> M<QZIYR+V&Y0,%]P&U>]Y6G\"#^A_T&Z]49,9W[TKE!R#+\"+M2I5'$O7%LI2)9\n> M]6&]<5:;1+2\"E=@K;?CW[@VZQ<X(,TT;2B9%`OU)+M_U<\\JE0'A^=GLC3GC;\n> MS7BE2A,L3/).O>%]D-[0;VIPV=_&&M0+0,5T(_/^%49'%AL_@'E#C^JW.#W6\n> ME1(@#<_NRK,:M++/)OS[=F52JE4HJ<(8:NNL4B,I3J&A)A5VQ49095JJ[WQ'\n> M002*:JPF?E>JB4Y-4U7<DW1W=<&6%]R3_@<Z9E^P$0.]92S4\\EX^MO)[`>B8\n> M[VPB)?:C.\"^.;1M>`+);4+%4Z^*?)*,NI*NL\\9S]`/J?DM.0?ZKD2,YD,G3!\n> MA]`SA\\%YG4$W$E64<L=(^/>6/:K,7%Z$A;ZFX*_B[!6'WF/_TD3$#CYFT*$9\n> M+<>O)'2X[[+EZ_-+FO%L.2)QN+^3\\>$%U.,8.)B!R3&QDRZ<3.\"S!(93K<\"/\n> MSM?SKW\".S\\!<'6=*5FB2U;1T,]EEE'D,'9?:::NBP$!0@F=E9$U]7Q6GME'(\n> MYC#DXXF<=;QW4O6H?W*LO$F&LR\\?*L<:,,=4J])S):[3=<G*?%*F$2O8.>Q[\n> M[T9*))5,,U+><J8IW=QQINJK;L`L+$8:4*:H;D#2:_%>(=N0^TY7/:8!94O*\n> MLV,[@?L[W2;32^$<%\\OHR-BE.\\\\'-?W)5DCIQ,Q-;\\*^_^ZAUIZH8E-7(M>(\n> MV49!#;)43>RR0C^KS.(PPC,IHB%F]]J?3W.)'->EP>7=]H!\\$,E*O[$4(6]C\n> MG>`%(-WCR-M1L<P+H-_X5*G?^(!Z3V0.H-]$-Z'?CFU<?@#]QHMA&H>L!TJ/\n> MYNP&D#&KWWBTWQ/9\"Y!4N?M6BN[J6&0_@&R2^[email protected]'(O,\"\\X#SGAGC<\"B\n> M`&$95E`3_KW/>%%F<L[\"S.2<1:').0M#,_8S/J[WQ[[H<LS8S]+4'_O^!SFZ\n> MF+?R_K'O!Q!-4?_MV-CW`^@WT:C(V]&Q[P?0;SSVS=NQL>\\'T&\\B\"&39_6/?\n> M#T!WW)Z4<?_8]P.0Y8KSQA+_/9&]`'*^-F_A_6-?5OP'C/VS1B\"93\\G8M^'?\n> M?]V'[W\\6\\ZRKE!B.PYV::41DQ`;]N1*0XE($V#`B9J)B?-WG,7)MUBE^IW7?\n> M^]!;3OFT?<_JZP6@^Z[AVU%5E1_`O.5.Y#@\\IFGU`^BWS$TJSH_M_'X`_5:P\n> MOCJ&3D?>1M=]-X#L?/HMC$]$]@+(TJW?>.F.BV/J.;UTE\\GY2_<9@ZBH4O6<\n> M1/KD@_#OOW3G)F<U<4S.612;G-72;8:O\"C,V?/72K8=OEN:L&]7#U_^@UV*]\n> M(J;.;Z-KH!L@*R)W.96WH\\/7#V#><B?R\\>'K!]!OF9O4\\>'K!]!O/(BD2L='\n> MH!]`&DR_)<6)R%X`67W-6WC_\\-6K[_G#]ZQ!I'9+/7QM^/<^K,=).<ECL5X@\n> M;O$D9)M1^LYGTIB<<;WGN.B?V/D.3(GG<F3GK4G>Z<SN?5`293A\\2X]M9WX`\n> M_98@RSC&0B-O8Y%9020!SCJZG],F_.Q\\MVWROKTA7@64LU@0X;M8%J$'W.>A\n> M89/NC3+5O8&;7'E';[@?1$PP;Y@W>77LVL8/(+U!!;6](6_'>T,\"G-\\;)]J$\n> MG^WWAUA;'9=,XJHD)A]L*G%*ECTIS@-Q4JGO;-L5JZ6]2`N<#>(\\5\\]5U%>I\n> MN%6^-N]1(C<O_@?]%O,Q)X8U19%DQ_9*5MB0T=&Y>^59U<KQG.-9AW__6TE*\n> M-85.,JHR:BPH?^-0K6I%@N-,1!YR!7&ZTW,:DSU`7T',VX=I4/UN&M3_H-^X\n> M045[7,1'[P:]`.3;@[?*F1M%<DS)*\\O<`WKCG#:):907B5PLZO#O?_.2ETHF\n> MA.M%2EKA-&7?2WQGA\\BT3$+_.2E[BXV^><E+T7BQE\"GO:R.$R@>RBQV^94?U\n> M!UX`_9:R/)NS[41Z5/PM(3%(@/,N3\\YH$WYVOMLV.=$;IW&^@#OB.V-Z9&D/\n> M<>ZDY)(!8L(O\\:U'4ED?Y<#!/']@6D7>3^HP5Y4\\?$VD80](B^!RX6U^G\\?_\n> M@QS^R6CSPV#EQ*2I_L!8.70U]J&P<I#6!\\'*X5+=CY43/VB(X'0V!/MX,>*6\n> M?\"(E.$FD`P\"'7;W?S*_K9Y.;S?9!+N]I_H%*QEXG]Y0,_SPH1;@:A5&_;,3+\n> M^*#&SS](,F08DT>#0?'09,AM+L\\'I?EQOIL?#IOM$(;IG/0(<7^0'B7&[)@/\n> M:?02W)CO6TU@@WS`:G)Z]U;S8<V6JVT)`^P#=0/2HRGP@;I!\"<TGTGM8^R&]\n> M#U=?\\G2.RGYJEBWA`4E!0ACIVNW-R[L;M6Y,OOWNX<G1*/:34_OQ@Z0-2!HC\n> M$^&+!T)2D+)G+!U08#XD'0)E&FTFZKTL?E#W<6+#Z;F]82\"XM'A04Q&6E1*!\n> MWG/-@$!6E`.I3,8\\_=_#QCPY_(T4Z_/MYF%2&0I61F.+#R;/0\\M%>^Z(3/8+\n> M!SPG5PTFXT,'/-G]L`?K^PUX>\"2/I//0`0]HF=%FPFBH\\H<G-A!>]8!7\"^J#\n> MI@]YIHZ,5*)8/OQ(?%[/'B8@\\J*5#]IL^\\`VX[%0#+:-83J$DW\\S7]6!QMH.\n> M=NIP_6_J03/!!),_U)NE)IC)RXB-F^;3+T$0P\\1$R47^;/::J(&^(VJ@6E-N\n> MUX8?%*Q*#B-H2XR@LYN)QXH3J_\\\"^AN!%0<4V&#%.4SS>%:\"2/5Z^OE?GG_W\n> M\\HL7LRA%SLQF]MVWW[_\\\\XLOB!$X(JZL+U_,-M-O_SIKIB__\\L6LFWZ']S^_\n> M>/[7OW[QXI/OU??OOOWJFY?JWV^_5\"7[^U=?_(-YO1)#[5K,BD@SCE,6+^KY\n> MNOLO)OIRV,#:Z5_F*NT?F;_P<QQHF8,3]/2:!A;,3>W\\1R8+XG.OY0M;:;ZP\n> M9;WO5AO#JQ5,J48./2D3\"@M9.Y(0]B(E8-^N#T*B\"X;F^6H%@B8BKB;.6Z:B\n> M3SUB82:G/5A67>+'XCZK!W1OQ)#&-*'$(;7F\"MMV<\"K8F`H&FC%,<S<UTWUW\n> MW:WG.ZX$\"E5HJLBVE[+#_?>VK;DVK4Y;>(7?3/](Q&(@<V1*,6*0NEW/]YH_\n> MSZ%=W.Z6-3'CK;>K;@'*QR^__1I\\I:!O&I+\"@K>>QH0AAE;SFEI3.)T/NUO+\n> M^4DDFHMZ)QS/VZO7]>+`R;0!DPA[Q+[/0)&5IR6-DJTJ9E3,U-)T99B#:2!+\n> M953<`<5N:QFYAJW32,NO>@4G9BU#J649U#4U('/@69X]+S*-?$+#*<+\",&GM\n> MA=]SQTW1!4(C9]G5T/V:=NNQDR9&F0DF-.E@Y&ONF(6+$QJR#JHD48Z<RJ'F\n> MRW97:ZKX,A]I3>(V#D>XC4$#C5FRO;W9.X4?&\\N=H6P[T:*!-.EJIU+536JS\n> M'VW22I7=-NG*M&>K:?DD.2(U)4)0(H]F@D@,P_EB`0YQ&N*&+M'E\"#>-Q/SB\n> M$58F6H0+M0QGQ&OY%^9H`P?T<UHZ5`=AW:BG7YKUHE-33)BQOZ)UG'8()@\\-\n> MHCC*>;G[HR;F?5L+#Z^E?*,YX%+68J7JM[0FQU-E=S)I*+T5<YWV)BHQ*ZHY\n> MO7T#]DYA;F/6\\B*G(:*3T52OM&I-1M9((@BUK.O@JZ?66V]5\\6IZ9N978IMN\n> MF%P5O.+U?.>2=6+!.LPIY161QZA@FF>=&#>_V1XT=7F28^539:8Y-UHBIT#8\n> M+GB-4YFN.ZPDZ'^,#1!:@O^=!XRLX]-K*F='8HD0\"\\ZY\\[email protected]*4LX#>&\n> M,;)1DI\\E\\8[-E/I+O9-]$1TUWVB2/A/:$BFJC4IM\\-ZGL:HVEG<>Q)4\\I$%S\n> M^+;M%JV0KX-T%K_M9PM-!WI[?07J7&P:-W-P!<M7\\!).J0=V4]*+\">FUWF'<\n> MU6BEYRQH%R^Q\"JLLL.>\"[7=?@Q+:C,]E#9(_O5]C'N2AWBZD*9II=!&!-)><\n> MH\\D^^4[2HOT!5'_[S?8QE1C5NJ1BFN`EL1:\"H7(EV>-[\"2Y,7A-`O-I2(I]R\n> M$:I,\".B?LWS&(V$U77=OK,3`=)<8$#VJ]ICY+@.?OO3JSE\"-W^Y)E]H2@_.J\n> MWO,^AAL<=^D-KO3:A3*H3MF#CQ&1.$I4]O>^P(VR`U'GKL/8V-_M#_6U<%^'\n> M^;U9K8D'=_=T>W-0LL9_J5P1BV@&[HG58QYE0N2D(A;U#?'U=H8`6RWWZS56\n> M,2Q;:M.J98:^IC6JUGS,9M%KU0#&NN3ON=P2Q+!=[X7^7>>@Y69DRE(+%Z@J\n> M:<6@@,(3JM=16A)VVYM=QVLVRX-\",5UD!<]E2*+3[[@KNNGW*`#D7E*/6F%5\n> M]U8M9>RF5ZK4M]UZ2?L?AMR<ESD4U&,\"E75&^+R'$@IU<7]+KX6&M-N0O,/?\n> M4/(J0@=<;6\\/E@G9+ET0$[@7YYJ[O*/FJ5E:7C+S]>6TNR#>8V;1Q1:'55EX\n> MJ>?84=0A7G465?-0[X6]52:8WN`[D-$;3F,N85%QC[@LW_5ZR0L&XG_NK*Z7\n> MLPM>-GEMN>85>=DU*N-&K:E$>LIY3WJM*_*)=,1!Y8>EVYZS.@H\"45[X<D.,\n> M7FJZUV,'@J:_UT%2&!!0&S9<73%()ZJEE90@:R$OST)CO*\\/PO9LI!@<`YBA\n> MW23():PJ9E$>W2V3,+#G+TH,0+\"UPWWM\\?72--S2,*+-@Z!G:T/J#0;UVIY3\n> M_.Q0ASW+)1MS\\J(L<,1=X3#(#,0933_,UQH2:\\-$L^C\"G5H%UG=<=YD?.VB_\n> M7*9S'F!ZWV;F\\1T$KGKE!@S4#!9N9,TO>V3;9\";BJ'0.=GP\"OS7\\S%:2IE6%\n> MJ)E!SAN#7_LYTS.K$0T^8`R\"Z_FR9LI?D-KZ>Z0[4?26::3@5N2A/788U1RJ\n> M)C0D:10V5-_N:!T2;&FF#BP1'9SR#WN.1;3-TA[@W*(V0;^L/%#=,GIE:T?%\n> M$I8)94XQB[!E^[839*1T@3XQZ#;BDR)U0$_UD:C_`OH;CQ`\"%T((3*H,G'-?\n> M?/'\\ZZ_^O\\]G3TB<?/G5M]_,HE1-E6^_5*.2`K73OZ@?J5?__M4W?^;M[?.O\n> MGZN,__;]%[,\\EE-_GJBS1T**VW1()9P+E3\"9SS\"5<\"I4PEF6,)5P7#Z,2ICV\n> ME%<L['^D%/Y(*?R14MC/^B.E\\$=*X?]Q2F$!=R!BX8^4PA\\IA3]2\"G^D%/Y(\n> M*?QK40H#^+\"(#:58(:;_82;<\"0A0G46#1@BVG!0>QX@<>@\"B9`W\",6`G5PC*\n> M:Z0C(\\!9F<=%:NI!CZ<S-]E<V\\AGQ'A8'D\"!=.\"Z+5JW4$A1@//XJRJ#6(G'\n> M,RBD=#;7-O(9,1Z4!\\HO%%*:B^B,&-R(B%'$!E\\;G-+2)O(\\;)13-M$T<,.^\n> M&<*+EP_C=2*+E!%RH5V]%I-2:,6]%'L\\<&&H:9&8HY`MWO&,R;7F,/)\\DHHN\n> M9$=!D]P1N-)!G\"JQ<:K3V,5I8O-)$U.\\!W5!'`&%=6!W]D!F)E#I(:WW8V9\"\n> M.J\")?2`S$P\";-3E9F)D^Q+,TC&#XG].'G!RW+2=WN@\\Y#O<AQSG=AZ``D7SP\n> M_$OZ$.8\\[]^';)K]_GW(/'&_H`_)10)+`K//XW%M>9+7'\"`^!VD]+TQ2>9&=\n> M@Y;LT#&;R&?$>%`>3$G,)/+T\"#3DDPN#C@%&[--4I:5>L!'A]#Y)$7AC-=Q.\n> MIR,`AEEBX/F,*%EBHYQ7\\8AQ$CD*8R\">BB(HT12#=U!YU(OXN6LXND>RI^?Q\n> M>3F,X[3D.>UBV;0XQU-15C[2;U1.2JYPGI1\"=\\5?`40>$<^=^VAP*[3';&QX\n> M>K$(C1%FC>&[GLH7C\\[7\\_$RCG%-Q3K'.$UUVOR5R?3(.<A][->4_'4-B7%Q\n> M;DU/YXM'Y^LOJ*DF3BU2@PB/YS.HA(22F.,(`PP]0_85-B1Y/C7@031<\"!=N\n> M>E81.$YF16PI`IXE6X0YKP@PQ`;7IV:#R4[SMR%.&=LXY3D$;NP<J?=I[=1(\n> MSX1YH$4%>3Y'5(!CATXN/[WM(TX5VCC5R>6%JSHDR\\$S>GW-8>3Y5+')]T%:\n> M#L\\2[:24*D6`E\"I%P+/$1YCSBL#)998F_(P.YS@R)8OPK`ZGC0\"4G[+*XGG-\n> MS]+'L7T\\2166Z9[CQ$[W7)F9*/0H^0QB'/<?C_)\";17@K(CIT!>EPM!(6T8\"\n> M--J8KN&B%&B]<5S&%*:/Z.Z\"Y5]KF&C&&P6>N_,N#MU40PMC$L7'8'N)#.O\"\n> M`%N=Y?]]3IVB-&<;N+43_KWA#53[ZYP3HK60G!/:%B1G]=VT9I(G8ZTIZ\"RZ\n> M-8ET[R*TK>F]\"X**M\":MAGA+CV$;>`%HCET8W%>2]2_N@RLCTJ6'=<59#9)6\n> MIBML^/?O\"@)?9;#0/$\\TLUO&D*P3`3MT'P>0J<R1D94RIIF$@U\\M*0>_:^R>\n> MTD%328\\!MWF_D\\1PH<$MD[\"\\'S*549S.ATP]W0IX=+Y^`,C4-(S)FP+`UW3D\n> M*QC:#]ZK9:2W%;KXIT<U]XJPU_0:O%=#II:,]UUH=`_W77!\\N>D%LNPXXF<D\n> M2!1GHQ><K$L85BH_U$6'?6]0CT3U1,P`%G&<QR1W\\DZBEHXXD5TERHU0Q.9Y\n> M:0\\`B]SE+T*]-E^;I8+?:2'Q/@BXDO\\F8W44N=D-H-\\8Z4,@H>7M.&:T#N[_\n> MPV#5I'8Z$CD68.KS`6_.:=\"$&C0!+)\\-__[PP97).2']?*S5O8G).<ERTY4)\n> MJ5J&74E$EVY7&@Q0W97^!PW!ZKT=[TH_@'[CWA``[>-=Z0?0.'/Z+;V_*V5'\n> M>4!7GM6@>6ZZTH9__UFI!DX2LX1*N$1JK2QYX*1*ZJY2GI6%^A[S<TZ[2I\\U\n> M0F:E-/:U!CV3=\\Q*]T-<N8!>^BT_1IOD!Y`W53@&?0,\"6Q(=W9B\\`((KE40\\\n> MVQ)@[0LVTO%9*2+%>;/RC`9-DI(L#\"%[FO#O/2N+R.2L!I')60T<D[.:E:8K\n> M$_)3'G:ES$K3E<(B8;O2_R\"(E[VWHUWI!Y`WZ4K!@CS>E7X`#:@?\"9!V?G]7\n> MRJQ\\0%>>U:!Y:;K2AO\\@\"ZR2H,UZD\":E60_2.#<+K)S$L<\"F:3*.=2IHRG3K\n> MQ2PAH9$YO`\\23I9*$5<$U7D4==D-(&NDH#S)&BG]>[PW'@\"Y=%:;T!J9QH59\n> M(Q'^O7L#:G90B\"=E0NS@`-[*B2NAR%ER(J83L&0DH`\\?41OB$%*)V\"U\\28F&\n> M>_<_:&(D_<8XDN&QXX\\?P%!6\\.&UN!^#TOL]]J0>/DRE1Q'ZW-_CV!%4$T;3\n> M\"N^AG;DP`.%G=?VI#HC+@J0'3$(=]GU[/5=GMS2#EBI72W2:0]62E3$QI%!.\n> M8(UGT9F09$5)XF*U,F)[).B*F0O@#J16YUU>>#KD3)YP%.Q,X/P%.>V,!CQ=\n> M%ZI`EM*\\,6'?FU>/+GFBLB2\"F0C7=B5`R:.\"CL,,\"1V5I\"LH(S\"<%:1#()9R\n> MKQ4KX<'1&)4\"HE_I-<S_H,&3A5:/WV2E&B5D=`-H?I<B==^2_\"B;HQM`WH0M\n> M1%#'A>ELE&+$.[><PQ)W3GOFI':HPA3/.OP'P)S+H#/MWU#^!]P?'H+P`%2E\n> M`985.P0^,)VA'02#PZT?\"`['#KP#8)L'HYL5^C[=2\\;U5WIH:L4`5TB<A!BU\n> M!9Y`#TB3&[email protected][U?R'2'&A\\H5=^;Z`S&!M4(U8I`G3&WJL/LHU(LWM3\n> MW#\\,U`V\\?1^FKG3CSZ3*QR!Y'HA]0PDR1?(Q'*H'`:?0MLY62^\\'5T:C8P0>\n> M2[RS'C0ZC*W#R)1_<$HP@>KA^JQ6OWB<5>%8BK^X-Y'@2&5_*:H8[G)'NN&A\n> MO1EE8V.,['<?O`D`/=!/Y\\_'5LJ'8NID2IB)2G5&LY@Z$9!MBF<S=8H#*L!_\n> MN%ZVM><&NB0WT$[<0%W'_]4HZ$X\\4UN^:N(JLA[!C5H3V,VZG7[/;M;D@!:3\n> M`Y;Q\"8X(),)X9)-/XE+[9+\\6#V;QB+R>OZG9GXYP$KKIHIUO5L:E-K\"^?$/?\n> M84#A`&VAX6>W0B@X633<D=?=!FY^VS6<_1C)1A`9*.FK.6/CT$Z]U[@&KPF)\n> M8Z$+!F>_'SO42(,][*C@XG8Y\\W$4&KC$M[8)#G\"0U?`;]5Z#0S0Z5[2A.E(&\n> M=U/T-7Q6E]JIV_@,2A'(6YG1!#IX%TJ9D4C*;K(;P\"N19_NWQK,=O^=AR1WU\n> MC[;>6`](U4.JKD%S/ZZ$@3/H!LVZ)+@1<K\\D9U;3#I>NOR_59DDN?/``!0Z!\n> MU,SZ\"*/C\"%+X3EQWD_F*G:#)5YC]M8VOL(9GT'TOSOSDAKG8U7/RS6XU@@7]\n> MIL>/.\")ON=^WY+0-3]K6`%0TY(_[_,]_YKZF'R]F2:8AGJ@=B[@<<T<VGJCL\n> MWCS?P?M;7*_'&K=U03NL[_,0=\\>%O-#.MLBE4:OZ\\C$#,@7L8ZP[F_\";[]@[\n> MG=V>%]N=&K$W6]4/FY7G3PP0(-L*JMV62_23>\"BW[`;?^`[<&(3S-?ON&F=W\n> M\\5CF++7[-0I4I@4C6LRYL[FQ7*P/;O:5T^QP7&U5=G%HG?'A*KNI:RHC(2;=\n> MN=,$ZYA@,[#+/X$@O-EH-!FIPIBC.TI9`;1H\"'W%>\"IZSK?L31RX^#*M`[\\%\n> M?\"K&T4E(M9NI-&GX=+J2MLEE>ABL\",PF3)`Y56`W_?.+[QC=A-LDCK2W.XW)\n> M'F1-;>&PVDD/``'[`2-:44T30I&[0SL2(`6!4NVW.YZ?!X$/$SB#`Q604.$X\n> MDZ`F_)C.]II7PJ,5ZH\"_HX)__^V+E^[DBMW)E9#-\"(8)@M$>AV%B<E>-KB8<\n> M5I0#EGX\"AL)<(/`30JWIIF_48B-M@906/*W-(+^<=@TFOEZ7>+78$2*&[>)#\n> MV^TUDA\"M34I2H!93'4S[+=6&0,J`AH)<!\"7;C%_\"<\"#XD.ESWDG083U(B[TS\n> MJ%8H06OG4F/G$L`*N!ZVX=!>,;#:1\"18JJV[I)YDE(*]\"X9!JR`7CA=(LS3U\n> MQ($1`#$7L:<AZ8`'MB`#`*BH&P,!,&MK.]<=%C13#:!P!&%!+S+7<RJ-@;Q0\n> MY40/MQK(A#$['%P'Y(4V88B`^=76`7_:[C`Y>4S\"MQ[H/8ROX8%Q:(&4BV.D\n> M(X-L,(;7T&T6Z]NE`>3CN=NO'H:@X)[POB\\X`33Y##:1!^B@%^^;[;X35#-5\n> MH$!7V<4M^$15BH:3@(;5+HJ&`+>Y0'P$(\\1^:M)T`[\"!5/T7T-]DB+.8_O-Q\n> M%I,R\"V*BV1Q@#J058PY$N<8<R!/&'\"#,`F`.)&2W\\`#,`34W/D(.?(0<^`@Y\n> ML/@(.7!B4'R$'/B?@1S`[6&>9!?91\\B!CY`#'R$'/D(.?(0<^`@Y\\'\\%Y,!'\n> MU_./KN<?7<__YUW/<9LHT!_<;9GN03%=SJ)SH3\\*XX:,[email protected],)\\?[8I@R\n> MGIG'1[?ACV[#']V&/[H-?W0;?H#;,-F'/`AZ(S'2WYG0&TF4/0QZ@R(\\%'JC\n> MCSF2GT8$>3A,\"=45U)?7TI/5R5%!4;(X-U&R^+2H2@5A>BY^SO*3O8B22124\n> MK#@O\"AV&=)0,OH]<2LX1A>?'4^>.*\"?'1P;V@EE;>;)I$\">/<A,'SR?B5(44\n> MZ9J?\\^RT8SS*QG&X;*?;DSP=)GF!^L!&',]GQ\"G\"W,3!\\SGY9+'-Y]SZI&%H\n> MZI.&3.N>FOC<MMDY78<%7A>!G@MFO#Y=[`1%X&(G(3-FER8^-P<_G]XECD#:\n> MR#Q.(O,XEM1'3(%^G3YB\"OS?C2GPT2_[XJ-?]D>_[/_#_+(_.O-^=.:=?'3F\n> M_?#.O!\\!#_[O`3Q(<H,#E)8&\\0=?!0<HSKS'4VA(.00%BX;DOI.VW815[6?!\n> M>D:=HYW?1?(1-*0HM\"A`8W,IBVS@LZ;2R59(<O_KAT!#BHN\"%JH2>TH>J><\"\n> M:$%QJIXSS%FUM5`]&(B63A%)$2=^^\\O*8B\"1^-6TO_<N+V7$RUK*\\GAX/R:2\n> M!\"!5X85QJ)>WXXZ]?@!YDRFA%BK,(.ZFXSG+$GH6'-,YC1EG%;43?]?AWWLQ\n> M)#^7C$ZJ='A2&65J'Z`YFN7T#-2G.%>K;I8F<+DODHC\"]+8F#13&AD<JJ3!V\n> MWFDQ]#[(DJ3?LA208W%^Y#`ETX=4$^<N2>=4*U-[?9:@06WX]_9YC\\ES+\\G@\n> MHTVZQC0%+$0$L'KZKI\\S5MLC?&:.<3V\\\"#ZK$DP;EFUY!TZ;^R'*\"D=1$&4`\n> M\"A/MP2A>A!M`P\":TTSM/D22*CX%->`&DH#(;Y$TDE;'($B`[W^G]K`;-G>?*\n> M/+_GW%##5,U]Z.UB=237BE#^7O!=5`H%J?.<5_VNY+E!_H[<E67JO$/EXWZ0\n> MF6+>H/-)B^HHZIX;@%1U%ZRQI<AQP;\\=`S^0>2@!SII8Y[0)/SO?;9N\\CQ!=\n> MJ\"F:L5:XJ$JM>>7OC&>1E-#(.L]9OS=D]\\T*O>/$I?-.6X[W010]^BV#.*86\n> MD2.]X0<0J9@*2I'5>0*_'>L-D<`EP'E\"]!EMPL_.]^P#]$861>4D9_*-C/I8\n> M],/\\G6^7U'H:^L]IOS<$X43)^7(ZY9$H[W0Z]3X(J(=YP_#.LV.]X0<0`!$J\n> M*/:>C'\\[UANR'4F`<WKCK#;A9^=[^@%Z(PU3E5*$[4X)7=4DYZV,41QC;'=I\n> ME,23/,$NKO:.B,+TU3X\\-_)0=G$9_/).<\\/[(.-5O_'AD^Y=[A6+2+M_KEAT\n> M3K4(Z4>=EQFI4H?_`'@G'W$M_N_$M?@(2/%_$2`%_(>R/IS0!`<OBV?SD!1Q\n> MDSVH)KDA/Z1#*P+*2`=&==^QT_GDL'T8Q`Z2*P>]25Z_$SA5/CBQ?`!39*&3\n> M)MOF(0FFHXQ4/]S\\,BPF2FV(#P4W^`>.\"[;Z^2\"%HN1@H^0GMZZ;PV%7/ZCY\n> [email protected],,IO<8LG3)J>2(#YDD*,KRZH/4F!2*:O1^L,Y`>GGX0<J&R9\"G97^\\\n> M:?_T!PU<&##U`9/>IQ=H+HS@DOVBBJ)#BP_6\"985H@?A\\^*[AZ%,Y=D'0)G\"\n> M;6:6#;N18!&\\=!Z*)$2ZH[C*Q)5;D`/(K;^T2$+?.9`I2W);MT@!G4%XZ0,'\n> MK0`<U/,BI_-(0'_37Y.R/F,TF21E0)PO-\"0(0R94#EY1PA5V<4/@IW^[KPU\\\n> M@@>\"LYO6*J\"2K2X)?Z81C)&64R\",`)5(S2@!'IS)_F+V9OI\\/UM.U?-BUUT!\n> MX:&>[]8=X=9,NP.C3)2J=)S[&\\$XN)X#\"L@4`B@%5*7O=MO%-T!8\"!S@!,!O\n> M+.;K-3+8J?+L]@;3@>!:&%;\"+V\\;3/>W5S0%]PPLD::5@);4%LL#.\"+3_:&^\n> M(1P@P4&J=VIP7&M@GZ#M-Q&@>IX!,2++8FKZ[?0RC`H`?YA^(*`,`7@Q`!@\"\n> MJ>*#R:P,ADPC^`P\"DC!H<=T6,D35P6T)P\"D-LM-XN`XM8\\(\\F25Q`F>INZF/\n> M#^/WP,%B/IC\\;Z9JSFQ5#@:LP0+)-!I%IITV]4&P<@QF\"3*WR0L\"\"UKN@@ND\n> M3K=WT\\_GZ\\7M>@YH\"-TN#MX#0T.H-;X=@QNJ'10AP2EJITL:8HT:;P`%4DG2\n> M(KP*;,&?S4C-7!09SZ3+.,[]CBMY`E'W[-!]U`;!6\"/PP/VIV40&3,H,5,)9\n> M.;2\"3=$8:(J@7S\\](^>')[-4G?Y+H-A8%(S90N.E#!N@&8)9M=/YC3K2:[\"5\n> M5D:'QB)3#<$H,Y!Q-_,U8;X<&+N)FPQEJ%`&3)4CB!R!KH*/K@)8E#F>:.\"L\n> ML%I?H,6K(@_&6QQ84OMZL=V@U$=&'&/MJ,:.36^O_,;>WAXL_EKKEH\"1L)94\n> MN2`*DUS53K<(5NE>F]32)IU:K]:T,<P7B^UN*2-JBU;485?3W>V:UABJ8T3:\n> MHB/#*@G,L.)X_BAJ.O6UD8X\\S-^8Q=H`4Q%2$,H%]**&UPT,/IZ*JG[M1#=1\n> M([!RG6Z&)USYJ\"`4+6X37CH:3K_;MW9_&,XV!YJ-9F^4A+'>B>Q>F9@:OZBI\n> M9]2ZJQK7WW$P!C<U`<?-=W=Z132P3Q8\\ZU[L-E6&(*(K:@;@^G%E<*T$(%!W\n> M%K!B#*;53FWA-.TL5%Q#`WFMPNA9N>=IB7'4>J.?BBE0.(P7\"TRX?<U[32WS\n> M<!I?9!<9-U(:QXRK]L5/AYIARAHH,9Q:\"4K?Y8S7I2B+\\V,CB\"##]*I$\\\\1.\n> MDJ\"WH=LEB8&*-)`7%]Y=EAGCB,?;*M#MM3C<8FDP2Q6M\"E%6TK*@&@?`3Q+4\n> M;:`ZD,YHCZY&&I/.G9TR.5<$T[:GQNP..CR/'3W9U%Q$0?*4\"@(?XB7!,=Y>\n> M\\XP`AB'U;;>]W=-22X-03\\TB+HXU;`9(/JP_K=NNR\\#;8=S51Q8&M67NZCE+\n> M5O.-OZY\"(:2&F&I6=UMJ24-D.2EXFJ,&#+^2@P$%4YD?7$CT2^@M)T%](\n> MEEB=NV5M025W]?YV?9!1;@<+P#4-C*$I?..N-#$7OHH%TE!UT6LJ,[;PVF!<\n> M^O.-%XPXS`RTED#^.8M5X*Y?N@)8P@@TJZ6AJU*]`JI8IR9=?;%2@G:H)\"P2\n> M>[\"$G[-JH!XT6FCYB\",5[\\X\"=`GX&8F3*@CU!F8XK0XK9W60@<:MI#:_P-W]\n> MW,G`>Q($78)PPXHA;1&KE?BN+U7J#M9HCYW>1W?S.XPFP&SQLD8RCFIMQG1M\n> M1@`61^4E7H2?<.W)2HM&H&H6/B,=&'R1<3HU'E=?@'71$(,1\"58+>*%&P\"/P\n> M+*Y\"S<!S[;3=4A3:0DG,GNOUG*XJ8P'P4SD!1(^%6CZDN>(]+UXT,OVY=<8`\n> M.!!8'@1=*W.V0TD7!4HA-`^Q$^UY9V5!V51)`)!'`(ED*G#L6(!J.<>/QC]^\n> M!/XYQN\"F$J3:>KN]82%+T&)%X%)UP,F%,1WW!VKGP$$2W-6'VQT=%E\"K#.B-\n> M5W<&*$^=DQA+CYKFHI:9I5)&89%1S;($9U2+V,\\9.?F0TH-W27V\",')OPQ.`\n> M6@A\"`2&V<G'R1(1-0@\".,5PH\\<W623D8)FVV8$&5_.9O7W_-_:M10QM>4TWE\n> MFT!CMKI`=#5V'C64N#!&\\FW,C%5K<4SWML.U^#FWT(:!;!O!I&OY%*25\"Z9U\n> M`].\\L:Q74<9'#?S.$MO(2;&U:'ZTOAO-DD:,I34Y+C.2ZU2%:W,:P^EU-7)>\n> M)DA9M:RA%?HG;ZL#P/%%MY(=(BSAXT0D0N4DKB(M5/)>FQ#PT;\"YOB2$W1TK\n> M0$[.VJ9WN+O4*R!J1N,@T)B!6%Y6!@5RI;8+M7#45!'('WI7H[(F858%=^XF\n> M7SL3O+U7I`ZT9L0_IQJ)422BFYO=]F;7<27<+6$/$3RA2]([%^?1PA9B`&DL\n> MRCCQUA\"&E\\6\\P8;W%I.@IF@T=P\":ZT_'D2P(;!3S9N4BO7:JQ8#JK$I.@CDO\n> M;)@8?44`:Q,$*%-:$(N\\UXB\",HGRC!U<!WJ$L<)HT5$?N49;G4N(,@E(IMO@\n> M%^KT_(_9>DI`HE@<@ZD!SE6C97FGY<S%E@X@*-D104+D*Q&2,*R<#61XLI.&\n> M2L,XD&VVX6%,\\]8*.1;VF?:ZM@_,>>]N)IMM3S^9J_\\\"^IL-42[S7P'E,LVT\n> MFK*W_7U/\\CK+KKPR:@A/3[3A2;0!0NM;K[_IM.6*L'L6`8X>67VUC]/2-++A\n> M=E@1TFP[9Z&')M+5?/&&IE\\(M;Z(JZ_-*8F12VF.+%G%;`6!SD-G;?A\\H/+G\n> MW6<O,IM,3F>G3:%]X'U$SSY:80'IWB`2BI-G4-_)EH<2OVWKC5EUS#:F/^BP\n> M'4Y42S6V%[PXXG!EL^J.-U\\S!N=-XRF+=</M9S&!]E.[26D%F94V^H2@\\U^J\n> MZ;=EN&O5%KO.#/#^>M7J-DE)^CA('\"DH3;GU6F_*>X.3[LP&:J:L3\"<#P3J2\n> M\\=>R\"&..PJ([HGRYEQC<7PGK@=[2-<QL[0O7.+BA#.J0U*U9U]TZ?3UGY2UD\n> M$DS`-.6CAI9-%JXH(L=P+<>MK`2DEVLC7(G\\+:V\"H3HI0QQL6UP_Q+.,6OTO\n> MO,J9@_IJ^A>LL*SO%OJ&CG5H7VG=2XW>#2J-JD_RZANB%.@6-&Z6]=S#V^UC\n> M=3<,IG]`L5F0<2<G(5[37`:0^=2#KS>D#HTC\\4,H9PA;5+.BX_!TP46O#1)Z\n> M3Y=.S?[##5B!K#:==><>'GL/S%\\(#\"!>(W3@;9\\:2IKD+@CYM`%_PVT,#B`R\n> M<&*Y,2I$NM$B/,YFCAI&CR2UJ,WMAJ-''-;\\>YJJAZ4]L=5=Z5NMWBV`K`1<\n> M3*@=S\"ZO*I*KTG[E:72\\\\K&`I`=V:]@*<#_CB,FF'>=OL\"BI42X$$;\"*48=F\n> M&MMONWVM)PMAX*O!1IZ[)+BN-E#F+%RQ@]ODIX.I@SE\"7M4.B#>)A#P3HCBG\n> MQ+[:\\*JTW2U)_<)39J!A=(;72@\\K9L/0BB^U6!T,GXC6%[>NK*(6>!%6KN=+\n> M@YA/R:,X\">GR_-,>.$[V![55TT6$VBLC<B$[?E;$\\6\"^V]<['+A(?]E8O1YG\n> M'DROMTM-LZ%D]76W:@_V7DQK1&FYNI'M3]1<=`\"GX]R&9YL[[N\",GU904VD\"\n> MEHEW;PJA1UJPU2VH=<0I^>]'1R_'8JX9;S?,*X*S$A.K0,YJ<<Q%#4DS.+\\Y\n> M6`1Y]#]NE$1^HZP!\"F^(33ISAJFM2&?ZFVD#M+:\"ND'4@_<45A@YGFX-\"<F*\n> M(?A%H!0EJLWMR#0.>%EJ1HHEC;@RAP\\]W_?4'X&H#;5Z]!C/!+8T7MYJ69/-\n> M\\M8&',1=R`TG#8YX(B=)'U9A-'(*.7;1;%3.IN.<D5E3F3M97P*S`#H2M-M@\n> M*W_=$]G@#:^QL5JUW!L80IA_PD.VRNE><6]V)4<HZIPI7_>)\"USYT#N$.WP%\n> M(+IP+F/=M4K+I.ZI1'59',81WP&,'\"?L#2J+BR0)[]4^NV-!S4&V'RS**RMM\n> M^!0RO!+&$:D8:24T<E+ORKL6P>XU'20;7EA71J\"U.=MU#TM<8^_:S3%E2A>'\n> M:U8]T0(E&BLE<\\Z[-8O3<5QFEGAI-?V.E[1.'0HH!HJ<E+F5.TB_,[^^GM-V\n> MA@LV*HMIV=99%EG/S11\"FJ\\`>FU*X.*.Y]NR6\\Z8IXG$M!]N.U9C]:F:6G^<\n> MH>1I2HLXEP&GI7:.X_&:#Y!FI*LAJ$I4V[OXOA!CIK>98=BEW)7>G0\"-F0\"M\n> M1WS5\\\"IN\\V\"DAS#G82:9-69+\"_@JICG'\"H$CBWC8!*XU`A./U&LU;KX_7!]X\n> M,S!L&)LE\"X_PN%4##P>\"MD=E(QWF#M.\\U$PR1%XQ<=2P5C?,2CU]ZMH[F@%/\n> M20:=&!AL-FS#@#*L;!D&@\\9N@>CH(M6ZO^6)Q6A,'=Y8=7@[:$HT%M^OBR)[\n> MOI`=&BOS?GY-RY-SWD+KE&'*IC-+-4V=065+[K#D=%:L&2]Z.[)]JS,D+TJK\n> M_@;N,Q5A`-UL]_N.-ETN7)Z\"/&:!:(OIU9;W6K*XH6V&E=%FV*[X.L6Y8D*E\n> M6S//ES+'.I?2H]=C<DVXD_5I!ST?%Z?\"F93T.*I-KDB+ILKP?+F$BDKM$J1T\n> M5GVV>(,E3S9'4+:1=G-G;J6&\\[;QI,,A!1O-+!5[7<]Q5Z*/[''%A]'39AOU\n> M87'!$CC5;.$8(-4!IPW*JCDKM7&$G&DNE.[:K+=VC`\\$%4QE+,X\\[Y(H3&3>\n> M00/AWL=;VCI7T)55+NAIY?3EOKNYN,J]/I$,]WW`9<BI##1IUC4N!29?OID^\n> MFY$BZID*_0R*LF#Z=+];/*6Y7&^63WDP/)VK'KW[K_IB,=N*\",<'LD/]F!=5\n> M=.62MU6C*-ZIU4OOH[B24*D=Y$8/)8+*5=N],$W/@J7.A9P)>;<6/AO11\"N1\n> M=;X4DPQIC[?M=NW1H9&\"3<DL7^@],R*+/V8E(YJJV]V\"HI$KCI0$^E5LRL+`\n> MI:;(&QX1+I?6)__^B;XX,*?8*ZKDJMOHNT)[C$7ZANK+W\\.YVZD>?G&$+PJE\n> M2E(2*G`9*\\5I=&FP?7SRFT^D(^EPJ(YL*]CD0)?(\"U=/\"D&S=KAE5!_VLB/V\n> MV/P0G68`=9>OY$)\\%(UYAQRAMQ561MOI@?3Z_&`DXNNMG`@18?^FN[FIEX^1\n> M/,S&I`B';C&G8^=R2TI'O;2IGJ]WHE/C03#0Q1;JOX#^YK^FK6A!%PBC2MB=\n> MW3F^0NEI*X>2#+=?51J(`-:)K&,6JII6>M5:>[email protected]);RU)\n> MG8S!R@W&XQ6&3+KQG265Q^P<=]1))MJ\\/<0.WD5QGKGE*XIU]Z8V]VR]H>52\n> M?*D%<KV^W:NZ'VJQ_TMS(>^30N'&&EH+U?DQ+OTR.A'JO-OI3S0N[A`YPS%,\n> MLW=%GZ+UR#/KCM99TGB^W;*BVU2SISAHS26'92;4!&U6#\"#JSI=8C;393RTZ\n> M5MDB]D:,T_VJ4N<+_@*$H=?.(<BAT3LBVP0#!9->4TY(B5JTD=W`L=*:$/OY\n> MG3?R_D$;-\\?&T(.$.M^#3G53&W.&^V04D2!I8?CJP\",5,]A(*+([46`4HJAD\n> M85]HL<,QFI$,(<#H//:/57FT<,-#C(X**[$11MC6$<H#$F\\<@;?E?%9\"6JC:\n> M^IIOA*JD0K]0CBSPH/ZN&25?;SI'F%K+/MI$V<H^=)?KR#PP.U/S73;WUIHU\n> MV0G'3+5[D7L>\\_XX/YR^`S!<L:N!%\"0&';K;1&'*I2D*K,S>FMYJ0KECQQ7T\n> MG5J*.T=SN/02AX$5M%K#R\\,:?7H-LR2V>H@B4K/)0*WU>7M^2@HFPS%R9>%E\n> MB%0*\"UXJ_D/*\\(C#*(%*3?[QU15G71!?J(P?W:`C'L\\J0@],BA\"G=!U8E?_1\n> MGB+-.%GB(+B;OINE*;W`D/[I(P@#3#6,F[*W+DGM@?>M1T\\Y3@:C]_\\0]?6C\n> M'R#A?49Z&94\"W_?CQ\\O9I[,@BK)J$N4A15$\"*;_E[EL1\\QO29D)651XE\\'4_\n> M0>JAZPLUG*01&U.,,J6@*O<GOV==M2I#>V1%T`T41&E<D@*JA*W?]>')[]\\Z\n> MX6:?BET95NE_GY6TE:K\")*&0(#8RMOZ#,V2%J7!<=FO,O^MZOF'MY9Q6TM]2\n> MV7[+HQLV4JPN__=9!).51]SS6L*KQ09&IM6%ZMSO2;_UEN^6L+\"H(T&'@8:A\n> M;:ZJ8U9Z]Q)>L%[.(854*?SVK=P)<9O^=K;2AAG[+>^P]HC#MF5A$0S+O)O_\n> M6._V\\S55<\\DWD9#&+RXN[#$!K(%/[email protected]`$\\R]LCX56RVHMJ79T;%.:T/6\n> M_8SV@WM[F!,J.:$H3,EC@CO:W6&HIS'Z1!,C8Y$U'O(6D-H!;VS%$V>],3'!\n> M+8<LG'S[P-/'*;5J[:O;@SY9SEI9:S=;2;3,AGTVAT<RB16/>9?80/@1>RR6\n> M*E4_D1T@+7)O=VRUA2Y$JJQT0*I!;XCQE&[-+M181=(U\"+_UQ3DM9*8<TFNL\n> M.NB5%L=$9K\\E\"@IUY*-U_3=T?^]<S[\"JM+?*<J+Y2**J4'SJJ#<0\\,VRJQ9A\n> M7R[0<X:'LO05G]O]X25G9]V#7:/D!#W(F%H;Y@J?T1!KU*!F4Y_?_8ZEQC\"5\n> M%`N*?#DR.M7GWU!<NI^]G+'T&\\B96/)\\Q^8;?$95::7TL5YO5Y?3+UZ\\^/;%\n> MX]^R<H1&.D12/FW0D/IQONZ6;F^A)7_S6[UDT3$SL(FR'I>/`I=VJ;Z<??/5\n> MUY^ZUC94,I7;SUA2Q=IE+')+D6G\\[.X^97,L-D+YN7\\Z48>1,J\"_Q=!2I/R?\n> MLQ1Q1$6K;^1#\"AMM\\\"'E\"]&AX2CM:90<H;LQAVS7;M7WO#AZ)\"+JB$PN[EWK\n> MC@4DCL`7Y5=&E->7^#3-WM;KM=HN52H$+>)+$8$K15\"0',S'HY4?ER&B,(L#\n> M0B&[<S6S5H:@1`MJ836<,Y5X`8,7SA1F`A`(,(N/R`04JR([:9Y%2M2\\5G+H\n> MR[L;*L=GM$Q\\_M<_O?K^BZ^_^/RE6J;C))I41<+K,KVH.>6]5?RFDB5I(_:$\n> M@N8,H2#*PC\"(8MCECLL$=`>4)/%P_1_(!,'#A0*Z@TNJZ,,+!;ALS482?G^A\n> M@)+.HNC#\"05(L(AZJS:PVY/H_84\")%2%9P@%&%+$VFD'&($`Z+<@(B(/\"`6X\n> M&(3L^@&%`B1:5A]8**`E+DRK#RX44+I1.%+:]Q$*D&@^DNA[\"`7@`X`YC3^\\\n> MZ`ZQ_$5\"085]IGRP4$#QTK#TA`(J7:HEX@\\E%#!?47&^4(`(9>$*!;12QSG.\n> M%2>%`MH7U&Y?0\"C@VY,LLM>S4\"R^J-DHHYU^ST894+.&9>*J\"\"MVG!(=D[X\\\n> M-D80UNW4O2<[;O>A^>GKZ7?;_4$-7I(O5MHZ;L3NAN\\1T@1N(_H*8\\&W@OO;\n> MFYOMSMIM\\FI+RF//3J35=B*8!]0K7<WJ!ZV*'-53&'O(%:O)=&ZLO0_U/<)2\n> MWTP[9/>K8?Y0&^F\\M2=N<RQOQ\\9&1,/G6,*-,:I<\"-F>22[2BX@OM5[.GM!-\n> M4[U0V;>;[@>RK$6'RP];-E'ZRC&7^OML/35F-:@WS\"=X3.YL/ZI5HEYPFN:^\n> M@)T21&/;BG&@M<6\"SLFTA>CU(-&]]!U\"5#._I5Y=P-GN-244U&S'21>^U@G8\n> MOPJ[S\\!HOG>O#IKYSO%R\\>QR_$YG1[WM9E&35\\FSOE!=J?\\\"^EO^FBK_ZJC*\n> MWWH,.DU-=QGNN$1[++>UO39Q=I$C$\\`8;AZL8&*GCV_EB%O;U9RN2?1-`R8(\n> MAH8NTPK.?F1RHO,>L8W:6_,!@AH?<Y%P#2ILVGY]X6C'V;\"][G%##?'7^6K#\n> M)R`4NL4X%%4D5=C<?P1':IR6B:YQ[=:8]^SKNCX8Z8V,QHGQV5RARDE\"7\\4U\n> MKI4.WPI?3X%Q$>296-E\\LS4+IM=KG;TJ7E$Q^BWS;)86W(I?=G3?H#U-6IWU\n> M>KLEHPK''M[:SM@E&(MB0&._B.\"J*ZMC/5A@(7V/V<#QK8\"TH%D*3<OY1C^-\n> M\\?\\7V^NC?4EE8HB%KVM[9=ISR-.Y!&U_^;[J01X\\4Z.NG%2Y7%2Q\"Y*:XVH.\n> M2^LJZ>OP\"L\\4,@IQQ0P+.7.SM-]LZ6[IAH1*U?Q1B+$B(K\\*I01T]9U(;A+6\n> MI+.;']F'3W^O@L2?HO.C.->'5;4O1=K);'#IJD5(V'/940K3:2JM:BH4`>D1\n> M4=/D1R583]YRC33X\"54FJ9+`O:\"+60+$;VEF*L&-8-L@BTC_X=:CG?Z_?J\\J\n> MDZ$B=(Y(?4`;(W>*F=I5+3M/@)7]P(X06RE6D2>30;'0R\"LTL@JDS@:Q4SYN\n> M9$0NRWXCMU,J6RP7YVFB>JA\"3^SAH^B6/`[S\\9+W<3(ZG`2Y>_R5R<%_Z0UX\n> MW3NJH^.(L_J[TS&-\\;DWD:_FM,N_N<+9+M!>63SDS;Y-C8>R1]#$<7EI07F#\n> M2BSA_#6GC51#4#DYB;FW[N:6<43JG48*D7X+M.7.>K]U\"V@N7MJ!.6:<\\CI$\n> M<`/&7Q+%+,4H2C*IS>!@P`'_0E4$3I-1-\\AHZ8'NM-:;4_?<%4Q(C$PF5W//\n> MO_F3>!C<U+LYW]:B?$G(CNY(Y;4QZ;3H\"F/3D4=N3!Z*_L@-S,B%)?TE#6`X\n> MBU-HN'\"/3+$@!B[1Z!3CJ'D4&7\\BKB$\"(2XC\\]C]WLUWA04GDM%>5.)!_R65\n> M`5:YQAJ:_#8-LL$*SBQP)69#]ENRKD$G='PER>X\"RR4U]6O>%,=&B;Y9LL9@\n> MZ.+)8(RVGI4U2EMB1;(#Q^)^.4Z=O=Z_#J;BL$QS82&WN9LI>R-X7@/6%M@5\n> MEX+!>&-#-IX(GH65*@7*645B95/+8/;L-;UYV[BC)PG3:+@<'QL]2426W8,%\n> M,$@B&%#?NP\"2\"COJ+8`<F1\"DQ\\<5G0;\"T^/J)G\"EZ8(X%0/22U<#%741_O-5\n> MU+#CTYO1]_;^>KYVC(;'#JC-P#YXU'*N'0[<WM!>!:,3P`XB/KZS/YZCH=9N\n> M$DZ^[C@?/5:*A$1'D;_@\\!4&8E<QQPA:K\\V!DA%RR!<'1@-6W2C&/01H(`:5\n> MJ\\`M&4Y<&:1_74(/<`T*2,]QQV[S^LI;+\\SC1W0S+;N!!,D('[5WZM3>Y>R.\n> M2:Z64'>/R9MOV^V>O3X]F=`[-Z%S:),(`$+E+1BN3_F@^8V#*KL%9;VCC895\n> M,F?9*#)^M/O;*Y[Q>ZO^TW@'>A,<%;&QCPO6D;5N66G[1O:,`3[#H\"3/R2''\n> M[LP03P!<[!CJ!OP-V,/B_<QF1&+D6V,$=125L\"2-9X5?#=>P9.!60YHPP6P0\n> M(R;X_<`AA*I@?8H;YU#0]BUZ#1(&7R\"]!5P)=`'PM7$]6@*M*O?ML6M6XA.Y\n> MPIV??3TRC[O>G&,SKZ5MQ-IKK59#@U@C36/>M'+%/7XGQR@X#$W*2\"[+!M*G\n> ML_X<.2/7@>,F.6:\\YEA`&H<_4U6VXH<SR$$<M1BMIF5,&92ND)75#-,6MR_B\n> M78OQR\\:U,ER<(N\\.;-8AE35*`>U005Z8-UL&*!!'<0/0X$$RL*$1RE,EJK6H\n> M,,NA_-UJ,\"+52AV;!I.^R!-\\;7YMX*($+;8[-1QNMAL!)>L7P'9'H'N1>;.Q\n> M3'H5;%7MY@<M^+YMC;F9-]?JWES#1M'X7:C]H%OM6R=^<!$[DHQ.]\\YZ@O`P\n> MM2O:/7.]]7:+P5QO!W.]L7=B49ZS0`R10DP.:>GSA/:5(]6U]I+\"P3%ZBYZS\n> M2)G6:VN^1X=Y&*:X[)7[7?&#BXMRK%G8R]G`W*W$_BH86X'\\5;\\O%C83;[JQ\n> MV[[`=JSO&(B1#3/EBK/D/70IFX\"VT`0&FV>AN?+;BEI3[IV-\"T_G^H5(8P;.\n> M\"0Y7!XZ4K#/CHC`V3B_7*`JNCF7L`IJZO:%]H.&1[QSN5?:U<S!%AQ`L_=$.\n> M>2TU8?BJ7F\\XZF&S-?NNEY=.PSM^4T'K^CQX;9`5X<@X#?OC-#C:!(W7]AC(\n> MING%<5)TK(XI[E@%-/J*7,B&<A&Q\\MH[&&_P(^T-7^:B#`-[0T0W!3AM]M07\n> M2ZM5[2`U#G60SV9!)@WSC8-28/6;JQ$EX\\UNJY;!:R-LPW6\\Q(&\";M4)D8KO\n> M*44F]*X5.O9'F!\\$N6KEN@L&KB+2B)9C)NFCKKIR%X,FJK+J7KVBNY]T(PKX\n> MH7*12`#CL(A[VL7)B'91!0W(Z_.8=K%WW().>7-[?:6VL32J5-1J<!Q4WXF:\n> MO7\\<%*4C1TOZBH0&81`U<14)K5^\"6B72T8&/?2S+4,Y8+*!#KK2J8J.B%.V[\n> M)[4`*9J5TR2\"DYD%7_6)4$LVL=M;VD2E/3-2FP0#K25^R[.>2L4TKCI[XEP^\n> MKK9DCSAV]=67HX6C_5L`.TM;?@16@>D<Y8'614<IBT[6,T[N[QF!#QC=C\"@W\n> MN&`E=82&'L*!O`K@;J7*CSTY,^[/?*XPZ\\T)+5WC:15]B<D;X9^H!/9P-J-\\\n> MO+4E8+@P,=#1F`^KGEJ/1)B5/L\"M],U3PRHF]'_02M>O3(^S?8%HHAH+GNW>\n> M3#5F>'%9<KUD\\MVC8_?5>F)T<[RRO\"KP<$NB8D0;S<,MB9-CPRU(&*!1#S>9\n> MJD9/LB(3EL&LMN<]69T)^:>YF)6$$L_PBH]T.PC\\34;%D=11-+''>/2TKXN)\n> MU'\\!_0U_Q9O-(K+`]U:I[LI%G0-?T]H;Z+V1+9R;ZH#G&Q_+,&@Z]P)[`/?\"\n> M-PI!DI=C74A:K126*H-KA+0L!FM%7XF6P5C#ZT%>6B-DFA.LCX%Q\\E1I.0Q0\n> M3FMH@Z**78A3-'(G6+6-W>1@FV*KK6V+.-@)$#(^[M_<U//=WAN!XXHKYQJN\n> M@A7?<Y[8A]')KU9*$0ETB>Q:SM=I:3;LF1NH/`/=+GNC\\XPBH'H-YAO_YD!L\n> MB>I2)?0I7V.\"L!(M^7S/&K*[[2W,$3:XHZOUWF27+:T6JCLRHQ)MC.<T,[O'\n> M`47DP]VLZ5U.^][([C*(<B8PA_.%L.\"XI*M!D`S*U-6=F04^]C5MF'1YID_:\n> M6B<.\\#^N<^W\"<WKK(:XG4]CFJ<WBV/%:UGA3>%-UJT'S*B_^VL.RC\\%2[YVR\n> MQQ,+NP\"[DMJ3I=&0:4XXD5HWI[U:V?H/U53%5#5,.G),9-N]3FWP`!*&RY(^\n> M;P:.MU9_@5'#AV66*X:U<^2!6@\"SN>TR&\"\\\"JR`R_LWM]*UX0&D_]GI#)G1\\\n> MAYI6@9@\"'&KR>33&(#MQ^M;W,N2`O+X59:;5XGC(MI0''9^V#+%AQ/'&.AZO\n> MM#>MQJ<3_H^H@-4AI@,6(%:4ND76ZH@;7.!?W^($7$/+0E8#RX#52YVNMEZ@\n> M]+(Q`\")SKO.TZ0$7!1AB=EM@\\9+<2&4]J<*1\"S@U[%]A+7$N4IQ%I<J=35P)\n> M6:V.W+#LW09]T1>:19/JC.P)XS`VJ31:_LX+\\D$:NXY9D?Q-T=2L&MLE$#4.\n> M3U[DZ:O$..\\G$]@\"RFV?P6>Y[&TG*\\<S6T_`[__CZRH6QH+-<KY;(D]9S.>W\n> MARUYIBX`Z=[PQ.]H&)J[L+:&O949:7*YRDX_!/\"5&_,&NVF@&.X-NHL_K*WA\n> M?%,:T0N8,6PO=<RU26`U'UP&;!3B)8\\E1(RTG;N+1M]=J*WUOVH'&=T3)UT<\n> M:25_4\"VQ-@%V$@0RLL1IQ+253)X>?P**Q4Y/MQN5/=I<JW,\"8]$NUS8^.IS%\n> M[O'<_EOCQL!@`>X:9NV!='F6W7YQN]]+0>`HQ>@_0\"XA:8SUHH'X;*\\(&#C]\n> M1WLW6XER72R29H!=@\\+8N5*PR+M7;*3.H/L`>:'QD_P_E)RH?__#Y`;!D1VL\n> MD*O@Q,L2`EQK34O`/7_#\"!M+ZC5KDFWP(WK+#%!7VJ&F`1I,QX\"5UT,4AKVR\n> M5)9OA(W!\\56>;_9O:]&%76-`4`&PP+.B=H0VXXU<MXMV5<KIPF^PRI-[\"4ZO\n> M`=^%YVR`M]WS7FW'(?7S!:.OLR)=/31S;'*J_FHEG3.,,,/5WN[UP$0=Q\"^5\n> M_8H[/7K?MB0S6LH-46[J\"<:7WG'JZC-7PXM1!P!&)(SY80[JC</N=G$0:*!C\n> MFK[AU4MMC#E%/3YR/\\M%@\\$)<FO<S%I6F@8#Y>[`YFS$GI7Z\"H!%`H\"\"N>%@\n> MCC\"]%/ND,2[*XC@JRHYMM?6_@/HDFUCRU^!DR\"(J0#(K)YE[4_DK4NEN.(U`\n> MXU'?GX8X\".V?DERM8H(`#&;`U!2X58;D:&Z5>>C=V'U[:%;(=FJ!:P3ACG![\n> M>]\"ZQHT&5T85(.5=?8@N<Z(951MBR*J*R(G>-_8M8O5?0'^CH7E\"_#_I0?><\n> M\\*E?`&_WQ_K%[;IV_><$IH*/*_8@9Q`^^N`>S9CW_LK1H*WZB!Y683?WY7SR\n> MJ3`^=1U/V,5VM5&;I,4<YXGH02GZ=MQJTA-T@LX7R*%-#P1)++U)SW&8K[>K\n> MO5;59AITC5I#+ONM_0*0764K;+$54CD\\O4_@'G,$N@9+-*MZ9ZV[#!HKB,'(\n> MA4M.JI5#WS;-OC[\\?;XC![P]XQ:+\"H0:Y36%^1Q3P@NS#/0YJ)-EI8_+PNOT\n> M4)M(/9'!.6\\(.GG$IDI?)/5.,<'P=ME)XCZ+<U;&BRV4(?AKY%X_J[0713>&\n> M-.IZ8.C:'3GE.I>_SMV@PW)WY`J\\=]:5>]F<+5LM*TA$^$3.!6DS9@PQ<2](\n> M5]X%J7B%V,O5@37$`&Z#_:2*A.`C7O)IS8YA(\\NMS`!V1J!!\"F=^P=&%HK8H\n> M=]H?@C=`&C4%G/.<C?#-Z=M.SR3'RJ:_L,\\P=4J&Y3@ZPACBSQYG_=.U09;3\n> M=B,&\"U;U]7-Q;;)'WMKP6FZFA`PSYQ2.9(W;^R@TQ@W>55()-*ONGJI[*J^!\n> M,MLX&NB%V+2/O<ARC!1H0?;5&]2'%9SV=*],/$.IX1U+:U9K`;8.A)Q&K,5$\n> MM@D=I8(ZX->TA7=[6H.%E@LKQ$'L(OF(U]GJX<8RA*+#8:)T1B[K%R[8#3>\"\n> M>A0G)7&/'%UE5T<65TA_CP%S>&1IC4N[_(HM0L'FE<^7R[\\8ST%>JN5>@_6@\n> M[-[(#BQ*5+MOBNUJW`1:QZTYFZJABC&@3=C#8R&X?Y@%'>\"UV*\\\\2I)BZ%C^\n> MX[9;RL^`<!XIPW^(SNZ13(,:%\"+B]D=!Z$+E$>VSCV=51?@S68ZN`8;JX56G\n> M3IK`Q&(XPEV]QK'HU5KEL$:4(,HBBO)\"?G)\"/<:B1BD](H3@1LG=XJ\\>9<03\n> M*0[K2F[-'8]UTKI3@5[-!?*_@N=Z\"Y_-3]FK/*]RC`LE,Y8V`MU;L3=[Q\"9:\n> M7S.QU:,=[&L>:X?A*'.=B0OWK4HR[;NN7F\";,QA7G&\\``8_\\+=NI4]PGOV<:\n> M9\\I:M4\\>$RD:):,:<UUO5H=6/.Z)FFPL>5,3<BOV(K%[;.+@5QPIFBU9X!?-\n> ML5MX+#ZX%2>7)71\\&6:79.^1G?6\\E=Q8$!O-#5T3IRZL0)SAXLR\\5;'I&K[\"\n> M#08SVFD\\-'V!VX?O7GP?O_K\\;R]>?/'-RU=JV_WF6_;>_7>Z&-*%T#%PG>V.\n> M_-!X[Y85.]BC,0:+2;\\Q[FEZ=M2N!`J`6V.TD)-!(2%;AU'DMZ,4%E/%E#<)\n> MB]!TWK\"M3A2XUWE!$B7A+RQN5(4GBHL>3F+'RU\\=*#5$%7Y+!**J=S9+U'\\!\n> M_8U_S>O*A-!9]*%L1]:UV)TN[3HW@U/Z:OJ;J_ER_\\.:EKTT9E@36O74497M\n> MH]4NP_N+LSXW?D?0<!94F4!/(/8?MZ`7V+9%HF`XU.G;[8[86CIK-R`N88&#\n> M\\6W1JGCF:==Y29A=CNAVF0WW++`O=$/K]5:P8%V#*$HM2T92DRN<QKG\"\"=@8\n> M5S-HB:DD#C.X2F@$S=ND6YET@V&ZD`W4`>,WS#V=>65@S\"%>$/Y]!N!U\\9TW\n> M3>_-4I7>9WK7$?\"@HN#(,;G]'HO&^`0,2\"@()5E%AJK'E@.&_RFKV)8VZ$L?\n> MSN@`O*!>.=2V?W5TT=501)%>+K@EG');3(5`EUE)7)%3R<M^X@X(PW__-X%%\n> M#),S^4;)\\7R_O[WZNMN\\(>]LFW=4W9^W'TGECUB\\;*A857DT&RH3S5G&\\)O^\n> M+(!^!;^0^HQOW_HZEH#%$A&8)#6#I1IE9=DW!G+`5,GQ.;#LU5=WCGFDI&BE\n> M1,Q&2X0D()>NJS16E?WM%4-,1CG#[<D%UU*CAY)+WU<'@L!40NJB=568ZI1X\n> M`(N5CG2</)#P6Q;:1@D8OS-C@RQG*BE&`>P,#WY\\P!;>(PJ/-7H?PV';,VWM\n> MZ)=:#\\!RMIH>`>XVIRHN35X:M5.ME0Z,Q2X4,F#=K:)HB/8(<1KCHL*X&.UY\n> M[)FJ+1_1J5T0'./007\",0^QV70,.6Z'A<%<2)=4HV0]CEC$\\/N5X,?:[_=ON\n> [email protected]('@)5@UXFQ$_(F$B<X22W8)^#E*RGF,_FQY)`,GI>F#LBCFC.E\n> M^Y;S&\\E8<4[6+L<&OC0`)`8Q/_.W*#5W&4PE3JK)U;VIX\";GD2RN<2%PB250\n> MZ25E?S7CPH(*RV`\"EGGA5*2*^0WUKP\">\\C/_E(0.L\"#AP#DP@PDLT`7R,HF!\n> M>B@(;*\"T,!U%@R=)<-G4!UHK4O5?0'^3H9HX_9]4$P_6&Z,A-BO7PM$0!T..\n> M&T^'TAK%S=Q1V[\".SII9[6<!;C0[XU/K7)8!@X1-&!W\\M;X\"&(N\\:'A%<9-1\n> M:_U1\"`'TY;\"UM.@.^WH-'B=8[%QI%VAC8TY:';6\"VD53%`IT1+>:8K/V)M09\n> M]KYYU5-5>9\"T:OVNV<+#8=XJRD`4)RARXKCK&_AD5T%\\;+.)!*^A(SV<IT.]\n> MUX/6:$\\]RC5'9P<-K*L?EG969_0`^PL5\"J=HK)\",8,<KOC^JAGH&/ND';(S!\n> M!_TR[D<T=71V[1_(H$?6$W5Z*$L'^O54W):NE@W<DYH'<M:N!I!);DK?;FHG\n> M%8@*I*,('>RD*(K=ZZT%7X]KNP+C`V2NLSQW3.L4;C16M:O/;USG&->A1:ZW\n> MF$*M_\"776W^=;[H;U1#V<NNHH\\NH!HW!\\!+95%\\,!`7LX-83U)445@/M:]<7\n> M#FK#;\"V5W;$3IQ46X-JZ66I[`,9[M`XI)UV5-EOB(1]Q5>JQ&QF%?/\"E(;QA\n> [email protected]\"Y[FNY*1G(.FBQ-\"LFFF':@'/[,E\"J[]YS7P9B:QH-;6EH(GW4\n> ME<:Q+!`K`09(XNRS2+2%?W8.)K6]:VCM;86:16LUBS3+&?\\JHA9[PK&ZTQN8\n> MPST!M(1DUB2.,E1],5L@@@[18N:5&%YIS640%;!7[H\\\\1S-!97J,*R\\P#\"*@\n> MK\"U1B=.<6ES2E!1Y$<M=JW&YB_!&HBHA%8[>SA&OPJ'C7KE+A0OB$(>N=Z*-\n> MBG\"FTKJIJ'#P\">,X,8B$2EP3Y0>)=SBRRT53QV0!>WT01X<*-CX^\\/VB\"[?$\n> M.REV[T2?>1*Q4J[T05@;,ZY8`\\`8#0&.\\GO1P@E`H,1]JO'N*CD90YKD4KYR\n> M!M`S45MEL1O]G<:IJYPSF#OL8%SW2%O<4G]<^L-21$EJ;-&SY8*,B/8K!)N/\n> M4SZA@KSD-GCR>[BA7LX>2QHYIT$BYI490UJK%[ORZL_RL2P#;HVK73U_(QJJ\n> MF/<1K9,*0^]-!%%6>0E@N=^:1CR'U!/)SN8T9)#$:>Y45R*(+?I>SU5I1OUK\n> M(_H[TX2D/'.3<7;PP_7-*_*M)C437HAQEM8.'3'GB#WAEIR>`_J;_IIZMHQ/\n> M!EK/IJNC]5[_,!?+#6-1RE+(U[1&Q?2C&)OM<7LF5`&,IJ#&&`\\=@Y))QTVT\n> M\\I/?ZY8!01/3*^B?X)T.U.;,2X9QT[663I4V.&/(#C+$N\"5MG^CD!L/6T;\"Y\n> M6C&G=52^?^W9)\\!5#7A?M!7)]@JYUM66)=`%27OX]:WY=LKP*XOI+2OP5\"//\n> M`*;K:MF\\QAUI5VT^;]D?7+M#8)QKG9N7U.Z631\"X&/IJT<&*L)BH2GZXI'<!\n> M>-<J.2\\YQ[%,UF':WAMVZ2;X:36[`E=)Y_6V5L9QZ[M3BM>^S]R!X\\PV5S/'\n> M<>F,,%^TEU,[2=$;7C=HU9/1TR'J.U_CEF04;XWNIA%W.?WBIQMQ_5%C3O^@\n> M\\Z%-[LGOU:`@SNTE:^,\"HXU3I:1Y[$:R(_537P?'&?<:8<+(T^M-_=.Q^.H\\\n> M'MC:_\"P?B]!?F>EC*M<0=F6F\\[SP9F@*BRSUZ\"WD)B%58TFX+GYF/1!378@\"\n> ML$S&A.*12V.T3EFYJK^E`S<NX+^\"->VY5+N6P00RX9AD^G#E8Z(XU;Y2[7OG\n> M^(34/C598\\5DQTC#,4()1D3ER4!4)OD8D\"]TH:!%X]9*B,S)`0(:\",ETK.ZH\n> M,\"ZY/>XA&FU^&D@J*U<*%:R5GB6-^.W!&&!^@`K3@+K?S)]([A`Q=VKQ.8`]\n> MC#N+:G`SR:\"34P$8URY@^#1D,*Z?]&1SH\\S@TNF3!&Z\\]QK@9:2,S/F1:,&_\n> M(PE9BL=F=+1*41JJA*4:K0C[]XEJQ;=$*^2:(?6YT:`(N84/SWQFG+S9LH_;\n> MW]AB#;H7.VA4R8'`]+-V6^_!C_#2YWF(V*J.V(`9UVU35Y1A`PS!**<.4'D'\n> M&L(.1=,&L]KZ>(GCC_1KS\"<G8^-DCT)=H&&Y^P;2/:)H#1;+0]>5-1NM#^G<\n> MHP\\SE$3B]X\"QR:1`<:IIK;1JF&ZH:<7H+PK'SROKY2O4^O&LI`ON+&'+%[;O\n> M?ZM_6XE]A1DCKVYOM'(YJZK`*)?SK.1#3GM4N4RWWJ6O7%82<%&>H5Q6<F]2\n> M.BKCLO)TJYGSEO#T9WZ/A&<C[B6#P;TDX^_;GAH><H9G'%!U),612T1]QJGY\n> MC---L>SR$6>EK_J2J,J=R]*CIQHFW\\ARIP+\"\"X(Q88YN[D\"\"'E(.-<8^Y;(W\n> MUB\"5K[14C@2+4>$Z5_\\%]#<;:H[S7T%S7&6>C!W>9]Y0W'O0$J$ZSR\"PVK&/\n> M-G+&N]'Z<F\"1<>S(_]31QNH._-FJ2OL',[%QE9-85CDO+-=>S#+2S$,L=0>\"\n> MJM\\S^DFN=N5P7[IV)V7EO+\"L=\\&ZA!!\\\"Z#&DN.\\PRVP>$,:.BSD/T)VGI-%\n> ME[GFGN5Y&H@,1\\[:+O7#]+?;VP/!2?[656TS9+B+@(TD6.)[--,\\8?N#OB9$\n> MS,=&!G\"W,'&+Y@0J3J#ER?G[\\$)%0:?/'\"![-_8*KFFA[!A(A,6_1ZQG7FUA\n> M?;ER]!J.2KS&FM]!B\\VK`%+@>UEJRQ'11]6GA8],!V<KR!567P)[.DZDT,68\n> M;[1[YLUZ#BOOJUMK]HC^F?D4>9P\"2Y>/]'6C<1MA*:9E':&QV-5$L:9RG`9$\n> M44V7D8DR@Z>5M3I0+?KD]YY<`;\",4\"M=\\]#5J)PX1*IT'B=0=A=D'2<B+HUY\n> M$7%=+89PNAD#N-2^!<+$QO-%+:\"EGC\"M=[NH?A1.MG=BN<4L:D8?YNI)B*U,\n> M:T8LTYAJC^#D,L-^9?Y%H\"PUFIOLK+7&$)`=7VR$>.SD^A?T\"N::^[$J+#Q:\n> ML&!8,/8SU*%'\"Y;'62#J,$M0YJL3_U';H6HU(2O1A&A]O*,)66FL?B:\\L11E\n> MGD)Q>&07D-^NKPJ!SDS8R'R]HJ83.[>_C^E#L+W$#]A?#+W8>)_3&!86,9XI\n> M0@WF;BO\"#*8U?,P<;=Y*0W.D21!^AEI:N`8&%\\2%^B^@O\\=X@O,/J4,3K*SB\n> M/I[@H8VS48TY1BW!E7NX5;*Z!S]]A(:^3W=:NZ3PO=.\"2^$8N/;\\5L-&Z_P;\n> M\"]+E'5`,5E;@I^7CY@PO+4:NHZ[N+&R6+:31Z5%AS3VA.;&56M1%4^%4)NB1\n> M+MB/WJ!G!NB;P*4M0I@XU`=227M\\7^%4V^C#G$A`!QR<)#G8P[@=T9J.X'-@\n> M[X@KEPYT04]LW,&1P:!]D]JYIKG1^)AL4J/6/*8A9O&+`5=N-X>.-O8?\\XOD\n> M(A:_R*\\$192/`L**S$A7!JPOL-UDQ([U=M4MC->&DH1SS>AL3F0%MJ%>X?DZ\n> MFG='QW:NY=5&'1%<NQ).IN3=C$Y!9<I>!RHHFZ,OMC=*DF(%NNH,-MBI<C)A\n> ME_LF9]C5S$[D',BB$'N5>R\"+(NPVE##4^2YG&2<?Q=@9J.^)S+7I6=KU`B?P\n> M\"Y=S(8\\7[UB8D^D;MH[15*#T>XWB2((I]@.U7-8<.</B?CQR7BI1(L-:+VOZ\n> M:R8C\"B!EO5*C=R$D8*10#\"=1CL6<?J3-\\W(\\;;+9+RHETF`MYWY839]$=&)E\n> MWL.H`$R_UN%Y%]N+@,?U-]O#R#H75<EQZ[U&K/>T&J\\^IL;S024=PNG1*>5J\n> M>;XZ2#%PDXW=&TL\\5E@'\\]R=V)T_L7U]C.6E#NY98AM'?=B1\"0.MGS]1&G!\"\n> M.[!5(%/CXL0_9Q7A-]^^M,L58QA,CC4OG47MRB&0+^R?+FQA>@'AI9$7C94L\n> M&G(_S9JKB%&02,4J4?6Z,3\"`Z2\\;EMXYCK/87S<F`FHW++Y>.P)_[2BK4(/=\n> M?<-HFH-%A*#](ZNK22L3>,5KB%CZY9%8\"(ZN'(&GR\\F+@:$@N^8Y2X=,N-:;\n> M<*JU]803[3^N\"7V#.M;YT`GE/I/C!ID9,N8J-$ISX6$U$ZZ&\\D\\5!F[5]=^[\n> M^JWU2XZUJ[MQ3([#P91KA_:R(VGIKLUY&B&=78VC'LYEKM[:\\;'LY+9+\"W8'\n> ML;^@P4.H>I8M@.(<QY!S@'P][/!+.<#W47'9\"\\YX?3JV%@Z6@:&-G6@P3P=J\n> MV$4:UK!'@0,M[/GB,FPRHR>1N(\\3])5CY3)F[.'A!FM<>X>8%I-O0>N%&AE:\n> M!`M:I_SZ1M`6>^4Z>/H`R<UH-1ROTIOV;O]$U&;,,+;P*K7'5J<7P`!.VX[#\n> M[SV8UP*\">`$NJ7$;'%?O/.KIC4&4:APA5^^G<8M=2YO653?7@3,>6:.MR<U:\n> M>^4(W,3>`:)4_P7T=TCE6Y2S,CNB)\\P^D)Z0SC9&3S@PG`;=(/1_8_/?F$U#\n> M/_[=84?;.,P2.L<]C^\"16-[7GGM\\NE[C=%U#2:[61^M!A!BL)F0%NG'`,<'[\n> M.G6J!8O0;)3(LJM>I;5*G193*2E[%&;$@G2?=9&2CUC7:(R+(F@<@U.V113L\n> MF/8Q&-4^L@^M=NH8<P7JJ=R#LW3N2+4P+AO!^3IW]IQ-O*BB_C*7R\">4\\'\"%\n> ME4MD#OM.[/PBUW[JN!(^&%7\"^SIXI%<XZ07CFY6KE&#KI,OI[_HJ;ZV5,#?2\n> M1<'N;3R844MW^#;.N.5X8C/%\\>S`Q;%E3!-$NL2@KQ`WBD+?5$F4@%JMQUK`\n> MBYX\"$$H^UO^-:L5%&_A..\\?FCMY0]'_F3?1_>4*>K3A`Z<E\"YQ>Z_W?L\\$0E\n> M^&Z65XE6\">KYQM-/2S&WZS4;,/P=0KCJ#9XYT)T>[FY@KHLDDSB'N2Z_0-^'\n> MX\\N(<J]GA==7VP>\"4NLKO8T!N*?EL_HR-0\"WNV[5\\<7ECP;ARH7;P,5H.^7)\n> M*#IM:+:T&M!3OUW5YL1!5C87*LXWV]TU%.`K!CAC!9_QR</\"I)6&(PF2\"@9&\n> MRE8C`9L>`V86B`>ZMJ3BBX.>ZM%-LF>YK/KM$SOH/R%U_U<;09<XKF>T,41X\n> M8ZLMW?A[@?3NIB$U&ZUA73.J<^1N8!.BN::<DBH#O.V@-:/<':;#T`E&5>F6\n> M[4?<?0-/SK<R@P!N^A4\".$$^DQ'/;T0!&8^D=SG]Z_Q.>M9UVQ3LT\\!#@B!D\n> MLA4S1GHJ35]+N[UIZ22U8&V*:S$*=WRLT?^/YKIGE6=/2TMR>6#GQSOYF(M=\n> M&=SQB(N9BOS*Z9G?Z^L*D5MI%9+(+#7\"(DA?B]/D-K>EU'S?7KU6PHY:(K2Q\n> MOXAZ.DM>X42?2FEC0O>DHDK]%]#?7Y.+M:@&IHFYM</J&1<.ZKYA5RE]`VHJ\n> MZ]@.KLRBQVJGRYE.1M8^LS(M`W&?U%>C[$#5[RK?;)\"U\\3^3;D?N2ZWKLC'8\n> M>OGB;U]\\\"E<JEFQ^IF<Q[G.,KT1\\$<,`([[@1?OLJW@LO8R[JHDT\\TX,ME@*\n> M$5<M$2RT,1<+%!<]UT\\,V[&]G(VUM!!!7F]79G^O]/Z^HMU]Q./LL<0NW-CG\n> M;?`43PL&'._4!@_CMK(*3M?'7BKVZA.,U,?>3TE]K,30KT]P;WVT2#%6G[$[\n> M-;ACQN[EYXG;JL\"_KAK>5G&2GGNAK/@X]XT9F!ZSVR6I)_\"70:CBM+'YB?%4\n> MB4'Y?>/IJ%FO2$X/'D\\B8SU@/)$@%@S;'Y;!+9]G]>F<+8,;:QF\\TKH2D=6.\n> M70VB25?:-'CE4')JHBT2(.C>4$SDS`7PB9M&-@[N'.-@E8A%=66KK;@OA4W8\n> M.K@]:AU,-RS6/)A3*4=2T5;#[<`\\N(&-'`!_Y;#^&WUS9L1,=U1961&+[DU#\n> M3F8CQKF.-ZR]%Q87?QVZ9PILS79-EQO!#YGYEK@,3=`=,T=V+Y>#(^;(_:[7\n> MY35R76@-J*QT!AN&H^;(-/7O-4EN9(<S$AJ;)`?WVB1;\\>L>P^Q[;)*ML!7R\n> M:8LV,7MC3!`>_HVQ>!(G=K.R-\\;P.=8WQG`LUC?&I#<^<F-<ANJ_0/TE`[B>\n> MPJ?\\%8@QC[L4CZEW!T[%@,QSD\"=)L2M874=4Q%\"[6J5^>P^[@#9YK8<(AC>\"\n> M^F5LS.Y!H3.LKD-@@::/5.<H)Z$LMMB(9RB+?7A)DG7E1MFJ*:WO[21+\"M]1\n> M;D+KP%TP:/BC%YU]Z*M*\"?@YC!S'5&O:_M1N.\"K\\I(@T%IB:H'T(KP*VC$9G\n> ME4\\<BSF+J*6DOLS56?$B4?HJ*SEKF\\W>H9?P;`,L2EM8<$)1U#<--5YE`C<.\n> M^4*C>U^9([-F6(L2+R&CM$JEG,&X%/\"[>WR.'[.1D&YX)!AGKN&*V_*\\8_F*\n> MH21R0SO[O`?OI,32U+3FN\"*+PZ7&5<@15JC%`D@K4&V-@MG06#5Z+I5.Y;H(\n> M'15;W%&T<DJ_XJ+#<3/+])TM7007>8].O4>>[:PA`JGB`*(P`E\\?._E^P&+7\n> M%;JHAA1SN^G?;I9J&K\\@M<%+JN>W#;7J=]1*JEFWQC4:=V=FQ7/O?Z>W2,*S\n> ME3%\\F2O-EQD,EI8AUL(99C3PS(QP,8=50Y?5V-IK;^/Z*-Q*,(JW8FY7!]X;\n> M[MU.X\",*1![P9*LQ$%L,!MR-+5T:$>U#%F@\"1Z?-1F!2_>8:(541T%E>2=7F\n> M'=]W8Q$QD-R]_;UHU?ZRG#ZB9OU&%96,$JI(\"<8X\"?,2W/'/'(-7R#AVU?IQ\n> MDEB/8;6M5J[_<\"J:?6#;93@+]RT^3>H,M_*8SOVQM1%,+-3;?5&EE[93-J[`\n> M2A4;!+8XKUP_YB)SW\\J(WW`-5.:5F<#L!TLL)-^MYYM-O7OZ[<VAN^[^B\\BF\n> MZ/<HC)T)7@:X;[WAL#RD8([0T8UIH]ERL+=2(`X!;/]:/&N@/PU(@4J+!$O[\n> M<NVZXL&;7)1R][;BW_2PMZ9C0C\"H)#;<+N+<V0<^&;>IXA&F;=&:45NT`57F\n> M8*J/D41\"-J4FBT$4(GF0[8.&WV\\\".S5^N/D!^Z+CQ^/.#S8R^_.?7:L-$NE$\n> M#OW;AGA7;C=JW!.00RDZ(P(UM1P48H#&D'*'MM-L\"F!(O@#AS/2%6D'`+(8]\n> MU+!D:2(.4L)%W)Q_V;[E-F?A]WF/<`JD-U^ZO`9Z:G_ET'\\U!@W[?GIU]L)C\n> M$\"QM9#;*<]4XN!QH_%2/5V9`>:,AK&]NU&;\"*E:VKIVMA'Q4ET=TN5=R37_*\n> M6'%.6&H]P&W=:8WMM,A8&/ID;:62M:*`_OZ:9&VE2];V5BW6V%SVMX2+`=+`\n> M3ISH:\"B!?X.Y<\"8C35\\'3MNW;#%D'21YW!F;P[D(!9;E7D-3!-HE@@<L$SH(\n> MIE#O;\"#V5U1D34R:!U?6>F*</GWON108Y'C:`4^-*)TCW<)_KCYTQ!PCM\"]<\n> M5#ZD>#+/0F.6J6'/I$4IT8[V^2SW[#MV/?]),_Y0T\"P\"C52/K2[(X+GF49\\9\n> MJCEX?9U#->>1/*#CS5IY7R/`B5(5LP>5SLQ(*RT@B`G(+=%;Q;06?-7P$#6N\n> MI*.S?65G.U%M,=F(.C;3$G,Y!<(&TZP8#IJ&U=\"=7->X0.7]_@]6/\"J7=APZ\n> M7(3035B/]4:OE+`VT4:[*`TV&0WO)XOMD89;]:P(J=VT\":PA-4/W,[),T_,=\n> MYE9MA?ED)8:[QDM%MAF0\\V&7.1A\\JD;3J9*?#6#%U3`/O`UN5^]OUYH&JM>'\n> M,N2\\N=@=9;63%FOFW5JH`D&\"]9+O.SVR,FQZZ&%ABM*V8BC8GGE7N7QOV^WZ\n> MJ&5Z8&1J#\\N^'I@0@EPM!.N6%#F*V:(1VHF3PYW7;L80\\(A+7\"OGD5%\\!^=;\n> M(<A8\"NO!M32\"\",L(S=SS*&841KS2'1]'LLF8':FUZ07&.YP'C[<E.5[%/3G\"\n> M0CAB\\:=YQ]/&FK_#Y(Q+\"-ZQ@PA__=':UHLW:GSGA+A-(=EIFQV&I3MN`D+Z\n> MP:_B,`W!YV92I97WF7I'[8-AEN.S]CNF2D'!U-I=?26P56/PEZL>E)6P0\\8@\n> MC56'9&'?2T!-_8^V(XRNS&T'AW5B=.Q#%CG99X&,HK9_X.V1=VRG3X3.$!2[\n> M>D2QI(U!!3:#C7_Z:Z3IA=J)\"7@&#$],C#4_G\"YN:Y=QW-8P'22(=F5%B&)W\n> MV>NOM&)!Z6*!B-N2(5(4(D1/1A\"C8&)-O%U#]58;C@33\">:4B&;B#=+5)I(U\n> MX!N[/2]D,?'VY,`@*)L1A#6(YSDL)6;\"LMH(,YYH\")E?-H)FGH8\\<9(NKS[[\n> MO6SB0J\\SOIMKTW>.\\X0\\+P<\\M%&1]H(,=OD`X8293(=S_#LM>2VQ=2-L3EC5\n> M7[QX\\>V+9^P&P2:[;/1Q\"[<$5J[I+I!9C_%)+%I,9TCW)&CIW_Q&F!1C6PC5\n> M!A#6<\",F)T02+\"IVI&F8N7)3VS4+<A&MW7MK0:''A\\/Q/@80*WKCUQQM.7W3\n> M&9*4`>4U)%R^;V,2.:&UM+=;XXB+HED\"W6)D-`^G1+:`[=.'/:L.WBX_M--+\n> M'\"V!AJ'?VT+EF#IZM9V0O&&J[P^\\]<ZQ']UN-\"7CTMM@Z3Y$V/:LW%K(#.ZS\n> MW@?.W&!J:=C+P`AG9=@+,VA;-9_=8OKZUC&F!HDGV%<:H:JFLKI\\YIJ]7,@?\n> M8>5M8@'JAUJ.Q\\#U5BXW(1NP.Q8A>4K?Y&G4ZYO@RHC#ZA`8C?9&@=5L3(0.\n> MXA*[PVD96NZEC#HD]=4A./:QAP-.P'UE\"$,9LCI$2P2HG:,.$3G:J$,:4H<`\n> M(4P/I'$99,]\"HP@6<O]%K5P;:_4!1Y'=G+3TNR);[`,,D(S9.CM/NWR]E!CE\n> MR7PHC5B@M?60<NP-3?F[M]L=RZQR`T>%U%*T1F0ED^F$3E9?;[=O6(Q3(2QJ\n> MM'.HFO6=>&0I(1;-/4MI-MY(0_F'=B,[email protected]/Q85'HLFPF\"_/G%=_9<[RHY\n> MO__VQ4MKE6]$0/M[_</W\"]W[$EDWNSDFRXT)S%-Y1W3.PW(QZ5$F+\\W$\"BPB\n> MRW-\"`W9=FC0GDC-J?#&`#RKB=,/HP$KRE@$,$=[O>%=%-3W<WJQK9O&4VU*V\n> MG]:\\GQJ^8'][Q<I`1^UL5\"4.:.?EM+M0\\MEK<9!TCRN-;4_J\"(-G1)O_:Q+C\n> MU)Z^[)453<?ZLCG6T.!J</L@)92:,/H,[email protected]]'D6)+N_EI2\n> MLM0*Q[46M$A]M>%\"L0KMH*],Q$E?5;.>;F]W_<'.:[;G<L$M<:-!D6B3M7HR\n> M._\"/GW/X5,_G/ZMQ(W$(!>0[;4/RU-@Q[?-D#Q4H<C:0(V(K1\\3>W7BL_@OH\n> M[Y\"5L8S_^<X0I,:4NW%>R%M[9C4'9N_(RGP`WFU.XQX)>SLJ'T_XDBH9N80_\n> ML!D/G4N9V9;]P1R8ZOGB(/Y^)SC^@E$]\\03WAH-<OX=&E6>\"&21'].1'3MJC\n> M\\/WF7%0/E@L-Q8$Q0^,+U^>XC;1SD7FJ/?^?$:6PPQK=U[.J'XX<OST(-V$Y\n> MZ[:W>U)*\"9L*@+0S+M&-[$2IL`@.FA#[/O4;<UJS4OE8-[G`+VQW#.'`WI#K\n> M5:4=O9+`(DFB'%S6:_>J`#1_D\"^]S71$`9TP5-?\\H!GSL$F*TCLXLFKM#5*\\\n> MLTEBX;5[H^/7!E47]D&0'A:QW4--G_A+O]X4W7W3V&H)8Y_=)+%W\\-[8N'R9\n> MO46QT8NBUK@1\"2%,6JG7%B.=Y-WK#.]<[email protected]+K?K9,!ZZ&YKL7T7FY\n> M-6=+:\\J.TE3.NFKVT2.U$*YJAU77[(AFG\\7>(Z266G\\6.\"(5+:T,DOSY7)\\C\n> MA,_;8'K1S1FK_.;6'>$X&6;;Z\\K>):\\9\\-U&7WB-\"6&P'@`L,X1FZ6I98XP;\n> M(A\\7#`:[[X;H+S:BHSMZK>1?X=G]G*4.J3?#Q>,.5P<@@6IM/15W+KB#B`U\"\n> MTJ(VUYW<,HC\"EGD&C6OW2B_K06N$%^3(H-,T2.,1$PJ0&+0@/=^+BE9*$+26\n> M%'[!RZ#C:ZKQ*F0\"ZN'J0MG3)O9D%G`A<'_LG0_\\=;G1!YVG6W-'##.$%2G5\n> M\"E$J+@Q?=N/29<^`N\"1V&RE3;WJ&'T'?\\,-D\\I2R?;JGVJB%L-E?+\"25O`I.\n> M\\EWW4KE9PXQC(\\S;81`!_1J;8T<:<P@9#&]E]QNOQ61+1$ORD4R#&VI$^I:O\n> MSG3#:XIX487H4.K`0K\"0*`2C9N]O\".X*(F'7>#EII,N9*,D?ZZL+G,(;HU?F\n> M;KK>XFANV>#Y^+R7T<:(VD=&FTKC[=;7#>K-VQ].GL^QH(*HQ6JW7=XN5(F#\n> M@3`Z#@R\"$C&,-Y6(+C`@UXUR*^AQ*-P*?=NATT.(QH_*@X80\\L=X#*(J'D./\n> MU=KM+(7FJS3:[<+5;I/Q+'[UM=LEN(CX<VFTVY6:8_0UD,!&N<VF-SXC%6[D\n> MK?F49^`06-EL;A?*5K-YCPIHOO1Z!\"Z'#%+\"5\"P8?>4;Y/`Y2X`CXIF;?.O@\n> M63BIUX*E2KWQTO@VV(O%CM1_?2K<21R!S$,*\\_J>O-R:8#=?#0!$]Z1-=]%T\n> M:AG/I/_?RU9:<^L^8=>R\"$;OMJGKGAG3J#+<NY@(A)CKZ`74:NS2.>!+5\"Q(\n> MF_$#':.RP27!C)0NT)(GDV#(#2J^].T:C!4$MD'R)=RPAZ*E/I.+)4#\"`HTC\n> M+O28MN=;Z,G87P/7ITB$*13*W-4+P$S$')?M5B0`<XQ@_]Q<=*S-(.UZ\\+@]\n> MFQF7$,1*)0-\"$(WOK'%R^[`',X<K\",5/8X;Z<>4.7\\_`*Y/A%1\\5,0=0+VC5\n> MX42P<FL@C@<T`(K)U6\"P^5=2LDSNK:[\",+_[^Q?MR<RKYAH,/D;QE?\"W]-K1\n> M<6[U*BR(EXWNEQN5&$0OQZ1:;&HZX_]CKI:U/L,4LM48,VP1HM$D`MT)XA;A\n> M+0L81BXY]M#.,NA-?VG;U]2D\"WW*]JM%-X`61UY&3-`[F/I+$WM=N!,?I[EZ\n> MMX-$>$U@QRNZ2>>FN-EU&[[F.FR-V+Y7IT*L8]:6W6*'.I)D&X@QRY7:U@FQ\n> M$H:*N!<:HWIBAXN`M[%7/&[`K_M*+4.O>.>!.?%R^F@AY+91G$=0<@=B:_X:\n> MS!7^,#7!2A?MP6EK06H0;2]#-8B_A3%@!X?%.OI4?LS-C^WTD2K;*TR>>LJ8\n> M.E]]+>%8RTA6[7DX<;1YK[K-*Q3@E72%,%-VT_!3%^1/0+:^VC^_G.HZU^2I\n> MJ>$<-`#@N[[*BN@'`OK[:Y+5E@D<5XRG*O?!%S@(&UH1ZAQ5@T\\=9EK[=48H\n> M0ZN`@<[)I;1;:M_41LA;;-\\K(91=6`==#M!%51)V4P5L[N?:+YO'JSTL3IN.\n> M\\>JM`$VTB>R]^HB#SAJC1J=%V/-;YZFIUMOE]OK)[]6_FZVDB&0JG8P#3^,>\n> MYK2^1$\"8V/OUT5/'_54WP$JH5P1JB5USN1(T\\,B]MBB``6*:$^6QY>(V9Q6#\n> M$)2U/%H))XX/^\\Q%R`ZT#,EF?69E.';[5[1`:\\@G71XU,?_[OV>OR3W@\\&KP\n> M8QY'VME6A2+BW6&86GA>U<(R^)':ATK\"'KKO4%UQT+40YTP%_5GXF,8087;J\n> M#Y]R>/;-'851MCM[H[G'?B0I*&!G(JUC)=9*@_V[FJHY?.@VM:BR6U*_X.0H\n> MTMK%Q04=4>2BH)5`S[_YTV-.2$,`-]-O7Y\"&BL+KK7)/GGX9D=X;D%_4@?UQ\n> MK</<8PPC<7/K&)?+&2/BUA8@+GOCON/.8!=;IY6$(5VE]37+6\"HMO;8!^B(V\n> MZ!YFR6O9R5RM;(M+\\_7QK$@U],?]\"SHYU=$!A_SBR/DC(B9N<9@=6\\CQ,WO.\n> M>@NX*1Y[RQ)^`/V_,R9ZE5OZE</*\\MEG5&[M,VL;!M(B!JV;(BU55/<0L2ES\n> MN<1_!X!\"`2:IU>GQDDT;'O\\68ZX6DX;.,VD00R:JH/:\"956B<Y.J;1M^B]I2\n> M'NS?*E.YX:F,DB3\"*TJ]+CZG\\L)NHQ)E9;J,UTN1XMCE4&0IC<NHW3T%:L7;\n> MDU:PKZY7:(J``T/2X*$F'IMFR5FKX;*WL:,P#0-QX+QOL&A.5D%^O)1,!?[>\n> M#GGV$'IL:%J=+<8;3.`=8*E@,)BH#M8!,S<.F&@-H74]ZGYI`7OA<:G=+WM;\n> M=*K^\"^COD,2U3/_YMTK'/2Y=10*%.*I'H!^':@3YZFL1\\#%`4+5K[8^Y<?Y/\n> MZ1\"$%M:H$#2RFL73,<K0$7U\"/7K&-UI(\"X])0KX#0'-#1$\"<N33CGBX#U*X6\n> MZX]H1>*[$FV`Y2H8T0JTH_3:GJ9@T(@KS_C2-&/K'1)8-=9#T;,4\"0Z_Z$JL\n> M\")V[6.?>6E\\3#IM0FZ22;5R/DF9I^&@V3B%P!@I6KN;`,LS\"PP8GF/ERZ1NA\n> MWM>/C6NRB82M=J`>-198!3V50,Y(G]ZM8NS)_J[B?S5V.VF]E`[F`L?GPQRY\n> M*A\"#'+K.N-%*=Y0'$I=&Z8'@3O8\\SY@PM\\P\\RR3F3A-WM1*+'X[..+Q`K4^Q\n> MJM\"LF!P#^BP=K<+J;$;1=;>YG-YT;+U.YH(1H^]$H'&AFGZUF57:5K-G**/)\n> MC_UK)]9BT%12Z?8O+OLVJX-5(;\"-*YHF(883VTIV\\Q\"'7=F&7]NKY)$!(]*=\n> MT7LLM(J*5R??$\"D0`R1])>-8'*V<GG`L17W?*;0=*)9ZI%>#FVR>C?8R2$TE\n> M`W5J;F/,0827#]C].&83N''K>AH1P<_3Y%/U6#-KI2!\\\"&;KZ9;=4*[JQ184\n> M?B-LLBLS2EOK@8=UQ06J?&R0.3R%OMQ71DRC8^ZZK>:$Y_SJ2!=*FXR9CJW<\n> M^T$]#K2_G9BH^GN&-W7%>SJ'=1L#G1(J\"$89*\\;NT\\8Y;OJ,US.F.)$3P;CD\n> MQ)LMZTV\"GN)$C@-:<6(WIE?[[:N&E`SL0<T'!=:+B/R/.`P-^.C`*@L!]R.T\n> MM!\"^W[[6(G#5%A.#]N=P?[RHU[,XKP0_8U>OF6)$I&KY=N%LGP9C9%!RB<DB\n> MN!S_:]\"P=&:S8Z(Y.0!*UZQ&,.8H,\"%NBE0.OA9;!OZ%006?,ED)@P.*F,RE\n> M8J=%?9!?6G'U\\:Q*#\"J,'$TZB+-CO0'0YM<&LYDR8W'^'6,ALO2NMKQ7A_6K\n> MFAV'+J>_DZ9[K#+F`G2NO\"QMQ<(^TU.(I&\\.\"<.V9ZIS$?)_9E_BPH%Q+[2F\n> M[Z3.((#2X(C.H#(X++]89T`EX2/`.X9K9('_*1VZ>SJ`X*@2P-<!Y+FA\\QC1\n> M`01C2@!$X1.$&B94#CD^N.?XVCG'!\\<.\\HC+APV&M)2SQF\"\\S`C))B@R`VI\\\n> MWP+!X#;D52GG<%\"2L&YQ9\"1^VC_+9.J_@/[^FIR]90:^`*-NE!,N`#5D[/80\n> M(^SR0/@ZF87789':''59K#UYTJ7<6>#D<RXK^\\9[@L[K4.X=4XAP1PC*G5'\\\n> MR1&WO>^(RYAXH]U$RH\">OF^L-525M?8/*5898<MK&+PP<D'Q-&`>\"!%8&_<S\n> M^[K$113XX!H;WHE=0TY/8F=DAL\"[%O_AE$V$2E(,.RX6ZC#\"3A!$R(!5='\":\n> MO=TP$#>Y*\"`('6<#.<Y*0I:G`>NI/8^F$P\\T7AO+X]`9B\"F]0`FQ!;^Y^/*P\n> M-5;]`Z+<9](Y);B'K0:)\"`Q?F''!Q%J<'1&OC*GY+%`)+VX/6L9K['$!O]-O\n> M.(V+<R<E61'9TU?-Z3M>XS;;3BT\".]W`+@4)-.!;;5-F0G_8/R?_/C0.76\"^\n> MIBHKF6YW6T.AANLR`KC+`=M#)RV&A2-C5(T,8/UP(,X=/0/A$N\\>*:[1-WR.\n> MTUBK!=JAI2BC_T7.75DKDG[#]'KLN\"HVMV@0;$5R.,:AHG^D<(Y]D/1$TN_X\n> MM`\"D]DNUZ31\\.+!&H&P??X,1R.4\"(!%97SO*FB!32QU^^KOOSC@ATAK]@ZNI\n> MR5-\\#GQ5#=$6.X&-KH8][NSIC,XFS$T,N0$SOG!F/+L\\ZBM5#';-PN==I+-.\n> MA`]Z-^3W2()D;O)WY^?*N*L(\\S9WF<7MH.Y4>['`*[+KU_P-^[@/[T-]/RH/\n> MX^OT5+!JH:-^EV().5^PU3H7JF3W^YVQ'+S>DHQPP9S?2L*DW[^1-GQ-A,LU\n> MC.GF(R:M4(88:?:PO1%/76?FL[$N2JR*L5M:3`V[4)D&:;PSLDE(N(8%XL!;\n> MUJ#W@NF73`E>W`,])RX$A1%=\\7RY)%.$C;7Z=0FIG`%/-<'A.=!U4X+8KF;/\n> M;[=.E/-)3U%!Z+'F:3/#^*)*5J9ZGY^S:<\"@*\\4ZQ8ONVGO4TYMZ1QZA1LDU\n> M;EL?,&D+WWYOII?JT&\",M+^3%(`!309_.S+Y!@'/?+.GG^9<FSJP%_&K477H\n> MT,['K\\KEM+Y8J1,E\\;VD3$RB\\^#]J.6N^/R;+]6QG4[;_<V9]1),%\\PFY[W:\n> M/(?1CCH%:JP+FP44-+UBM^<4VU'?:4L:5\"$W5&EJ5%Q@:=-,8H[C#_O^!8;-\n> M9]3NG\"I4A/T*4>9?_#0GSQDV;>(4'=?_T<DO-EKLOT!*PG.<L+72)-$.L(`+\n> M4?^BJ@5ZBV^)Z(#-'JBZBJ..[D%$K$J(^7=??<\\P<?J7TFX+45(1A;.-(6[N\n> M:55YP8TEX,KJ3%N!K4&!R4LQ$5V,MM`9Q65QRHY=7%L^,>K*K-7:-#'WQERT\n> M:EEK6W[!I-E5-#HF43#K%6Y2Q=!P>9X8<(@;56XXM7:@Z=FCN>:=]^'LH/NJ\n> M0CQT_96OG3+2N3:*706N=929%`,PF[XS/+>:[email protected]*;]S9FX1Y/-)2[.D3\n> M`F79<M\",CMT!#,72`<KQ_(WN41&\"#T<P/\"?BB8U!3F,HB8`:0N`.&73HUS=:\n> M;28]I*]B[26&Z3735U@T+>($BGIE44A@KNPH99D62O(O<;3:\\I)A\\\"-20AI3\n> M/_5G59D4^\"'P)U559%YX#1U!O2\"?O3DE/I(@Z=0FE3)J=7V9=84$S%I;RI.[\n> M^5MIMQ@G?NW]7.O)P\\X1]SK:80XZW(DKN`YJDWB1P><.O[%6TO.M-G)/H#=P\n> M%+1+G98#G=\"8%4R`;.8]_(2V-U)>PQ\\QUMAAK?9#A,Y:+-@\"8\\)&DRY)@`8\"\n> M7[BS3=D\"SX:M9\\+64WODZK^`_@[9U,O\\5V9)(D`\\SYX.>A\"<?E\\9,9IAX&H!\n> M<-6&;YG8O:6I*$*@AV0LXD>B:]5\\`G*\\LU#O?)^CP?X9W%3CK<)!S;B.:P*,\n> [email protected]&MI(:X7#0&#O*K'(IV25%)Y&M8W?#W4E+:QMHD9*&&?T&T4C@5MF\\\n> M2J?YM:G9([+KT*P@GW%AWKXBN`+J=_Y%E+D&P9\"5-`)9:'@)P*0DO`36A<H2\n> M6VCMN'],%HQ/-KMR\"!Y<]B7&5-5-6_.)N/$)3?B&VP6X-9<0FFY)2#)T!UG_\n> M8JL$)_YR0FLPM,VV4PSUDI>,1<Z]T#=47JBG&IHV=&HRTA8./>%OT`NBZO;8\n> MF9+2:1S-EY\"FY81IF?5XA4&E@-*ZW`CW:M\\\\LJ2RROG><U\\/2,$M+\\+*YJJI\n> M10SK`1=5DR<)I8&FX!!^I-CC3AJE2R<UK]R8Z.[_7.NC'3\\GJX&N?U33G\"$4\n> M\"YP[<<'R2'=MGV5+?!DT5#WII&@_V%\\@MMA#D=Z:]/%\\'<.N?:_$3QY]A8W`\n> M+C5H-_+\"B(PQE#2E0.2^O+N1W[7%DVE('8TO5EQ[-+L\"_2\"\\DH)Q&CNHT'*?\n> M8A!/4WZ#^I_O2*#^_PKWLVZM784.L[KA_-VYZPLO+Z30ESL35N@'UCRJ&9O>\n> MR-P8154,8Q^P\\6+KV2I2QJXQXZ=R#Q+*O44W_9S=<@T!*XXR6IZ\\PLTV>W=^\n> MPM1(,V84*\"RL_2/ZT33W)\\*9H&^VK'BL'46Q4A<6Z9[B*RGM$^,R:ER$'=0G\n> M?2<-#5QA8>X?0>J5ZVM99!SG*Q>(FXP;]S5G;:Y<6MJ/?V2J28CA^W;[EI*Z\n> M\\91L@0Q2'L7.[0M=FO#MRV`4JW)]QQ?B&&N:4'BUNM1[5I3EX<3`1YCH9E>@\n> MZYC`,K?SU8M^XPN5@1A1J/\\\"^GN,K?V?<GM2C-,*F15&;DRQEEAAS/$S[$.F\n> M4Q^ZE$1,AB#).408C;Y2URSD?#A#]PL`G65C[R7C',P:N:!U>N')[R'AN^1&\n> M$ONI1UND:NJ23NC+'9>U2-.MX`K&TD?X^=%5C-$\\R!6\"%EM<DB8E=@3#+?RO\n> M/:[/F@\\D>'.:J[8F`M1A+'8<W8#YCIM0#1_/>B>BZUN9TYQ.,9(..8(\"'D<?\n> MB([;8R(1EG3<?5[55HL\\7%N[M&DADB)J28=W2&N<VVON5;^U<49BUHS`D8P2\n> M1[Z)W8[3C%;0.PU2XCM21-=R30IT`C%H/J\\\\0L&D91[0J-QOR*H)I0(+O>\\1\n> M^V\"?TX0C^M9/F)'40N9%\\TO8*YY$*0QK))'\"T(#5;@=:DN?%C:_`1%RR&7QM\n> M)\"IO!^X9S;K,4GV!547]\"UL[0Z.SW;[![F#.!C)`W,7%VMNA=;6XY0]75;`+\n> MO9/K\"230,*1'Z!I!%GT,25BZVDAGV;'U\"9;9*WMHO86^?\\[^57*69F0*CZS*\n> M2]!%=9K*S-3X;JO`57B_E8++\"NNQ6;D3*S\"$5'T!VK=UASN(QSS%B]E@EI4R\n> MUKKQ678Y^^PSX:`*70XJSS(>B9`)QJ`(<&?\"/;G'8\\5E%]8>34_%R\\#06![T\n> M#QX$H''>AI)(3B=S24U(JXAJ@[@9T:Z.B9:8SFNZ*B=G8RX07&D1`!3GL*CG\n> MJP)AIN+B_]SCE\\)^(I?G5A;EM])]T[(HZ\\]#IA$%!%J>\"K+]%_JN%I^+-'$0\n> MW`HF$'=O=-E_T3&4;GP4-ZB[;K2?>^,`7>G;,Q\\B@0TJC;$;B!(:#9^WK)\\P\n> M7F$):4HC=RU$,Z2!?W%#3F!NHPK?`1\"WM6Z5W\"PFF`AT&N)]I2'>J>PO$>.K\n> M:T8D59(M%4GPDAWCYM4`_%V5X$N_!,8B],K#BV+$#-1[9V[OE^::JX^^95I?\n> M,(J>:!`^!M@_#]B_!Y+K`N_C+K7/_A,(.B`I+5W5J-7H43(TJ$`T\"/VBJ%ZI\n> M(ZN2`6>@)#-QZE&D6D:Q)5&$$6N=B_-Q5\";//E:C_I`0.ZH0#ES/;+>B3S0:\n> M'Q7U%FOT&VTPL:O9!H%M:XU=I8!_:=E9]=2;O9CH4D,3;%DCBM,>,7W@HT2X\n> M:+4NQ!WC]ELPQ8[=B.%;R+V#G<1!CP12W0:\\$P*`)J#2M?:ZYRXU,\\^IVTI?\n> MW\"7Z!@/-:7-:\"92BYIB@+'5#GX2,YT9GN5.40$'_OG1E+4[(A%[PO<U8H'MW\n> MOF7O>O41B%,Z,Y%\\I\"_IG3MC02T#WD>B*2/4EMY).W@CWB`:.$:SVMH\\&&\"K\n> M-HQM0(N[T(&1_0G/)DY<S_!NSQXF(G/4%JC(@<`S\\V-!2=5J3CVGHR'.!73C\n> M<[!XPA;$H%MMMCL?E6ZV,H82K5C2U\\(W9+#3!Z?$4OT7T-]BJ&PN__D,;0,4\n> M.EQ/;&HJ[7QGV)X92(<G\\$9]6-3LH6(-MH.I(UO-@:AJ[OMDZ1IR,CWA_'&R\n> MY\"$E6#\\NV<4:MV5T7TV5/H:R4LO.^904)B0';:<>OLHDJ6#'UK,C<WN;]^%#\n> MWU/%V5?='9H]D>K#7B_S\\\\TX>!?N-9SMD8793K`M6)_,]#U_Y&UJ.7*]QG`U\n> M<HUCJ`1:ETJ`I&%'?XB+0H&&\\DP_5M;R0OM[D%X>IZZW)%N]GC;U038PUP\".\n> M6L\"@:%J#&1@`G<'?<.#4[>BPM\"9\"LI[AL,>%OE`+SW,<5Y=8Q_6Y-]!'/J!0\n> M,5+P8,>V=@5Z_USY'CU!#J4UHRUU@_9NI;VQT#C79HW;WG(/BA+@/HE%5P\\@\n> M'>92VI9/F_)=\"'!>V._RUJD:NHM:ZLX:!+2GJ$(<BSYIDQ7:Y+&>4,XH!$A>\n> M+GQ?GB\"Q%*E$#]&9\\(@;_=\"8L]EJI!$]1(K]3/MZ[6BU8#-JN4W!L8B.>V8M\n> MUZPIK5F\\'YL^-8T4B.&?U+4U=7487Q&'=R1A\")&%VP!CU4Q^5N)\\NS^00Q2V\n> M81=?A#;([735,[email protected].&QA@Y]O-H8.F=&:1U*_K^N!T)=7BL.NDXW!-\n> MA+.P:T_:6PV/>CY!@L1Y5KQ*PCP7K`::2^`C^WZ]U;=B[/\\10;']A5[_:,E<\n> M,=.XL(]%D>BU^=HBAF);7V(DH:5H5&]0/[,N/DIC46FOF*=YO87IE^K'75>C\n> M](&TRMXTE4`6-AU6%)I(I(J/4M$L-PP=@9KS3:\"CH>.@&:N1Y4XJ#PU+N:KQ\n> M<BL7.+FEF-?W-(E3C:)TW\\HTT?<T=%:.G7N:YS0O29VLYL<C_[K!<W35TG_!\n> M1^58%-VR+'O@;H]D47ZDMD8>,@4S/<=<+;[T\"$M'\\:Q.NRX%7!Q&IKQQC#MR\n> MHG<YU#\\=GOR^7OQT>$4^HUHR^FQ6P8N<Q#<0&H/)[\\GO%WL\\7*A_7Q$X)D80\n> M!L]V*G<6\"6A6>!;BEY:OBF@L?;?;OE:GNDN2>%Y_M6FVY)_RNV[_)]6GAFLN\n> M\"T/G@&T:=8]]#O/+,/B9<8,9RPW3LG-?8Y!<XRP/ATIE#!DC=D.&;+J%MKDC\n> M44[,D?6*$#2&'S3.XY$4M8P'VW<]K]GP739,E\"8OJIZ2FE(L$O>BD+`,^N,8\n> M:I'?D(+%N2\"-B\\J-QQ>D<9FYRC@H(@66T[DBC:NH<JY1[1K?=#^]VMYTR[U;\n> M!A*8?M\"D]52-JF#53H$#EAY*NA.3,\"F]BT_^6+%J:N6HL53/8Q!!72O<U?:;\n> MI!9EI<,S+JG%42'S]2U1I$`?YK?9;SZCIIJ!3A-H.[_[G40M.&H%>?;2;:+?\n> M?,;6ZO\"D2-(@21(';X'I5,W8!MVH.BC`-W/@[%^I_P+Z6XY<\\90/N>+YVS?J\n> MG;8N):!_\\>+[+SY_J8;!\\V_^I)KLB__U^1??O50R;\"57/)6E3YM\\V8&IL(@F\n> MPJ;#?'IJK/^79S_Z-[(80?J$K[9GOA`:[4&K!L:BOA',[4Q0<YE^!!L0\\2-L\n> M=P=CN::6:IA\\BQW2AFD3A(9!323BR)1SL9%*;HS#!JRR\\:4S)6%SWLU2F]`#\n> M:+1CV$[(.B2E4?'H^OB.M0]4E-?!\\>`63UP+,4;%AMKRQDGM@ZHR\"JIN1'K@\n> M\"FQW>[F],?X%C*&\\W6S0B!JJ4C1[*418<$<L',]?B$\"&*W5_P3Q&7V@32C[3\n> MF]\"-%]K(0!K4AAL;U'$6_-4]A:+HK%[)<`$BU;BJ#V_YV,\"\"W-'R44HUD-/%\n> M`5V..-=0!ZTL\\KH]6I,V[V!Q>364/>SEF3X#Q:GLUL'1UW,6RNE7J,=96<JP\n> M\\\\3]F!HCQJ]5O7'0Y8VA9Y%]8-L8B&J-*T8]F^51'I2IMM1DQ_?/H__^//[O\n> MSQ-9O_[8^UJ$^43M^6I->O+DW_&?*.W-:Y&H$-AEH_^6RX;_IC:>)2!]5A\\Q\n> M^]5')9<B=!1\"5Q*3TZ,-J\\)D@[!!%&*332CL@L(N)6PB81<4=LGILAR04MB:\n> MPC92AE+\"UA2VF54$$AICJ_R\\5]TH3,CL)@[=\"BLQDCZ6(2?N)L4_I:\"O*BE+\n> M_9THFS)-::SY$8%/4!\\^F5U/]SB:&!-\"Y^8(?@^&RD7+^(UK\\=WK<?$L9T(6\n> M)=Q%P8`LAZ0MD>.?\"],1^H!7(_Y0DE><X5!R8OR1Z5:BD@F>#YZ^>,\\`[+PB\n> MTHQLW.7/KG@]I'Q9+7BP/D.[*R$OE+'I]T<,6]Y>?\\1L/(OA%9@A(S]!.HRI\n> M/[SO;#>:T'<9-O1=[2\"AC)O&Z]>8#2RS03II3$W5ZV\\BY0E]^NVC2YEQO(<-\n> MQMU,L+^%:H<,@==8H0W'3C+DV/D\\(E6`:B-13K2J9Y,(G#N6Z:B9?AY#+&VF\n> MG\\P_87Z=`JUI>SXNT9*FYR7AV\";\\1THXB,O22SC1\"5]]HOEX*MF68/.(X6[0\n> MFW!&#/K*+_3_,VY63??\"W9_HCWFBYZ+Z`I$&9Q6]KESIKSBS8$WA3YH;-\\(R\n> M(Z'81\"7C3[X0HP9W%0;J+RTO/0UD%?[*/!C?;`\\U@#F<6U;N(&PQB^WZ]GK#\n> MQN0S[8EGR;V\"'K-;MR36B`4T,9OY=;U_++H>T@<TVE:I#SFQ(B1]0$*S8A*,\n> M*#QY03.N`MW=U`3B09;+B3D7.+P.K/X/+'O*VUY%X\"S&7>ZXEN!#PEJ^'M#T\n> MCM%T[%0@FN%,6LV*\"IWHY)!^,)I^;.X\"=?K\"`HI=6B!_,9.A\":9F\\U=A7II%\n> M\"R$=PLH5ZDJVT&$T_Z6K0VE'Z^$+'D#B%JOCKT<WC!6(X:RI`1/%]?C;J.)6\n> MNM0Z8+5*Y=HLZN@>H4(%.;K=)D`Q\"X:T'-\\EJ-!EEHVO`6J61V$QJ;*\\O\\83\n> MHG04`MO$KO$$3Z-^R)VIKM=@_!0E=+OLKN4\"WI^[O+I0'+%8.!-76RU1<1M!\n> M^*M9%'8:*$J2\\HP6BI**#G\\2'Q'3K+BW@2911D>\\>UI([>#A2!.1RVX5#)JH\n> MP\"$['FNB$B=M=UL#O'V5NML4'9C6<UQ_V^8Q.N5FJF1_X6KN>`Z[_!I6I^C;\n> M.[+QNLCG<GX`0P2#BBMQO;MFV_A#RRM,\"%6(ID=_2T<+?:,H=%K?:*X[6-5C\n> MY`<KEI7V6Z0VV#YGK4'=67F$:RPW(7S_?G?O#H-8*&%/#(.8>5EY2T4\\O@T<\n> M'042H`R#_MR*F09U+-;G/';B-(WN'3MQE@Y$*'S/?1D>`R3.(0GT15K\\5$`F\n> M*'MCA[W>O+'C8EJMC+N,>(WNC7(88R%P1$/NI]N#N<E<\"T/A&]6H![J?^M_T\n> MSQO58+N`?ISQU\\M+_OPY_8-[?CJDTE45.PZ\"V/%F3BI$ND_92[DANLQQJ4-C\n> MU%H/NN=S.=\\NA*317LU<8=R351ET<OOMHE-Q=+W)-0E&`#$PDJ@\"+/U1'>!B\n> M^D<\"RJ>WFHO>.6]_8LQLX=NQ.R[UUT$<]\\#[WG(;V59J)`=II0FWDAK0YBL:\n> M:2V.C/1UQ3G^(+Z8M'FR</V2>VE^;7LQL#Y,ICU%D4\"+Z;J>&RT`TV;1M#NP\n> M+074$,$4]FF'F;ZJXDG%ULDG)Y7(>792B8AW>6Q:D>5R@F9SIA5]9*%O=%H-\n> MP5Z4\"%M%`?T-?T5=5A5979:=/)$93ZWI:CM-9%\"I\";$,O+YV?'2%)\\FL$72<\n> M2I)DY)05)+#L]0]9:9:,[+_T2Q8E(T>L(\"O&3EA*WHC'#EAYU5]DP/\">0()H\n> MT=P1,^;^14[0M!%`A=<3AVI]X\\XJ/[Z4_)O</OV=>T<-\"5RMD^\">7R07L3\"W\n> M\"\\W'E\\`(6K*(:M4FV&1F/E\"%9\\AI;D+=2\\#`G'M-9S3V+&SNB@_FVH<U8'M.\n> MW90N``F)[Z<@@K[88&!YNMVHU4H)CU;U%?2+U0C-O*=N=`3B@4K0):YD7[/K\n> MZ87F22<1YSD[/#NG>RU]\"DK\"?\"]<:QX#.`FG`*ADKJ*]H-O$Y+@2]/G=8(J[\n> M$_R;T,6_$5H/&A6[[D![$?GTWNV5_,S!4T:4Z2<X,`@DHQP-/^.F[5Y5BTXS\n> MRN&D0BV&[*Q[G^%)6YEV(_TP@[*X/N*$)+E90?Y$]:FG<(GC7)(U#F,8!OR7\n> M)+3L>'LCTRZ1U*@N`C(`T$`N(FX@5=/?7M<,2M-HHYL9XRV+'->8O5I;L$&-\n> MPJN:[DECSC1ADP=>0_>&`'K$VF'/8\"]Z^>'X*%H1\\W'\"%\\-@1/ECAY[<&RLY\n> MUC?Q46$CK)#?\\6#HIM^CY1\"DTBJVYS@,VS$;C+&MRG#2;<\\,P19I:(E)WRUK\n> MAFXBDQVMU09C$P_5..0\\>R,+ME0QFZ3I?`P:9[>&UY)P'>O[@]5N?GT])\\%X\n> M=\\O*:Q]$\"2[F(,Z87U_<65,3E\")*ARQ)NA1,A'R`2.L@;-!&`#04`G03(PG9\n> M:X3!C?!ZM@>]AAF3)\\/DN)FO[U@&<$Q+P(>XLA6%WTT,FVJ_QBO=,#2RWTS_\n> MNL5ZJ^UQ<*#8Z<*Q?U.W%$87)KZA[/^KOE@X)IY*:);5B&Y\\[+([N#J`4+?7\n> MFOD:_4I>N_H<L`H<U**>LLY<<_@0HTV/45EMW[EP-7$A-%B'<XO;4]TZ)]*.\n> M46[55!9-;IQJZX7C<A+I^S(@,FH8]<^CSR*.GL$TV]'SY8`P/\"(]D8*O@#&#\n> M2:FG.22$C*RO.<R.Z(P_Y[*5I9>B41DVGVC[7)?!N^B=)YK>!F/[<V7[<Q5H\n> MI<RNEGE^Y#*G=;JQ<3FQWD!)U6?&=N'5,-RZZ3]FZZF`OQG)JF'\";6W?%UC]\n> M3CTY-J1F8JC^_+$_E40#14A'FUH'^N/,>/#K+4*,=!F=NMN1I2H?N6$K_;E6\n> MK`5\"8&RI$'L;QPHP9V9Z\"JD<%(I&;%`I!)8!A\\[=L$\\@#%NVZ7OFT&+SQ1)N\n> M@60>HOCR'*BB&90.AQ#Z96^!9HK+Z0#>O'560VW:6SMF^:^%8YV'0NM,'>Q)\n> M=B@\\5I(WW1T)X!U+MXXI+).B4LSO#]=\\V>88,H(XS-O6!YH\\L0AD`WB+\\:Y\"\n> MB?J<Y27/G%>5:HR;$V;/:2CWT?5A<3&#&8,:#+%5NOJD6\"NF_Z20K,3<<%B(\n> MC\"[,G\\9E<?6:>A#ZL!$\\SGIGI5C]%]#?(75T%?\\:1KL$M8WM^`]7]:K;[&]J\n> M=3)?$UCCY`_K]4]DB$$/=Y-4+41_N-W]I!YB>K@+DJP,)W_8O>TF?Z`;6XZI\n> MJO9'2NA/VP7L\\IY-FFYUJUKV*0_-5XPX0?9U%S?[X-_^.E]MC#7-LTET$6;!\n> MTW_],K[9_ZE;'$BM-5G2P[)N`OL9977>GUX?5,G4)KWK?IK<W!Z\"IXOM^DDT\n> M>1=.5-'4@+_[>7+5;99(AGX+U4\\7I#)S_^YWJZM>N,@/%QT+%^MPT?WI)7ZX\n> MH^FEDW?1.>7+_'!'T\\MUN!/E*_QP1],K^^U'AE,CX:I^N+P:;^=!AY3Q>$#T\n> M\"%T4JB#1/26,8ILB%>YX%R<V()7N>,#4!J32'0^8];,^TCI1WL_Z6/,4_:R/\n> M-0\\ZQM;W>!G1,[:^1P/&Z!E;W^,!HW[61VH=Q_VLC]0Z3OI9'ZEUC)[)$\"2Z\n> MKPMC]$R.T4-H(\\<#HF<*!\"&[A>,!\"SL#,_-W+&!I`W(!Z.]8P,H&Y`+0W[&U\n> M)+0!RU+_'0L8.0'3(Y4)ZLTRV,]_K-6&H`),TBJ[R\":`YB/M11!-U'JZ7\\S7\n> M=:`2O9F\\6ZRW^_IF?FC=W.IF\\J[>-DJT<[^N=I-W*S)35N<$[_M>?:<\\W8_[\n> M^>1=_]M.!1R)OYZ\\6ZO#SF'K?KR>O+O>_MC[N%-?=\\//F\\F[3?VV7PF5V?ZP\n> MV[[QB]6JK^WVK?=MO5`?2<Y00O'\\QO_IM?GI];;;^+^]-;^][9:]S%6WX%?U\n> MK^J[[<ZKQO:@ZD&RI5\\V*@9UC?=QB626\\[V7?J-ZB*RCF^WFX`5O$'SP>=%(\n> MVH-?WJ*9E+0TJ,-!];<9.=X/&U7^Y>W-A/XG=LJZGL%DDDZ>Q).=.N#A]VA2\n> M_[1H)_O;JTE\"@P\\_7-^N)_/E\\I\\7F.>%BN)V_?)HJ9V4*0'OS<N7?NW/30@P\n> M$)<F[WK\"S81__()TC?7R>^KQ\"<T*BOBS$_L+E=:[85\"9+!,UJ3ET8/-2$YRD\n> MHNM.A5]WZI]@,PDGA9)\"K]6_X60]R=7FJ?^E[^N)FO\"+=7<3T-J1QU5E_MTO\n> M@H*7O/7;X-\\FWVW7=S2L59*I6K7H?]>3J`SY<3U1,YXEL_U$+0K#\"+%:H24\"\n> M/:[Y,4H0F0+(HRK/J:2BL#29X_ET[B102`QZO#]\"1$*(5#!+3,EZ49Z^[*[K\n> M_9,7V^OY9J+F7I31ZCNA:;5O@DRMQJ1GO0Y4-%EA)\\[!:4*GD,GE]+DZI^ED\n> M6Y7NB60K]5BHK=)+=E+1LDXKR/1BHO[_(0E2.8FFRR\\G;,T?F`K,!OUD<\"(0\n> M+Q0O-;]_$C(<(S#PZTF:DU%;A0%R;Y\\BG,1!_--QU(F:#.9BFT]X7AP,&XF#\n> MY[573DZ7GT^,7@[J5I6>SREV4MHX]'RB>=`D4M7CS7-_IR9)`:[W?J\\.!_'G\n> M#QO$24&UB*OLPPUCE!6P7.\\WD#D=NF9[^$A6X;&4DM`1Q*2\"@GO4-9D\\T#5.\n> M1#W`W^,4SS%9R+K/$7K6K,D;NK3(.\"HE0VL?)^-WI\"J'*LA\\M]N^;>OY4L4[\n> M*WM^=K[;[%7ZW\"Q\\QIXHZ4_EH_,\\6NN$;D4EVR0K3/+\\G;--\\BCSGP>UQJR0\n> M6B=Y>':MS\\J>GYWOOZ36&^D-WHGBHC([$3_SCL%ASMK7N)J<7!JE)CE^YB0X\n> MS%G)Q5EL2X?GT3W,+T)9F#C\\_+`X;E%/QTFR?E6/Q_''6$7+<)Z14)-4U,!%\n> MD5+KE$6EOO-S!2,O7IXK`E$DV!)OC&$1TF.L4EN7K.PGQ]A9V;.M6L)%D?#O\n> M-;.BLJ!KCPI\"2063ZCB$%$7?,QYNH6I,[YG\"#]<31+WF_4&2.57K<[*79^>[\n> MS?XA,PLREVR2+\"^>WMLYG,1!_#-D\"))*0T<L/2T.0/ID:8\"E3QXS;BF1ZEG\"\n> M`$*ZU3PM\"R`*BP(<Y;0HP*TAU3RK9;QQEQ91HCL^+8FRC#N8OW/'IR4V7^=Y\n> M,.Y2.DC(N$O+/#QWW)V5/3\\[WQ\\P[N#K=C/Y_F9L\"X_22:DF-+;0<E+&.79*\n> ME2D!^ZQAJ3CA12:.TW12%,7(;\".7FNO@7]1J%%X0P5)X04M365ZD^H56\"`JE\n> M#J\"DQE!AJZ)0/U1)I?[&89$IN:B*Z),JDPJK*F3\"DH?O13SA!.,(8<L\\4UEA\n> M]Z'3C@[KMW8P*C:<JG*4$&-4B4<)^_XM'<7$+U6H[KTF3Z6<W:SIKB(F[O*2\n> M\"@$G!?4=SU2*B(:#+S;0;.`E7;5+$F47Y!\\>EA<$V%6$>\"N+\"SJ`AAE/!MN,\n> M85A=$#-B5E*79&5$P;,LIH_L+QX[P:-$]0T!P*4A_1,7T059F)09O9';/*;G\n> M^0U_5@O$X%8/*SSK\\&>U_OU\"+XQK!Z?*Z7=;V$--#ML)B=&P97B`*(U4JZ@O\n> M26]5N/UAPH>(!Z47PLJWZ\"?XOL5$LGG1K[TJ9[Q9_H)RXBQ\\SJGICP\\_^G_8\n> M0U,&<>&]STQ(Y@%'IOM34VMW=J0_DMTOZ0].\\!>,&]()W\\Q7=:#5<<%.K6/_\n> MIA[T%6@P^0.QGLC-:IX2+YZ:EW.VF:3+T1GAUE7/9J^G?YH?YK-ZNC_L;A>'\n> M6UCC:_R5@$%I5@YWN'/C#\\>(5!O;.U?OKP-[[W[)M]36'N)S@Z?FW;6#']LR\n> MF`26PL28='SN(+(X5]_:<@-^$BD\\1)%WQWG7I,DB.Y`O.^9G:9A3J`7P&4/,\n> M:*.2P.%\\,UPH@MW_'.5VL60TB*T!)X+KY_6\\`\\X*NXW$NCA[0B2TV'JP+A*&\n> M;$#S;,3'\"'T3U-0Y;.M-*[F86\\3@_G.-D9;C-F.OR68L,$9CC>`MBD$A-]N2\n> MNGWE=/O*A]V9K1QR>A2D\"$.'<<L0]Z4P=X0A+MG2:Q\"S[8U8-%`:7BNVPL#F\n> M43\")\"1(XK[S!U)C!)-AT<RE-)O`_3\"`1B3$=4R?%N>,'=H1UYK$=TL;\"(QA:\n> M>)#5\"-EO`(O!]\"#\\RV`$A,$$.J_`JQ(*60+$`IU-P`.2BF93XT1:`Q]M_.(%\n> M8T</I49&$(U&'Y^Q[4^.6M\"RND&#<WG@>.06<R%342R*&.U+D)_FZST[\"K3\"\n> M,R-(MH`?;!A77GK[[L8X\\C>!P+KN/1.5$2L9%*E*Q!K4F=)D;QIH@U,-<*G;\n> M@*G3ZA]`[1!5&O8@G5QI6RCXA_&P,%AFVE.DG1H&&AGNG5X\"C%%5RSX5`9</\n> MWF($`EI;F\\!5;Y#6-IM.(X4);F<=#.C8V<10%[&U<1NS3JY@$\";#2JW]XGS)\n> M;E=9R3:Y*Y>,B@FO!-*E-TJ\"_BB1]FXGPX8B\"S<&E`@#)9R6UIJ7#/79<+N=\n> M?J^MMDGR\"<6#F(B!R'9)PRA2E[.9L;;X[L1JMM%87Q;3<>7B4MD.]%I6&A)V\n> M8$!E`/:O&3>.-6C;,R/D@K+'%UF=+<D_P+%B-D;,C3&KTRB\"G3&IK3V>0EU\"\n> MLT!99R0QT$7[7X(MH#-XL0W!0EQ$W+Y9'/&2^G+V)`-0C2I,N^E^(`!-N&K)\n> M#UM^^TH[\"*BR_'VVGG8U&Y\\Z%GHTR+XU$&+:O6KE((MR&642VQH%9A\"*>RUP\n> MX6!US3M17*01CX;OQ\"+_6\\<BGQRV<F,82N195&XQF`<.G0\"6LBD;#U[#GHA&\n> MW-X>!DL&;!@U>&OG`0<O+2!F$QRQ%!>.53(89Q<MV+-?D9F]CVO'VUWCL(.-\n> M<-3\"YI`<IVGAOM[N.>E]S0BT-.UK;ET\"E3&FAG\"<4`L(:.*F=*[<6UEB9=%?\n> MJ%L94A+(80LVM4:;,0.&,3IMM=.7II-X/?U*'(X7=BH,_\"8^4:THZ,*'^?Z-\n> M[A!0U\\&%^O8`#[email protected]%-44+(B[?2BQYYG@\"I3:\\M<K?B,F6JL7*5G8:/H\\Z\"*\n> M;[,>5)0T69/3JK+L>[T`2K4U5N%[X_*]4I+Q7ITB:[(_1&'8FA9H?U=S=ES9\n> M:MR\"5=\"WV^Z6]?P91V0W-<A4U)7L].HC*M<L]-;$8FRA7LE.D[V[%HYSWA\\!\n> MOT/%%$(#(3$28W%&&(4I[G5]?07;8UY]Q4UL@U:LM0N+ME\"EP2#6MA(/`T]'\n> MA?L@J)-IDG^_Y;F/71G^4$)B8;@#.0V-S\\@IO,%@V,JP:?TPJP`%9#[%VJQN\n> M\\S6&/AJ2X1J6<+X'Z#V<;#2R)*:0,#U(PG`AO>9>VCI-(/Y6QO'R(.X%2Z=(\n> MG+WQVK%,481,TBW8AU\\\\!S%>V/%&I`+X-A.(PGYKN&[%`YIYB1IV%F6Z47+Q\n> M87E'B!65..88N1OGA^7MS9H1D2_4'N(9VB;JOX#^QK^F4V+B.\"724F5%<B&G\n> MPYF(;<P=3X;58-V0?@?2?:,]1:BM@6)G93H7/=R<:(3VT9QH:->+0N%6M-!1\n> M<\"=(73G>\\)_J8P*7AHI`IE^[.9Q5]6%2(^VH[A+4_98ELQ^^7V@:=!%JL*DQ\n> MR+:MMGM2`((8P'QAB:Z%M$%:=@N]N05PYF&KEN>V9C[M@Y:81:)WQ+)5CZQS\n> M9<0RJJ,(1=C]YAM!#-/6^3R3C:1(YMA$=`VGRYX5=9JP%36!*+(5=5ZQ%36I\n> M%=F*F@PBSK>BMH1]'\\VG/YI/?S2?_F@^_=%\\^O\\P\\^F$=.U9FGPTG_YH/OW1\n> M?/JC^?0O,Y\\F\\V=M75QD9YA/`TE=8M#C&3;/E(=87%,>8G%-C]I,+3:/ITS)\n> M5$@QGZ;XH^;3]]_`Y2K:!S5:3I0@,32N_AL4>KA=?6AB0PMH39-WC]TR^I&L\n> M$G1'TO,Y/2EQ\\)R>M/FA;&`R)]G@63I3HJ,D_#CLS--]`POA#]TYP_MO'/\".\n> MVKW[=8XBTTY)FI_5MDE4V3AX/J-M*1]I7.0CC<MY2N.B++^L=7&[_F&;%[8K\n> MI]MW8%S@VY&%=AQF2796^V9A:>/@^73[(A]I7^0C[<MY<GPNRR]K7T3[L.V+\n> M&IYNWX')NU=O0H.`W;Y,<SROY3E#51$D.U)K?]$`,I6DAN?L],`F6T*)@\\D@\n> M)>!GCL]ASBH\";`8E.7X^780T2TT<C!<I`C]S?`YS5A'2,C+)\\?,917#BN-F>\n> MCI-E_6*?:2]9$1H;489<PTPO(8:0]81&/;[*8XDD$;;L&[]S-$P9/*9G&4J>\n> MS!=?Y3$QC^]AD1S3F5.RA-&[Y!D7N4F>G[F*''Y078E:&#OZ,RM\\7O:5?3;A\n> MWZ_6Q(2:)#!(C7-U=I7B\\G>6Y6\"]Y#TG`SOLJ)\":7LL\\/+/6YV3/S\\[WY!?8\n> M8?>M@$UCIWEI&CLECBQI5'[F_N7P@[Z6J$5HDCFOUN=E7]IG$_[]:IU4IK'3\n> M+#*-RM]3\\5S,0_]YT-=)F)N^Y@7OS%J?DST_.]]_25_+1@V#8=FH\\7R&0*#C\n> MX+D\\[8N)?-@:G?-AHWM^YO@<ICSG;,)!\\\\@6.W_X\\01\"!-NF?U`A`JO/B$TD\n> M7T,\\.+5X<.#A^XJ1`\\]0QA5W&\\BUY[CHD)B@X^\"Y.GWD1#Y1;&1V/(LL+?$Y\n> M3'76L1-!B\\@6NWAXUT+^AN7VAY6_(<CUY,/M[O#@LR<G5IY]^KS/6X&X#:(P\n> MA.U^029540CC_3R/U/<4)OM%&9&):L'[(3U'/1MZ$'9HEX6\\2\"_(-2&Y@)M3\n> M=E'JES+MN2S$843.\"N31$!$+!%G'5W%R4?*4I`.V#9NG%9G+J\\)>Y.KW*L5;\n> M`D-]N,_1R'F`U\\(Y5<]+\"L/^#\";\\^[LNI%FETJJ0=TKHY'&(O-6:$A$&8XFM\n> MJ2!>-C2[VIKHN=_L\"0#22V[W),NIW8E=3K5UDN?4\\/H-;FNEV_1JQ:_@4X*V\n> M3]5\\)2>$$HT/OQL,,!,\\S1*T?H763[,2K5^A];&)8DD^O_7/:@$\"OXVJF+_K\n> M\\._?^GE4T(Z7JZR+!!MA!.U<1?LC948#/DE3<ID@-/B$\"'*\\AL_4%\"23*S2\\\n> M>DG(BZ8LJ$VR.(HO8G7>22)2^T>D1ZL*IR6S$\"-7K04I-7P%=)5$C3>5!A\\C\n> M:.>W#9^JU!.UGD9T@9\"D*G)!*%S5144V-AG_<6:)\"D\"_1/!<*:),94)!:/XE\n> M@^\"1ZK:+3`>H,O(62@CP)9D0[X`?N$@CU?=(6A6\\**A8:I.A#$A6(7\\?.[QP\n> MZY%$.552M90J#&T)U$ZP\\4G]F4T_QU5.)8]1?K6+J%9D\".K\"*8/:N^E[&E*C\n> MI!7*D$;4)N10`)G&MG114E)I3.V=)R@HO^`8[88]/6!/#IJ<92U^E+!GC=6'\n> M>@P42HJ-:!F=N^X\"4?B,X6/)_H_-E1KF\")AU%BZ040<!\\TGJ/^'$<B!YXR0N\n> M+(#DF[ZM';/M=@`%U$;KK4\\.;*TR5JYY*(RHM-E3&[AQILMZ;8R[!\"1>LX\\=\n> MME*LDD!)X0O!>+^PAH%Y\"RB_6GYFTB]C$AA9FH-(X/JTY<IVO^^NUC\"MA!UB\n> M;:V%9JVE&3,`]L::J=-&;*U?`A0S3<'XYQ0%/:$-33R^$0<(W[5?6I+%(#5^\n> M[7Z>KX7R&_9<V@#.@\"\"R%MLQ%&[$=%P0\"+<[*6!&H,]L-O@:YG\"!9Q`+:ZKK\n> M^1NF)(=I_';W1F57S_>=A:PF*LK48HK'0TQQPP9([+K\"`Z@]2_[&`*33?\\R\"\n> M]922;XPEK1CKH*CD&&G9PT--J]$8:]V5`7AEGA[F]NFA0\\/<M!.,=V.'JG^`\n> M9?%C!MX>0EF[I(*>Q:FQ1]X*4U5)+\"FN\"20:@0F>T<]JL@@6,OE3U'/RZM@3\n> MMBF&TW:W9&!E:GZA13P.UKH7RR@-8YT9*S)8:SZ1(E4Y;!,]P'*,V1'`\\AZ'\n> MX1#H]5*[#WB46T,*$)K6K25;]'E`'\"SJ&+0U9+FXF\"[:[7:O*=L\\BULL&+OM\n> M?$%^/6)%2E\"NG(6&4'7-V*D4#IAQPH0W;$RJFHS,3+6GQ-`J==K,%TP/-3_P\n> MF+!-U1@0[AZV^X\"'C$WGU3IVNUG.`4W;;1AT-\"*;,**%?,R=9C@NL%3#2Z('\n> M@NJ6`$9\\VUV_``,@7V\"2]A8Y&!.3T>WN^G;-=,),^*@.&\"\"R6M\\)Z&Q%1:Q_\n> MN.W,PD.I+ETJJ,#\"F&M\"BUY))Z,E1?]?>*BP/B-C9F\"WV4B]E`8Q9I`&T'EF\n> M\"\"1K5,3G:6I&8/EI0VI!X>B[`L%OH--@M9K9PFT6XH09-,@J&&*ZDVE@;]0(\n> M]4T/Y355_P7T-QFBO*;_?)17,.R,$5:-,N`EFJ[,\\-1%`)U>3:-9`!J1;,AJ\n> MQ61=FJIOG`%/E2(MDB&SWFKZR>(3IN?)M-,\"C89DK!]J,O$4#JYL!-W;5`E<\n> M6Q$-;H\\419U/73QM6S7>0)6P/H$EIUH`=5)8^-DGT-0PP*^`^\"8_][(R&9%[\n> M)7VJLG$H\\,4G1)D\"_\\@PS5T7C<0?_8TVX'4&/X8<UG0[RQJ7.<*.^IY+'P_Z\n> M#?L\".%.4(C]9LUA'@86]BRB,CA$FQOWA$D4@-1[K5B;YBD'IY9)\\)2`KZX^7\n> MI#]>5#'R(6/BBN#/A0LLS8OS1PPQAQT=,C$'(;&^-V:B'`36O;[D^MEA$Q7@\n> ML3;C)L:XV8#5O(,P[A#OF`I'840<9%7ICA_^6&;%Q(-^;Y@=755>AI!#WZ`9\n> M,VVW&WYR4(IU;`/ND@DV6(%;XT@@7C'<<AV<<$9XRK04:1Q%1*H7$K(2W7K@\n> M*(:I:EG#KZ4[*IA8S._)<'RRGPCQE0;80O[DID;>N9$$*\"/;&/[\"S,<AX]@(\n> M2@JF&9Y8MPH1C(S30T>N:5)T2)SS_3.U6BMIF\"`][GS&K4:HA'ID4F^FSSZC\n> M,X$2<6@_?$.<:O3VFIU5?F)>J5@[LB#D?W(B/]%QC'X31XP=O;$O<YR4#@5=\n> M,ZRJMEZGF%J<I.IVG(VJA!RQ)E_2XE=F-LTWTT]GY#CX3#41_Z]57V@N5)6-\n> MI48VL1Z53E$P`U5$VV(-6JPB'@,0(7&T*B3PE1PS\\8J:K*RJ(\"9#%03HJ$57\n> MT\\^B644*L2S*N53/57XAH(,2'5`5@&)RB#6U=\"<M[;3M:[6CD_;&*4$4AUE@\n> MB_\"?](%`#H9%B$`58LKP$Y.#Q=Q!JD0I`];8$J518<*;WLJ+L9&[G!))$3OS\n> M)7%OY`YHR%<R>ALA]T-WOC:CQLP>&BG=R.A=\\>A-DT`=7J+^Z+VATFXT+98W\n> M>)>_</\"\"6RN/>0SZH[=B=HB1]DC$L2U,`[\\]'%(2,X6/-4)PJA'BTG#ENHVP\n> M\"L;HX+P9/!EKA�\"*_Z,WC2;X,D#N/`G\\&]*DYZ4U=6JB+,`Q6YTE.7J,XI\n> M,3-U)V-3MRP*$PMSD#16L>9$4*WT9M!*-&V3N,@#9]HF,1,47ZEF4K.6E%[%\n> MV*Q-D@04G6;6*OD_<V<M0AR;M8&=M50`=]8Z)7C%L]8K0N#,6IV#Z3!_UGHE\n> M4K,VT.&IAWKR>Z;^\"^AO^FLZCV76>6PP5^!)YE+:_ZG;O[[=+\"RG(K.*?T.\"\n> M`\"G>6'=P#:_7/WWSI3F5ITR02P/UC0H4IT34L5:\"1E0P==[GLS0F!72D!QPY\n> M;HN?&`WT!05ZYNP8FKGQE690S+-*QZ=N+,)0QS^P\\Z]'4Q(8B:=W-*:DHHBN\n> M[W5:JN/B\"(D%>A@[`\"#P8;_V&?](3Y/1]5&N:T/O`1B0),6L+&WU_I,S?N;,\n> MJC?\\'8N,2JXH2<,=<_0WZ@A+5\\VLR)I3<O1[$I8F^9*XG3AYK\"Z=)WBH!*NB\n> MP`$1$5[SNTJ`2!FM6YS4ECF,;W9;=?#'N$!70X#;=40,#E63B&\\M?\"\\#XABS\n> M?'9@YKLG@0L^E!N=JB9MV\\_O>$'Z4C@[K0#>.?YR--1F:]$A9BRK6A4`J;%S\n> M,X*_8KHGTW?':)Z\"5CO]H[WH$B^+,]->>,_ABDQ,K-4$V08&XL!XM/-YARE$\n> M9;>21?Q_L?A'*AHP@6UQHG>&D0G><'#-&VFDU23(>:UU^4;[R]!21I*;.F^R\n> MG5,8NQ&-RX;\"4BH,KPN>>9U?6C>]/[&D^I^ZXF\\<@>:+V9[SHJ.+]\"DKCD21\n> M!%HW]4K=BEKF([5\\/:SE*^XN6Y+@2,W<4OLU7'$Z2VJMX&@X,+-R#9BS5>K`\n> M!.\\X`KE#DKF?7PH+FP?;4S-RBX;/\\%N@\"WJ3IIZNMRMUU.FTURGE(E</2T-%\n> M1;K>!=VA=,817?1PS#0]O[E9WS']T.@L`%X'CZ\\J,RW/[?\"\"!U&DJLTS:C!J\n> MOYREI8J8F[6/K(4D'6I:39;ZY2PC794.%\\_R*'?RFPEK*II7IRZ94>1\"[<=.\n> M)DK8#$PF_\\G]QP&)1MO)I5*3ULT%O))AQ&=JKX[QJ3I&(2\\(NI*<#!?`KZ4-\n> MR=74(<^L)T</=$6=C%[U:^KF1%6EH(%35^)J'ZEK8NJ*RB&/,@O,7IQ)/+=R\n> M;G.0L1G%T%FK2@<VQBM>$[Z<Y84;+)GE56(2YK8(!FU!\\<HT<N*I1BA2KT!U\n> MP`V015XIJJ*7?&-6\";L.?JDD125CNLFK!3WQTF^X&!&)MK9A$O4!+>7D@\":.\n> MDI$F3M'$:F=S!A$AN$:IR3=5NSA'EJD2FT&41X$-J=JW3$Q(6[/_U)ME7NF$\n> M`Y2SB',G898'!NWK%$2U;V##RR(N[6N\")=R^7BD\"OWTG3OLZI>?V==)O`J=]\n> MG0RX?=T<P\"8*11\\V\\P-$2>@C.]9'.@!KN&_1=Y6#=9=U^IT&#F&0(@*O>Q$Q\n> M)M`6JM@7*=Y$;Z72J7?7'=^%18F2Y^^F&[\"T02B9-P?<`LX%!V'3`?5F<PLD\n> M#\\&Y(:2M/6\\,$<TL\"\"O?`X[D6TMR#KRX;OH=W7.Y6T`S_1I;`&3M8,K;Q9[+\n> MD^7\"^OXWU.VUN0>W(CT4L*0`K\"V#H]H?`&$EBMT5TYG;S6C/NY%W)=<.U6^!\n> MR]'.)+>`7)G3X=@DI6%0;M361B2G;M4\"V=[DINT-L<4N/?I>QE5BGN/:`'(X\n> MB4,S'>!>&03G5\\QDK*_8N5@E$S(*I!,.`D]8\\R$W<;MI._]1;J\"&_'^Z(\\TU\n> M^5$>2[?5&&+&@QRS_,$!$^E&(Z*K)8W]\"QU8(+RR+G6_O1;$%:ULU2S-E9DQ\n> MY]'0:X6BY29_30?FYH@ZPA4B1X3-5UIE,5!+(.EE8)<J7VA$]JRZ2`2%S58B\n> ML'J/826L5M0N/)]S_O][M@A\\D8UU+7[E5KW*G=:YN*+\\4,8\\+CL:&;,>DYI_\n> M\"(R,KQN\"P=_ZO5GW%,%NO(UN`*\\W5V,5[A6T\\WIS1--F*QP</2[DL3.:3\"TB\n> M==2_<W-SU8'VZQ^Y9/^;AI)3!5.)8*P28X67<\\_J?DG^CT;B'U;2GG?<WHCX\n> MV,V],>R/7O/(I&O]@7EP>LD=F(U,QM8[)JD5SQNO_CE(USKP3GO#L=4;HZ_L\n> M9`F.C='_U*WB=:M;X#HX=EA$Z:G1V,X*L#5BUY,H\\[email protected][K'(\"(:\n> M$Q_>T?2!A4U]U#K_^?QF?M6MU8:<D$(C!?'[H:OWVBQ)-D0A!*8K[[=SODRR\n> M9E)08<@7OH>BA;G3N'@]4Q2P4,,L:*/OJ5:!T/H2)J;9YK&CFRT1*K'Z8B5L\n> MV&JK5J(#L*A(]=&SR0%Q+7[LF4!H6Q?\"%9HQ%N2ANZ[IE@^XG6\\9RM=HYP1`\n> MCMB`9YH0=PZLNNWZ5JR=!\"]/[(.DY?(X!A[=;D?98M,A[,PU[_U,.ZY;25B&\n> M6>`25<YSD8@LU?O*HXQN-7A>P.9GI&%RRL2XD#=DA=C:MN&2E3$K6IR=%J/'\n> M7!7RE=!:\"3%+VET%_ZMA0<*UE0NXSY=&!;C>WLCEM[40,G(DV:_M89M\",@B;\n> M:`5Q`30TED18)^FIV!<R0&E,2M>7>:RZ_OD>`'(6+@O@S\"2EK-=J3!'HU?ZQ\n> MVJEHP\"]X&#%X,32/<\\++U!DQM*<9:!KTE(?F\"-SG7D\"X\"(A2#<J#AE+50,BM\n> MSDZ;-(D$;>%]Q0Q0FUMJWNK:EK,3M>@0%ZOIP95R+[\":F&Y613;26-#-*2QH\n> M*I8`60+#-!-@/!C^`?)LM87<+$S@5P95V!P)R!#1Q>SK'(/6)\"0$S,$Z`,TX\n> M3?'.E7H;T<L$C'K(C7\"!<\\0WVT,-D75^,$5W>M5@<JZ\"WE+@+\"#7<VU.1;II\n> MKLEB\"TEQ._V)%@&8?KT!*.KM%8NE>VUON*\"AN8+0?7MC#-K$#G&^6K&]VX8T\n> MMS3.:8H!!?D-QM.\"^AI'JWIY(0==YA)GD]YCBBU+2CY<3%?#%A3-G]&'L0DC\n> M&0)L<\"\"CDY5.#FW)].0X:/#A!:?`^8+V#=RF@YO=`(8Z+;N6<Q2&X\\\">H^F-\n> M$5[\\!:GV1B,4MATM&%II9ZG2F1B=3BO+WO$+1M6/:<32M6PC^UQRD6!5Q^'`\n> MU_DUMJ`;!O:E:UNYAOF1T1'E]@5C64C9<?MBK>(\"[J;7_6Y:'M$_BBE23V<?\n> M^&5Q](^M/B'7UCRPTP!]EAK=GHH;[U3<Z%-Q[THL5_\\%]#<;FK3EOP9QN3%I\n> M$]L72GF^Q##9,.2L7HNZO0%'[$&SDRVX@_)K=-[Z&*AVA\\.:%@>U@\\TA_&SM\n> M&?%MNUW#NG#OP\"_C/HU&CKG*:+#]=0NVTC3SS.\"9\\XXM9]ZC1LD:H'6A43(M\n> M@K`%1Z9E'[\"W@!/&+3-C*&HD_]:QB%Z)172M3]>=Y13HXT1+#;65O4HX+4A*\n> MPKD70/A\\MR+G7ETV(\\&T<V-KCF,_Z;D[;986:`3_O=H4%KQR+AW,[L;\"6UJN\n> M@R!+8'\"(*K`V3G0$-O216F#]7&@D36\"S+N>T)#V;T95PGHCMQF48V1/^\\Z7J\n> MTBK0MU/SY=)>EWGD\"$8BE$VQ$4Q7O5<#Z3]H+=3_GC=21Q*@:A\"73%YE;*SV\n> MU4;5FK'-\"=&5PKA6249OP459R]V8>VEY0\\J0I_O=XFFW6:QOE_53B!M/T:`L\n> M>6RG%ZV^P$K5L\"E5=9$[B<[+NN%+;BZVRI3+BX!06[^;91$9N-)%Y3<JP9=S\n> MVHHI\"L7_E'^%E=OGU\\N7+(T3^*RJ[$L$*#+2VP--_8)>R.O0?:'#G7JA=**8\n> MK!.1S2P.J5<>L5;H/V[GZT_1=E$,D\\1_GY54A*])S\"$5I@K9ZR\"4+(@2V-I]\n> MS>+0(T>JX9)'*>P5]>]D2?YJ)QX?KTB(>B7II(4;CI#+!N&HS:(,_@,_,UHP\n> MVA(_!!&,6DV;-[;!G4&`%-BV]1UGRU:M8^V.QF.+5FE)ME6]X`:OL$!PL\\9A\n> M!#SN[78MC85V4-(%MVD<D@\\(VC0XT:94PC@\"[*W4T1;_4YIF02RJXI%YEB?F\n> M%MB@!;^!^>%;)73L6;,GUA)FPF#R6:.*E4R[WL10U5#;1E2P78@W0R9ZAA`!\n> M0;U9\\MS8/7U3<[87J@`+/N;Q!(G)!Y:-2<GC8B'(STK6^?]P%+40.V_[__W_\n> M4QE]IM8A[K.8?\"Y5!_Y6';YVA]^2V\\SS/W[[XN6KER^>?_/]SX^Y2[)<`L&N\n> M[[=D.?+\\<[(I40%*M7O&.4A-+NB%[,PJ^U(DE>W:HJJ0CJH8$OGBFS^9C-`H\n> M)=V\\Z?FBPN%XKX)BUG-C<Y9JCXM,+NJL4I0FRR1,2I-E$F*VO_MMMU'+[IRS\n> M_>J;[U]^\\?Q/DFD28;J;3,TPTOF:_I2LDQAZ?<DMEB4!+TG\"+]0S;!(T_?G3\n> MOM!2J/\\\"7$_^FG8\\!:XR15KIC7?7@XCV47$\\PV4`Y$B63!FK?VN8.-14,CL*\n> M6HWET5TM!`Z.<`')W\"#R:X#H)$ZL$8XZTK_6LH=!D&:<=R9Y@:3T_/O/O_IJ\n> MF$F#@RK.X6Q\\[FKXR3BEVQPLR+)&S%_LZGJ#4K\"(<BF*A64-;.O5BD\\A-[<[\n> M);36>ZC:$(`50T#GDYL#(-S+!#=T1JOICQUD\"6ZM_?QBEA0,+JTV97;!,>U+\n> MQ6!SE5%_/2,@R-$C$\"'&44XM:^85(!(!WQIW[Y@V>Z;W>T/9DTQ,N6(IUQ,N\n> M$[PE]/'+F/8#3/]VLYP9<HC:-,!^9.]72U3@+VV\\^6]O#Y37_F(AFBX1/53.\n> MD*Z\"DTLDIT/J'$[(72(3=0;(BVRX0OZX[9;XM<\"N]$H5`QO@Y?1[P&%^M6FV\n> MI'U2.]]C(VBTTT>L1^\"8%>]]X-$J90NCEZ\"*[(N:[/R2$N56B-V(_5%M1I><\n> M33W]+0WQ9YXZ8_I;,K/GR)#KJ:BTQTHD5KD\\^;T3B2($U(`J_\\0N;4=RQ6GH\n> MMV0`-)`:`\\X\\84YF)ZE>&;!B<3%Z:7!1\",X@QC#Z&2T7)57,<C/UAV[=9OJ(\n> M?TTS^O45]:CTB6IQ3B6+8MOF$7GJFW:.\\L1[(PM-;G8EF630'&W5>,'VIPY5\n> M^QOU>CG]YF]??TV[WN^8H,BT=:EDW#MUG%W,UZ\\&#:QZZC-U5J;O+U0AU=*S\n> MNZVEI8C]-K(MQ7FB@4YDB[@$=V7BNKGW^P;*P?%\"!%2!F%#A[M0RJ:1[7AYM\n> M:LA-M68<12'Z!#M#G,7.UK`P1]K46:58P\"'7[1<U2167<JLO5H_DNH;#P-NV\n> MWEC#2;GQ-F*42[.\"`[\"07%$@UPIS9;A@L%$DD1@*DX/K:]Z-L#\"J<;0[L*I0\n> MZ(#0WKU\"@&<`+!C]];7A%/35MV9\\H<2WJD\";#6\\#'M&;5H9RR=#C\\-XF==A*\n> M[9VD7`G22*BHL-0OQ/F<*5LZ=W>%5IZFYU[<:>FDON:[WS2T3K?P1Y;-B_WR\n> M22KM9+V,R3#G#JIG4)C<7FD5?`OE=B`7P^*N*/I^32TF9]0#F;\\ZNGM64WG$\n> M$`$II#(FK.GKS*EZ7!;,OP,[=<L)'.QA8'+Z\\VY[>^,R.1D\"!7;O)W_T)-9$\n> M7J4[!,TE_A7;IAW9@[@4A.L!*7L]^V&ZW![V<@C5>T<MSM5X>'[=71CK`DP0\n> M,D#\"UJ&WBR`NL*2Z4P#0!QAQV!T@V/YTL_M<\"5'U3Z0I?%33Z^&GP^-9H9:5\n> MN,0BAFC=])&2B)08(WM*7!;.`A=7SI*FI-S*6>\"2,..W5,5(U#&5%E0U1*%\"\n> M9W(]=0J[%,6W2EZ%4[(N1JHDG\\1)Y\"085^Y;DK%(WV,P*=5_`?TMAAJW\\E=P\n> M(L7129Q(_WT6%14F%BV(D)I?86`]XY-BQ0?%I%2Y!TZ7(0S=D?(HQ%+Z\"#?(\n> MW'\\)<\"?U*9,<N&T'(AO=97PB365G1#[3*[5MO?D4IDGB:'K!=DJL*](O%;]D\n> M:3K),]X9Z=!01*2*_1DZH\"I)1F1U-A,ULZ'V%V1=M1G;J^\"2059EMOT<6;]K\n> M5N7HBSAJX@K%<R_DDF2XHJZ8,4W/YI7#\\<DFI_6!.>JT-.SXA0[FK3D=!]\"<\n> MA.PG=4KV,_.7\\OTST`YN;]0D[L3IF^)^=>\"3%!..7=60`9B8:FEN/<W=2K=Y\n> MPKJ;\"+-4WZP:OYYFQ!&?M=ZT@ZW)*7L%=A<#^J%.S;=\",[JIZZ6^=6C&%W*R\n> MD\\+]L+Z@D=)@7=#,<4[/T$8#G4V29\\Y:!>U,\"FJSP>B085_:%:L-O\"5KY2Y9\n> M-`RC%/QC6+)6SI*%C#/P?['J+<IQQ8X<`&).P5>[F[T\\DPS$`HW^*\"JK`E<T\n> M^B-^?J7D')\"(OL3&NH2X\\^7SK[__0M1A14GJ7S^.X)20AL,-2AB,_:\"-DW:`\n> MQ\"^G?ZGG-_@D%+AJ6?CFJZ\\ED2JG/GCZ\"+,V#LD15(FL'?-K+=K5=_/=_%J?\n> M=U7_R5TH=GHV_=[<$E`%=3SM8Q!M5%I!'-*EM4JKU89L5[PG/]DO--41[JE0\n> M9/4+=>EW:J?\\!LP_2\"+\"1?.CIZQIB7$#V#78P.G_T74-ECES&_KD]QJ[[<GO\n> M3?%5,U/CD8\"JPF8%)48DB8&_;/A)=H,4'[.#91*3&&T'4Z/CK1!O-=7R*'19\n> M\"2YX?^9[E3P2EZ/O++<P7Z<;T33F^W1]E^\"KI-F4+O`7%W&D69[2MZU44US<\n> M\\9)&]P:OS;T!67@L8>ZGR3?+4&Z`6W'II8NN9ZQ?K+(Q/3X*3[L.97,]WT'\\\n> M$^V$OIR0PSZ^!8V!9]&+Y>V>611Q.<;S41W>.P9-L:*Q50]`U16RPSNM.,V=\n> MZ^BBH6\"P]/3E<\\>>L0UP)]O(G>QVMZ?UAW:K)\"K+8U4E\"7;!DT&3AK%(;7C#\n> M&KZ2\\&\\D-!UK*^S#@;-L]OO=M)]4O(.FE\\1(,,`Q*=U?F(..[R.YW33%+E_=\n> MF@ZIQ1J!EW\"\"2MK?;#=\\;RE;AN/`W_0L.SL1B>VE\\\"K@!L!U&%]UNBW14DMX\n> M-,Q^2P2F/@-4CTK]%]#?\\M?4)E:J@TLKB3F\\M\\CO&4,S<?.\\>OMJ^PI84-AR\n> MZ.F5<*TG1<%^:\"22,ERX:$[HA04K_2*R%$$M*%$JP$+<3?^5,#/:[7JY-R,)\n> M8)1JS*C&!5;,C;O?IE41L/!%JRU9=:L?/ND5]!,*)G*96E$I2Q;,OF'IYA'9\n> ME`!\\X9(_R<*J2O,I0I>0V['\\?K5_?JF^/Z9.M,U$>R>%JUC:SV*\"HZ1EQ&E)\n> ME8_65ERZWP/D1EE1M\"B$Z+]Y\\GLJN=;<8`/\\UU@'*6@5_M=_Q?=6%QI+\\(8+\n> M'$4)+[W\\@I-`O58IJ7=\"S@A-,57`D'?!9OJYP^XY_6VSAOZ5+IA^*W;<8EAA\n> M5HM9EE$*Y%BHVG]UK*M6(SU2!XQQ00FD22@=N(:5N006VN.;;:=-^%H2PU24\n> M0`DPE6R/5`=XXXD,^,J)?:G;9DEML[UY3,X)61#!CV[Z.R<D_Z+DG1R_Z#NJ\n> M5Z*5IBU-Y41R3361H?JB5AL^U,%S[/$L_CU_19+7K#%\\P48B1KD+6E6GCYP5\n> MUVFV.SEJZM@7W#QE2O>?5%=ZJX\"&`>.LSTQ^KPY;_'LYI4K6T]\\9_9);@P`S\n> M3\\D]I9V'2NXIW+=2;D155DH&*;BR[?3/VOR?R;^;@W#C>N-;2!TO9;'-5%I*\n> M<BFDPG,EV).)!`,B81&6LEG*>A:(=+*20E[(\\/K>P:<RPS%@[<;=;S!CI964\n> MH$([(Y*3`0$ILLAS$DORX,Z?@RV-CS6\"7[K##_U.Z:6$JD--L9H^ARD;0_[1\n> MU%PS_2D;ZD$>=+GM91R@)EF8!S)1:\"R[&DC^G41^4X4<7F!>'7RM)4O03FFE\n> MK(625&1&Z[)*4=FZ9#4R]E2!5!$\")?RDTEU]#>7*%`PDT/V\"]8-3X90,FVA\"\n> MYDN^VWV#E95&/PJKF9#1L$VWX:$AUH(S:\\,!26FA%T,N2\"!,Q7XYG%63M;I8\n> M-;-\"6Z^A6;[9OH7>D,]G.-J12/(;D[:V,5-K[,I=8[%4NQE^BAU(*'Q_AEFR\n> M\"\"<_TY:JA8/^I6%$T+@AH341/O%`]:(^*U']F.XEZ.E>)K],]P+8K1Z\"5V]O\n> M?S;+24/YR>7T$['4Z(52K??)Y4SMKVDUT1L_P1;QOJ]:K\\.V2GU`05@\"^-E!\n> M[?IOEBR^O[V2*4_!6+7\"*?'FKE+\"-0\\2BY!8H/4LC,D5V\\0&5;&WS700[_T:\n> M4(Z%J.D8J2OF'(<B`=D$/Y>%5DE2CZ'V)V$@`+@2;_]NH&^^?<F!.&1\"FCMN\n> M\"B4>!+KTLN??5WP(<YBYAU=BU!@,0E*Z)!R8JH@8T#48P2RZI`/1A<Z%\"%Q)\n> M3+4'BHPP*KX$_MK)XDOZ*<=C<6(C*Q6KR-%KR:>42R#2PL^<)>_\\)]OZVQ>F\n> MJ?\\UE3L1B@TA0-J098#[VM`QY#C2=BP72-NQ*#`H6W#?0*`>YOX524$*I[;]\n> M@%8!AMH&IJ8=\\BLM7W__Q==4.HQTZ6@R(*QQ7GEUF.]6-2_T\\:P@!V1&OV2&\n> M9+(^WFVOM21>LY^;?E7!@YA/BRO2Y>COK7B^&0&^(KPMF!N]PR2+!6VR-PZX\n> M2:@5^-K0'4Z?2D1<I_)`@!DDUE%:\"7(\"PX@S-AC\"2Q*9%W4(*3*YZ:OH\"B3E\n> M5+B<CCRLAE3YJ00BK([Q4;3YE.N49:GI#59%3!DO+RYB1P51>S9!C$/,G2+6\n> MTZQPY;5$8%]Q<N3Y:4VL7;,A^8K4L)2VGN\\A%E`<.>?:_T&\\3EE%`ET2FV+L\n> M[ZZOMNN]3:#IG=8DJS?.8?:&\"5\\HG4`U<SSY40W'R=OI6_HE)F`IRN'O:EBJ\n> MCUM\\S#/SD4)B?KB5Y0_D@)IZL640N:VP!6*G&O*IER0O81*0M2YI)&=/,R\\D\n> M2]X():D)X><CL.0IH\\OM'CIBJSP)`%V'I#QY_J\"\"D&+L+SRE=+$#3H+FE%L3\n> MW\"6JR\"3]N)%UI0-;ERJ4NC#HS,V$Y:!P$$?JGP2H(&:F^@BT%PE]I%G%7U75\n> M,B8<M&'22V\\\\;4@13$;MT$OK421(JJF^E^.RY[AYDB#LXF)'O+1R8$:\\T:-W\n> M_IBOO?SIP$!BW7Z_771SV+%C\"%_7\\XU6JHB'T\\K!B&74+OBMD,L1X!:#)&*W\n> M61]N\\9&Y[17XR\"2\"V1H/-,2+8;IFP!*=&'_D&#%LRTP,NC.C57LLQN=#@0XX\n> M#0'_&SY,=Q,8W<WD?-T-@RJ1K!B)TQW@54G?#$7@%;O!ZVL1`M]B@31A\"]Q+\n> MBQPYTH)J5*7D3:>;0^T?:9X<:S^Z[8R(M<\"+,<F@T!]M/_'MS\\L>TJ2[<!'U\n> M$T+TERU\"WZ#OSJ*5J8,=/@7XM!?HZP.KA%9&.[?76LKYAJ]^'&QI.F]H0S4:\n> MG8)*;$[RK%>$E3$5DQQ:HT(=3IVBX/0N\\^J-@=96%2T@G7BHXCTG0,9@W2E9\n> MAPMN;]Y:?;OU>*9=Y3P4JU$T;MX::,A8.'=VL]2&`H,T[\"T8P&-\"Z1J*L]2)\n> M03914BW&F'-!Z&H_1?%#<WE@UJ=M!ZWG23#F>B)>,!=L^?!5PPY8<%VCHN65\n> M&E<_B`T7U,:D55X\"7I1N9@S6_JS1OBZ=I^AF\"TA]@L<8X?,@KA2AG=:7H:ZD\n> M:UJ27%PJ\"'BFAM9AM34N2UORDWO;&;#3AMW7=*.QF8S1?[M#<66&8FO@AMCX\n> MAZ']A;-A0>Y%$1<'FB4'W7?.HY7T+6QZ<\\]0UTGKBREQ=W\"%B+@L)I2'C/2W\n> MM/F0GZ_S#=)#HH[\"\\LT(#RL]$\\D.HR183AI;?P\"ILB:D(!R0/ZS7/Y&NA1[N\n> MB+MF\\H?;W4^3+,7#G5K#XW#R!]6BDS\\0>B''5&LP*)0US\\6S28/&V3_%['IE\n> M56$]F?_B9A_\\VU_GJTW7=.PYH];QBS`3%F?P/A,8X)(>B*^Y1P?MO#^]/JAR\n> M7L\\/N^ZGR<WM`13O3Z+).\\/7T:-_#XG*GMG?[=\\1FOC(#Q<=I;S7X:+[TTO\\\n> M<$?32RT[_;WI97ZXH^GE.MR)\\A5^N*/IE?WV(YOVD7!5/QQAY8RU\\Z!#B+-Q\n> M+\"!Z)$L0)+JGA%%L4Z3\"'>_BQ`:DTAT/F-J`5+KC`;-^UD=:1ZVNO:R/-4_1\n> MS_I8\\Z!C;'V/EQ$]8^M[-&\",GK'U/1XPZF=]I-9QW,_Z2*WCI)_UD5K'Z)D,\n> M0:+[NC!&S^08/:1A.1X0/5,@\"&$A'P]8V!F8F;]C`4L;D`M`?\\<\"5C8@%X#^\n> MCJTEH0U8EOKO6,#(\"9@>J4Q0;Y;!?OYC'41@&\\M```?_50(`\"H2@;KZN`Y7H\n> MS>3=8KW=UT3%Y>96-Y-W];914K#[=;6;O%NI'8$,_+WO>_6=\\G0_[N>3=_UO\n> M.Q5P)/YZ\\HZ(P0Y;]^/UY-WU]L?>QYWZNAM^WDS>;>JW_4JHS/:'W?:-7ZQ6\n> M?6VW;[UOZX7Z6!^H$(OYC?_3:_,3>8SYO[TUO[WMEKW,5;?@5_6OZKOMSJO&\n> M]J#JL:43FA>'BD%=XWU<(IGE?.^EWZ@>:M1;HR0T+WB#X(//BT;2'OSR%LVD\n> MY*Y!'0ZJO\\W(\\7[8J/(O;V\\F]+_%[8[L+'0]@\\DDG3R))[OM>HW?HTG]TZ*=\n> M[&^O)@D-/OR@A-;)?+G\\YP7F>:&BN%V_/%IJ)V5*P'OS\\J5?^W,3`@R$I\\F[\n> MGG`S>2I$8@<BT\\+-VX1F!47\\V8G]A4KKW3\"H3):)FM0<.K!YJ4,R2477G0J_\n> M[M0_P6823HHJGERK?T-0G\\7F7_J^-JQYM';D<569?_>+H.`E;_W69_$DWL>(\n> MT->OF840SVM^CF-PSE?V\\00E*$(R=RPGP-RQ]W*GDRN?CH/G^\"31/<I<AJ$I\n> M,ST_D'@TBL%>F@TH6_EF^@$4H6!A!66JGY(ZHM[#^4HU3D'?\"KIV(8:F1R@B\n> MUQQ\"GD\\QL5+0*C2)56/-X<=(\"Q.#'B6?DYGDA2&8SXN1;CI%IJIJ2?OI^S9Y\n> MG*B!4@[[3AV]CS<YN!YEJ\"=5;H8ZGF74(<QY8YV#\\L#E)$X/]B2+;!P\\GQ[L\n> M7&P>[5SL7S#:D[C\\0*,=%)$CHYVT'2?:_J$5=QO+Z9=[XU`]P0%Z/4DC_;SF\n> MYR0'\\S;\"R/.I/D90(0:/RB/$X+TXY%LG<?B9LSJ=#]T%Z7RBL:J>Z!EJ+S\"#\n> MOG<?YZ%P@9X_OSS65QJA,3%S7D]RNM/(P2==9\"43?:@M*R67:5\"]YF7.)!P^\n> MRS$MSW&44RO22A,7(X-^R!QZ,F-\"G8XST(6:L.]%04_4.)SX-;'@9#K/.\"Q-\n> M\\G&$BQW4,28`!:FZ6]\\\\-_4%Y_29%3XO^S#2E;;AWZO6$>FF(M(PTZZ=DX]B\n> M3ME&69K3!Z(4CM(H8MHG>DYA)E_ZM<:&`),JV?QE23Y5Z[.RS^A[6B5XUN'?\n> MJ]8):I1F)1C/\"<(ZP[:MI@H!S:)V:A8FZKGB9RI.EO9K780\"7XW%D\"SS\\G-J\n> M?5;V\\)M(RP2\\[CK\\^]6:&'SC4AWS94&+JQ3)%Z%\\U\\]5B3T4X?G9)=!&5![A\n> MDLR(S#%2Z[.R3YSGW#R_3ZW)T#$-HQCRCIJN:<C2`OGH)%STF\"!ET[!,\\5Q&\n> M:NV.BUZML2AG&8M-A-X3CJSL([4^(_N$0$&2JHKPK,._5ZU3XE!*9/],RU!G\n> MF])F(,FG9%`B-4TSM:A)\"[BU)CE%:@V\"\\S-K?5;V1/8LM;;A'UIK8D77)62N\n> M;GJ^?Y>F&+11ZQAQ>48,K.,20]9TCEU\"6DS-TPE1A`U/*SXGP6B\\/\"ENIC8&\n> MLJ].G\\:0LEO)T\\TB)NS90QK&IGU^;8;YG!&']A,]_<KDO/H@#K<!QSFC/EEA\n> MNAIQI*LY/I>3TSVKLV'C*EV'YW.KJN.4R5G=S6F[53W=/!S';=+3S6/3/K\\^\n> MPWS.B$/GPK3\"L1I\"+)[/B4-2DX[#$M3I.\"'D=8[#ZQN>LQAG!(21YU/'#-\"X\n> M)GQDH&>)=KH($@=%2$X?39\"/6]73S<-QW\"8]W3PV[?/K,\\CGG#AE8>(HL>N\\\n> M^B`.MP''.:,^96&ZF^-P=^-9RHDPYW5W6N:FZ_!\\=E4Y#A?A='=S/FY53S</\n> MQW&;]'3SV+3/K\\\\@G_$X]Y]3,YR4^P?>2:7\"3$@'/[V8J/]_T,F73IWE0+GQ\n> M??T#>8!.+J?/+V</4E1%T8<N8E2F)\\KXQP>6D=)*LS3[<&54!T^59#Q0(-@R\n> M?OZP,BI![T.7,8W2$V7\\T\\/*6)&38]Y/SG5%>6!B=*CQ$^N[CSQDV!!:W8<K\n> M'R?W04L8)VHZ8UWX(\"5$<EA:/E@)DRKZD\"5$<@\\M(5UNWLQ7=:#OE8*=.MC\\\n> MFWK0ECW!Y`_U9JD-A@B:(\";S\\;FVA@+8<O2,<>>^=:R=:FM-1=9.9/^V]%E:\n> MVU&B$<U8$0QY1BXOF2W!T\"-HM'<'/AAD1>R3VK,D8R1)C<V\\U_[*XU9@@8'.\n> M$RN[`QO262O$+(LY:]<,T>!W,#4-?*GG@JH/GZQ^;LM1\\T=&&\"&G1(ONLV(K\n> ML;E@1,T:\"T;;.,@^@2Y?FA:<IV<FJ<H-/;;]KE*Y41,P3\\S70D)?\"N_+07P+\n> M!4`>Z&L&-I?8%0B)9L>>F7N!.7(L+F%>Z1I$PI%8FU[\"P'9@+3G`_`-JTTM\"\n> M?(*S-4!!5#-23]9+AC07*'DJGG;%-PP4[,0O5`UPP@CZ+8=&7VR:KF&P-!2*\n> M:9`(+9K9I!PX]M>C2-2\"$NC1_=7B]F_=[=&H,/V\\83/=^?AH%9M%@U\\86*/!\n> MFIK1069G7\\[1@=]88&B&T*\"V/V[\\*.!-$?F\\$<H_&39WE@R`#3=].\\4\\#;D<\n> M-'H\"\\7/(R\\)\\-):*N,)P/EJC8;8$12.S9Q],0D&TV(IEI33'I;%A-ZA87#1M\n> M`DI&J4#H<*U25Q:!G7FUK%4J`?@PK!D('3QTLT##NT_N>/1X?I*\"--!8I`%K\n> MEWO09L\\TWHW)J?7DB$\"OG,8]+P+7@AI@DXSJ%0BN^[YEFV1D<H!_:2UCG]T>\n> M`2>VJJ6?861\\Q`@5,WL,?*S14TM[6`=]_DI`R?=Q%**0`!;C@/^-1CPLXU_#\n> MPY($VE&(5G&U!VH(+:.J09/Y07L3\"T('O(D#?P(Y&&N=0TO<GT*=:5K,N$M9\n> MZ&@,LH^RP-;5#K2667X(7XL6(\"[CS80PTF)HY=D.'Y-%3SC\".:`?9;J)V[W0\n> MUL(G6T8.C]65[S7=0U,&!4C0HP`Q^%U\"!=()N(=Q0Q]=9#2V$A\"N@-M&`T7#\n> MFZ3Y&!C7EX38M!3:!YM^K1[G[$/)$#-C-N&HJDL4=^]8OV3(0,V.<GO8`J8J\n> M8!<*ZR:O698U/F$C5!0>B(N'[$70),;K0T;35;T0'VX3J2&2`F$H0,:E0..9\n> M7L\\*>'`80V[=X00+KK^[':X!\\U5&>SW`%AI1PG:Z]$\"05_%(#XC#>BUC0AS6\n> M3UBO7TX9@-YN<)W;>;*;=P=5A08H12.C#C!E!6-F#@'WU1GMRAM]K9-!;X:V\n> M>D3P2\"OSZ'@]D]B@__=A8BRXCBS>*R;[%6X3NW;3;0)EXGN+K9RE>Z2Z0(6+\n> MQ4W-\\9F/(E-1O:M8M(>Q:68K&H5).%+3OX\"X@MPXQ*^BGVIPS,D\"2!($TKD:\n> M7^/L*!#G.RYQ/5A#&8&N\"@?@-IKX0*^G+/!9YP?;+NRB8TB,N$D[#&;R(HTB\n> M@(/+.:2C<TA,+B?:&T//.>R;<,)#=Z_<O3J8*D'1P*7\"3X><I5;W'0W,%'!X\n> M4\"%?&)XIVO28095=16J-PM3H_C\"C*ABLHWH,/0,GL!(XBS&WC8#=-M3\\%[>-\n> M7-PVB@IN&X2P_OYN&Q^]-3YZ:WSTUOCHK?'16^/_,&^-)+X@@3#^Z*WQT5OC\n> MH[?&/\\];8_+T3[OYV^]O:!Q\\7^.D/GD7_,O3NX3+1V'^Y>E/WMM=[/WFO=U%\n> MWF_^VWRB/JCPZB_56RT_]'^%J;8D,9^H5%0F=_<&^^EJHHI%J27WIG8U476A\n> MU.X+IDJD<N.)K-Y4\"51954SDH6*J%O\\1/XG?R?_=3BXPF8>?QC5;YN!YS<]Q\n> M\"/LQA)'G4[982(Y-\\CFY49/\\D3ABJ%N$1TQ=^UX:B<V'GJ5X#W.Z()/+$=/_\n> MOGO*&1?943:XJWJPLPQY#HRX#\\QWJ_U][C)*R(\"CRO4DBO3SFI_%5CZVCZ>L\n> MJV)K%Q?%YQE*Q=:2+XJS<>ML/PJYIT@N]/A++/HS<H<:WO0/S/#/2\"4?]/\\O\n> M<`H@CMU^,B<Z#A---S9<PKBQ\\2QM@C#G=1VNG75RN((>;5:_[VCH:L.X/#/]\n> MC6<]>-A.XZS1DY0VN>2HK\\;Q./G1T>#'R;.'YP/_,?$5(F6$+$QXEL4#8<Y;\n> MYY\"C3BZJLB/K3\\_VRSKGI7%BUD8\\2WR$.;,(5<_[Z9PB.''<;.^U1XJ3!^;C\n> M6XMG:4*@LVP0F--SDK$'`#WGF398]Y_#OHT\\D.^JA).I0IW,26OQ<[+G9_>[\n> MR?XA=M-1H?9LTA)<L_L&/:ZU)P=LQ1!\"GD_-9]H).#'>&<+3QGAQJJ,D1:GS\n> MQZ/$1HCS\\H=YOB2616?EGX:JV3@*#)(X?SQ*;(0X+W_8'TEB]'A._DX4)]/[\n> M[<#3A^:\"#58:)HQT+?$HL=EO]*Q:$M:N'C+%T2*?X?()K[J>M9;6P_X\"PS>(\n> M5'%_1_MN\"Z7JY+\"]3R\"IK$0\"PV212/\"L-Y7J?)&DLC())W>&3%)9H83CG!9*\n> M8$^M\\\\E^H5B\"^34BEPQ<0,])Y@,()B24/%PR24,KF:29E4SP+,V\",.?U(.QQ\n> M=7+)>9*)%\\?)]OZY7#TX'\\P=&%5?\\^*(9UDHQ9#8VI.?(0'!^CTSR>'YG\"*4\n> M//!B_7QJ@!<F'WX>M7D^,<D+F.</1L<#CS\\0VHK!D'^PXWH6ARQK/&BL0GZ3\n> MED!=I,7YF5N%PYS5@;#;U!V(YS.,R>/$=$9:%*;3^9GC<YBSBB!>89Q<=I[[\n> M@A>G.#H>_#A%\\<!\\?+D.$(\\1G%\"2E+S](O9QQ'?R4::&)_MI]SG*^G(=K4Z(\n> M*G*Y)'-2KCLG>P[C?+?9___;>_?G.(XC?W!_W?XK1H[[G@8,$NSW@SHYCJ:H\n> M%7=E2A9I>R_,#<80T_,0@1EH9B`:Z]/_?OGY9%9U]SPP#0+KNXL`'#*[>^J9\n> ME965F57UR<^]!1B6B:M6EMS$%:_?M=HTQOF0]O-NKWEG07NMK-&OU[VJUS2M\n> M[W?N=82J$FS/XWX9PB.'>JE8OU=Z=0\\[O)WG).SV&N`DS'EAAJV6<JS3O6K7\n> MY_9W7_MG=CHN<'=8+8>XK$IG(=CW7/48Z-.=YVW#A9>US'!1([2?X=*K>GUN\n> M?_\\,PZ4[U`CN8<2.P[+T1-7O)':,@]G=Y^VAIE?-QEH]\"#W'ND?U]MS^?N>Q\n> M3N)R@#/G\\\":F\\IB2I/Q*Y35.I>KV8[0]S.@DLAELC)9P=)\"/ULO'UM?HCN.;\n> ME)FK,:D*5[9^U0O%81QV'K=[FN2IZRD7V7X]/5JO/K:^WKFG896X&A&LVLKF\n> M5ZT1$/+MQYTQA2*O/:5BUK.G1^OE8^OK77L:9;FK,<I+5[9^56=P$6>=Q^V>\n> M4AQK3^FBZ-?3X_7RL?7USF-:-4N!<*<7^?:]4LVO68CL>5LZT8%DTDD-F7[2\n> MJ4_U]MS^_AG22?WJ::I7NM/2GH^[XDVIT\"S[E(K=+&;^,(N9/YJ]B-WN@3T>\n> M<UE!`3%+AL^6[6@#+`L;T.,Z-XMN]_,X:31/FYS':;.;YWAWFO;<@@3`.\\DR\n> MO7N,&^-\\[I,GB9H\\2=0K#]MC>:QMFK_,/+2:/1]%^4+W[%HRGBW;\\298'C:A\n> MQ[5DEMWNZG'R:)XV28^39S?/\\?XT[>E/`W*XY<G\"GOTIF^'6/#WZ4S;#S3PV\n> MW)I?VZGE]AKNM&R&F\\]]N^JN,H?]AEO+;G?U.'DT3YNDQ\\FSF^=X?YKV]*<!\n> M!!R]NB9$^GB88^<39H[R^,R&B[.1H;'N([(VV\\V/_6,?&:[HBXH#F?3JH>6@\n> M]-N?X8?+>C%87^[1O7$'/(J3DMLFHN-&29+0LD\",Y;@D-A<PW8D,CV?L9R3)\n> M'A7F-&2C3T/38YKW\\^!?NQ^PB=5ZBW)]XRFTG6,RP;]B$^<T=)5(Q9W^!?M4\n> M^S[=$G,6X7*(?N;3]]05#A*T`@H0H+4N!B7\"R65<Z:H\\!306H(G*&`E(3&C:\n> M29XD75I:;Z'%G#K8M5/UIRHMNQ^,>MBB`XWRLA`BR@`=H&4W@8T`,/5.%?$,\n> M;RF3[,E<9OQ=U)2>XW\"4&MA<DZ9@#'S:NPX!CCU&:57QA`,CG!)6KZH<[%HI\n> MU!>&B#D:8.UD\"PD-4ZD]!,!R.\\5&1HQ_P,Z=#X\"(PT@DS`7,4O)X=6@(.@DJ\n> M$A-;O/*/$`GE91R.@P-PBXEPG!8%$_\"K2WOG`0@1<S-,!T`[Q!94AJ$N\\DI(\n> M1-2W'\"';4\\X\\F\"YQM@W`9Y2'=^54E_<$A#&Y(0/0_8#5'Q35B8$=6[[%!]BX\n> MFT!,)+Z0IF6D(YD?&+N<(P,G0]\\!.$:+@B'42L\"&^;1W'0#,X`%#L0'HIHI0\n> MK2+O):@WP1#SPJ@,@CXSRMK.MK]H$)U1*,.X.PJ=#]B':$8!GHJ&KGM(V4V`\n> M;>Y3Q5*E\"$I9L,ZH?9F3VPY$+YK@8J&,2\\%GE_[.2P*#(!64@3D<:Q!V622\"\n> M(J0$`OP@AS\\'!.:V*\"I))672\"_>:)S8\"G7=[23@9X`2DE#DDA=J_`^83M>@R\n> M6^!3?FCRM'Z&.G4*`!3DU2_)C9(KCOL.U\\U$PZY;Q(73I;OK(,6027F8\\RQ`\n> M$@T4E#/&G,T3J@MP\\:;$SQ2Q7S\"^37?=SLG!;J3LM4QLR>B\\VTL1D^=S2OT\\\n> M/$!QG$TX)2I17^WG:%\\0'3./(SY:VCO+G$R8&'XF'/6)DT%&GDX0V#/70QK`\n> MRT5L3CQBSRC?PM:U.>T(*%,2!%&)#&'3?@>0,?B>.9+,;C$<$A;Y+0EXO\"\\X\n> M<8JHS^=-VKL2,$U+!&$#RZ>H1<5Q\"MS6K\"3\"9)YC+',^)N#`J$M`0`>U\"`@O\n> MBU_2A(\"==VQ\\>0*FZ<T$E.IN1\\#C?<E\"Z0L]1S[MG:=P%4/%3\"INVLBHR`)+\n> M\"%1$]DU2W3$IB`\"JWPM<3=_1OXU,7OE+(RH#C?[=_6!$-_U;V\"6^4?_N)L`&\n> M;*-_QU5*W?R@_BTM#F^G@/>B244LV,C.!EGZ.X\\&3B6D>4JDV+Q(!VD1ZE89\n> MYSX]G3$F5UI$J<JD7)ZS;:!:[:L-BIX.(OM2&$`L=#Z8&@%<KL9$@DOP@&3H\n> M),\"6`Y<S5BFL@A*%?P^9I;G67/1>U'K1I,@JR$8*9Y_^[L(%`B,KL\":D:2J4\n> MSJGHIPE!UO1442IK18I8AS8ITV+;=Y\\F>=8>#?>>Q$[&=#ZX-R6O#0I.\"AT0\n> M,^EM\"=JO6[+FI!EM_B9]+X(>\"SF0'#WZ=TL\\.>+S]S_Z=P1)+HF/MN^66'+$\n> MN[NO]A7^\\-\"]8>>%>T]#W>DP)DX09-$V'-^=4`B!HY]6.T>_[L(XF&GWW$P]\n> MQ'6DF;?D'PQY?K^-)`_=U,3;0F.*:8&S,O<#]D?\\AAU@OKN@$18J5N^O?5#N\n> M[K%]N)^0I?=&0!97W\"<%%1'WWEJHQ=VRA;=%2\\P2(*`5Q19:8OQ,L5F^`1S)\n> M60<ZS'!\"IH%#/_MPO1\\9CB@SV_!;\"?X+]-_X0#CL_#[#81<N'';2\"H?]'>*O\n> MLQZB]`%Q95ROSU;S2\\79:Q\"T&A\"E0+''KC2HL`/%6I^>9+D%2@=R&PL]`ZJ9\n> M@OI-EE<+`,,PYK`!E0&$Y;PVK$+B5WT</@.ZEOS_=/A,L:2>KE=G3T%A&:RG\n> MBLGW=+0X'UW_=WUZYF))BWHW2!\">?`M>IPW7%.9,@BMAP:_;F$VAYF\\%O&]`\n> MNIZ=X%2>Z#3!M<$:2MDM&$=%6\"/PT6IY-77H3#,#'6Q`:!S'.'0<C2;L<(2D\n> M;Z.+&AB&Y*)+R3IS:(!3Z\\>:34'DD&O\"HFTC^1R$(`+B%(<`M+)A(K@:2YCH\n> M\\/OVM1C:T+I:K6,+BJ(38U[)BZ\\M\"KK!F;\"VNDTQP\\#2$-AG5ZNUQ_5!DR^%\n> M\"2^):^1@O8`#.`(ISR3-#&@_0I?SZ7(UW\\PN@$BV-\"S+OX\\NGJ\"%`<SW:^&V\n> MNHV.U(D9?2X*K@$3.JPW8=/1V<9#%08[4%:#`U!6@.)\"9'D9A<M+M+1I.AM3\n> MY6R,1W?J(I`AZG,\\*!W<U*_+^1A?@K*4+Y;6!6=6UGR-.F/,E$?\")T!V^EXY\n> MX=&C5KK')U46!E6*4CXLEX\"U>T09+*SRWL)R(S1]\"0S`$*G^<9*F>,E#X?7Y\n> MY-WPU?KYNR&KF+0Z_NY$LC%E),HOLF7`$8Q*9&NW0')]#;`Q4&4Q?C?LMJY&\n> MR>].OD+N(,(1-*T4_VL+T>&C=R=,^>3WKOGR\\6N4_/:G/[^4MF0Y\"J@0<'RG\n> MAU+-U])SI/Q*VYED&ID<\\)Q?H1]!E$;X])MV\"L;Y]?#I(\\GUDW+GK_7YM91#\n> M;,\"YQ9A7G,#1:LK51.1?(&:5!@\"25D@#G\\M_N!DS&=:X+`\"HRC61,<E2\\NF+\n> M$\\D11'G&'$^U\\B(J20:\"9^)_%L5KTJ(\"<\\N7+T\"\"UW_^_GLA0<\"NP86ZGV5:\n> M18VWBWI\\$H51&$0E)TUGC&;#78[Y2EM:`M,K.%99N]VK=F55UK,R\")PXI,S[\n> M#9(^B*.TC8TW=CB_B:Q`_*T%C\"N?L0V'ST$+15=D5>33EFT0W9;(%S,B\\C(_\n> M;<O\\>EML$8I0UX'9UCHPWX8ZG':A#CW>F%L5VD!V*G=5I,4QH/*&%,(?W1*A\n> MP&M$D/2POA[XN(QSS=7&%6[A%#<Z\"W_QXDR8.8E$S7C>@0[&<L'F4=\">2;NG\n> MMB#%B9-=4#KJ]J*C`'6&+-RI6M<UH@=/V^C!16#+$S0(X:TU%1&=<P2%G+4;\n> M`))A[9)TURK*2:J$HE/6.RET,_++WE]/SH>V\\,TO+L_K\"\\H'+`U>F<1X*44G\n> M^*Y5R7*DU%'T0$-XGDCS@\"HGJY!\"K<J0K0F-\"73;N2-.2@$L+1L+MXN&VV!7\n> MUK]<.0X:G:,I$Y^3H,?$RS5XO94H=\\UH\\P.$2UQ)7^27R.`$UU<?=$K)&,89\n> M2\"E=7JK*TJ`KLUVX%&/M(LSRT*\"J9RLLD2W,74#B*K3GI4[0M<Z2QZKC>-*M\n> MF_9U:;=658`0H%0`EPO4<+;$&\"\"1!P2^&&*E^>1:R'6H!7D;8<E3N$8'B.-G\n> MBJ/<5,&\\B3#ZS7S],^?LK^CZ:R`_0VA_>_($]`!JKLC]#PHX^LWK;P%7.!G^\n> M>2',MKE:2.$R2@%;DG.=DXIDQ!5E&151E#=\"`6,R;E\"I#75QWN#_:OZI@9J\"\n> MJSMY6XC6U/P4RELTP]5R?'7FV`EA.:[16N5C8@I/%1*Y,R\\<8O>TF713G8X[\n> M0-];TW'BIR/8\\$,MY,!'<\"WXG*THL$9N-=J!8ZHX:C=&(;R!43R:*L\"S_)\\;\n> M.)'T5^>F!#F`^)6J.$DH'/Q(?R$FNZTJ[S?+]RJQF:PV-:A,(J*>!XV^T\\`D\n> M=A4>0Q]7A2>!'7:M94D>;=!7LJ#C-\\7L7@OESF;O2*ZWH^D[55U,#3(4;*I!\n> M@0%0GXT`$#U\\^UY;_(RJA^$^_V/''$SQ7Z#_)GO0F-.3/#R`QIS=%QIS+B:A\n> MH3&OW.I-\\?((IL;7@.=TGYU*]M6)9`H(/FP4HMHR>O)[2`^A#LI,(R5.7F$K\n> M(G&TF8$[G^%KD\"6:`D#*685G5XLI.E#@+D8?:XS0.UVK13&0Y$&>^9&3Y.<^\n> M?9DF@R**]W!,2S>I@SV:4%Q%@Z*(J;#O\\(Y66B:QJ_1GL$N[4FS[]*]TU:H4\n> MF^\\W5!J%!7[[email protected])`PBBHJ\"E+TD]]OKB_K'^9CMD\\LFC_\\\\,/W/[SZ\n> MQA(\"N\\$E7%Z^E:10K*7`'WYZ__(_?_S):L9Q4Y>,%U:;$?F>^I[1#FH1'V;#\n> M)Y%O=U*$5+<QJ:SX=W[$@-[/DJU%N)DOLU?6_8]?D3<B!!V$Y@?VB;!![/GG\n> MAY^>Z=<\\K#S_1'E>!?T9\"#=&JD,<%$0X.W3S:.[G(`3K*6\\835QE*P_QD-Y2\n> MZUUMBX>\".(R+FZH5E;?888TY6*/#&?\"@%_LY0R9MPQJB(N?%'M8(^O$&%,?\\\n> M$&\\$.\\PA6EW>80[H5KEG#F'F//#,(<K1,_N<9YX[9\"'/]HJ7X`![R'*;M=BC\n> M.]=Q;B3XO'$JP_3&<=+5[O@X:8R//>,D_6^-DRUW-XW3RHU3,SRV+MKPS`Y/\n> MW4!3,^!&:WAL[?R-RX`MDO8BBV3@TV8(9ZI+XV\\[2R,.066!_IL>\\)16]^DI\n> M39VG-..:[9?$<3T9\"1UT)=>0`/^`YT'!]QV5XDIJ!:&X/\"*IKG_J6]#U[S>J\n> M(+KJV;-(.N>2H%-,-1\"H6KK\"_48W:5$6\"O/_>KE1,/).5`W\"GF\\N-A>2>:[6\n> M%G4QZGN7]*;\\L?D)?NDUW1+>)_E!\\?&IM)FW=FH16BR&C=/%X:-\"B\\JTT$@Y\n> MZ]E\\LGE*PQ;*ZDD`^XTFTW*!Z`US-4GRMO-Q>8[@**X=(PM;P9I'IB0N`C./\n> MN(2MKQ>;T=_)A?7IU(K;.&-1:CH3I7TQ:IS7B%%2491M/BWIP)<$]-?(N$,9\n> M7E[!0322MR^_^I+!:3[4FT]UO:\"Q8AIXX+Q[M*`N5SYJSD*=>`NV6XR7TY./\n> M[-U8PRI\\@G;+WFG@CL\"9,4*XBH)3-6\"1`$OJS-BLAZ:]O+AP9:.!0@X0R(.8\n> M!Q-IB1+KV0D.-$9A:<C\\S?#C1!\\\\Z5`<VY_!GVOUJZ'+.2)5TZ&B'J__>S?#\n> MP&=X]/2D2')<F(>\"T4JJJ5\">_(X-.?S^E;9-;?$._T6,6&-MZ_Q\"IZ9\\F5_,\n> M_QNFZIM.2P/&\"F=+YZ[Z/ME=N]7D;3)VDC&O[P+@EM$%<%\"$`S1\"W@6M:8:E\n> MPA3P<80PGE$9^=!6\\!<5)'-]2?:`W7E.;WH[IE4[^-)\\R[7LG2CT-UX1>%_M\n> M6AB.2/PG&LR-\"WK\"N<[FE@4$TGA9KQG49#(\\7_IM`M\\\"9]LWL:D8]FA#/_L4\n> M>+IPX:N=;_C\\J/:3VK#X_LN53$OXN15O'CW:G,FD1-\"&Y[9-`CZ6WVJ5%E'%\n> M;:=V/(ZQ.G4`]3^B.\\?'6EJ?!+ZS>P,@N*`[9Z.%Q@^\"*^V\\;D>L8G\"C,-&*\n> MV_%9NJXPG;>CS9<RAS6P4511P'^:74L+:5R/YS[XP=`YOUIQ8O:'7PB:_08_\n> M>^NM$`2ZT7BJ^V`1%XM7NE4`N@1&&`2V(@LYCID.)RICVLWIA`?Z(&(,OIM%\n> M?5:OUZ/5]3-NE,5QO\"_D#J4RUL\\NATAS$>PB<'$I-\"Y--]0#&:.]NTO\"SB%C\n> MO;`,5%J>BK:K$G(ULM:S2/6W4IEQKA7\\*LOH:+W6;5;G5L5$D)DU&GL&4\"GJ\n> MN\\GH/ETN'S,RD_2B\"42&\\%\"(:6$>3.I'&D.L'E[BN(W&BI\"RYA?UJ3J>TZPZ\n> M0#D&/>N$6L,P;]1S,[8(-8@LPD5Z>^O:T0N+X'K9ZK50=;.\\(J7IU\\RX3^$H\n> M80H`,Y)+C(ER&E%L%L2R8X])8+N['YK%74>U$R\"'/N?I:G0Y.YGN82@9B='E\n> MY?F\\V<FTK3.W$VQM</%$_%\"\\6JSKE>X?P5L41$7C+IIW4K9VFI\"R;N+:S37\"\n> MEHL4)FU@T*^XH%V(R'*Q:00SM[T\\T6WEL]:F<G!@5WET;KO*N@?`R;(=GR;X\n> M<%.GW.X!'595YM9DQ\"M29]:?+.[9(_-\"9;`N]I;V(UI%(&7ZMZBY/#ZIRC!(\n> M`,8!(>$2\"UD>K;DOIZ7BE!*44TE7B=Z=X.V4;T&BBNNI_::JZ^F.TIWCOT#_\n> MS?;XH_)_0G2P+&KYHV3%3S#%_OV*NXJ8'ZV=%B>7VS)(A4.0X284^_Q(XSRV\n> M)]ZID/:O-55!^KXO5TO($-V'H:]:LIM_ZY%;!5MQ\"&49@7_[_2MOQ_U48SN@\n> M)C,@KRK_HG^@-ZK]_[*ZMHU,-_OIO\\$`=KY_A3SF&V.>;:E1M_)M_29Z2YQ$\n> M@3K33OEBAH2]J&_LE,U2=Y89'_#RF6JA)EZ`RK\\Z284(ZMOZS<(9I>T086<W\n> M36!ZPI^Q\"&P2;\\V*07M6,`VW@H/]Q>V?%L*EV-@-L5_0TA6:><%RDUQWJTD!\n> MV_-5>HAB631O;@M8R2,3Q>\\'W\\A^P1[^`_NY'=ZC_!?L9T\"W*;R/`X.C+.@V\n> MA8T'HS(I@ULR(?9Z?:;;<Z%M]1IM;2O7OQ6%IW0<<0/Z`\">*U>,Y,8ZJW/:#\n> M<9DP2TP]G]GVC-%`<KZY%EWA@E&H8@!74$47_2I.M]9.3]SYHE%^)JV5AL,Q\n> M]=$A;]S4W=*/3#QMEI<\\1\\+&X#K`3L\"M<N`#;@E1J0E=K9MB+-#CB!H.@Q?Z\n> M0+=*J<Z6SJHVKY+?G:1KP;8LQ5\"VAM!:T/VW,P2+.X6ZDSG+&@^=?<OI[K[E\n> M-+#\\/LJGD6WM]5:_>=@*U;M118];=<T&++5E(2F'-D\\L$F^S$3OF[ZV(H!IC\n> M+\"XHIM^J,M(-ZNI.:.U&SU2+WBF%SRPF9<E(7MMZWIO61J1S/&P'0K--OZEI\n> MQ$*5;\\]'4V]'44J.5F)>\\<`.]3[3#+FQ3?L)IB-\\%ZZ.B6<<#48)GVDSY&U[\n> MR`4R=&'P?,:9LQ$O1M?>8(!1<[5J1P7TS(%I[631'+N_J_J)U4X?K$59Q!3%\n> M\\-(5J$6X&F^*6X>#,#BPMA(E_IJ#4',[T,E[=QP\"[B8&'DS\"M-/GLR:JI`;7\n> M4]_CEIJV<<=W6J<T*!UHEV^?T]B9TBT#KMEW;QHIV=H&_.S0T3N-SVF1?Y>Z\n> MA3W2Y+L#J$&E_?!U(QXW4X0QHRV.K[-J<>*%0N$RB'`2SP?YE>^FVG))8%\\:\n> M.T!;5\\+]Z'Q*=&2-S4Q;\\@B=L[?]D+7':LIPG;K+:Z,_:6)^FG&O#K1&FDZ:\n> M+6`$6]1=T^TA^'E+JM(-<G&U5L$!MOZ@;-[A'-UCGG5F`UOQ:7E%CZ-^1N'+\n> MJP]PS[$%ZH\"6G&>=$X,^:+2.S/ER.C]S@9<O@Z[:7.\"_0/_-_ZFG>HN3-,'V\n> M#E25[1\"$HG4R!&$D:K\"&($RJ0$,05H6&((SS_\"XA\",>+R4/\\P8?X@P_Q![M5\n> M/\\0??(@_^/]Z_,&T(GQ%]A!_\\\"'^X$/\\P8?X@P_Q!_^)\\0=S'TP'3X8VB.NT\n> MK4<%0#P\"7)@7'F<Q#7O!+\")'&E:9U<U+J%IA^S'L$^*$^7DE5C,E^3'0U^T<\n> M^1Y0^#TY>%&V7QU=/#1@*<2*\\QN&I08=``P='A5)'=?@N\\_A-JA[B(`8#-$Q\n> MR-+0%7(,-?EXW?K8^NIKOA5B<A6[V^(.>33T@8>R//:!A^RY3^\"A+,E\\<7SN\n> M$7@HBQ(?>(C//0(/N7KXG.^##.X1,4A14^\\<>(@HI_<0>\"C+=FYU]PD\\9*1@\n> M-!\\CN3XK631-KQ$D^H0KCL][*7LX3ZO:&_/XIO:M9Q?4/$MSCRJ>I0V2>);%\n> M'LB\\\\YSF^T#-]5*Y$4F+Z05J?J1Z2]/ZWE1_JRF*B*LF]C3\\4&*!+,+0R4!-\n> MH\\_'IFB:-,7Q^;@852FA>2PPD`8!Y+,U@6EZ-8$QIEQQ?#[>A$Z>5K4WYO%-\n> M[5O/=\">829H6H8LFPH@_#E(N2Q2$\"3*]_9P66SS&H\\+,:L%,K)AC/-:K>GUN\n> M?6^JOPV/<95*+\"Q!Y9[/]=FHIBM9KP'6I$IX+>+XXJYYHC3T>?!\\)$_>#+`^\n> M[QW@(S&$D\\I\"5MTEI!=5+]6,[A;16+&DTUNN`AHE!Q?*;!'ELZWCC);F0K7J\n> M<Z\\`@F'1!`,\\'O/4HL6F?NWG<Y\\`@E8/GY-]@]XG\\E^X`V_T.>NXQFR]^SK.\n> M`*:?$4!02:%1^93D^JQDT32]1E\"#`5IQ?-Y+V<-Y6M4>7<=O5<_N.AX3`$D7\n> M4J(LV8(9QV92(-QP^SGR\\*[M=9Q9;1VW8GJMXT>JM^?6]Z@_NNSV.F[6EH9L\n> MM\\BG[0CR3-,+=%ZCN5MQ?.Z!.E\\UP/H:T-E\"W/+9FE#UQKW7@,Y6')^/-Z&3\n> MIU7MT77\\5O7LKN.1QC[1J&`:]$V_QVD3B*S]K+%.=H*2Q5$3E$R+Z;6.'ZM>\n> MGUO?F^IOO8[GB5]#\\R1TJ[C1C\"EZAC5@4@O#4\"6]_`/,4V9^X2^/1YG)F\\'5\n> MY[V#VV,-9]UW7L/+'7OPLY;P/?!A-RX`'8[-1)PF%?'WZ3,(<R`!(Z(TO^IC\n> MJIC*6>$?6\\R:I:'C52UA3U37758]7F_:/!;^\\0[!J>\"Z,+43;@Z->\\6/&@P+\n> MLKS]N!V$*TM]#\"Z,?K\\87$<JY5/S[:[QM^(P2KP>#_1KIZ_SN^GQ6%(ZSSMF\n> M!#439T;0W.MI1O2I7I];WS_'C)AN10?,P\",)0V%40\"F.-1));M_=<YE;I#[_\n> MW(FJ437&DQ:S1SKL#1+8H_JJ>?;I/T/HYH1'#/W\"D!=[E(]]>=(FPB6?>^1A\n> M_RV/BS_%_#9+6*X^'U,@$(0X#ZG[Z+-F.]YLR\\-JPZ-YM.QV5X^31_.T27J<\n> M/+MYCO>G:4]_&A`_UO)0^^W1'\\VC--`\\Q_O#/#;<ZBW3X=;\\%A:1Y?8:;BKG\n> M-G3ZW+>KEH?5'A]N+;O=U>/DT3QMDAXGSVZ>X_UIVM.7!KN2W&ER$*5.8^-W\n> MT^0@O3O/.XHD);E3)*GP]U0D^U2OSZWOGZ-([DKR.(U*)TKCC(%W*$KYW3UK\n> MT#%-K\\_;DMSU6HO9XQ,Z(,F/5E\\USS[]9TIR@CF;\".%S#Y&81(U4YG.//.R_\n> MY3%::/Y<(_:R7'WN(\\D!,N^DF&4[W@3+PR841_-HV>VN'B>/YFF3]#AY=O,<\n> M[T_3GOXTH'BS/!0-/?JC>:)&*O?H#_/8<*N_1(=;\\VL[M=Q>PZUN%ATZ/O?M\n> MJLO#)AP?;BV[W=7CY-$\\;9(>)\\]NGN/]:=ISA`:'`Y_$>35(BZIDS)(L'J2E\n> M#D2<Y/),PR9)B@A(?WQ&A*<4%L&^D%<6?``[email protected]\\(/,.15^X-%CC\"L?+@<\n> M\\59&-\\=+PJ9LWQ@H?;HE(E2>J1`WZ>\\<JR\"L@-9:`M0_J<I(]!7&Z$@1US<K\n> ME(@5=1KSJ^62OBSWQ_'`QED[CD>6AT;0[@>+XY$5)\"%\\?WP[%)-0*F6\"LG<H\n> MMC[=2L.B!)@T0Z3X]'<.S8.AB1!\\Q892=(7,#66DL6O(H5%N(;&`])6'R5X.\n> M]4$>C2&C)\"XZ'&H?C$,M.)YQ:)0E!X(_N!B$>=R7H+VZ!0Z-=*%OTM\\+A\\8(\n> M86-#&6O<0@ZEBRM6459KI%=P:)Q5Y5X.M8!IGD/C)\"HZ'&H?C$/AHVTX-$ZK\n> M`\\%BC$.Q1-R&0X]TBQP:IT7A.)3I[R&:!AWC>;*]:W*7X`\"(\\R>61K+M/^/%\n> MEMN'@B#2^_VVD(O-#2V\\910(TI\"AD.^7AD\"2.M#\"6T97(`WOMX6Z97.XA;>D\n> M(2+5JI9S+^$!M+A\\9S/P+@$,$)3P_EJHQ=U_\"_-\\>Y3OU,*\\W#D8=8<6<ECN\n> MKX5:W\"U;>-L@$`Q:F815$P2\"\"!O):90<C@.!\"[C!M^>&D/L1L-^XA_@W!9[X\n> MP\\EJ@!\\)#;$<+C3-\"Y9&H.$)$$?.MD%7[/HC4%Z)0I%U$*_=]<:W)XNAOXX8\n> MI>Y*J@.?NNF2,NHFGM\"$E0!1:^S`<ALL?H,NLIO'O&D+7.@?EVM9?VK`.R@D\n> M+6_.&=X0[N4120'%S'G?<[1@3(@6WBWN-FJ#\"-6Q`SG#ILVL:?ZN\\+Q;,E&2\n> M#;\\ES]$$7SZ(L85.RQN$O)T\\<Q=^%1YI'ARH?SI\\(?7/A31GR\\46:6I'&L*R\n> M%,\"_]C2:#M\\L)<$G`I4`B@CHU'/>G`;<GL8_\\'A,O((>M+'%\">D[1_Y)%T6W\n> M<^'4H_@2*ABXO@K,2[0MJ57A510#N(&\"VP;<YC5)`OS6@4?XG6V13K,3SAMY\n> M@834W`5M`<DP=H.BJ6ACRC32\"_?O\"/8\\<C/*8Q_-,+DBAZ/D/P>*B01B?AS^\n> M%>0XVS>:O+R/1O/:IT*-,RZ&36B4`^RYT?2QMJ<B;IG'LEK5N$3R?#'F'6T6\n> M*9J5M28@H,65H3>1]1288&V56/`-`\\2RF]JX3$#\\;@U^$N@%<^#,36>&<E01\n> M'$T&OY8?0>XY^&LA,LSN/4]K`$6?$&/<P3BO&RBNM</K,NX0I?1LOM%!\\X/M\n> M`)\\-]@@]=;PS/F4[Q-@@4BT%T'@;?UJ1Z0D1;RA21N>9\\J7'-'.W:B?MR_5S\n> M!0;:P:-V&/`>PASM96.B.%9.T893V/(>KR'*M)#;UQ_GEQX>2^]R6R\"5-4#=\n> M]79R/=IH<(_VE`D<$'T'-4W1KYOV-/>QM64E@@#877&@\\46AAQCJP`PT@6!P\n> M7=YNR\"\\(:-Z*)T*F'ZV4ADVE#?8^X-KJB8]\"TL'T@K`1.\\_PY`V/<..#B:`K\n> M^Z#VW47H5CP?#D>7_\\E@O%\"/;BN*M2V#-==`P'PHB!E%`5I7NQ@I8RR,T];\"\n> M\"-\"YY<<UYPEEGJYTAR1^T\"QXK%XO8RL';D1H0L0PO`&IOO*(!5V@!X.A>+:#\n> M9T2(Q$#_+?;@&97_!#PCXI8X/*-W+G2.(FRC5ZOEQ0FBAJ2E0RQ2O`QY=Q#;\n> MFF723@\\3\"SD4:*A6B`Q^*'\"G?U^6%U\\1,R$K]X&9?4MD0V%RZ78'WX\"B[;I9\n> MMCP:W-3DHH8.,I&UO7H8K`CG]83E#[;*WPZ3,>O>_5?\\+QP>R8FL#\"\"+6K%%\n> M/*R`@OZUA<S,Q4F8-*`AU#O:L/<4]7OP4XA2<5[[Z;?&NM.\",`!J0!$JS+.)\n> M9(?[L82BEQ?Q`+^W$1J48EBR30?QP!G[($L4,ZD,<X6W>1?'N5?^7LSJLX\\*\n> MSS+W(S+81G!<6\\\\G%M]A&NSB3K\"OC%ND(E.E$Q?LS1SA(F8>%281KD`Q62C-\n> MRG-%;-BJ$#-$EP\\OM]QZM1Y=<#E:7%$XUM:NC:R9JN3IFJ5`&XJTN%RMZO7E\n> M<F&QFES:0!.C'<1#L49[WIQI74+<Z\\O:X3^%81JT\">GA^U8&?7BA\\$X?**,W\n> M.OH:88NXK)?+]=QT0$9!VQ&V#:5^1OYQ-[];?()F\\5&N)DA&1O!_A#\\P:(O+\n> M57U6C]VL@J9*]@8W>S:E2HZI]$+CG%%JC9J9Z2.#M7A\\UM8/W@T96`7CT\"\"2\n> ML\"D1$$N\"UJKDVZ.K;6=F.3823G3=)-`L$CA^F*]I1UPMQIQ'IJZ9Q;+$#*+=\n> M,AVI'D0*98B*$!$XNHD%IB`I;IRW8FA8O#?)E58#9K7@-SZHVQ8`RKR+>R*,\n> M$BA^[3Y&8;P0WQ3`WRQHK34M)^;)PA#,=T5E&TBX\"PK50G&:M,!?I0@.!T/L\n> MN$FAMU;;H$HS+]\\0[\"8!G(B3Y`2<8;\"JI>F5;O!;0?FVFCD:&PASX$+/:\"-*\n> MT^=]F!GE.+5]MR+2&(YVDW1*]FCDY);(=5@Q73X#W&3#;*:CG$P:Y$DVC+#`\n> M#3SMS\\$>>-J6]C/?`JHY5?!?Q#L2SHH04@,86^VF>QBO65,-JR:F\\#XXW`@X\n> MN.S16./=<!83/T8U9Z?[L5.J0<XWJCM[1+GE:CZ=+T;@F:WR3U6F946YQ:KH\n> M^W,&7&M!2.VGMU@[5Y<G[:B,PW,*K,!T6>MWQ]R!Q)JH[34#@O>T\"13(Z9HG\n> M6!(O-6*Q6A9N141\\=/[>\"1.H6#JBD@01H)?W`*R*\\@_#F>J_KB1N\"KJYQK$-\n> M6E5!V`]87FO^&XR0!FND$6^>DA9L5;M`H\\DGZ6=]XF(4\\0!K0?PY(B[#H%6\\\n> MMP:X\"5>/,>$\\OSO@W(:-/\"+L+GINX%9LPZ?;9M:/PS?SQ9DB2T5EEGL@<>O.\n> MM`%BFKJAOIC#`)V(\\0[LZ@G!NTP?U!GA.:WF:D74+UWT`9G%I`H/S-2=]8OM\n> MJ`A*^FFT[E*SB4\\T)0+PJO:+.Z0753P/.C8W*#X7+LJ@E[N:!!E2Z>VDEH=0\n> M$PO$IQTK=EA4$0W5JVBS=E1)!>ANH=@Z\"C8T#[96]L29?\\[!TD9_\\V&Q.C!O\n> M!@I'_#2<?;WV(]X&AR,C`0'.0,T\"GAS:G0W?C=HS>[;;Y'TZ'L3/Z.RCQ4IX\n> MK>N=GT\\.HUP].+\\8Z&)[;,PMM*$QMM:N1$075:N,PNZ=1?=BP(\"I6G2&R8W0\n> M4K6:CK8,-/&83`M'2Q8>:1K,(44O%O7JZ5*1VNM6N+NIQP2LSZX(31?$<1;3\n> M,+9D#2#QQ*$&.Q<&!UQ&YV1J,4@U`M7/@2F#XRY2(`QUCZ+L8)#%`AE[&$$=\n> MB3.#>POBA/\"J;OJO/6>O3Q76,*798-;U!-9U\"G9LK&LO1UIHC1.ULN<M*[MV\n> M<@+5F^V;Y`!U#!WX<,O,?!'!?_0B;EN;P$2U$\"D-OAR+R...S6F9DXX=R<X4\n> ML6D%=-@(`T9P$IMV,_J@T1U$5Y5IYY'?S#DT8@0X>/M2&(<6^N^LV^^Q]OOG\n> M3K\\]N*43YA/?_T0:7T;%_O[/T']#TIRA_]A$+@E\"JHLZE#HF#\"0E50V@L(:5\n> MV.18G!K+?8ID#3TFV`<C18(DU(EK*F-F*B.$/954^'<(U3<SVZP5O^/-U8?S\n> M^>)C9^9R+5T[0>75^YFU>`N(,/\"6^G:^F%N#5:*4^<OSGQ1:NUGP?#NVXP!,\n> M.PUM!'SMU-1W;D5XKA[PMS,?XMFO<;3NNR4IN+]UF<V+LCCPCD``R&L?9EN*\n> MR(2ZA]F$CD0T*27=')HKM0$-]-,BI?J.NHZ-MOW9I4,;L74@&GB\\Q^!U^LO'\n> MX4M79)0H2OA6[1;XCX\"NIL`ZKSA'-;!!F>X=E+:AL-V$B=\\#T(8R?)E;H[9L\n> MZ(D;^VG0D*6E['BE5X,AS/9TV()N0V5^LVRO(HUA-((?1T/CNEZTB<Y9DB06\n> MM(1T.KN)X5V;)_O;?!-IFJT/&U.+MC+MM''BZB0!D\\J[6)UD;Y.P=B,UVS=2\n> M.R3<TRHCX;:;LL)_@?Y;_E-C'552<`D6SW?Q(T7L$3_2PT<6N<)'Y@X^,DG#\n> MV\\-'.FOU`3CR`3CR`3CR`3CR`3CR_UO`D7$.X,@J?P\".?`\".?`\".?`\".?`\".\n> M_\"<\"1R),3E(HU$)4V/.1>V*EGEEW6?C<(POQE\"Q+<OP2';)D59,E.WH#J`I]\n> M7_\"8A<?1HM#[3+$IV/N,X!1\\MNQ(8H^[E[!N/DN:9FC/#@*%<T/<XE0JH1>*\n> M;3@CG\"WY[,/6<>&P`SJ-:[:B;EM6O'.,F7Z86Q:#NPY;Q<\"+!>_N#4A)A(\"Q\n> MH2?%C8WY;/RI<&]A+Q0;H,NXXH@8=YS%.WE:U=Z8QS>U?SUZ#%WS$%+*NLIG\n> MR\\\\T_;K*6P%.`%1EC]G<O9`4IF+-\\[@T8J.4#B5.ORMDAAC[8?>YW`'20:0Y\n> M`YO3L:QZW7_N5;T^M[Z7GP6DH_RC:'0ZP(I&I\\]IU/!8V@=H5\\?;BN.S9NN=\n> MIU7M41Z[93W*/YI'^4J[JL^:7]/TZJKRE</Q*\\/C3=CEL31._2\"G<>$',TU\"\n> MSU>=9[W\\OLUCS.J@'^,]-_T/\\-BQZC5-ZWM3_:UNFV._*XMS?R.>ST<&BWFR\n> MI,F3'<<-89XP;3!40AM@XJ*P3YI&GX_=-@?2GFL\"G\\M]``\\];EYD\\0[HXJVQ\n> MEM(2N*X[MX@T:OVSP9=??WG[XHJ=^R7'T!>!^&HD27@O5$=%GY4\\FJ8?A7$W\n> MVU&8SWLIW)WY<>E'15&4K0EE,T*:IE<3%%39BN-SCR:T\\Y0'&:.;IXQO64]7\n> M6D1).<ARE7>(\"Y[EA8IF?\"]\"O5D/];#]G*?;V!2*<*TPE\"12O@?N<H^TZ%.]\n> MIFE];ZK_;$2.(G'5QFF5N.+UNU8;@Z:=Y]U>`WO$>JW<V:_7O:K7-*WOG]/K\n> M!6\\C\\WX\\`1K=\\S&(00!I$Y!;\\_\"Y1QZS0+0>M4`TOR)J:[GZ?&SZ8!VV)NCS\n> M7AST(_?G`&^0%L7=`4FQ_E4[T+2?*2-9G\"Y;M\\(WA<G@1J4H0C<J^FR8Y4S3\n> MB\\*T9EQQ?#Z.-$\\:6![*'&N\"/EL3F*97$T@#5QR?CS>ADR<[R!C=/%ERZWIX\n> M,50OIZ?J0.!#K&((/^KC,3[&O5\\%`2Y<EE[I6[7=.\"+:NO[E0SBG##^!6_A\\\n> M/-?'RF`N_..QKB%EH0`BD3W>=H+B@FNU8U6OZO/%Z$(FUN^>_^YV-Z+#D!$K\n> MNL7]]/;E#=.*BEO2`NYN<+NSLJ79]4'M-B0C!\\29'96:7$$T!U%L$P,P#4/+\n> MK$M,G\\H)86M%I4G8I_)6CJ;&F_6J\\%9U[()_&;Z0P2@.'%IBV>`VMA[C`PB.\n> M+0#'/JOMT7H-O-$_QI^ST/*P,[&;+_2?'MC/FB=2@B;N^=R><[4<T^;YMBXU\n> MQ:B*=Q:NG^KU>'EQ*[T^(ZK'CKFQJM<Z55]$Q\\#5C39M0/6C>%1&&\\VCM-%G\n> MI8>F^3S:$%?I7FA#6(<;:1/?Y(R#T,HJ78<AS?ELRTT>J^)$P1;W4IRP#EAQ\n> M\"MP1'U_K6GG:U1Y=>&Y9#R6I`HEX=&4/KYQG7+PT3=9G;54L32N.S]EQC'W(\n> M#<M#:69-T&?-KVEZ-8%2T!7'YQY-:.=I57M4\\-ZJGEW1FZ>A!SXT>$C[GGNL\n> MQ>YSN`]WT0&1DDC[4\"8/2.!CU>MS^[NO_M9BN`2$H8E4/O<0PV44>M'-9Q/#\n> M94852=-D!^)R]1##95QN\"XC/%,,E3A%_CAC&YDP<^^V;^#AAX'VL?(9T#P;V\n> M;H8B]QF*X^ZX,C38<<W!Y^/[5MH+/'%<CM?A-\\8H8UA3ZC>L[/'&N=[UFB3<\n> M.PMC(G!!B\\K*F-Z1W+Z[YRSU7A-[;@.+,:OY\\;68/3)TC]>D5_5)ZSGTSW?P\n> MFD3P'%BU#$UBU49ED;GB]3DSTD;9GEYK5H,/UV)Z];I?]6GK.<KNI=?42JO$\n> M59N%1%)C\\?QNSS(K,]<<>][NM?G3K9@]C+NOUWVJ3UO/A7^^2Z\\!(V=;15%>\n> M)CZ>$;_K5E&$+=;.\\\\Y.52<L$K?B^NU4]:I>GUO?/V>GJM-KPCMFD3JQJKQ4\n> M!_NYPCYF,4T$H7_8?8ZJ+0YG$\"7US9L[)-YC&^_!H>M1O3VWOC?5?VZO0R&P\n> MJS9&G#4KGM^M6NS0=)YW>JU:I/6:?JB>O>Y3O3ZWOM^YUTE!J.+2;PXR5$,\"\n> M;.LD]8^9;E0B;;8=(H+9HL2[*/<MCWN$]_%ZD^8Q=X^W]OM\"J%*9MGT!*ME]\n> MHMM8GM(]]\\F3YTV>_.BBKWFB)C@6G\\VTLW9JFJI70#X$3[)FZ_/QKG+)=WD2\n> M]]PGCW95\\QSOJN;1KFJ>J/\"JI;53T_3J*K25/(Z\\-L/GX]I/GC9'EO*TUY&E\n> MO&B.+/&YA^(7-D>6^'Q<C[.^X+%(>ZFC9G+JF2.UE+5\"S<YVI,?C=.Z*@EPT\n> M)C<G<X,R+4+[[IYUBFMZ?=Z6!WD4ATTQ>TA]0\"(<K3YI/>?^^;/$0I&4/EY<\n> M41U'(,=4='E*]]PG3YXU>?)^P;6*J(E]RV<3\"]9.3://?<2\"-5N?CW>54]'E\n> M2=QSGSS:5<USO*N:1[NJ>;2K^JSMU#3]N@JJF-%+:O4P>C6/&KV:1XU>?59#\n> M5],<,GIW=P-,SG`[P)A?GRW81MP\\]RK.`DU$[OD844N'`JN./3[WR1.739[X\n> MJ!><>1SR/P,G*!2\\YK<SD2PWZ^.QBRL/#:M'RWHT6_/$99/G>+.9QYK-/-9L\n> MS:]-U7)[-9L]K,+04X'//2A7Q463Y_B.%?.41>%'B,]&[<HL39:;]0O,*35:\n> ML]GS'LW6/-ILS7.\\V<QCS68>:[;FSRR>:!'>U.R#.-QQ`G,G+E.*CR@:)&)<\n> MH4\"<%$OB-*<H\"2L`^\"EGYC'V;8KN@H4P#Z=Z:N'4A7TXU7C&0,X._K7[P=[@\n> MC\"8`-W'+#;Y\\#PXW!NA4XQ?TQ.'NUZVR`**Y2DN7_LY(\\66!:Y8YE^`R1,TY\n> M:J-YR>N7'+E<OD<%GW&4--H.MH5(NJ=J^\"BPN;[C`)\\2M/O!WHR@:5@PP,%!\n> M@H)U;D?07MTJJX(H]7QVZ>]*4.&U`M9<K@9\\*L]YH88Z3ANR!AG<'$<\"2SYC\n> M9TZ-O2W_`4BC@/$7#H/?WH6@W0]0XEMO6<8`!_$AZ'UX))`\\+OHBQ??J5H:`\n> M$!KZM$E_9PY-,P:UH]V8<1NST)@%D=26Q1S*+)7G7,-28&]6TFSKJ$G2ZK+&\n> M#SC-'\"<BED'G@PN?D>K,3Y5[A;\\/0.]G*B^RJC>']NE6FA38GZ4KTZ>_*T$+\n> MR)=49.2%Z!8RL;,89>89OV),\\S25KS*6YV+-X+&(MIDSPEV\\TN)L1%$6G2+\"\n> MHH79Z+Q'<8FT%=G97B*I9S\\=NPG<6\\GI7U4)0B.D\\0&F[OQ>A&A\"K!$1\\K)D\n> M0(KL0*VY\"G0=PSZ#=YR$)1_QU:>]AXF0()8E&49X)=)3FOR:ZV8IHH>V'G>V\n> ML52H)'G8EBEZ.Z01*?IN$B4BZ2,;RT,$[/QNX2HBK2N,,O^R;^+H$J\"_]YPW\n> M1ZC`Q];7_KMIAX5ZS)-F=$7CY$[*R1$!54<%321R!8%Z^)7HI^5>TN>)FS(I\n> M2)6'A9LRK7<$@-5(,Z0,8ZMDT2'2F^3OO30>[TM<18,L2_EH:>],P%*F111J\n> MN%6X2Z/(;J!H\"%\\N)65128VZA5'%,=)46U348#+:V0NWSN69$SSM=V%&$*;D\n> MI)95EM_\"Z(``CU28P:/2EXY]N@3-*0JI>C3I[R&Z3PS<X$B7L0@U<[F(DUAQ\n> M,AG=!]'\\0HONDU!(;RG!<8MSH'<R]$Q>&4MVWFU.5ZD*\"/Y3Y0=8LO.[Q1\"J\n> M2%[1O,C@87E`AKN80/V'H1<Q1&4'D@D56)_^KL-`Z+0X+*@255$ZB\"/R;HS(\n> MP,+&NA4GFE&L(93!8I(FK;:'0:,F94EG'.R]&0C[@.,V37+W5A2'QJ*3P-YP\n> M]^(T=`Q@;X<-&9<\\+%J9<4&G^7AX1N'P4E\\KJ`=!Y9](_H]$;-+??2CC2MA:\n> MA:*L98BDKH<A$>\\H+W5_4<2[?-?G2EJG8=#V#&5I$=UB4V/T'4/9^8#S6TUR\n> M>Y.^'1K*3@+W9D-D;P=C0W43N+=\":]9H4_:V+W,G@6,'?8LJ7?1USM_`!T5O\n> MU;C/:,0AOU/#:M+?6;)FC(^8J3\"I4\"KC7B6XJQ#G*D\"P+147NF^#!53X(.WR\n> M`0(4M?E`+,VXPP?=#PBQWB2WM\\-\\T$W@WDR8V]M!/N@F<&\\ZE'\"`-F_[I',G\n> M@>G+[DU-I,-\\8*+]%GS0:S0R,3'C@B/0I+\\?/E!)I#5'9H2A9I5$R@<JVI4/\n> M=D2[XP,GVD4S+#JBO?L!-T::Y.[MH&CO)K`WFYSF,3DLVKL)W%#:FPWE0='N\n> MAK*_:.]%4`ZEBO8F_;U837FHSE\\8#+EZ!O6[JFRTECK/X?8>O]??TXX_!F<_\n> M.\\:3?3\"M/(_RK#&?<K$2;K2?+($I6VAH8T'9VV$3RA+TMJ&.T42?6]_#_@</\n> M;C:DQ$XNG?61IVIHP/S(X4LR6RI/Z?2D,96GVX$IW6CD%IC2F4]9Y5T0[0]F\n> M4.'H:&-1Y5EX:#1L\\-+>1FFO;L&LRK.D<G85T]^+854PGJ!:(86:OK1\"BK\"(\n> MG6%51&J1P+`JHG1KQ3*98UUN+*L\\<09JYX/95GFI(3[5N,JK0XN.J0#8ZKN-\n> M<76L6S\"NBI#BODE_+\\:5E)H[>T*(%3E[HHC\"TAE70M#2&5<%,`GW&5>>H,Z:\n> MRLNX:U[I!S?E*]7J=1'-J^P`AW83F(F%_<7&QBK\"_.:%^!:CT8LFL+%$'XN=\n> MC<7T]V)CE6$>.9.@5,.:)H'H2X6SL>1[Z&RL,DK2[='@XBBJ06<T[+T9#?M@\n> M9M+66Y$?&HU.`GLKP[9^;F^';2R77-=<>S,;JPRSFW5K;&#?QL8Z0E#:6&5(\n> M(C;I[\\7&*@N5D=#JRT+E(K3Z,N=A,-I899&5SL8JR[#</Y2%;8R8267OC8UE\n> M'\\S2Z;Y)WPX-92>!>XO+EH%6IH<B1'<3N+=<:U83RMX.VU@NN6IB]F8V5ED<\n> MW,\\Q/BAZ[^?T&0W:6&5>)<[&8OI[T:U+!$(P5;#4Z-=4!:7#7K<6/O\"ZM?#!\n> M?MW:\\X&95`T?=#^8I=-].\\P'W03N3?G`O1WD@VX\"]Z9#:2;483[H)C#%W+VI\n> M8GZ8#TRTWX(/>HT&%'-9A+UBSO3WPP<JB;1FE41:LTHBY0,5[<H'.Z+=\\8$3\n> M[692-:*]^\\',I*VW@Z*]F\\#>3#J;\"758M'<3N*&T-QO*@Z+=#65_T=Z+H!Q*\n> M%>U-^CL/)2*?RV*BI[\"DADH47]0`9;/2F_))(155D;:BR\"MYSK=.9601^=H?\n> M(K!W4XID*+L?[*W,J82F%?W897G(3.HF`';`J9Z!.?6[PE5XR*5M0UG=8I^J\n> M#TT0*JRRVYT^_9T7VE2T^RJR8WDQJ)[0=9HFI=1<\\6!)7!1N-.*X3#`:6QIL\n> MHANI_HR,Z96(76X3J_/!F:F)>OMTRQ;G9P\\L=YT$IL\\64=1Z,]8_K/Y:`K?=\n> M$&K-JGX)50^MTJIPW6(H>Q$T3?%=#QWY]'?>JD^Y#8B*L:.<5=2,()6SDO90\n> MGLIC15!'7I346;[G'(F>8[AP.ROV3DNY_4&F5\"MY%.L*5A6'+.5.`K=OF:K7\n> M2_?:1.H=L@H[\"0H]0)1S]2N4]<KP@'#4#7N<@NXYA$<)B5WZK$Q)2)>VU^@=\n> M@79D1*3MZX\\:\"^<6MQ\\S?Y;^7G$=JQWP*\\3TNFTIN[@3&FJ(@05N65:Z>\\&3\n> M$5MN=2L]YJV&;)OJ;[]_>>O+[7!T[-XX_9QK\\KRJUBVIOBUB6<&;+7?M&*_2\n> MWDO'M*2[=PRS!*[2^Y@F>K;\\7J<)UK0[<R4+RG=`2S]_JD`+NN/LY4PIDAWD\n> MV<^9*46VT[?/G\"E%N=.QSYHI=^\\8^?M>.J8EW;UCQ.ZXEPOU)/6-%^J36Q5'\n> MKS?OK1Y`-'KQ&8A&N_-E!]+H&``!]I&J'2`T!,NYU?#AIG6T`YGRZVBU6#X;\n> M1+<O*]T90BEKM-G<NCA\"R]V]A\\2;O*<>:EDW]3\"^]1B6U;98_ZPQK':5L<_J\n> M(8A^]R:14/?4).U>NK.(?BY;L6DW%;<UA@A\"<3F:UH$#L@]68MK\\+WEP\\9:\"\n> MP?]9+\\8NBE,6%8,$E[5&+NQ@%#/\"?90^._EY^`TB[-4[D06#;S\\.GR.`U7+X\n> M[AUB:7T<OF#L24;5FB#B&L-6^I\"1[7B1ET$GNA7\"]$1AP'_#BM&MN*XRNM5F\n> M*-]/,H;4O!B^^.[YCV]?_J3!K!*$&@L^#'_\\X<W;?_OIY9L_?:^QK[[]Z60Q\n> M_.&/)Q.&PYH/?^3[O_WT_(]_?/G3EV_D^X\\_O'K]5O[]X=N3Z?`OKU[^]80K\n> M>9RD&IWOE8N=B)C2]<JB=5EG&!5WP8AERW.-_ZY!ZC7\"[,0BR2*T*8,KCRS0\n> M5Q-%42.OKS2>'L,=GM67&PNC:9'[&`]ZIA''+Y:K3J1>!$F+*VNKB\\+=BGIF\n> MX>,M:N1\"6ZSA35MQ)2VNX@\\+Q%5D=1;@7(,FNM!I,Q?`[]UPI/$:78#LZ7#T\n> M8:FQVX(A8JD3#SY++?[MST9!!I=>NQ#75H=P+$#K:Q]E&F'N]H02GEM$5Q<9\n> M;SM&[C38CI%K801IQ$2)A;O[N!,J-]&XI(S#6(*Y6@-D13'<7BN\"+\"F,@-G3\n> M`S'3VQU<URXX-$-4__G[[SEL:6$!(#_49ZSDC.U;#J\\9=XZQ>A%=[GRYF&I(\n> M[AHAKT^5ND'&R*IH-(/L^1#GK.02$6/S!(Z(9`!`L,&O&DWY<K1\"#-J@J\"K^\n> M9H&/,4B7@RHLW$>F/1\\MFFCHC$NH`>[6<RC:,D@:*;&9`Q\\9`M;&9=7$[*V'\n> MEZOEDQ/=84^\"Z^%9/1;N.]40M:\\VTO6ST4)CGL\\LY'B]'8[WG%&//PZ?2>IG\n> MDO:9QA-_NEZ=/47<6A%D3S=GR\\NGES*XB\"%T>N;C1B=A`).(D009;I<A\\AXQ\n> MLF>%,*A\"L?>DS7LAQ7OM^=EL)-T+AH\\XV=]K!)O'PH<_S,&KCS;7E_6OCT_*\n> MO!Q4+C@JA@G0J?P<1\"'CHY(_WJ[J^GL-2_E(2\\3;CYH_\"DLD?+&\\$,D__J8F\n> M[PLG(?0M`'YPF@()_B&B%$>6XQ!OI^XM;[\\E<<(WL9\"\"*\"GQ\\O21T/\"G6@=C\n> M.OPS)@G\"G+[R<J@F\"\\R'+U7^3-CFN;!R&E6!U(U2'KFHN\"X0L!8D'UQB!*[4\n> M(.6O7ON0O8C)SE*RT)4BOTZ;<!4068^>HKVXKA0/KH/A?/).12Q\"H#[Y?7<Z\n> M,N)N/?SB:Y$KKU]]+R1BUCR.`U`H32M<AT)X2U^$U/\"UU*N]>^^[_=YHXBNC\n> M^*H1HY5%%BE*^<T(6X9QB\\S8%G5O053%^L:QJLJ(V;B0P#K@-/\\.TIRKE4YN\n> M$]!-!&)$-1YMNK&L?23AT7BL4;$Q`)?UBH*]MDC;KI#1^7QS?1+4+<D\\;R3S\n> ML+;(PL`Y4-$!83-IYM[$YIX*XJ`5L?.<,K`]^7[9F7PK):;[%[3\\CI&=5S(3\n> MESZ\".Q\"DJYV8U2O'>I<!PM4SC<FAADLO!V6>=7[S`T@NR((X32HE]T]>_*Q/\n> MHLKTCSF%<WO1)KLV2[8L1Y3X:Q=Q^^?N>CWVO\\\\\"\"QHZ&726[FF3@HOX4@-A\n> MIR)P$2E9YO=F/D*4Z6L-B%W_<C6W\\1F=:YA1U]*Y,L0$FH<&#)>)XX.5<NG!\n> MY&*T<?D%RAI^;$VL^1#!ENO1^%1;D<=98(M&HBL=`[U(U0R)/=PPGK<%'9ZO\n> M?11UQWH,F+M_<92^4>I?G9N\"\\H9M>+.YV.C(Y&5F(:`;AF3LY$9=84!O2&'?\n> M#A!YZNC#^-;4I\\BPK=BLIR)AH.[-$5DV1FAI!I-?MR?2>GE1M\\97^)J5DBX%\n> M%H=@;Z!VU5UJ\"\\<N0Z-KQP11V4%$M-!%9$^Q$K]%B'+\\HH'I?70SZ^1T/]4A\n> M=[0M96CJG-![//SF];=L*IIS.?QFOOZ9<_U7].BURN'9\\-N3)YGT&2*!K8:*\n> M`*;X<*UL<[:8S\"?7G\"1N1=00[]*J.&0@[M%&(^BVDG)<N7!KN[C*C*_.:D3`\n> ME4POV#0-BHSA@/S_A&[J9&`L]+$V@\"%EN_K19J8<_$,KF+'[A6J@UVA8>Q6G\n> MNGP_%PF1BLK#+R8%B*%^.4\"(<O<96HT0\\!+'O?@QL(](R^Y%A4Z7>K0ZGU/!\n> MXG@AY/N$9&?L[(5^9VCBP/5*2;75J9D+.L\\^::.QJ@<^Q9E&/W=ZKDI7%U@^\n> MCC@B$%JR2@3&/6>K6EEG1`60OX(=VL'/I2^U&21S#+EKZ^ABOM#HUX[/DY!:\n> MA)4SV>;$N9__&BR:NC<C4EN+015JI<A^@?C.X+)O(4E7C&C=C9J]0R1ED*GK\n> ME<UC#47.&.H0%&+(.UV]'GC![.*9>^E'Y7/2J=#Q4Q-R>=*AHM;7EJ->5)H%\n> MI/-,Q2JG^7ZQJBC=.9J)M7QN%8R=3CQQ3`#RH\"PL^YY3(\"DXVK0X4HO\\OJ+.\n> M9V'9@T8HSAL:U\"T:4!U':\"V:0UB\"F[E(*L94`S_4DR6-2%,M+/BTEXOCUGJE\n> M8F9+\"C+ZN\\HW6D?L,&UKBX#GE1(MVO$6XZ)35Z2)I6`F:!&%SM@DCLT!S]C(\n> MXZU,JB7!<#R?R*^3&J,]62TOO#V^7'2CG^^L'O/MU:-6(=4.ZLV6)=3SG!WV\n> ML2D3W3#YK=J!K'U+$P_GPO\\S&<.1BX(N$^8:YGK-D?%T`;$#OPKYB/8M&K,-\n> M:1BKV`>QMZ-^1[A#%P7Z;W@@ZG=^GU&_\"XOZ'44GB>C2Y@]QMBX9^Y'^`N_#\n> M0<T:]M,[9R`/']GP0,5F9G41B%43XX9]I(8+7M0ZMA=>=\\5+FJI%:Q;-F]&O\n> MX!-;V\"=S&.>/U3F@IF0PY)80O\\P<P^J\\NCA)\"YP#;QDV$V^X0B*K'4,F#ISP\n> MHC*FB6K*(#]+45@1Q5:87[KD<U`4_/R4K2\\1S[X]]FJ8>+H\\^7WSVU?($905\n> M<JCJ\\2V4CIT<S6_,,:BR*+A63IU*VDD[+;YJJBB,8*7,UW^8B\\5ZS:1\"L'9:\n> M^\\FE+S3]C])`6G];S7\"_?'42,'U413I0T^%;S\"%S1>G<;2F'4[_TL\\DSXY9)\n> M8-]!V@A@'$K;T9H*\\14UZ-EP#\"DPH:9C<Q7U-,5SJ1^[1?Y42TLB+>WG(4=%\n> MS.0B5/N]H]C6TD4@/40`N;X6/J;]+O4\\>G?BT[Y'[]^/Q/:Z_F^Q0TX*T42C\n> MM*R8X37X9LH,W4%HU_-8\\P`O8RO/^00*[;NMP78.KG<GEC,/#]<&6Y9S*,KS\n> MJIE141%WWLHR<!,L`I\"GS;#=@8.^X-56:OQ>(,[`Y#KV55@>H\">`7R+`B5XW\n> MJJ;J;],N71UM4,\\5NX&2XS`NM''SX;>J`/RJLL6T\"*Z<SG&YW8\")+J!SR&@P\n> M0HS3>&\"$J:A]\"W0,:;BN!%8@1+Z?]\"3^>Y;![%%:-%,[CL-\"B(CU5BRF=TWE\n> MC^G+ZK1$>%*SY`7EGW!9$\"<BP?=T;4J5IJ,X3:#(T(TR:47]G%(Q\"K($XKK,\n> MG2S:+0H:R515]48S\\M]1:M`4^_B$!>+N#[N*\"1'CR&)K?)4DD`AS>&2^TOYD\n> M=(II#4SQSO%HW69N7PKXN<RC`%:J9/S?N\\6#`U@W[4@ETUN.,LLQ,0)\"T?4:\n> M8#G'PN`4$Q4)YMFZUD[1[\".5D/>Q\\^FJGQH3W7G4H?3,&O%2JWM2.)_%T&(;\n> MHD$M';$6;AE+*7.X46A*H[`O+`=M*4=-VC8M:IK^`VJ^:Z3A/I)U*4-S8^^H\n> M!,/S1?WWO3DI`,PP,'%@&KBMQ*;HGK(.4RNWI'I-NLUEY;P8J<TQV1+J09LT\n> M((%I@X^,:R=P<.F\\O7#KM2E@.$$QG\\S/1NHF:VE\\+$=UMT>[(VP,(8D\"5:Y`\n> M[VV]BG@$@?X;[=EOBD_D>?]^4W9?^TU9D3;ZE1A=Y_7V.*F+_PLLN.KPE\"RZ\n> M[2.R(T]2TZ,6NH@/+T8?:\\RS=XA7_/U\\\\1'CC&2B5`4<O(EZ(\\U>GRXWM%Y5\n> M/D\"A:C,[Y*AM8#!1GJ>!*F2/G*.WD4'RHVEH0FQ420U+';JOUL_?#?>SL\"C'\n> MF/UC';IW)])%R1RH1@7U4'0)U:-.]47U*/=2Z0NN1:CRLQ`]RK=I3J(<GCO(\n> M!FU(\\M%D^5K^^]W7OW,_4.W1`D%+1%DSY>?YZ__K_9L__^'[5Z__@XF%CQ)+\n> M+#*><A?))J-S*'.@ABE%O^F+JB';PQ'LC`?DLK>*M@8'!(?Z$NAPS#?.23_R\n> M&66(P,1;HP2E)O##9!I++2W5=]5&C/2F82BY@Y8F@=^<)D%JB2(QV\"(_[>BO\n> M3XI2Z*.JA=-4VB*M40/P]:3*G181'!RY=R=/?H^]%C^&JE;0A.08_A^_UT$,\n> M3&O8/XC??]\\91*<4=`<QH!:B@VC+O@ZB+>$<Q!]7]256\"K?D^.C7,FIS5<V6\n> MS7:RQ1)?/^-PV.J/8IYSNU<U&\"@ANK/2>$F#]@FU>GAQ195&[!TMIW3ER(Q[\n> M+:2<#=D))@,7;6Q?$H89&&3BDFFSNES2K/OLK*S[;16G1A,ZZ@W7KB>_UP9^\n> MSV'B3#9EX!]*85W@R00Q/9&/>.C*.*69JW5KBIH#>?B61;\\TG]&C3<U--^1Z\n> MM_W;NQ,4:[EUD991]:.\"7)6L=+8:GX\\N+\\6J>]=.\\QCV*ZIX\\GL4Y>2HK=C*\n> M`;9V>H5D9@H)M`9U,JS]R-/;!BV#LWBF]BL%JPV?+;V/O&KF/%TV/%Q'P38<\n> M7_GD-HNGWK5SX4K*74ET>ECV+9[\\`N.#U+JNVSC;XHRT[VN=E%_3#&XHHX2P\n> MU7<UG\\XLY42GNODS.*W-^J<&\\ZXC&W0.-\\SRU<[Z#)`DRE6NU?O]'M5]^CU2\n> MY_=(3K`7Z=?E;9[GBNP)A\"4YCYLE.8YL258>S\\GCY\\>8/(X\"]8&T<ZWVYVI(\n> M[K/28W*KZ4'WLTX1*4`7=0IAJWMY^=[_;+X2^Z0>(#@7H&R\\7UZ^DU\\>*X/)\n> MSS3;NM,FCLQ!(@S`:3!C]O:LP^?'-.$NWW?R%5M,-E-2J$K;I02&096#WSB`\n> MH@\\X\\=PZRK-?59S*C%S3:S2=KNKI\"%N'6>5=)8_41SWC=AQW9,U@<,*ZI>)B\n> M\\]3IL%I$X8KX[OE?7KW^-_F-1X,GNN9.F^33X5^_>_G32Z?Z:G95+&`PY-ZU\n> M`IUJ6_!*!YY/I[`^OJ;:3]U#N=,Y490]3?EX/A[_22K9+N@QQ0O94.?MPF@[\n> M,%>)$M<<(^R!OI>5JR$P%4)J^&Z$;=;]]=2NGGF[FL`T#JM&50P:':^7GYPQ\n> M/.529O(0AH&2)S`E9`]I3!U>BZ+Y-0X\"<4EG#:*8!#NF4TT3[*#E1->0Z\"@!\n> MV\\F7?%_%>TXOH/Z?6Z:RZB<M]\\*$&HR?',TOC]MN#59@KI'`]);?O#-BGY7L\n> M_23*;ZT::YZQDKEWM5+GABHVC^B>_V2N4--\"^;LI+.I!,[5#7:>F67G;M:WG\n> M-37Z=JLS@?XFW6UN>V(WG:9Z1Y*K5M4*K;;C'*/0\\X.WIU;[email protected][M:VY-Y\n> MOJDOUG3?_XI=S,8-!)]NRW/PH9[.%V;S>LNTY0*>>/^0:\"!!T]8M'_!\\N.7X\n> M=6I)JV_.`RPZ\"3)L^WU-*W$9;*OB:[/!K=OJ(_!IVM[<^;!QX:(\\4VR:M.8I\n> MMNH;]_\"NTV#@G0;Z)MJ%]R^:<K&J-U>KA4@WW4J2*K[2+8'$#MULJP*PGM)`\n> M_TWVF.KI/\\%4U^,KZ9[3*^9(NQSHZ96T.747?!J:\\^O$=H5M?_^%9N'A%+C?\n> M[,@C%7VWI6P'0*#A!=-A/5?'5+,1T]E_=CN/]..U?Y@LS\\^Y:0O'50!G^5QJ\n> MY+93UMG7@0_$9;33>9`>9]A:N^%,QMP=.]RL='?R;+G8C.;TUG\";[[P>4?C0\n> M68_3-@O.#'E;#*?4<EM]FC8M0!-5)7H[XXZG&<%S+M.75QNWVZ,+YP?L([KI\n> MV-X:)&V%_3)S7:\"R7Y?SL:HQ'1>E<U!*7Q[3#,TJ\\.[WZCA^]&A['4\"AIAO]\n> MXR03EE,MYWL=S$>B<GV%SZ:4>#\\(BF^<'<BG7HQ_P+?3:\"U3D]Z@%_R1M5)K\n> MXA06$J5U+M9OJN9B[GB]I2VEN\"%W+@,D=C,Z])05.OUD_PK(U&PR'9%[EL(B\n> M=3J*SFOVN7%W@`\"F?G0HP,,8C@\"FDB@%W);,I.T:T*V-S0I6@\\:[A^\"BU_F'\n> MM^]?_N>//ZEOS\"DF=(X%II50!W0$G#:'-K@CJ`NMV\\S6-0\".+E-B'M$4'W-1\n> M<XNR&DZFJ#@WBKI&=L@(B5FFB=-D_+J^E1\"+.[OWE39<%1E/471,/2B_Z8MN\n> MLZCZQ?>\\#)Q+T'0*;[<((SYF^XV8LQ8MU8NBAKIJ$6:HJZ>C,S_<0BYYX+*;\n> M['.HH1Q50#I.DG9'3'WXS5Y,>_\"L$YCCX3=.+U,1?M/C@_0>>`$\\'K2/I)U<\n> M!@!*89J_R'#@%!\"VJ2X'L/_=9SWR;#MV[N2@/Y^&&9?@XH-CD1UAQPU\\ECOE\n> M*4#:W_\"L3#M[!)/A=0VIM^:9S=7RK%ZO&Q^W'I5EU84[FVS'LP<9\\2S3S!UP\n> M:C?63G\"-[<!Z\\,$)Q78SI[Z9JO>)W9Q*2:[,4A8E\\_)P!8'=(TO\"N0ST^)K^\n> M(9;1.OQ%[PDZ4,,I@BZ))!@O%U]*?9`3C4=RINYB=]3)GP^S!9_G^X>72PZ9\n> MV4E.:6H:8-2BYJ7'>^3_FH-HPN5K/1&SO]-^^]/.KKM#D%@`5`7B.@FZZ5&&\n> M0]NY3'(F=K1\\?7Q2%DE@6A%_^'GXB.1][[;*)MZ6:9]T0*VF'ND28=J1\\X(Z\n> M]8<_J2-F9_DPM<A+3VW0I+61[N6HN5[^L7.Y!K'MLD#_3?^ISI1,-Z@ZATC\\\n> M@7BZ4A[]@AZI.:2N$]HNKZ`,M<[EM:QD.]<]G?]:^WLM[='0G5#5'AXU_#=I\n> M,X?C#*<LV,%J3$SL&ZE+Y9$[Y_\"^/0UKMWVG;A,,C&V(?TU_<;LI&)C0'5WY\n> M!RU.U1@Z]I)OEU)!]0BU+-4#HH(^=*=(M\"!5&[BX-7-PZCS%5XMF)M6F7\\W]\n> MMJ;Y$U3;^,5(\\O5)@:-&JCML]=LT`66^6C>C<\",\"EUJ\\Z:ZJA'7-^9Y>JXR2\n> M-><7/<C`?]9_\"_]+>VO:@UJO7F_@RY;>P,J]XJ#'-YSB$'H'!<W)%R/>PEG5\n> M9U>KM?#)^;5Y4Z=^0]SY:*E</.66HBD/W95STJR<]%3HU/9+)[C%>S_HL5'%\n> MH;-:4IT\"JH4J#<'PD-1I<D&?!8DCQ$.#=@'W>T?D3(>=LQ[L@&H9O]D+SW;8\n> M<*CG3)FL==C#**WZA)\\H+;,L,-_&KEF6X[]`_\\T:LTPFA4B57X9OKBXN1GI1\n> M35D5`N%'O:%#6PQ'76<\\U.TW[+</]@UY70N+QN75^6CEU&&Q,BXNZA6N+F+T\n> M?OKF#W]\\LU9+W5V,$\"'UGZQ:6?S3<G4^QB4(O77$(Y\\R/[\"\"S-T1WA,[O[Q1\n> M?=`V-[UA9H<)_O1]%=,;OQDMQJ/5V'G9?S&/Q_EH,;T2\"CWFLJK&UYHML9U&\n> MF4C\"E\"MPWMEP/-J,/HS6:/'H#+,5FEJ[834-Q-G5E+KJQ?(*]P1L3P&9<16)\n> MF7AX$#;I\\+GT#_^:L-3_GU\\897&#S\":JNXI9,3$OVHWM>JBGX_SB\\KS&%XH1\n> M8;B-[D@)!8P4>D\"X-;#P7IU]5(WZ$QP14@:<\"U(J*]?Y/A'-\\DKX[!E5NPB1\n> M>K=M:U7(X/V'X3BC+_'$G5I5<_,3^4>WMEI-Y743D0XW%JJ7*SBXW8L<W0L>\n> MC8;)&F7<-[7M[;#ZB>E_E\\LUK\\>=:B]5E+`NT-<W;F2^(<?D;(/1R1Q/G]K;\n> M-A?+C;\\ELE'-I[4K-/PT6YYS_WBY^HA+2%+57WD-@6V@6'IU$J.WTO#5ICFT\n> MZQOD>:1U.P-'\">0GJ4XR!+4?K,<G'Z6X,7H\\/].R_&E>JL(Z'J-SSENV0:4A\n> M[S,ZYG+]D/^;6]TD\\H+NJ.:6\\=08K<-;[@#S:'[.(YY3=\\#Y;'EUCMDXJ\\_A\n> MKAQI]6JDR8R:KD:0&;AK\"4T\"!Z^Y$ZC>A4\\C/8<N),0%AS=+/>#^2CZ.Z[-Y\n> MRSZ<+\\[.K[BBM7V&'_5H_'D]QAT!R2?-;JZW<O:K@<A3'JI,8<AHE^N=\"$>*\n> M]?)J=<8C7W:HS]QY&VVI(R+F_&QYJ048\">:8!LVE856XA<LG5^=P(H%OV!1=\n> M)43BX188V5//84,*0\\@*4Y$FRN2LO5[,1HLSG*QN#T>M8B>AV!F/YWH5#1><\n> M2E+/DXMB$B)U)I)`1T9-W/%\\?7:UIDI/Q9,'_FVD.3[H@EYJNVS-G$XKGJ!Z\n> M/8I-PCINY1;#^?RCG1#?K.;3*6\\A!ZV3[5@14!,E@EZ65B+I^MA<S#WCR;L/\n> M=A]<1`ROB]OJ^&(VNH15$Z/^\",.D_:K7GD7D_T6OD'DSXB&U=\\.+$>S!$;_P\n> M.A3.\\D]7O(<W4PF_JO7\\N62IF[6\"5[_5R+\\0S6&JTY@=6%^O*42@>.FRR+T^\n> M6='43^VN6=M`B)[8IJU>ZP?Y[;Z@>@%<YV@P0>(WMQ:<$18,:[N=)2WW%S-^\n> MG>OQ?BO7J.M.QXNR<>F&.1B.QKBT(I-?3R4VP^NN+M*K\\%I7:%W@UDNWJ$T.\n> M\\@/3*<H!-$#'&6[K_%)>=''CD,Q/(;[M@MP%SH)?NMDZ7XB(FV\\X\">R\"F%[0\n> M6-<K'1%U=8QKBA8I45=W=!=K=\"<3[LKY+W2+>N9T'QV#?C1WP9EZBKG2<>T@\n> M)4:0XFM=<LR7XD:+\\SK9.UI^L*9[!BN8ZB3<$LS3EF\">Z'AVY3)GJ]%S;5@*\n> MZLK1R_>R_DRM7J0D;5K2?MU(<G7,X:YFT$98P*S`O:SFXH\\)%31@XH$91JM5\n> MTTT<^P[4E<T&T9WDKXNLC#_0,D=X=_W#1@DRBW-*5_]W=C>JUKM1./1(T^)*\n> M'19ZY-8LL*TE7R<!ZC9*K]V5O<<F^SF(W-=B\"T637-2KITN9FQ?S_ZY7S070\n> MUJ5)<H%8.1N%DM@8\\9IKP.XRMJP*5SJ[0'V<FN=!L?.E*73@?\\.&`#]!F#PG\n> MA`?IIN=V.#XMW5+OQ-$[7T-Y'>D-)@4'V-(?VT1PROU\\O5:5KC4O]2;2KKJG\n> MS?`W2:5/XX[6]K3M\"=2;D(Z3FDD^M?5(2X8VVSE6ZBY;X]\"JKL(3KL*XZRVJ\n> M59)EL+C%XMFQA0K\\%^B_>>-@<;;0'T0QG$,%N9R=++CFEHU-]+=OY)MSM;RL\n> M\\O\\ZJ3!U7YPJJ?_]5-KUC1[^ULN3P^_TT^I3O7AV\\A'4>BY*V;]=S<=VJJAE\n> MK6`Y55-%:AD[+I_IW;VKE917!\\.7NG0_/@$$E5I,6,[7R\\63OYZ<RSPZ=ZL@\n> M(*G9_\"CD><N_B?U?EO\\EJP\\6TG^O>6-KI:FY(GV#NU)_/C^7E>J9\"K\\?5S((\n> M$*QK90?AO&_<TN:X]3\\62Z],/9'__J`VTAM;WJP7TEK<C$)S_R+T@_IP+LJ1\n> M5AQA+K]87EQ>T=TX?',VKR&\\9SBMMX;>+MTHT9<@TGOQ?B`&'(BT&0A>ZO4#\n> M(29+KC;)<[OO^VK!/;+E^.K,V)RZHN_6M&EYP)9_;!ILI4>0`KD,Q[@[&HQ.\n> ML6<X`E*8.:LJU2%1!\\C?OO_AQ7]4,B:E7DI>7HA2SCJ^7YY]G`G3HP,!U[@?\n> MI3A*<>SP@5N8[D=V9LKKH5\":OY17S'+E,.O\"3*W!<7OO&RO\"$VVVVC_.UL>`\n> M=-1_49_8Y(RJZ-_>O)6Z?GA=5D)SK+9_%\"$[JL]U(_+-1NR0#ZO11QG'L:R+\n> MIR<_B_)]^HSWO=&+'Q::\\\"<LJX\\YPN@`L4;0,7Y[,3J3.F:8^F-;@_XBO9_S\n> MRNPGO3&O3<]H-A'TB`W?QWPR.<,\\\"KQ(*+Q(>+L2TZ1>!:)S!5`.QO.SS>\"I\n> MO#V9+9<?!V(I?%K\\P[W^-I\\`E^F';X-_>?C[__W?Y6AS-GOZT\\OGW_SQY?]4\n> M'5$8(B3POPP&@Z*(^&]D[WB,((@'.4+>A/)[.(!K)0W_91#^3S6H_7<%)\\-@\n> M\\\"\\.B.=P.EFJ_QD-^N?^?3?_(@C\\KLA`A,Q`=T0&\\_5@]*NY+`:B'PT<A7XY\n> M?_)K?II*/M%1!V2@`3?0UIL!?D!.9T,/1IN!R-F!\"(_!<L)'\\R0.-O7?-_`4\n> M#*ZN9)%;(C&V`[XXD?8,<.BX*039D!Q%7%P/_C@2J;;Z<CV0^M?S]>D`RO(`\n> ML!5_'Z@2M'X<H\"<CMEF-Q<&O,GY8Z$X'KP:SY:64**KAX'IY-9C,)2TL,X`Y\n> M3%')?#-0'P2+^00WS>!#/1`SY?)Z,)\\,J(990N:6%LY7@T\\C^76Q6;+!C:D1\n> M=):1P>5J^3-N\\2O]</H$YH<CCTL[$$7P63`8/#E!-P?THP_.U%`:7.`8*5HD\n> MF63]`+WE<RTI1FNT\"?M,4J(H^C*B@X#S;/DK#:9/\\GD`&S5H\"A<5=\"F=<*5K\n> M[M&@96<CT\\65#+14V'3LR[46[73A`5T7(-U*,@MA,':P)`>P#,E<L^6G@1!(\n> MD@SH2?1Y3UKMD7%8-<VA%W`]@\"*)WC3]Z#9EX.P^;9.WTC3C1,@*BI#&:M=)\n> M#VF`R;\"NT:;+D1\"2]O1`3:W'6M*.5?-X4#O[Y13M79!6&ZRAXR[1.*1:2M>J\n> M<*.MITT':C\"00/CJ;Z8_U:OJ@_/E='Z&>:5E<1M1!O`'L<#G]%X)3\\,W)_5+\n> MV3()DU/1`JXV@P6VC@8$2AK`H2X3<7DUG;&[>@&$'(YI@MF\"N7PZ>\"UC)%/P\n> M\\O)\\;I//K\"9DXPR7?Z?UIFDG\"[&VFH]0R8[,\"QDKFWRG`S/U!XG-Y<[4\",:U\n> M3/PFIZNV)5VD8X]EK@X^UO6ESF\"1(!>8@K\"S@S-1GD>`/F-^/TU9F(@8Z32X\n> MA)URG+%Q0JP>.^D12!;T4TB,NTZ8BI0/`]P97'WT?&3T>COK)*(466OC<&LB\n> M6&_$D@7-/F!Z+VJ.C`G&BR4G^RO\\\\*6FUX&2Y&,*DHO@XGI=GT]$S+WZ\\D(^\n> MHG=\"N8OY^5QFQK7,7ID/9W6WQ)/!F^5*?H3@1DN\"9T^&02#SZQ7F'D2C(5^)\n> M8#V;\\7\"FI:T/L*H.[]-FS(41E2_I7W#EK9E8)Y;T830P&\"FB2`T`(/4KH:,L\n> MB6)^H\"P16\\+@BA(R,!GJVJC^C`',]`'F_OKQ@/XHRI63QVZ*/N;\\<).3/VJ?\n> M_9S@2D*)9+T:;3:KN8R'??FT'.A>Y:#9-$&9(A\\7=(P//EPS'T\\4\"!=>\"[N+\n> M5!NQR1=<\"T7=^5^G5F?:U(AB*)$!C6C<,!EA%$7_4+$SA]@75IZ<+V4PST38\n> M29J+^=_%OC\\=O)'ED,S%DDQL#L;S\\6\"QW$BGYS)!,`<ZHF!'D@B5D7V\\E`XC\n> M'Y)*[V1)/MN<*[ML#_3`(+X&?_WNY>O!G]^\\_&;PZC4*>?'#'__PZO5SG+48\n> M_/75V^^4*&^^L-YP105&+&GP:;7$K!1*8)C)@X[?4);'L1@0&1V3\"FK$>F!`\n> M)X,KX:%S:?*G8V.(TC9^Q=P>39UA.#+4;<-._2CFKR2GH\\,:8_/4$VC-&<]\"\n> M1`2)X/D@HI@3AF*%<(-K1V]=L&<\\2O=81<,-;;1^+4V\\ZU1!,<@O2Q.F^FC!\n> MB]\\0W32]T8AFYNF4Q+X*6$AG1#VFJ)*NCL0L)Q.Q\\TIGU;&<*.C09E<>(/,.\n> MFVB=Y\"RWLMK\\%9U.,KS^X>W+9X-7G.;R//BP6GX4^E*U!\"OKE'+,QF4)G[UR\n> M^M044UT#\\6=IO_CB\"WQ3Z693<30>&X7K\\_&@.;M@*ZVDH*R&2[=1%71]A71V\n> MTTJ5.Y/'G@ZZC.#4_-H3\":L<*SM%&>1>DD)[6H\\^JG97?[@2Z8TC+:R'$U;F\n> M^3[AZJ9),ZJ:>WFUN416&US.`#V+9O06?5(SJ/03=1GG2=J+./)P=U%$CPI2\n> M=./-R^^=#?#J]9N7/[W=$H+-U/&^TP&0]`8\\VP`=5G@0Q9Y>LY\"NVBLU4/*(\n> MX!)-:7V\"Q\"([A855%M+.,-60,L^S`I5*Y!U:\"T]/3XU;VE^,/?3+R4G#./9)\n> MTGPU($4XK[E'H<489SUM<L`L((%%S7EFS/OJA;!OC6M.SP:_^X;+-3PCL`=X\n> M1HG3%O+M[W__.V4#T_Z._/#*<Q2)Z%D*;*HZ\"]2CP?KJ$N<..!HCZ`QD+\"#I\n> M:A6/'7\\J+TU&(K.$=G`\\B^;BY(:[SJ$-&(@4AKZPPMJ[TM'<:M,8K<64N(+?\n> M>:TRX0(C/*V_\")2?1YLOUR8I(3Z$<Z_M%38A6[^!`8*F\\P:+TF\\B8_V,AAH7\n> M>2M496?7R#/YQ;RGPN,RF\\Y1^Q.H,A?UAZ7PMW00(V-P?4WZ]Z8`#$^:SE-7\n> MD!68C/-*J'SIM-GF1C(Z+1/WO;ZH^BVR]9,8!$(IL33&^*CKDU1QINO_2/3U\n> M>K2XNMP1U*V[SJ?::6DKMCLY?V208*2JK3G0W0+/Y$Y&-\\HN]!RHNJ]<?TU+\n> MPIZKICK=YBO'5C7Q]%0VC.>3B>BGHCQ@,=U\\POPUJP3JCFGF*H[(SJH4K`?/\n> MGE!LOU*:8VL=15,BRO#(*K:Z-A%[7KMN/$6KWVO73M>_G$,'(<_*8M54!4(K\n> MU]>MQ7<]$)I)BQQ%/G&*B;4,&$<ANNIAU%0LAZZTUV!]9V\"8CNNYBJ1V)<Y;\n> M:[email protected]#A@)_?FUK&+N]I3D]Y8+>+$_2='=\\!F3[@O;K0*_)2&.Q'+C=[\\'X\n> MZN)\"1:*>\"I,4<Q$KR/\"_#>SJ]`=-I=\\NT2+_SH>O?S]X-_=DME/NPBCUBNUO\n> M%L7!X/3FESW%=4@1!,&;I2HIL*IEI.DND1Y0673KCCI(9(+B]NEJO&9W!F\\V\n> M(I46#Z[AA[^'OX>_A[^'OX>_A[^'OX>_A[^'OX>_A[^'OX>_A[^'OX>_A[^'\n> MOX>_A[^'OX>_A[^'OX>_A[^'OX>_A[^'OX>_A[^'OX>_A[^'OX>_A[__J;__\n> )!VTKBA$`N`L`\n> `\n> end\n> ------------------------------cut here-----------------------------------\n> \n> -- \n> +------------------------------------------------------------------------+\n> + Simkovics Stefan +\n> + Student an der TU Wien (Informatik) +\n> + Tel.: 02239/3367 +\n> + email: [email protected] | [email protected] +\n> +------------------------------------------------------------------------+\n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 17 Jan 1999 19:07:36 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: EXCEPT/INTERSECT for v6.4"
}
] |
[
{
"msg_contents": "I would prefer, for simplicities sake, only handling four digit\nyear dates. Handling the two digit year dates is what caused this\nwhole mess in the first place.\n\nD.\n\n\n\n-----Original Message-----\nFrom: Thomas G. Lockhart [mailto:[email protected]]\nSent: Friday, January 08, 1999 11:43 AM\nTo: Massimo Dal Zotto\nCc: PostgreSQL Hackers\nSubject: Re: [HACKERS] postgres and year 2000\n\n\n> it seems that the year handling in pgsql dates is not very consistent:\n> The problem I see is that the same number is converted to a different \n> year depending on the number of digits and the number itself. I think \n> that this kind of things are the most likely sources of Y2K troubles.\n> A more consistent approach would be to treat the year literally and \n> let any smart hack with dates entirely to the user under his \n> responsability.\n\nI agree that there are some cases in your examples which should be\ngiving a different result. Not *every* example you gave led to an\nincorrect or unpredictable result, so could you please identify those\nyou feel are a problem? In glancing at the examples, the ones with zero\nvalue (but lots of zeros) seem to be a problem, and perhaps some or all\nof the ones with an odd number of year digits. Any others?\n\nWe do need to handle two-digit years, and we currently do so using 1970\nas the break point. I've read recently that some industries are using a\n\"50/50 rule\" for 2 digit conversions, where 1950 is the break point.\nDon't know if we should try to use that (rather than the \"Unix rule\" :),\nsince it really doesn't offer a magic cure for all date possibilities.\n\n> Only then we could declare pgsql as full Y2K compliant.\n\nfwiw, the date candidates which are failing are outside the range of\nnormal usage or could be considered mal-formed. But I should be able to\nget a fix in for it, and can post patches. Let me know what cases you\nwould like tested and fixed (but let's not bog down in discussion on\ntwo-digit issues).\n\n - Tom\n",
"msg_date": "Fri, 8 Jan 1999 12:06:47 -0500 ",
"msg_from": "Dan Gowin <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] postgres and year 2000"
}
] |
[
{
"msg_contents": "you're all wrong. 6.4 was late because of me. i didn't know the postgresql\ndevelopment process or political process, and that combined with some real\nlife issues made me unresponsive and/or unavailable when d'arcy needed me.\nyou all have my apologies, and i promise that i will do better next time.\n",
"msg_date": "Fri, 08 Jan 1999 09:19:32 -0800",
"msg_from": "Paul A Vixie <[email protected]>",
"msg_from_op": true,
"msg_subject": "6.4 lateness and CIDR"
},
{
"msg_contents": "\nUmmm...why are we rehashing this? Who/why is anyone tryign to assign\nblame? It happened...so? Let's avoid it happening again, but I think we\nare up to, what, 4 martyrs now? :)\n\n\n\n\nOn Fri, 8 Jan 1999, Paul A Vixie wrote:\n\n> you're all wrong. 6.4 was late because of me. i didn't know the postgresql\n> development process or political process, and that combined with some real\n> life issues made me unresponsive and/or unavailable when d'arcy needed me.\n> you all have my apologies, and i promise that i will do better next time.\n> \n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Fri, 8 Jan 1999 13:54:12 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "The Hermit Hacker wrote:\n> \n> Ummm...why are we rehashing this? Who/why is anyone tryign to assign\n> blame? It happened...so? Let's avoid it happening again, but I think we\n> are up to, what, 4 martyrs now? :)\n\nNO !!! NO !!! I AM NOT AGREE !!! THEY MUST BE BAAAaaaDLY PUNISHED !!!\n\n:-)\n\nAs a punishment they have to solve that damn problem of missing -lcrypt\nin libpgtcl Makefile on RedHat Linux 5.x i386 !!!\n\nI didn't count the number of messages that I received from PgAccess\nusers claiming that PgAccess + libpgtcl does not recognize the crypt\nlibrary! I even made a crypt.txt file that I am attaching to all my\nmessages regarding this problem :-)\n\n-- \nConstantin Teodorescu\nFLEX Consulting Braila, ROMANIA\n",
"msg_date": "Fri, 08 Jan 1999 20:15:45 +0200",
"msg_from": "Constantin Teodorescu <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "> \n> Ummm...why are we rehashing this? Who/why is anyone tryign to assign\n> blame? It happened...so? Let's avoid it happening again, but I think we\n> are up to, what, 4 martyrs now? :)\n\nWe are all falling on our swords. Better than everyone blaming everyone\nelse.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Fri, 8 Jan 1999 15:12:52 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "> Ummm...why are we rehashing this? Who/why is anyone tryign to assign\n> blame? It happened...so? Let's avoid it happening again, but I think \n> we are up to, what, 4 martyrs now? :)\n\nYeah, we're all goin' down before this is over :)\n\nI'm not sure how this came up again (besides Bruce wading through 3\nmonth old mail) but this is all old news. Next release won't be the\nsame, and in hindsight *every* release could be done better.\n\n - Tom\n",
"msg_date": "Sat, 09 Jan 1999 02:41:04 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "On Sat, 9 Jan 1999, Thomas G. Lockhart wrote:\n\n> > Ummm...why are we rehashing this? Who/why is anyone tryign to assign\n> > blame? It happened...so? Let's avoid it happening again, but I think \n> > we are up to, what, 4 martyrs now? :)\n> \n> Yeah, we're all goin' down before this is over :)\n> \n> I'm not sure how this came up again (besides Bruce wading through 3\n> month old mail) but this is all old news. Next release won't be the\n> same, and in hindsight *every* release could be done better.\n\nThat was kinda the way I was thinking of it, which is why I was sooooo\nconfused that we were having to many ppl volunteering to 'fall upon their\ntrusty swords' :)\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Fri, 8 Jan 1999 23:12:49 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "\nOn 09-Jan-99 The Hermit Hacker wrote:\n> On Sat, 9 Jan 1999, Thomas G. Lockhart wrote:\n> \n>> > Ummm...why are we rehashing this? Who/why is anyone tryign to assign\n>> > blame? It happened...so? Let's avoid it happening again, but I think \n>> > we are up to, what, 4 martyrs now? :)\n>> \n>> Yeah, we're all goin' down before this is over :)\n>> \n>> I'm not sure how this came up again (besides Bruce wading through 3\n>> month old mail) but this is all old news. Next release won't be the\n>> same, and in hindsight *every* release could be done better.\n> \n> That was kinda the way I was thinking of it, which is why I was sooooo\n> confused that we were having to many ppl volunteering to 'fall upon their\n> trusty swords' :)\n\nIf it's any consolation, Marc, I was considering nominating you :)\n\nVince.\n-- \n==========================================================================\nVince Vielhaber -- KA8CSH email: [email protected] flame-mail: /dev/null\n # include <std/disclaimers.h> TEAM-OS2\n Online Searchable Campground Listings http://www.camping-usa.com\n \"There is no outfit less entitled to lecture me about bloat\n than the federal government\" -- Tony Snow\n==========================================================================\n\n\n",
"msg_date": "Fri, 08 Jan 1999 22:37:02 -0500 (EST)",
"msg_from": "Vince Vielhaber <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "On Fri, 8 Jan 1999, Vince Vielhaber wrote:\n\n> \n> On 09-Jan-99 The Hermit Hacker wrote:\n> > On Sat, 9 Jan 1999, Thomas G. Lockhart wrote:\n> > \n> >> > Ummm...why are we rehashing this? Who/why is anyone tryign to assign\n> >> > blame? It happened...so? Let's avoid it happening again, but I think \n> >> > we are up to, what, 4 martyrs now? :)\n> >> \n> >> Yeah, we're all goin' down before this is over :)\n> >> \n> >> I'm not sure how this came up again (besides Bruce wading through 3\n> >> month old mail) but this is all old news. Next release won't be the\n> >> same, and in hindsight *every* release could be done better.\n> > \n> > That was kinda the way I was thinking of it, which is why I was sooooo\n> > confused that we were having to many ppl volunteering to 'fall upon their\n> > trusty swords' :)\n> \n> If it's any consolation, Marc, I was considering nominating you :)\n\n\tD'oh!\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Sat, 9 Jan 1999 05:29:13 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4 lateness and CIDR"
},
{
"msg_contents": "I was wondering if any one is responsible for making \nthe RPM for postgresql? \"rpm\" could be added as a \nmake target and the sources required to build the\nrpm could be added to the CVS repository. In this\nway those people who only want binaries could be\nbetter supported. \n\nI don't have a ton of time, but I could do it if\nthere arn't any other heads available. I'm currently\nworking on SGML integration and would rather spend\nmy time there... \n\nClark\n",
"msg_date": "Wed, 13 Jan 1999 04:05:43 +0000",
"msg_from": "Clark Evans <[email protected]>",
"msg_from_op": false,
"msg_subject": "RPM maintainer?"
},
{
"msg_contents": "> I was wondering if any one is responsible for making\n> the RPM for postgresql? \"rpm\" could be added as a\n> make target and the sources required to build the\n> rpm could be added to the CVS repository. In this\n> way those people who only want binaries could be\n> better supported.\n\nThat would be great. The RedHat folks are swamped with maintaining\n*lots* of packages, and I'm sure it would help if we could do the\nmaintenance of the rpm for them.\n\nNobody else has stepped forward for that yet, though it has been a need\nfor a while. You're welcome to it ;)\n\nPerhaps once the rpm build is codified, it would be easier for others to\nhelp. I've got the rpm book, but have not waded through enough to figure\nout how to build a from-scratch installation.\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 06:04:15 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RPM maintainer?"
},
{
"msg_contents": " That would be great. The RedHat folks are swamped with maintaining\n *lots* of packages, and I'm sure it would help if we could do the\n maintenance of the rpm for them.\n\nDoes this mean that we should consider maintaining other packaging\nsystems, e.g., the pkg systems used by FreeBSD/NetBSD/... ? I have no\nidea if there is interest among the \"normal\" maintainers of those\nthings, but I suspect that like all of us they have too many things\ngoing on also.\n\nCheers,\nBrook\n",
"msg_date": "Wed, 13 Jan 1999 08:20:15 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RPM maintainer?"
},
{
"msg_contents": "> Does this mean that we should consider maintaining other packaging\n> systems, e.g., the pkg systems used by FreeBSD/NetBSD/... ? I have no\n> idea if there is interest among the \"normal\" maintainers of those\n> things, but I suspect that like all of us they have too many things\n> going on also.\n\nI'm not certain about all of them, but for example someone (Oliver?)\nmaintains a Debian package for Postgres. Distribution packages are\nalways a good idea and helpful to some folks.\n\nWe have no normal maintainers in my experience. However, anyone\ninterested in maintaining packages would be welcome to join us as\nanother abnormal maintainer :)\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 15:42:56 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RPM maintainer?"
}
] |
[
{
"msg_contents": "Does this seem right?\n\ndruid=> SELECT COUNT(*) FROM acctrans;\ncount\n-----\n 0\n(1 row)\n\ndruid=> SELECT client_id, SUM(tramount) FROM acctrans GROUP BY client_id;\nclient_id|sum\n---------+---\n | \n(1 row)\n\nIf there are no rows in the table then shouldn't the result be no rows\nwhen GROUP BY is used? Further, What about this?\n\ndruid=> SELECT SUM(tramount) FROM acctrans;\nsum\n---\n \n(1 row)\n\nShouldn't that be 0.00?\n\nWhat will the NUMERIC or DECIMAL types do in these situations? It\nlooks like INTEGER has the same behaviour as MONEY (which tramount is.)\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Fri, 8 Jan 1999 12:40:23 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "SUM() and GROUP BY"
}
] |
[
{
"msg_contents": "> Thanks for the info!!! I hope the dentist went well!!! \nIt went well, thank you for asking. I got a new dental assistant, and\nshe was gentle and sweet (a far improvement). She said that I have good\nteeth, and that I must have been taking good care of them because there\nwasn't much for her to do.\n\n> I have only one question more question(I need to sell the \n> Microsoft guy in the office on this.) I go to Microsoft and \n> they say 20,433 transactions per minute on a quad xenon 450 \n> bla, bla, bla. Is there a place where this statistic can be \n> found for Postgre.\n\nI don't know of any such statistics, but maybe some of the other guys\nwould.\n\t-DEJ\n",
"msg_date": "Fri, 8 Jan 1999 12:20:29 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>] Question s?"
}
] |
[
{
"msg_contents": "\n\n> -----Original Message-----\n> > I could be wrong (I don't have a copy of the standard), but I don't\n> > believe that the above syntax follows the standard. Let me know if \n> > I'm wrong, but my understanding of the syntax would be more like:\n> > SELECT * FROM t1 JOIN t2 ON (t1.i = t2.i);\n> > with the same result set as you listed (t2.i isn't suppressed).\n> > This would have a difference in approach from the above. \n> If I wanted \n> > to join on columns with different names I couldn't use your \n> syntax (as \n> > one example).\n> \n> The standard allows both syntaxes; USING is simpler to type, and ON is\n> more general, as you point out.\n> \n> In fact, the standard is annoyingly helpful in allowing \n> multiple ways to\n> write the same query. Makes the parsing and parse tree transformation\n> more complicated :(\nSorry for the extra work load, but hey that's cool.\n\n> - Tom\n\t-DEJ\n",
"msg_date": "Fri, 8 Jan 1999 12:22:28 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Outer Joins (and need CASE help)"
}
] |
[
{
"msg_contents": "I think one of the questions that, and perhaps you have, needs to be asked\nis:\n 1) what kind of transaction test produced those numbers?\n 2) what kind of user load, sim. transactions did the server have?\n 3) were the transactions non-conflicting?\n 4) were the transactions read only, read-update, insert?\n 5) and a most important question is over what time period was the test\n taken. This is where various logging, and checkpointing issues\nstart\n to surface. Additionally, some folks do not run the test for any\nextended\n period of time because of memory leakage issues, and other stability\n issues.\n\nIts very easy to design a test that produces high numbers. Its the\ndefinition of what they mean that causes the problem. Ask the MS\nguy to give you the structure of the test to get those numbers. If MS\ndoesn't release it then they are useless.\n\nAs for the TPC tests, they should have the code available. A, and B\nare next to useless.\n\nJordan\n\n----- Original Message -----\nFrom: Jackson, DeJuan <[email protected]>\nTo: G. Jayson Stangel <[email protected]>\nCc: PGSQL Hackers <[email protected]>\nSent: Friday, January 08, 1999 1:20 PM\nSubject: RE: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>] Question\ns?\n\n\n>> Thanks for the info!!! I hope the dentist went well!!!\n>It went well, thank you for asking. I got a new dental assistant, and\n>she was gentle and sweet (a far improvement). She said that I have good\n>teeth, and that I must have been taking good care of them because there\n>wasn't much for her to do.\n>\n>> I have only one question more question(I need to sell the\n>> Microsoft guy in the office on this.) I go to Microsoft and\n>> they say 20,433 transactions per minute on a quad xenon 450\n>> bla, bla, bla. Is there a place where this statistic can be\n>> found for Postgre.\n>\n>I don't know of any such statistics, but maybe some of the other guys\n>would.\n> -DEJ\n>\n\n",
"msg_date": "Fri, 8 Jan 1999 14:03:01 -0500",
"msg_from": "\"Jordan Henderson\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>] Question s?"
}
] |
[
{
"msg_contents": "Very observant of you. The postgres function system doesn't have a way\nto tell which parameter is null, so the function just returns null.\nI don't like it and we hope to get it changed for PGv6.5 (at least I\ndo).\n\nIs it on the TODO guys? Jan, do you think you'll have the time to get\nto it by beta?\n\n*** Blatant flattery start ***\nI'm still willing and looking for that backend mentor to truly get me\nstarted, and I want you, Jan and Bruce, because I want the best.\nBTW don't give me license to bug you unless you mean it.\n*** Blatant flattery end ***\n\t-DEJ\n\n> -----Original Message-----\n> I wrote a c func (ATTACHED BELOW) which takes 2 pointers to \n> type text, and\n> a bool field. When a field in a row passed to the func is \n> null, espically\n> the 'first' field, Postgresql does not return the value that my code\n> produced.\n> \n> For example: \n> First\t\tlastco\t\tnocompany\n> ------------------------------------------\n> NULL Walt Bigelow True\n> \n> the above data is contained in the database, and when passed \n> to my c func\n> like:\n> \n> SELECT lastfirst(first, lastco, nocompany) as name from \n> tbladdress where\n> agencyid = 691;\n> \n> I get:\n> name\n> ----\n> \n> (1 row)\n> \n> Not the expected output of 'Walt Bigelow, '.\n> \n> When I update that row with first = '', the result is correct, but the\n> value is no null, so it works.\n> \n> What am I missing??\n> \n> Any help is appricated,\n> Walt\n> \n> -------------\n> name.c:\n> \n> #include <string.h>\n> #include \"postgres.h\"\n> \n> #define\t\tCOMMA\t\", \"\n> \n> text *lastfirst (text *first, text *last, bool nocompany)\n> {\n> /* this function will take in the first name, and last name\n> * and based on nocompany set the output to either return \n> * 'last, first' or just the company name.\n> */\n> \n> int32\tret_size;\n> text\t*return_text;\n> \n> \tif (nocompany) {\n> \t\t\n> \t\tif (first == NULL) {\n> \t\t\t\n> \t\t\tret_size = VARSIZE(last) + sizeof(COMMA);\n> \t\t\treturn_text = (text *) palloc(ret_size);\n> \n> \t\t\tmemset(return_text, 0, ret_size);\n> \n> \t\t\tVARSIZE(return_text) = ret_size;\n> \n> \t\t\tstrncpy (VARDATA(return_text), VARDATA(last), \n> \t\t\t\tVARSIZE(last)-VARHDRSZ);\n> \t\t\t\n> \t\t\tstrncat (VARDATA(return_text), COMMA,\n> \t\t\t\tsizeof(COMMA));\n> \n> \t\t\treturn (return_text);\n> \t\t\n> \t\t} else {\n> \t\t\n> \t\t\tret_size = VARSIZE(first) + VARSIZE(last) + \n> \t\t\t\tsizeof(COMMA) - VARHDRSZ;\n> \t\t\n> \t\t\treturn_text = (text *) palloc(ret_size);\n> \n> \t\t\tmemset(return_text, 0, ret_size);\n> \n> \t\t\tVARSIZE(return_text) = ret_size;\n> \n> \t\t\tstrncpy (VARDATA(return_text), \n> VARDATA(last), VARSIZE(last)-VARHDRSZ);\n> \t\t\tstrncat (VARDATA(return_text), COMMA, \n> sizeof(COMMA));\n> \t\t\tstrncat (VARDATA(return_text), \n> VARDATA(first), VARSIZE(first)-VARHDRSZ);\n> \n> \t\t\treturn (return_text);\n> \n> \t\t}\n> \n> \t} else {\n> \t\t/* Just the company name is returned here */\n> \n> \t\tret_size = VARSIZE(last);\n> \t\treturn_text = (text *) palloc(ret_size);\n> \t\t\t\t\n> \t\tVARSIZE(return_text) = ret_size;\n> \t\tstrncpy(VARDATA(return_text), VARDATA(last),\n> \t\t\tVARSIZE(last)-VARHDRSZ);\n> \t\t\t\t\n> \t\treturn (return_text);\n> \t}\n> }\n",
"msg_date": "Fri, 8 Jan 1999 15:07:53 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [GENERAL] postgresql v6.4.2 c funcs and null text pointers..."
},
{
"msg_contents": "> Very observant of you. The postgres function system doesn't have a way\n> to tell which parameter is null, so the function just returns null.\n> I don't like it and we hope to get it changed for PGv6.5 (at least I\n> do).\n> \n> Is it on the TODO guys? Jan, do you think you'll have the time to get\n> to it by beta?\n\nAs some stupid spagetti commerical used to say, \"It's in there\":\n\n\t* redesign the function call interface to handle NULLs better(Jan)\n\nI even assigned it to Jan. :-)\n\n> \n> *** Blatant flattery start ***\n> I'm still willing and looking for that backend mentor to truly get me\n> started, and I want you, Jan and Bruce, because I want the best.\n> BTW don't give me license to bug you unless you mean it.\n> *** Blatant flattery end ***\n> \t-DEJ\n\nMy suggestion is that we set up time to talk on the PostgreSQL irc\nchannel. I will be on with Thomas tonight after 10pm EST. We can set\nup a separate time for you too.\n\nYou can poke around and ask questions, and we can both look at the code\nand discuss it. I will assume you have already read the backend\nflowchart and developers FAQ, so we can go over other items you may\nhave.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Fri, 8 Jan 1999 16:43:38 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [GENERAL] postgresql v6.4.2 c funcs and null text pointers..."
}
] |
[
{
"msg_contents": "> -----Original Message-----\n> > Ummm...why are we rehashing this? Who/why is anyone tryign \n> to assign\n> > blame? It happened...so? Let's avoid it happening again, \n> but I think we\n> > are up to, what, 4 martyrs now? :)\n> \n> We are all falling on our swords. Better than everyone \n> blaming everyone\n> else.\n\nI'll blame someone... It was DeJuan's fault. He kept sending irrelivant\ne-mail to the HACKERS list to demonstrate his obvious lack of a\npersonality and sense of humor. He should be flogged.\n",
"msg_date": "Fri, 8 Jan 1999 15:10:14 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] 6.4 lateness and CIDR"
}
] |
[
{
"msg_contents": "There has been a discussion in the HACKERS list (check the archives on\nthe website\n[http:\\\\www.postgresql.org\\]) about PostgreSQL vs MS SQL 6.5 for the\nlast 2 days or so. \nI think these ideas must come in threes, so where is that other person?\n\nI can't really give a feature comparison of MS SQL 7.0 as I've never\nused it. as for a benchmarking, although it is out of date, check\nhttp://www.tcx.se/benchmark.html. There is no direct comparison of\nPostgreSQL and MS SQL but it might give you some idea of something (hey\nI can't make heads or tails of it all, but remember a smaller bar is\nbetter).\nOh well, hope this helps somebody somehow,\n\t-DEJ\n\n> -----Original Message-----\n> Please forgive my newbieness but I am extremely new to both \n> Postgresql AND\n> Linux. I've been a consultant for 6 years working mostly on Microsoft\n> platforms (NT, MSSQL). I have developed many data-driven \n> websites using\n> WinNT/IIS/ASP/MSSQL. Now, getting sick of MS and being \n> enamored with the\n> whole Open Source movement, I wish to learn (and develop a \n> project or two)\n> on Linux/Apache/PHP/PGSQL.\n> \n> I was wondering if anyone has done any benchmarking (even roughly),\n> comparing PGSQL to Microsoft SQL Server 7.0 or even the older \n> MSSQL 6.5.\n> Specifically I just want the reassurance that the two \n> products are in the\n> same ballpark as far as performance goes. Has anyone done \n> this recently?\n> \n> Thank you,\n> Bryan\n> \n> \n",
"msg_date": "Fri, 8 Jan 1999 15:19:43 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL 7.0?"
},
{
"msg_contents": "\nOn Fri, 8 Jan 1999, Jackson, DeJuan wrote:\n\n> I can't really give a feature comparison of MS SQL 7.0 as I've never\n> used it. as for a benchmarking, although it is out of date, check\n> http://www.tcx.se/benchmark.html. There is no direct comparison of\n> PostgreSQL and MS SQL but it might give you some idea of something (hey\n> I can't make heads or tails of it all, but remember a smaller bar is\n> better).\n> Oh well, hope this helps somebody somehow,\n> \t-DEJ\n\n There appear to be significant benchmarking hurdles with MS SQL 7.0\nHasn't everybody seen Oracle's million dollar challenge to Microsoft? It\nis somewhere on www.oracle.com Basically, if you can publish benchmark\nresults that show Oracle less than X times (was it 90 times?) faster than\nMS SQL, Oracle would give you a million bucks. The trick here is that the\nlicence agreements for many DBMS prohibit the publishing of benchmarks!\nEither that, or Oracle really is 90 times faster than MS SQL!\n\n\nTom\n\n",
"msg_date": "Fri, 8 Jan 1999 23:56:32 -0800 (PST)",
"msg_from": "Tom <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft\n\tSQL 7.0?"
},
{
"msg_contents": "> There has been a discussion in the HACKERS list (check the archives on\n> the website\n> [http:\\\\www.postgresql.org\\]) about PostgreSQL vs MS SQL 6.5 for the\n> last 2 days or so. \n> I think these ideas must come in threes, so where is that other person?\n> \n> I can't really give a feature comparison of MS SQL 7.0 as I've never\n> used it. as for a benchmarking, although it is out of date, check\n> http://www.tcx.se/benchmark.html. There is no direct comparison of\n> PostgreSQL and MS SQL but it might give you some idea of something (hey\n> I can't make heads or tails of it all, but remember a smaller bar is\n> better).\n> Oh well, hope this helps somebody somehow,\n> \t-DEJ\n\nThis was hard to find, given the subject line, so here it is:\n\n\n---------------------------------------------------------------------------\n\n\nRE: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>] Questions?\n _________________________________________________________________\n \n * From: \"Jackson, DeJuan\" <[email protected]>\n * To: Magnus Hagander <[email protected]>, \"'Hal Snyder'\"\n <[email protected]>, [email protected], \"'[email protected]'\"\n <[email protected]>\n * Subject: RE: [HACKERS] [\"G. Jayson Stangel\" <[email protected]>]\n Questions?\n * Date: Thu, 7 Jan 1999 15:46:02 -0600\n _________________________________________________________________\n \nMy experience with MS SQL 6.5(my first database):\n wrote an application or managing risk in natural-gas commodities\n[PowerBuilder] (~1 yr.)\n wrote several shopping cart and invoicing applications [ASP] (~6\nmonths)\n\nMy experience with PostgreSQL v6.3 - v6.4.2(my favorite database):\n wrote several web sites for the advertising firm I now work at [PHP]\n(~1 yr.)\n\nI also have no formal database training. Therefore everything that I've\nlearned was because I had to (you always learn those best).\n\nWhat MS SQL 6.5 server has over PostgreSQL v6.4.2:\n1. row-level-locking promotable to page and table\n (therefor faster concurrent transactions)\n2. referential integrity(RI)/foreign key implementation\n (not including triggers [Postgres has those too])\n3. stored procedures that can return multiple rows\n (a design flaw in earlier versions of Postgres that hasn't yet been\ncorrected.)\n4. fully implemented dynamic precision datatypes ie NUMERIC, DECIMAL\n5. a fuller implementation of the SQL92 standard\n (only slightly)\n6. multiple database access from any query\n (to tell the truth I haven't missed this one at all)\n7. server replication\n8. automagically clustered tables\n (this could as easily be considered a drawback)\n9. service-pack patches of the server\n (don't let the fact that it's larger than the software it's patching\nbother you)\n10. all those nifty Microsoft interface programs\n (yeah right)\n\nWhat PostgreSQL v6.4.2 server has over MS SQL 6.5\n1. PRICE!!!!!\n (this includes unlimited licenses)\n2. stability\n (anything in Windows is subject to crashes)\n3. support\n (the Postgres mailing lists outdo Microsoft support by a factor of\n1000)\n4. You have the code\n (if you want something added you can do it)\n4. clustering on any index\n5. multiple PL languages for functions (a.k.a. stored procedures)\n6. type extensibility with full operators and aggregates\n (I've never even seen a MS SQL 6.5 extended type)\n7. a larger and more versatile base-type set\n (have a look at interval)\n8. database size is only as big as it needs to be\n (you don't need to set up data/log blocks in Postgres)\n9. regular expression(regex) matching of strings\n (I hope you know what regex is)\n10. functional indexes\n (you can index upper(my_col), think about it)\n11. unlimited tables in a query (limited only by your ram)\n (MS SQL 6.5 is limited to 16 including those in self joins and\nsubselects)\n12. multiple index methods\n (one method might work better for certain data distributions and\ntypes)\n13. a rule system that by itself could make up for the lack of RI\n14. a developement/release cycle that is faster than Microsoft patches\nand contain more feature improvements than Microsoft releases.\n (no joke)\n15. cascading triggers and rules\n (try triggering another tables trigger from a trigger in MS SQL 6.5,\nit won't happen)\n16. 8k row limit (settable at compile time)\n (MS SQL 6.5 has a 2k row limit, just try it and see)\nI could go on but I have to get to the dentist.\n\nIn PostgreSQL v6.5, scheduled to beta in February 1999, the locking\nsystem will be better then MS SQL 6.5 (already implemented in CSV\ntree[see the web site] if you want to compare speed), therefore\nconcurrent transactions should far outstrip MS SQL 6.5. PostgreSQL v6.5\nwill also have a fuller set of the SQL92 standard (including NUMERIC AND\nDECIMAL) then MS SQL 6.5. RI should be fully implemented by then (no\npromises, I'm not the one doing the work). I'm hoping Jan would have\nrewritten the function system to allow for multiple row return as well.\n\nI replicate a PostgreSQL server using a copy. And there are so many\ncool PostgreSQL interfaces written by many people out there I'll let you\nreview them for yourself.\n\nOh well, off to the dentist now (pray for me),\n -DEJ\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sat, 9 Jan 1999 07:59:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL 7"
},
{
"msg_contents": "Thanks to all for the reference to the earlier discussion about comparing\nMicrosoft SQL Server to PGSQL.\n\nre: Oracle - it's always been clear to me that Oracle far outpaces MSSQL; no\nneed to go into detail there. I was simply hoping for a comparison to MSSQL\nbecause that's the tool I have been using (and my clients generally prefer).\n\nHowever, re: the long message that someone quoted, with a feature-by-feature\ncomparison: I appreciate the laundry list, and it had a few eye-openers for\nme (was surprised that PGSQL doesn't yet support RI although I know it's\njust around the corner). But what it lacked was the real meat of my\nquestion, which is a basic benchmark comparing the two. Again, I'm not\nlooking for 10 significant digits of precision or anything, I just want to\nknow whether or not the two products are in the same ballpark in terms of\nraw performance. Assuming they are, then I can feel comfortable recommending\nusing PGSQL to my clients for future projects (I build data-driven web\nsites). I am anxious to get off the Microsoft wagon and onto the Open Source\nwagon, from a professional standpoint. I know that Linux can meet\ntit-for-tat Windows NT on performance, as can Apache meet IIS, as can PHP\nmeet ASP (I think, not sure about that). The last piece is, can PGSQL meet\nMSSQL, or at least be in the same ballpark.\n\nI only briefly considered MySQL before turning to PGSQL in my new platform\nresearch; I saw that MySQL lacked a lot of advanced SQL features (such as\ntransactions), and that although it was blazingly fast, it seemed kind of a\nhack. Also, it's sort of commercial in a weird way that I can't quite figure\nout. So here I am in the PGSQL camp checking things out.\n\nThanks,\nBryan\n\n\n\n\n",
"msg_date": "Sat, 9 Jan 1999 08:39:54 -0800",
"msg_from": "\"Bryan Field-Elliot\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "RE: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL\n 7"
},
{
"msg_contents": "> The last piece is, can PGSQL meet\n> MSSQL, or at least be in the same ballpark.\n\nYes. Depending on the size of your database and the loading on the web\nserver, I would expect that performance is acceptable now. If you can\ntell us the approximate size of your intended databases then it is\nlikely others can tell you they are already running something that size. \n\nWe have a major improvement in table locking for concurrent access\ncoming in the next release (v6.5) which we expect will offer significant\nimprovements in speed in a multi-client environment. afaik we don't have\nany numbers yet, but I think there are a few folks out there with the\nright setup to simulate a large number of clients. When Vadim sez \"go\",\nwe should try running some benchmark comparisons against v6.4.\n\nI ran some row inserting benchmarks a year or two ago and (I recall)\nfound that I saw ~50InsertsPerSecond on both my Alpha/DUnix/Ingres\ninstallation and on my i686/Linux/Postgres box. These were basically\nsingle-client tests. There may be some more info in our mailing list\narchives.\n\nAnother performance consideration is that linux takes fewer resources\nfrom the machine than does WindowsXX, so you can donate the extra to\nPostgres ;)\n\n - Tom\n",
"msg_date": "Sat, 09 Jan 1999 18:52:11 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL\n 7"
},
{
"msg_contents": "On Sat, 9 Jan 1999, Thomas G. Lockhart wrote:\n\n> We have a major improvement in table locking for concurrent access\n> coming in the next release (v6.5) which we expect will offer significant\n> improvements in speed in a multi-client environment.\n\nI'd like to jump in here with two comments:\n\n1) Yay for fine-grained locking!\n\n2) Will PostgreSQL _ever_ have support for foreign keys? That is the\nsingle biggest gripe around here at MindSpring. Lots of people use psql\nand love it, myself included, but the lack of support for foreign keys\nis so annoying that it boggles the mind. I know that you can simulate\nthe behaviour using triggers, and so the question comes, why can't\nsomeone just hack the command syntax to translate the \"foreign keys\"\ndirective into the appropriate trigger insertion? This has been on the\nTODO list for as long as I have been watching PostgreSQL, and at this\npoint I despair that it will ever get done. I would be happier if the\nREADME just said \"PostgreSQL does not and will never support the SQL\n'FOREIGN KEYS' directive.\"\n\nThis is intended to be user feedback, not whiny complaining. I'm very\nhappy with what PostgreSQL gives me. Thanks for the good work.\n\n-- \nTodd Graham Lewis [email protected] (800) 719-4664, x2804\n\n\"It's still ludicrous that nobody's ever made a run at us by making UNIX\n a popular platform on PCs. It's almost too late now.\" -- Steve Balmer\n\"It is too late.\" -- Bill Gates _Newsweek_, 6/23/97, p. 82\n\n",
"msg_date": "Sat, 9 Jan 1999 14:11:04 -0500 (EST)",
"msg_from": "Todd Graham Lewis <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL\n 7"
},
{
"msg_contents": "\nOn Sat, 9 Jan 1999, Bryan Field-Elliot wrote:\n\n...\n> question, which is a basic benchmark comparing the two. Again, I'm not\n> looking for 10 significant digits of precision or anything, I just want to\n> know whether or not the two products are in the same ballpark in terms of\n> raw performance. Assuming they are, then I can feel comfortable recommending\n...\n\n I don't if it possible to get such information, as I would think it be\nillegal to tell anybody (publish) the results if Postgresql and MSQL were\nbenchmarked together. This is the whole point of the Oracle challege.\nThe only MSSQL benchmarks available are those produced by Microsoft,\nbecause no body else is allowed to publish results. The only thing that\ncan be done, is to try to build an identical test system that Microsoft\nused, and run the postgresql tests on it.\n\n\nTom\n\n",
"msg_date": "Sat, 9 Jan 1999 11:13:50 -0800 (PST)",
"msg_from": "Tom <[email protected]>",
"msg_from_op": false,
"msg_subject": "RE: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL\n 7"
},
{
"msg_contents": "> 2) Will PostgreSQL _ever_ have support for foreign keys?\n\nYes.\n\n> That is the\n> single biggest gripe around here at MindSpring. Lots of people use \n> psql and love it, myself included, but the lack of support for foreign \n> keys is so annoying that it boggles the mind.\n\nWell, if it were *that* annoying someone would probably have fixed it up\nby now. That's supposed to be one of the advantages of open source\nsoftware; you can \"scratch your own itch\" rather than having to wait for\nothers.\n\nI used an Ingres database for years, which had neither foreign keys nor\nouter joins. Did OK with it too. I had a choice of what to work on, and\nI'm poking at outer joins (don't know if they will make the next release\nyet though). Foreign keys weren't as interesting to me, so ...\n\n> I know that you can simulate\n> the behaviour using triggers, and so the question comes, why can't\n> someone just hack the command syntax to translate the \"foreign keys\"\n> directive into the appropriate trigger insertion?\n\nThat might be a good way to do it. Go fur it dude!\n\n> I would be happier if the\n> README just said \"PostgreSQL does not and will never support the SQL\n> 'FOREIGN KEYS' directive.\"\n\nLiar :))\n\n - Tom\n",
"msg_date": "Sat, 09 Jan 1999 20:07:40 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL\n 7"
},
{
"msg_contents": "On Sat, 9 Jan 1999, Thomas G. Lockhart wrote:\n\n> > I know that you can simulate\n> > the behaviour using triggers, and so the question comes, why can't\n> > someone just hack the command syntax to translate the \"foreign keys\"\n> > directive into the appropriate trigger insertion?\n> \n> That might be a good way to do it. Go fur it dude!\n\nI'll see whom I can pester into doing it...\n\n> > I would be happier if the\n> > README just said \"PostgreSQL does not and will never support the SQL\n> > 'FOREIGN KEYS' directive.\"\n> \n> Liar :))\n\nOk, maybe \"happier\" was an overstatement...\n\n-- \nTodd Graham Lewis [email protected] (800) 719-4664, x2804\n\n\"It's still ludicrous that nobody's ever made a run at us by making UNIX\n a popular platform on PCs. It's almost too late now.\" -- Steve Balmer\n\"It is too late.\" -- Bill Gates _Newsweek_, 6/23/97, p. 82\n\n",
"msg_date": "Sat, 9 Jan 1999 15:11:18 -0500 (EST)",
"msg_from": "Todd Graham Lewis <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft\n\tSQL 7"
},
{
"msg_contents": "> 2) Will PostgreSQL _ever_ have support for foreign keys? That is the\n> single biggest gripe around here at MindSpring. Lots of people use psql\n> and love it, myself included, but the lack of support for foreign keys\n> is so annoying that it boggles the mind. I know that you can simulate\n> the behaviour using triggers, and so the question comes, why can't\n> someone just hack the command syntax to translate the \"foreign keys\"\n> directive into the appropriate trigger insertion? This has been on the\n> TODO list for as long as I have been watching PostgreSQL, and at this\n> point I despair that it will ever get done. I would be happier if the\n> README just said \"PostgreSQL does not and will never support the SQL\n> 'FOREIGN KEYS' directive.\"\n\nI was going to soapbox this item, row reuse without vacuum, and buffered\nlogging as things at we really need. I am thinking about row reuse now,\nand want to post something about buffered logging this week to Vadim.\n\nThat leaves foreign keys. We really need to address this, because it\ncomes up so often. Can we hash through the issues again, and see if\nsomeone takes on the project.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sat, 9 Jan 1999 20:47:46 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RE: [GENERAL] Benchmarking PGSQL against Microsoft SQL\n 7"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> > 2) Will PostgreSQL _ever_ have support for foreign keys? That is the\n> > single biggest gripe around here at MindSpring. Lots of people use psql\n> > and love it, myself included, but the lack of support for foreign keys\n> > is so annoying that it boggles the mind. I know that you can simulate\n> > the behaviour using triggers, and so the question comes, why can't\n> > someone just hack the command syntax to translate the \"foreign keys\"\n> > directive into the appropriate trigger insertion? This has been on the\n> > TODO list for as long as I have been watching PostgreSQL, and at this\n> > point I despair that it will ever get done. I would be happier if the\n> > README just said \"PostgreSQL does not and will never support the SQL\n> > 'FOREIGN KEYS' directive.\"\n> \n> I was going to soapbox this item, row reuse without vacuum, and buffered\n> logging as things at we really need. I am thinking about row reuse now,\n> and want to post something about buffered logging this week to Vadim.\n> \n> That leaves foreign keys. We really need to address this, because it\n> comes up so often. Can we hash through the issues again, and see if\n> someone takes on the project.\n\nIf my memory servers me right:\n\nVadim _almost_ did implement FOREIGN KEY syntax just before 6.4 (or\n6.4.1 ?)\nusing the trigger version in contrib.\n\nBut he stopped the work as he was unhappy with the way the contrib stuff \nworked i.e. the keys can't break the rules even _during_ the statement.\n\nI wholly agree with this being broken, but it still would not hurt very \nbecause the same is unfortunately true for current primary keys/unique \nindexes, try for example\n\nhannu=> create table test(\nhannu-> k int primary key,\nhannu-> d int\nhannu-> );\nNOTICE: CREATE TABLE/PRIMARY KEY will create implicit index test_pkey\nfor table test\nCREATE\nhannu=> insert into test values(1,1);\nINSERT 18285 1\nhannu=> insert into test values(2,2);\nINSERT 18286 1\nhannu=> update test set k=k-1;\nUPDATE 2\nhannu=> update test set k=k+1;\nERROR: Cannot insert a duplicate key into a unique index\n\nIMHO the right behaviour would be to allow this kind of updates, and I \nsuspect that Oracle does allow it.\n\nAt least the current behaviour is non-intuitive.\n\n---------------\n Hannu Krosing\n",
"msg_date": "Mon, 11 Jan 1999 15:45:35 +0200",
"msg_from": "Hannu Krosing <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOREIGN KEYs (Was: Benchmarking PGSQL against Microsoft\n\tSQL 7)"
},
{
"msg_contents": "Hannu Krosing wrote:\n> \n> Bruce Momjian wrote:\n> >\n> > That leaves foreign keys. We really need to address this, because it\n> > comes up so often. Can we hash through the issues again, and see if\n> > someone takes on the project.\n> \n> If my memory servers me right:\n> \n> Vadim _almost_ did implement FOREIGN KEY syntax just before 6.4 (or\n\nI did nothing for syntax.\n\n> 6.4.1 ?)\n> using the trigger version in contrib.\n> \n> But he stopped the work as he was unhappy with the way the contrib stuff\n> worked i.e. the keys can't break the rules even _during_ the statement.\n\nI stopped just after starting -:)\n\n> I wholly agree with this being broken, but it still would not hurt very\n> because the same is unfortunately true for current primary keys/unique\n> indexes, try for example\n\nYes, this would be nice, but I have no time/wish to do this - \nI would like to do \"right\" implementation someday. \nAnd for UNIQUE keys too.\n\n> hannu=> update test set k=k+1;\n> ERROR: Cannot insert a duplicate key into a unique index\n> \n> IMHO the right behaviour would be to allow this kind of updates, and I\n> suspect that Oracle does allow it.\n\nSure.\n\nVadim\n",
"msg_date": "Mon, 11 Jan 1999 21:03:31 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] FOREIGN KEYs (Was: Benchmarking PGSQL against Microsoft\n\tSQL 7)"
}
] |
[
{
"msg_contents": "I have posted this in quite a few places in the past without\nresponse, so I'll try it here once. I am not on the list, so please\nsend any responses to me directly. This is a minor bug that\nhas continued through the latest release.\n\n--------------------------------------------\n\nThis is a small bug in pg_dump that persists.\n\nSay you want to have an text field with the contents\nof \"\\Help\\\\\" (not including the quotes).\n\nIf you do the following dialog:\n\ncreatedb tempdb\npsql tempdb\ntempdb=>create table \"testtable\" ( \"content\" character(32));\ntempdb=>insert into \"testtable\" values ('\\\\Help\\\\\\\\');\ntempdb=>select * from testtable;\n\ncontent\n---------------------------------------\n\\Help\\\\\n\nUp to this point, everything is cool. But now if you do the following:\n\npg_dump -d tempdb\n\nYou will see that the insert statement it generates is now invalid:\n\nINSERT INTO \"testtable\" values ('\\Help\\\\ ');\n\npg_dump needs to convert a single backslash to a double backslash\nin order to be able to reload again as input to pgsql.\n\nAm I missing something? Thanks for taking a look at this. I have posted\nthis to the list a few times without response and you seem to want to\nget everything nailed down (Thanks!) so I thought I'd try you. If you want\nme to send this somewhere else, please let me know.\n\nThanks for all you are doing with this great database.\n\n\n\n\n",
"msg_date": "Fri, 8 Jan 1999 16:19:49 -0500",
"msg_from": "\"Frank Morton\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Minor Persistent Bug In pg_dump"
}
] |
[
{
"msg_contents": "On Thursday, January 07, 1999 2:17 AM, Tom Lane [SMTP:[email protected]]\nwrote:\n> Magnus Hagander <[email protected]> writes:\n> > [ Why is the server-side libpq so crufty? ]\n> \n> Apparently, that set of files was once used for both the frontend and\n> backend sides of the FE/BE protocol. It no longer is, but no one has\n> gotten around to ripping out the now-dead parts of the code, nor to\n> fixing the comments.\n> \n> I didn't bother to touch it when I rewrote the client-side libpq last\n> summer, because there wasn't any functional improvement to be had there.\n> It pretty much does everything the backend needs done.\n> \n> If you have the time and energy to clean it up just in the name of\n> code beautification, step right up :-). One thing that would be good\n> right off the bat is to change the name --- I think it's confusing to\n> call both the FE and BE modules libpq, when they are no longer the same\n> code or even very close.\n\nI guess I shouldn't have said anything :-)\nI'll see if I can shake loose the time to look at it. I might have to,\nthough.\n\n\n> > Finally - is there any special reason that the backend still uses the\n(FILE\n> > *) method to talk to the clients? Using the global Pfout and Pfin\nvariables?\n> > Wouldn't it be better to be consistent and use the same functions as in\nthe\n> > revised frontent library?\n> \n> The main reason for rewriting the front end was to satisfy clients that\n> didn't want to block while awaiting backend I/O. The backend doesn't\n> have any comparable requirement: when it's waiting for the frontend, it\n> has nothing better to do (AFAIK anyway). And using stdio does have its\n> advantages in simplicity and just plain standard-ness. So I doubt it's\n> worth making that kind of change.\n\nWell, I see one reason to change it. Which is why I came up with the\nquestion in the first place. I was looking at the possibility of putting SSL\non top of libpq. I have a project I'm working on that needs to transmit\n\"lightly sensitive data\" across the internet. Right now using SSH\nforwardings, but that's not exactly the \"ideal solution\".\nAnyway, SSLeay has functions that replace read() and write(), but nothing to\nwork with FILE *:s.\nSo if there are no major objections, I might take a shot at changing it to\nworking directly on the socket, and put SSLeay on it.\n\n\nA quick check shows that the \"copy\" command goes out of its way to break the\nabstraction of \"backend libpq\". Takes a copy of the Pfin and Pfout FILE*:s\nand writes directly to them. Darn. This might take more time than I first\nthought... :-(\n\n\n\n//Magnus\n",
"msg_date": "Sat, 9 Jan 1999 22:49:59 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Libpq functions "
},
{
"msg_contents": "Magnus Hagander <[email protected]> writes:\n> Well, I see one reason to change it. Which is why I came up with the\n> question in the first place. I was looking at the possibility of putting SSL\n> on top of libpq. I have a project I'm working on that needs to transmit\n> \"lightly sensitive data\" across the internet. Right now using SSH\n> forwardings, but that's not exactly the \"ideal solution\".\n> Anyway, SSLeay has functions that replace read() and write(), but nothing to\n> work with FILE *:s.\n> So if there are no major objections, I might take a shot at changing it to\n> working directly on the socket, and put SSLeay on it.\n\nAh. Now that you mention it, I recall someone bringing up that exact\nissue last summer on the hackers list. Was that you? If not, you might\nwant to go digging in the list archives (I forget what was said...)\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sat, 09 Jan 1999 17:13:11 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Libpq functions "
}
] |
[
{
"msg_contents": "On Saturday, January 09, 1999 11:13 PM, Tom Lane [SMTP:[email protected]]\nwrote:\n> Magnus Hagander <[email protected]> writes:\n> > Well, I see one reason to change it. Which is why I came up with the\n> > question in the first place. I was looking at the possibility of putting\nSSL\n> > on top of libpq. I have a project I'm working on that needs to transmit\n> > \"lightly sensitive data\" across the internet. Right now using SSH\n> > forwardings, but that's not exactly the \"ideal solution\".\n> > Anyway, SSLeay has functions that replace read() and write(), but\nnothing to\n> > work with FILE *:s.\n> > So if there are no major objections, I might take a shot at changing it\nto\n> > working directly on the socket, and put SSLeay on it.\n> \n> Ah. Now that you mention it, I recall someone bringing up that exact\n> issue last summer on the hackers list. Was that you? If not, you might\n> want to go digging in the list archives (I forget what was said...)\n\nCould've been me. I'll go check to be sure.\nI remember fighting this same obstacle in the frontend library before you\nfixed it - because Win32 does not handle fdopen() on sockets.\n\n\nAlso, I noticed that in backend/util/error/elog.c, it checks for \"Pfout !=\nNULL\" to see if it's running under the postmaster. If Pfout == NULL, it\nsends its data out to stderr instead of the client.\nShouldn't it be safe to just if on the global \"IsUnderPostmaster\" boolean\nhere? Or am I missing something? (I'm starting my work by trying to get rid\nof anything other than pqcomm.c using Pfout and Pfin).\n\n//Magnus\n",
"msg_date": "Sat, 9 Jan 1999 23:14:48 +0100 ",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Libpq functions "
},
{
"msg_contents": "> Also, I noticed that in backend/util/error/elog.c, it checks for \"Pfout !=\n> NULL\" to see if it's running under the postmaster. If Pfout == NULL, it\n> sends its data out to stderr instead of the client.\n> Shouldn't it be safe to just if on the global \"IsUnderPostmaster\" boolean\n> here? Or am I missing something? (I'm starting my work by trying to get rid\n> of anything other than pqcomm.c using Pfout and Pfin).\n\nNot sure, but it is possible the coder didn't know about the variable.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sat, 9 Jan 1999 20:43:26 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Libpq functions"
}
] |
[
{
"msg_contents": "\tcould soemeone explain to me how the tcl/tk configure stuff is suposed\n\tto worl?\n\n\tI built 6.4.2 on my HP-UX machine a week or so agao, and had to do some\n\t_really_ strange things to get it working. Tonite I tried to build on\n\tFreeBSD (sort of a reference paltform, no?), and still had the folowint\n\terror, during configure:\n\nchecking for tkConfig.sh... no\nconfigure: warning: TK support disabled; Tk configuration script missing\n\n\tI used teh --with-tcl and --with tk configure flags, so what's going on\n\there?\n\n-- \nStan Brown [email protected] 843-745-3154\nWestvaco\nCharleston SC.\n-- \nWindows 98: n.\n\tuseless extension to a minor patch release for 32-bit extensions and\n\ta graphical shell for a 16-bit patch to an 8-bit operating system\n\toriginally coded for a 4-bit microprocessor, written by a 2-bit \n\tcompany that can't stand for 1 bit of competition.\n-\n(c) 1999 Stan Brown. Redistribution via the Microsoft Network is prohibited.\n",
"msg_date": "Sat, 9 Jan 1999 19:41:21 -0500 (EST)",
"msg_from": "\"Stan Brown\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "TCL/TK configure problems"
},
{
"msg_contents": "\"Stan Brown\" <[email protected]> writes:\n\n| checking for tkConfig.sh... no\n| configure: warning: TK support disabled; Tk configuration script missing\n| \n| \tI used teh --with-tcl and --with tk configure flags, so what's going on\n| \there?\n\nThe configure script wrongly thinks, tcl and tk are to be found at the\nsame loaction. On SuSE Linux I've to use the following *ugly* hack:\n\n--- pgsql/src/configure.in~\tFri Oct 30 03:146 1998\n+++ pgsql/src/configure.in\tFri Oct 30 14:52:34 1998\n@@ -790,7 +790,7 @@\n \tAC_MSG_CHECKING(for tkConfig.sh)\n \tTK_CONFIG_SH=\n \t# library_dirs are set in the check for TCL\n-\tfor dir in $library_dirs\n+\tfor dir in /usr/X11R6/lib\n \tdo\n \t\tif test -d \"$dir\" -a -r \"$dir/tkConfig.sh\"\n \t\tthen\n\n-- \nKarl Eichwalder\n",
"msg_date": "10 Jan 1999 10:34:50 +0100",
"msg_from": "Karl Eichwalder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: TCL/TK configure problems"
},
{
"msg_contents": "> \"Stan Brown\" <[email protected]> writes:\n> \n> | checking for tkConfig.sh... no\n> | configure: warning: TK support disabled; Tk configuration script missing\n> | \n> | \tI used teh --with-tcl and --with tk configure flags, so what's going on\n> | \there?\n> \n> The configure script wrongly thinks, tcl and tk are to be found at the\n> same loaction. On SuSE Linux I've to use the following *ugly* hack:\n\nOr you could add configure --with-libs=DIR.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 10 Jan 1999 05:03:14 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: TCL/TK configure problems"
},
{
"msg_contents": "Karl Eichwalder <[email protected]> writes:\n> \"Stan Brown\" <[email protected]> writes:\n> | \tI used teh --with-tcl and --with tk configure flags, so what's going on\n> | \there?\n\n> The configure script wrongly thinks, tcl and tk are to be found at the\n> same loaction.\n\nWell, it doesn't exactly insist that the config files be in the same\ndirectory. What it does do is assume that they will both be found in\nthe \"auto_path\" search path that's configured into 'tclsh'. Maybe\nit'd be a better idea to look for a 'wish' and ask it its auto_path\nwhen looking for the Tk configure info.\n\nThat's not gonna be easy, however, since 'wish' won't fire up unless it\ncan connect to an X display. I don't think we want to have a configure\nscript that fails (or delivers different results) depending on whether\nyou are running it inside an X session or not.\n\nI'm also worried that we risk selecting incompatible versions of Tcl\nand Tk, which would be bad news indeed. In any reasonable installation\nTcl and Tk will have similar auto_path settings; you'd have to go out\nof your way during configure/build to make them different. If we don't\nfind the Tk configure file on the Tcl auto_path, I'd take that as a\nstrong sign that we've not found the Tcl that goes with Tk. (As a\nperson with several different Tcl/Tk versions installed on my system,\nI take this concern pretty seriously...)\n\nI'm inclined to leave the code alone and tell people with weird\nTcl/Tk layouts that they have to provide --with-tclconfig='search path'.\nThe code and docs do need improved to make it clear that more than one\ndirectory can be listed in --with-tclconfig.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 10 Jan 1999 12:28:43 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: TCL/TK configure problems "
},
{
"msg_contents": "Tom Lane <[email protected]> writes:\n\n| The code and docs do need improved to make it clear that more than one\n| directory can be listed in --with-tclconfig.\n\nThat's okay with me. Thanks for your explanation.\n\n-- \nKarl Eichwalder\n",
"msg_date": "10 Jan 1999 18:49:26 +0100",
"msg_from": "Karl Eichwalder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: TCL/TK configure problems"
}
] |
[
{
"msg_contents": "I've just migrated a database to 6.4.2 and was looking at the new\npgdump output. It looks much better than earlier versions, but it\ndoesn't seem to include any information about sequences created\nimplicitly via the serial datatype. Pgdump describes the field as one\nmight have before (e.g., int4 default nextval (table_id_seq)), but the\nsequence is never created. As a result, the script seems unable to\nrecreate the state of the database.\n\nWhat is required to fix this? Should this be a TODO item?\n\nCheers,\nBrook\n",
"msg_date": "Sat, 9 Jan 1999 19:50:50 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "pgdump and sequences"
}
] |
[
{
"msg_contents": "As of 6.4.2 (and snapshot, I guess), postmaser does not check if the\nnumber of backends exceeds MaxBackendId (defined in\ninclude/storage/sinvaladt.h). As a result (MaxBackendId+1)th backend\ngets started but failed in the middle of its initialising process.\nTypical error would be:\n\nNOTICE: SIAssignBackendId: discarding tag 2147430138\nConnection databese 'request' failed.\nFATAL 1: Backend cache invalidation initialization failed\n\nThen postmaster decides to re-initialize the shared memory and all\nrunning backends are killed. Too bad.\n\nAttached patches try to fix the problem.\n--\nTatsuo Ishii\n-------------------------------- cut here ---------------------------\n*** postgresql-6.4.2/src/backend/postmaster/postmaster.c.orig\tSun Nov 29 10:52:32 1998\n--- postgresql-6.4.2/src/backend/postmaster/postmaster.c\tSat Jan 9 18:14:52 1999\n***************\n*** 238,243 ****\n--- 238,244 ----\n static long PostmasterRandom(void);\n static void RandomSalt(char *salt);\n static void SignalChildren(SIGNAL_ARGS);\n+ static int CountChildren(void);\n \n #ifdef CYR_RECODE\n void\t\tGetCharSetByHost(char *, int, char *);\n***************\n*** 754,764 ****\n \t\t\t\t * by the backend.\n \t\t\t\t */\n \n! \t\t\t\tif (BackendStartup(port) != STATUS_OK)\n! \t\t\t\t\tPacketSendError(&port->pktInfo,\n \t\t\t\t\t\t\t\t\t\"Backend startup failed\");\n! \t\t\t\telse\n! \t\t\t\t\tstatus = STATUS_ERROR;\n \t\t\t}\n \n \t\t\t/* Close the connection if required. */\n--- 755,771 ----\n \t\t\t\t * by the backend.\n \t\t\t\t */\n \n! if (CountChildren() < MaxBackendId) {\n! \t\t\t\t\tif (BackendStartup(port) != STATUS_OK)\n! \t\t\t\t\t\tPacketSendError(&port->pktInfo,\n \t\t\t\t\t\t\t\t\t\"Backend startup failed\");\n! \t\t\t\t\telse {\n! \t\t\t\t\t\tstatus = STATUS_ERROR;\n! \t\t\t\t\t}\n! \t\t\t\t} else {\n! \t\t\t\t\tPacketSendError(&port->pktInfo,\n! \t\t\t\t\t\"There are too many backends\");\n! \t\t\t\t}\n \t\t\t}\n \n \t\t\t/* Close the connection if required. */\n***************\n*** 1617,1620 ****\n--- 1624,1655 ----\n \t}\n \n \treturn random() ^ random_seed;\n+ }\n+ \n+ /*\n+ * Count up number of chidren processes.\n+ */\n+ static int\n+ CountChildren(void)\n+ {\n+ \tDlelem\t *curr,\n+ \t\t\t *next;\n+ \tBackend *bp;\n+ \tint\t\t\tmypid = getpid();\n+ \tint\tcnt = 0;\n+ \n+ \tcurr = DLGetHead(BackendList);\n+ \twhile (curr)\n+ \t{\n+ \t\tnext = DLGetSucc(curr);\n+ \t\tbp = (Backend *) DLE_VAL(curr);\n+ \n+ \t\tif (bp->pid != mypid)\n+ \t\t{\n+ \t\t\tcnt++;\n+ \t\t}\n+ \n+ \t\tcurr = next;\n+ \t}\n+ \treturn(cnt);\n }\n",
"msg_date": "Sun, 10 Jan 1999 23:16:27 +0900",
"msg_from": "Tatsuo Ishii <[email protected]>",
"msg_from_op": true,
"msg_subject": "max backends checking patch"
},
{
"msg_contents": "Tatsuo Ishii wrote:\n> \n> As of 6.4.2 (and snapshot, I guess), postmaser does not check if the\n> number of backends exceeds MaxBackendId (defined in\n> include/storage/sinvaladt.h). As a result (MaxBackendId+1)th backend\n> gets started but failed in the middle of its initialising process.\n> Typical error would be:\n> \n> NOTICE: SIAssignBackendId: discarding tag 2147430138\n> Connection databese 'request' failed.\n> FATAL 1: Backend cache invalidation initialization failed\n> \n> Then postmaster decides to re-initialize the shared memory and all\n> running backends are killed. Too bad.\n> \n> Attached patches try to fix the problem.\n\n\nCouldn't postmaster just keep # of backends running\nin some variable, instead of examining BackendList ?\n\n\n> + /*\n> + * Count up number of chidren processes.\n> + */\n> + static int\n> + CountChildren(void)\n> + {\n> + Dlelem *curr,\n> + *next;\n> + Backend *bp;\n> + int mypid = getpid();\n> + int cnt = 0;\n> +\n> + curr = DLGetHead(BackendList);\n> + while (curr)\n> + {\n> + next = DLGetSucc(curr);\n> + bp = (Backend *) DLE_VAL(curr);\n> +\n> + if (bp->pid != mypid)\n> + {\n> + cnt++;\n> + }\n> +\n> + curr = next;\n> + }\n> + return(cnt);\n> }\n\nVadim\n",
"msg_date": "Sun, 10 Jan 1999 21:30:24 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] max backends checking patch"
},
{
"msg_contents": "> Couldn't postmaster just keep # of backends running\n> in some variable, instead of examining BackendList ?\n\nYes, you could do that way. I just want to keep things simple. There\nare three places in postmaster.c where element is deleted from the\nBackendList. So you need to insert a code to count down the variable\ninto those three places. Seems a seed of \"maintenance problem\" in the\nfuture IMHO:-)\n\nHaving a counter inside the dllist module is another idea. If there\nwere many codes in the backend that counting elemnts in the dllist,\nthis would be worth to think about. I'm not sure, though.\n---\nTatsuo Ishii\n",
"msg_date": "Mon, 11 Jan 1999 00:00:01 +0900",
"msg_from": "Tatsuo Ishii <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] max backends checking patch "
},
{
"msg_contents": "Tatsuo Ishii <[email protected]> writes:\n>> Couldn't postmaster just keep # of backends running\n>> in some variable, instead of examining BackendList ?\n\n> Yes, you could do that way. I just want to keep things simple.\n> Seems a seed of \"maintenance problem\" in the future IMHO:-)\n\nI agree with Tatsuo. Counting the children once per backend startup\nis certainly not a performance bottleneck, so there is no reason to\nadd complexity and a source of potential bugs to speed it up.\n\n> Having a counter inside the dllist module is another idea. If there\n> were many codes in the backend that counting elemnts in the dllist,\n> this would be worth to think about.\n\nThat would be reasonable if justified by usage --- the tradeoff is\nthe added cost of maintaining the count for every dllist, whether or\nnot it's ever asked for...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 10 Jan 1999 12:07:54 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] max backends checking patch "
}
] |
[
{
"msg_contents": "Vadim wrote:\n>Zeugswetter Andreas IZ5 wrote:\n>> \n>> >> Shouldn't we change default transaction mode now?\n>> >> And use option/SET TRANSACTION MODE to switch to\n>> >> un-chained mode if one like it?\n>> >\n>> >No comments on this?\n>> >I would like to make BEGIN implicit...\n>> >Objections?\n>> \n>> Yes, I object. I think the default behavior should be the one\nthat\n>> is sufficient in most cases.\n>> \n>> As to the isolation level:\n>> Informix and DB/2 default to read committed mode. In most cases\nthis\n>> is\n>> sufficient.\n>> Therefore most clients don't need the extra set transaction\n>> isolation statement.\n>\n>And this is the same as in Oracle & SyBase.\n>I don't object - currently READ COMMITTED is not\n>supported by DELETE/UPDATE code and so default is\n>SERIALIZABLE, to be changed latter.\n\nGood :-)\n\n>> \n>> As to transaction blocks:\n>> Actually most clients never use commit work, and remember\n\n>In Oracle, COMMIT is implicit if DB connection is closed\n>in \"normal\" way - by using PQfinish in our case.\n\nYes, but that is not really what we would want a client to do. \n(Not give a commit for 5 h)\n\n>> that postgresql will never free it's memory before commit. This\nwill\n>> hog up\n>> memory for otherwise lightweight clients.\n>\n>This is the bug, but I believe that it's fixed for\n>most cases ~1year ago by this\n>\n>#ifdef TBL_FREE_CMD_MEMORY\n> EndPortalAllocMode();\n> StartPortalAllocMode(DefaultAllocMode, 0);\n>#endif\n>\n>in xact.c\n\nYes, if you do a commit the memory will be freed alright. I meant a client,\nthat does\nonly selects, and therefore never does a commit. In chained mode the memory\nallocated for the selects would not be freed until the disconnect.\n\n>> I also think that the begin work; commit work; statement block is\n>> somewhat part of the postgresql philosophy. We do not stand alone\n>> with this approach (Informix).\n>\n>First, BEGIN/END is used in standard for grouping queries\n>(in PL etc), not for transaction control, and so currently\n>using of BEGIN/END is ambiguous.\n\nSorry, I am not so good with my wording, I meant to say transaction block\nnot statement block. I have been crying to take the begin/end syntax out of\nthe PostgreSQL SQL syntax for very long now. The SQL statements should be:\nbegin work;\ncommit work;\nnot begin and end which is unfortunately still allowed.\n>Second, Informix is using locking => chained transaction\n>mode is way to lock-escalation.\n\nI don't understand ? Informix has no lock escalation (the term meaning the\npromotion of a row to a page and then Table lock, if soandso many locks for\na table are already held) and does not use chained mode.\n\n>\n>> Actually the other DB's start the transaction with the first\n>> modifying\n>> statement, in my opinion a rather weird approach.\n>> I have seen programs that do a dummy update, just to simulate a\n>> begin work, what a waste.\n>\n>I believe that this is not how Oracle works.\n>I assumed that any DML statement will start transaction.\n\nNot in DB/2.\n\n>> But this alone will not work, because a select statement, that\nneeds\n>> \n>> a sort would have started the transaction earlier (Since it\ncreated\n>> an implicit\n>> temp table).\n> ^^^^^^^^^^\n>This is changed - backend don't create temp table now. \n\nI was referring to the other DB's.\n\n>> I could go on, but you see I don't really like this ......\n>> \n>> If you want to change the behavior, I think we will need a\n>> changeable default\n>> at database level.\n>\n>Ok, for compatibility reasons, I agreed that default\n>must be un-chained mode, with ability to switch in\n>compile/run time.\n\nSounds great !\n\nAndreas\n",
"msg_date": "Mon, 11 Jan 1999 10:58:46 +0100",
"msg_from": "Zeugswetter Andreas IZ5 <[email protected]>",
"msg_from_op": true,
"msg_subject": "WG: [HACKERS] MVCC works in serialized mode!"
},
{
"msg_contents": "Zeugswetter Andreas IZ5 wrote:\n> \n> >>\n> >> As to transaction blocks:\n> >> Actually most clients never use commit work, and remember\n> \n> >In Oracle, COMMIT is implicit if DB connection is closed\n> >in \"normal\" way - by using PQfinish in our case.\n> \n> Yes, but that is not really what we would want a client to do.\n> (Not give a commit for 5 h)\n\nI would say - it's up to client -:)\nAnd remember - this is standard behaviour and so this mode\nshould be implemented in any case.\n\n> >> that postgresql will never free it's memory before commit. This\n> will\n> >> hog up\n> >> memory for otherwise lightweight clients.\n> >\n> >This is the bug, but I believe that it's fixed for\n> >most cases ~1year ago by this\n> >\n> >#ifdef TBL_FREE_CMD_MEMORY\n> > EndPortalAllocMode();\n> > StartPortalAllocMode(DefaultAllocMode, 0);\n> >#endif\n> >\n> >in xact.c\n> \n> Yes, if you do a commit the memory will be freed alright. I meant a client,\n ^^^^^^^^^^^\nNo! This part of code run after _each_ statement...\n\n> >> I also think that the begin work; commit work; statement block is\n> >> somewhat part of the postgresql philosophy. We do not stand alone\n> >> with this approach (Informix).\n> >\n> >First, BEGIN/END is used in standard for grouping queries\n> >(in PL etc), not for transaction control, and so currently\n> >using of BEGIN/END is ambiguous.\n> \n> Sorry, I am not so good with my wording, I meant to say transaction block\n> not statement block. I have been crying to take the begin/end syntax out of\n> the PostgreSQL SQL syntax for very long now. The SQL statements should be:\n> begin work;\n> commit work;\n> not begin and end which is unfortunately still allowed.\n\nI'm not sure but it seems that COMMIT without WORK is allowed\nby standard (at least, in some levels).\n\n> >Second, Informix is using locking => chained transaction\n> >mode is way to lock-escalation.\n> \n> I don't understand ? Informix has no lock escalation (the term meaning the\n> promotion of a row to a page and then Table lock, if soandso many locks for\n> a table are already held) and does not use chained mode.\n\nI'm sure that Informix has nice lock manager, \nbut in SERIALIZABLE+chained mode Informix will hold lock on selected \nrows untill COMMIT/ABORT: this increases possibility of\nlock escalation and remember that read/write concurrent access \nto the same row causes blocking (not give a commit for 5h\nafter simple select... -:))\n\nUn-chained mode seems more suitable for locking systems.\nBTW, is there ability to switch to chained mode in Informix?\nSyBase has it.\n\n> >> Actually the other DB's start the transaction with the first\n> >> modifying\n> >> statement, in my opinion a rather weird approach.\n> >> I have seen programs that do a dummy update, just to simulate a\n> >> begin work, what a waste.\n> >\n> >I believe that this is not how Oracle works.\n> >I assumed that any DML statement will start transaction.\n> \n> Not in DB/2.\n\nBut in standard. And I read that DDL statements also start\ntransaction (if there is no active one).\nAbility to run DDL & DML statements in the same\ntransaction is implementation dependent. \n\nVadim\n",
"msg_date": "Tue, 12 Jan 1999 16:51:48 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: WG: [HACKERS] MVCC works in serialized mode!"
}
] |
[
{
"msg_contents": "Hi!\n\nIs FOR UPDATE OF _view_ is allowed or not?\nCould someone run test in other DBases ?\n\nTIA,\n\tVadim\n",
"msg_date": "Mon, 11 Jan 1999 17:02:41 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": true,
"msg_subject": "FOR UPDATE and VIEWs..."
}
] |
[
{
"msg_contents": "What happens if PQfinish() is called from within a transaction PRIOR\nto either committing or rolling back the transaction?\n\nDoes the transaction get committed? rolled back? left in an\nundefined state?\n\nAlso, in the normal case of a begin/commit sequence, does the\ntransaction end at the commit (i.e., is begin/commit/commit an error)\nor does the transaction continue until the connection is closed?\n\nThanks for your help.\n\nCheers,\nBrook\n",
"msg_date": "Mon, 11 Jan 1999 12:18:10 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "PQfinish in transactions"
}
] |
[
{
"msg_contents": "So, could someone send me the SQL92 constraints syntax as well as the\ndefinition of what a deferrable constraint is supposed to be?\nADVthanksANCE\n\t-DEJ\n",
"msg_date": "Mon, 11 Jan 1999 13:44:24 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "CONSTRAINTS..."
},
{
"msg_contents": "\"Jackson, DeJuan\" <[email protected]> writes:\n\n> So, could someone send me the SQL92 constraints syntax as well as the\n> definition of what a deferrable constraint is supposed to be?\n\nMaybe someone emailed you a good answer already...\n\nThere's a whole chapter on constraints in \"A Guide to the SQL\nStandard\" 4th ed. by Date and Darwen. Constraint syntax includes\nCREATE DOMAIN, ALTER DOMAIN, DROP DOMAIN, CREATE ASSERTION, DROP\nASSERTION, CREATE TABLE, ALTER TABLE, FOREIGN KEY, CHECK, and SET\nCONSTRAINTS, and others.\n\nA deferrable constraint is one that may be deferred,using INITIALLY\nDEFERRED or DEFERRABLE in its definition, or later, using SET\nCONSTRAINTS ... DEFERRED.\n\nA constraint is presumably deferred until a) the end of the applicable\ntransaction, or b) the next COMMIT, or c) the next SET CONSTRAINTS\n... IMMEDIATE, whichever comes first. A possible use of deferrable\nconstraints is to avoid Catch-22 when setting up tables which have\ncyclic dependency among foreign keys. [SQL lawyers, check me on this.]\n\nBTW, AltaVista Web search on \"sql92 ~ constraint\" yielded 8 hits -\n7 of which were PostgreSQL items.\n\n\n",
"msg_date": "11 Jan 1999 23:52:34 -0600",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": ">From \"A Guide to The SQL standard\" C.J.DATE:\n\nFOREIGN KEY Syntax:\n\n* base-table-constraint-def\n\t::= [ CONSTRAINT constraint ]\n\t\tforeign-key-def [ deferrability ]\n\n foreign-key-def\n\t::= FOREIGN KEY ( column-commalist ) references-def\n\n references-def\n\t::= REFERENCES base-table [ ( column-commalist ) ]\n\t\t[ MATCH { FULL | PARTIAL } ]\n\t\t[ ON DELETE referential-action ]\n\t\t[ ON UPDATE referential-action ]\n\t\t\n referential-action\n\t::= NO ACTION | CASCADE | SET DEFAULT | SET NULL\n\n deferrability\n\t::= INITIALLY { DEFERRED | IMMEDIATE } [ NOT ] DEFERRABLE\n\n\n* column-constraint-def\n\t::= references-def [ deferrability ]\n\t\n\n\n14.6 DEFERRED CONSTRAINT CHECKING \n\nUp to this point we have been assuming that all integrity constraints\nare checked \"immediately,\" i.e., as the final step in executing any SQL\nstatement - and, if any constraint is found to he violated, the\noffending SQL statement is simply rejected, so that its overall effect\non the database is nil. Sometimes, however, it is necessary that certain\nconstraints not he checked until some later time, on the grounds that if\nthey were to be checked \"immediately\" they would always fail. Here is an\nexample (involving a referential cycle): \n\n* Suppose we have two base tables, Tl and T2, each of which includes a\nforeign key that references some candidate key of the other, and suppose\nwe start with both tables empty. Then, if all foreign key checking is\ndone immediately, there is no way to get started: Any attempt to insert\na row into either table will fail, because there is no target row in the\nother table that it can possibly reference.\nThe facilities described immediately following are intended to address\nsuch situations. \n1. At any given time, with respect to any given transaction, any given\nconstraint must be in one or two \"modes,\" immediate or deferred.*\nImmediate means the constraint is checked \"immediately\" (as explained\nabove); deferred means it is not.\n2. Any given constraint definition can optionally include either or both\nof the following: \n\nINITIALLY { DEFERRED | IMMEDIATE }\n[ NOT ] DEFERRABLE\n\nThese specifications appear as the final syntactic component of the\nconstraint definition. They can appear in either order.\n- INITIALLY DEFERRED and NOT DEFERRABLE are mutually exclusive. If\nneither INITIALLY DEFERRED nor INITIALLY IMMEDIATE is specified,\nINITIALLY IMMEDIATE is implied. If INITIALLY IMMEDIATE is specified or\nimplied, then if neither DEFERRABLE nor NOT DEFERRABLE is specified, NOT\nDEFERRABLE is implied. If lNITIALLY DEFERRED is specified, then (as\nalready explained) NOT DEFERRABLE must not he specified; DEFERRABLE can\nbe specified, but is implied anyway.\n- INITIALLY DEFERRED and INITIALLY IMMEDIATE specify the \"initial\" mode\nof the constraint i.e., its mode immediately after it is defined and\nat the start of every transaction'! as deferred or immediate,\nrespectively. . DEFERRABLE and NOT DEFERRABLE specify whether or not\nthis constraint can ever be in deferred mode. DEFERRABLE means it can;\nNOT DEFERRABLE means it cannot. \n\n3. The SET CONSTRAINTS statement is used to set the mode for specified\nconstraints with respect to the current transaction and current session\n(or the next transaction to he initiated in the current session, if the\nSQL-agent has no transaction currently executing). The syntax is: \n\nSET CONSTRAINTS { constraint-commalist | ALL } \n { DEFERRED | IMMEDIATE } \n\nEach \"constraint\" mentioned by name must he DEFERRABLE; ALL is short-\nhand for \"all DEFERRABLE constraints.\" If DEFERRED is specified, the\nmode of all indicated constraints is set to deferred. If 1MMED1ATE is\nspecified, the mode of all indicated constraints is set to immediate,\nand those constraints are then checked; if any check fails, the SET\nCONSTRAINTS fails, and the mode of all indicated constraints remains\nunchanged. Note that because of paragraph 4 below, the checks should not\nfail if the SET CONSTRAINTS statement is executed while the SQL-agent\nhas no current transaction. \n\n4. COMMIT implies SET CONSTRAINTS ALL IMMEDIATE (for every active\nSQL-session for the applicable SQL-transaction). If some implied\nintegrity check then fails, the COMMIT fails, and the transaction fails\nalso (i.e., is rolled back). \nTo revert to the example mentioned at the beginning of this section (the\nreferential cycle involving two tables): We could deal with the problem\nusing the foregoing facilities as indicated by the following pseudocode.\nData definitions:\n \nCREATE TABLE Tl CONSTRAINT T1FK FOREIGN KEY ... REFERENCES T2 \nINITIALLY DEFERRED \n\nCREATE TABLE T2 CONSTRAINT T2FK FOREIGN KEY ... REFERENCES T1 \nINITIALLY DEFERRED \n\nSQL-transaction: \n\nINSERT INTO T1 ( ... ) VALUES ( ... ) \nINSERT INTO T2 ( ... ) VALUES ( ... ) \nSET CONSTRAINTS T1FK, T2FK IMMEDIATE \nIF SQLSTATE = code meaning \"SET CONSTRAINTS failed\" \nTHEN ROLLBACK --cancel the INSERTs \n\n\n\n\nJackson, DeJuan wrote:\n> \n> So, could someone send me the SQL92 constraints syntax as well as the\n> definition of what a deferrable constraint is supposed to be?\n> ADVthanksANCE\n> -DEJ\n\n-Jose'-\n",
"msg_date": "Tue, 12 Jan 1999 17:35:25 +0100",
"msg_from": "\"Jose' Soares\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": ">\n> >From \"A Guide to The SQL standard\" C.J.DATE:\n>\n> FOREIGN KEY Syntax:\n>\n> [Good description of foreign key constraints - tnx]\n>\n>\n> Jackson, DeJuan wrote:\n> >\n> > So, could someone send me the SQL92 constraints syntax as well as the\n> > definition of what a deferrable constraint is supposed to be?\n> > ADVthanksANCE\n> > -DEJ\n>\n> -Jose'-\n\n This reminds me on one of my personal TODO's, because it\n show's to me that an implementation of constraints using\n triggers or the like wouldn't be such a good idea. Especially\n the part on deferred constraint checks would mean a lot of\n buffering to do the checks at the end.\n\n My idea on constraints was to use the rewrite rule system for\n them. I wanted first to implement another optimizable\n statement - RAISE. RAISE is mostly the same as a SELECT, but\n the result will not be sent to the frontend. Instead it will\n produce some formatted elog message(s?).\n\n The syntax of RAISE I have in mind is:\n\n RAISE [ALL | FIRST [n]] expr [, expr ...] FROM ...\n\n Anything after FROM is exactly the same as for a SELECT.\n\n If the first result attribute of RAISE is a (var|bp)char or\n text field, single occurences of % in it will be substituted\n by the following attributes. Otherwise all the attrs are\n simply concatenated with a padding blank to form the error\n message.\n\n ALL or FIRST n means, that not only the first error should be\n shown. A bit tricky to implement but I think a bunch of\n NOTICE and a final \"ERROR: 5 errors counted\" would be\n possible.\n\n Having this, a foreign key constraint rule could look like\n this:\n\n CREATE RULE _CIconstraint_name AS ON INSERT TO mytab DO\n RAISE 'Key \"%\" not in keytab', new.myatt FROM keytab\n WHERE NOT EXISTS (SELECT * FROM keytab WHERE keyatt = new.myatt);\n\n Similar rules for update are simple and an ON DELETE CASCADE\n rule isn't that hard too.\n\n\n\n For the deferred constraints we now need some more\n informations on the rules themself. Currently all queries\n thrown in by the rule system are executed prior to the\n original query. If we add some syntax to CREATE RULE so we\n can tell\n\n CREATE [DEFERRABLE] [INITIALLY DEFERRED] RULE ...\n\n the rule system would be able to collect those queries (they\n all would be RAISE statements) to a global querytree list if\n they should be deferred. This global list is drained out\n (all queries run) when either the transaction commits or the\n SET ... IMMEDIATE is executed.\n\n Well, the information to remember isn't a small amount. Per\n constraint that is to be deferred, there will be one\n querytree. And that for every single INSERT/UPDATE/DELETE.\n And if a table has 5 constraints, it will be 5 remembered\n querytrees per operation. But the information to remember\n doesn't depend on the amount of data affected in the\n statement (like it would be in a trigger implementation). So\n it will work in a situation like\n\n BEGIN TRANSACTION;\n SET CONSTRAINST ALL DEFERRED;\n UPDATE tab1 SET ref1 = ref1 + 1900;\n UPDATE tab2 SET key1 = key1 + 1900;\n COMMIT TRANSACTION;\n\n even if there are millions of rows in the tables.\n\n\n\n As Bruce said once I mentioned using the rule system for\n constraints: \"It's a tempting solution\". And I'm glad to have\n the work delayed until now because yet the DEFERRED problem\n surfaced and could be taken into account too.\n\n Comments? (sure :-)\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Wed, 13 Jan 1999 12:01:19 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": "> For the deferred constraints we now need some more\n> informations on the rules themself. Currently all queries\n> thrown in by the rule system are executed prior to the\n> original query. If we add some syntax to CREATE RULE so we\n> can tell\n> \n> CREATE [DEFERRABLE] [INITIALLY DEFERRED] RULE ...\n> \n> the rule system would be able to collect those queries (they\n> all would be RAISE statements) to a global querytree list if\n> they should be deferred. This global list is drained out\n> (all queries run) when either the transaction commits or the\n> SET ... IMMEDIATE is executed.\n> \n> Well, the information to remember isn't a small amount. Per\n> constraint that is to be deferred, there will be one\n> querytree. And that for every single INSERT/UPDATE/DELETE.\n> And if a table has 5 constraints, it will be 5 remembered\n> querytrees per operation. But the information to remember\n> doesn't depend on the amount of data affected in the\n> statement (like it would be in a trigger implementation). So\n> it will work in a situation like\n\nLet's look at it another way. If we didn't use the query rewrite\nsystem, what method could we use for foreign key/contraints that would\nfunction better than this?\n\nAs far as I remember, triggers are C functions? We can't generate these\non the fly inside the backend. (Though compiling C code from the\nbackend and dynamically linking it into the engine is way too cool.)\n\nCould we generate generic triggers that would handle most/all\nsituations? I don't know. Even if we can, would they be much faster\nthan the needed queries themselves? Seems triggers work on single\ntables. How do we span tables? If it is going to launch queries from\nthe trigger, we should use the rewrite system. It is better suited to\nthis, with predigested queries and queries that flow through the\nexecutor in step with the user queries!\n\nMaybe let's go with the rewrite system, because it works, and is\nflexible and strangely designed for this type of problem. Similar to\nhow we use the rewrite system for views.\n\nI am basically asking for a reason _not_ to use the rewrite system for\nthis. I can't think of one myself.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 13 Jan 1999 06:23:25 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": "> Let's look at it another way. If we didn't use the query rewrite\n> system, what method could we use for foreign key/contraints that would\n> function better than this?\n>\n> As far as I remember, triggers are C functions? We can't generate these\n> on the fly inside the backend. (Though compiling C code from the\n> backend and dynamically linking it into the engine is way too cool.)\n>\n> Could we generate generic triggers that would handle most/all\n> situations? I don't know. Even if we can, would they be much faster\n> than the needed queries themselves? Seems triggers work on single\n> tables. How do we span tables? If it is going to launch queries from\n> the trigger, we should use the rewrite system. It is better suited to\n> this, with predigested queries and queries that flow through the\n> executor in step with the user queries!\n>\n\n Generic triggers in C that are argument driven would be\n possible. But the drawback is that those triggers have to be\n very smart to use saved SPI plans (one for every different\n argument set). And it must be row level triggers, so for an\n update to a 2 meg row table they will be fired 2 million\n times and run their queries inside - will take some time.\n\n More painful in the 2 meg row situation is that trigger\n invocation has to be delayed until COMMIT if the constraint\n is deferred. I think we cannot remember 2 million OLD plus 2\n million NEW tuples if one tuple can have up to 8K (will be\n 32GB to remember plus overhead), so we need to remember at\n least the CTID's of OLD and NEW and refetch them for the\n trigger invocation. OUTCH - the OLD ones are at the head and\n all the NEW ones are at the end of the tables file!\n\n> Maybe let's go with the rewrite system, because it works, and is\n> flexible and strangely designed for this type of problem. Similar to\n> how we use the rewrite system for views.\n\n And the other changes I've planned for the rewrite system\n will improve this much more.\n\n 1. Change pg_rewrite.ev_attr into an int28. This would be\n useful for ON UPDATE rules so the rewrite system can\n easily check if a rule has to be applied or not. If none\n of named attributes gets something different assigned\n than it's own OLD value, they aren't updated so the rule\n could never result in an action and can be omitted\n completely.\n\n 2. Create cross reference catalog that lists all relations\n used in a rule (rangetable). If we have a DELETE CASCADE\n constraint, the rule is triggered on the key table and\n the action is a DELETE from the referencing table. If now\n the referencing table is dropped, the rule get's\n corrupted because the resulting querytree isn't plannable\n any longer (the relations in the rules rangetable are\n identified by the OID in pg_class, not by relname). You\n can see the effect if you create a view and drop one of\n the base tables.\n\n Well, we need to define what to do if a table is dropped\n that occurs in the crossref. First of all, the rules have\n to be dropped too, but in the case of a view rule, maybe\n the whole view too?\n\n And in the case where a key table to which another one\n has a CHECK reference is dropped? The rule action will\n allways abort, so it isn't useful any more. But I\n wouldn't like to silently drop it, because someone might\n want to drop and recreate the key table and this would\n silently result in that all the constraints have been\n lost.\n\n Maybe we should change the rulesystem at all so that the\n rangetable entries in the rule actions etc. are updated\n with a lookup from pg_class at rewrite time. Must be done\n carefully because someone might drop a table and recreate\n it with a different schema corrupting the parsetree of\n the rule actions though.\n\n 3. Allow an unlimited number of rules on a relation.\n Currently there is a hard coded limit on the number of\n rules the relation can hold in it's slots.\n\n>\n> I am basically asking for a reason _not_ to use the rewrite system for\n> this. I can't think of one myself.\n\n It might interfere with the new MVCC code. The rule actions\n must see exactly the OLD tuples that where used in the\n original statements. Not only those in the updated table\n itself, think of an INSERT...SELECT or an UPDATE where the\n TLE or qual expressions are values from other tables.\n\n Not a real reason, just something to have in mind and maybe\n switching silently to another MVCC isolation level if\n constraint rules get applied, so all tables read from now on\n will get a read lock applied and cannot get updated\n concurrently until COMMIT.\n\n And it's a problem I've came across just writing this note\n where MVCC already could have broken rewrite rule system\n semantics.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Wed, 13 Jan 1999 15:16:55 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": "Jan Wieck wrote:\n> \n> Generic triggers in C that are argument driven would be\n> possible. But the drawback is that those triggers have to be\n> very smart to use saved SPI plans (one for every different\n> argument set). And it must be row level triggers, so for an\n> update to a 2 meg row table they will be fired 2 million\n> times and run their queries inside - will take some time.\n> \n> More painful in the 2 meg row situation is that trigger\n> invocation has to be delayed until COMMIT if the constraint\n> is deferred. I think we cannot remember 2 million OLD plus 2\n> million NEW tuples if one tuple can have up to 8K (will be\n> 32GB to remember plus overhead), so we need to remember at\n> least the CTID's of OLD and NEW and refetch them for the\n> trigger invocation. OUTCH - the OLD ones are at the head and\n> all the NEW ones are at the end of the tables file!\n\n(Note that now in the case of UPDATE t_ctid of OLD tuples\npoints to TID of NEW tuples.)\n\n> > I am basically asking for a reason _not_ to use the rewrite system for\n> > this. I can't think of one myself.\n> \n> It might interfere with the new MVCC code. The rule actions\n> must see exactly the OLD tuples that where used in the\n> original statements. Not only those in the updated table\n> itself, think of an INSERT...SELECT or an UPDATE where the\n> TLE or qual expressions are values from other tables.\n\nTwo things define data visibility: SnapShot & CommandId.\nWe would have to save them for deffered rules and restore them \nbefore run rule actions. But there is one issue: for what\nscans old visibility should be used? There are scans from\nuser query and there are scans added by rule action. Ok,\nlet's assume that for added scans current visibility will be used \n- this is what we need for RI rules (actually, something more -\nsee below). \n\nSo, first task is enable different scans in (rewritten) query\nuse different visibilities (SnapShot/CommandId pair -\n\"snapshot\", in short). We have to add new stuff to Executor\nand heap scan code and so I propose also new feature addition:\n\n1. add \n\nSET SNAPSHOT snapshot_name;\n\n statement to let users define some snapshot.\n\n2. extend query syntax to let users specify what snapshot\n must be used when a query table is scanned:\n\nSELECT ... FROM t1 AT SNAPSHOT s1, t2 AT SNAPSHOT s2 ...etc..\n\n\nUp to now new requirement due to MVCC is taking into account\nnot only CommandId (as already noted in last posting I got \nfrom Jan), but SnapShot too.\n\n> Not a real reason, just something to have in mind and maybe\n> switching silently to another MVCC isolation level if\n> constraint rules get applied, so all tables read from now on\n> will get a read lock applied and cannot get updated\n> concurrently until COMMIT.\n\nThere is no isolevel in MVCC where locking would be used\nimplicitly. We could use LOCK IN SHARE or SELECT FOR UPDATE \n(FOR SHARE LOCK ?) - using rules for RI is like implementing\nRI on applic level (!), - but this is bad.\n\nFortunately, there is a way without each row/table locking \nif scans added by RI rule could see uncommitted chages \nmade by concurrent xactions. Unique btree code already use\nspecial SnapshotDirty to see uncommitted changes and\navoid long-term row/page locking. With this Snapshot\nHeapTupleSatisfies returns true if\n\nt_xmin committed and (t_xmax is invalid OR is in-progress)\n\tOR\nt_xmin is in-progress and t_xmax is invalid\n\n- so, caller can wait (just like the same row writers do -\nby locking in-progress xaction ID in transaction pseudo-table)\nfor in-progress t_xmin/t_xmax xaction and decide what to do after \nconcurrent xaction COMMITs/ABORTs.\n\nBut before continuing with this approach I need in answer to\none question. Let's consider this case:\n\nXaction T1 with isolevel SERIALIZABLE inserts some row\ninto child table with deffered checking of primary key existance. \nThere were no primary key P for row inserted by T1 at the moment \nwhen T1 begun, but before T1 begins constraint checking another\nconcurrent xaction T2 inserts P and commits.\nAfter that T1 performs checking and - what? Will be the constraint\nsatisfied? I ask this because of all subsequent selects in T1\nwill not see P, but will see foreign key inserted - so, from the \npoint of application, child --> parent relationship will be broken...\n\nComments?\n\nCould someone run test below in Oracle?\n\n1. In session 1:\n\nCREATE TABLE p (x integer PRIMARY KEY);\nCREATE TABLE c (y integer REFERENCES p);\nINSERT INTO p VALUES (1);\n\n2. In session 2:\n\nSET TRANSACTION ISOLATION LEVEL SERIALIZABLE;\nSELECT * FROM p;\n-- empty\nINSERT INTO c VALUES (1);\n-- what? waits or rejects insertion?\n\n3. In session 1:\n\nCOMMIT;\n\n-- what in 1 if it was waiting?\n\n4. In session 2:\n\nINSERT INTO c VALUES (1);\n-- Ok or rejected?\n\nSELECT * FROM p;\nSELECT * FROM c;\nCOMMIT;\n\nTIA !!!\n\n> And it's a problem I've came across just writing this note\n> where MVCC already could have broken rewrite rule system\n> semantics.\n\nHow?\n\nVadim\n",
"msg_date": "Thu, 14 Jan 1999 22:51:07 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": "Vadim wrote:\n\n>\n> Jan Wieck wrote:\n> >\n> (Note that now in the case of UPDATE t_ctid of OLD tuples\n> points to TID of NEW tuples.)\n>\n> Two things define data visibility: SnapShot & CommandId.\n> We would have to save them for deffered rules and restore them\n> before run rule actions. But there is one issue: for what\n> scans old visibility should be used? There are scans from\n> user query and there are scans added by rule action. Ok,\n> let's assume that for added scans current visibility will be used\n> - this is what we need for RI rules (actually, something more -\n> see below).\n\n I addressed that problem (different visibility required for\n scans in one command) also in my other mail.\n\n Anyway, I just checked what happens in the following case:\n\n T1: begin;\n T1: select ...\n\n T2: update ...\n\n T1: select ... (gets the same (old) values)\n\n That's the result as long as T1 doesn't run in READ COMMITTED\n mode. And that's fine, because it doesn't have to worry\n about concurrent transactions of others.\n\n So the only problem left is the different visability. I think\n it is possible to change the visibility code not to check\n against the global command counter. Instead it might look at\n a command counter value in the range table entry related to\n the scan node. So the rewrite system and tcop could place\n the correct values there during query rewrite/processing.\n\n The range table of a rules rewritten parsetree is a\n combination of the range tables from the original user query,\n applied view rules and the the rule itself. For deferred\n rules, only the those coming with the rule action itself must\n have the command counter at COMMIT. All others must get the\n command counter value that is there when the query that fired\n this rule get's executed.\n\n The deferred querytrees can first be held in a new list of\n the rewritten querytree for the original user statement. The\n rewrite system puts into the rangetable entries\n USE_CURRENT_CMDID or USE_COMMIT_CMDID depending on where they\n are coming from.\n\n Before tcop calls the executor, a new function in the rewrite\n system is called to set the actual values for the command\n counter to use into the rangetable entries for one query and\n it's deferred ones. Then it adds all the deferred queries to\n the global deferred list and runs the query itself.\n\n At commit time, when all the deferred queries have to get\n run, those RTE's in them having USE_COMMIT_CMDID are set to\n the command counter at commit before running the plans.\n Voila.\n\n> > And it's a problem I've came across just writing this note\n> > where MVCC already could have broken rewrite rule system\n> > semantics.\n>\n> How?\n\n Yes it did!\n\n If a transaction runs in READ COMMITTED mode, the scan for\n the rules actions (performed first) could have different\n results than that for the original query (performed last).\n\n For now I see only one solution. READ COMMITTED is forbidden\n for anything that invokes non-view rules. This check must be\n done in the tcop and SPI, because saved SPI plans can be run\n without invoking the rewrite system at any time. So the plan\n must remember somewhere if READ COMMITTED is allowed for it\n or not.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Thu, 14 Jan 1999 20:32:34 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": "Jan Wieck wrote:\n> \n> > Two things define data visibility: SnapShot & CommandId.\n> > We would have to save them for deffered rules and restore them\n> > before run rule actions. But there is one issue: for what\n> > scans old visibility should be used? There are scans from\n> > user query and there are scans added by rule action. Ok,\n> > let's assume that for added scans current visibility will be used\n> > - this is what we need for RI rules (actually, something more -\n> > see below).\n> \n> I addressed that problem (different visibility required for\n> scans in one command) also in my other mail.\n> \n> Anyway, I just checked what happens in the following case:\n> \n> T1: begin;\n> T1: select ...\n> \n> T2: update ...\n> \n> T1: select ... (gets the same (old) values)\n> \n> That's the result as long as T1 doesn't run in READ COMMITTED\n> mode. And that's fine, because it doesn't have to worry\n> about concurrent transactions of others.\n> \n> So the only problem left is the different visability. I think\n> it is possible to change the visibility code not to check\n> against the global command counter. Instead it might look at\n> a command counter value in the range table entry related to\n> the scan node. So the rewrite system and tcop could place\n> the correct values there during query rewrite/processing.\n\nWhy you talk about CommandID only? What about SnapShot data?\nThe difference between scans in SERIALIZABLE/READ COMMITTED\nisolevels is that in SERIALIZABLE mode all queries use\nthe same SnapShot data (array of running xactions) \nand in READ COMMITTED mode new SnapShot data to use is created\nfor each query.\n\nCommandId defines visibility of self-changes.\nSnapShot defines visibility of concurrent changes.\n\n> \n> The range table of a rules rewritten parsetree is a\n> combination of the range tables from the original user query,\n> applied view rules and the the rule itself. For deferred\n> rules, only the those coming with the rule action itself must\n> have the command counter at COMMIT. All others must get the\n\nActually, not at COMMIT but when SET CONSTRAINT IMMEDIATE\nis called. COMMIT just imlicitly switches into immediate mode.\n\n> command counter value that is there when the query that fired\n> this rule get's executed.\n> \n> The deferred querytrees can first be held in a new list of\n> the rewritten querytree for the original user statement. The\n> rewrite system puts into the rangetable entries\n> USE_CURRENT_CMDID or USE_COMMIT_CMDID depending on where they\n> are coming from.\n> \n> Before tcop calls the executor, a new function in the rewrite\n> system is called to set the actual values for the command\n> counter to use into the rangetable entries for one query and\n> it's deferred ones. Then it adds all the deferred queries to\n> the global deferred list and runs the query itself.\n> \n> At commit time, when all the deferred queries have to get\n> run, those RTE's in them having USE_COMMIT_CMDID are set to\n> the command counter at commit before running the plans.\n> Voila.\n> \n> > > And it's a problem I've came across just writing this note\n> > > where MVCC already could have broken rewrite rule system\n> > > semantics.\n> >\n> > How?\n> \n> Yes it did!\n> \n> If a transaction runs in READ COMMITTED mode, the scan for\n> the rules actions (performed first) could have different\n> results than that for the original query (performed last).\n> \n> For now I see only one solution. READ COMMITTED is forbidden\n> for anything that invokes non-view rules. This check must be\n> done in the tcop and SPI, because saved SPI plans can be run\n> without invoking the rewrite system at any time. So the plan\n> must remember somewhere if READ COMMITTED is allowed for it\n> or not.\n\nREAD COMMITTED will be default mode when writers in this\nmode will be supported...\n\nThe solution is to use the same SnapShot data for both\naction' and original scans.\n\nBut there are other abilities for inconsistances in READ COMMITTED\nmode:\n\ncreate table t (x int);\ncreate table tlog (xold int, xnew int);\ninsert into t values (1);\ncreate rule r as on update to t do \ninsert into tlog values (old.x, new.x);\n\nNow consider two concurrent\n\nupdate t set x = 2 where x = 1;\n\tand\nupdate t set x = 3 where x = 1;\n\n: two rows will be inserted into tlog - (1,2) and (1,3) -\nby rule actions run BEFORE original queries, but actually \nonly one update will be succeeded - another one will see\nnot 1 in t.x after first update commit and so row will not \nbe updated by second update...\n\nOne approach is to force FOR UPDATE OF \"OLD\"-table\nin action' INSERT INTO ... SELECT, another one is \nto run rule action AFTER original query (wouldn't\nit be possible having ability to directly set \nscan visibilities?).\n\nActually, for non-locking RI implementation (using dirty read) \nconstraint checking must occure AFTER data changes are made \n(i.e. deffered untill query end): this will allow PK deleters\nnotice that there are concurrent FK inserters, UK inserters see \nthat someone else tries to insert the same key, etc, wait for \nconcurrent COMMIT/ABORT and make appropriate things after that.\n\nMore letters will follow, but maybe after week-end only...\n\nVadim\n",
"msg_date": "Fri, 15 Jan 1999 12:58:58 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
}
] |
[
{
"msg_contents": "I was wondering who maintains the postgres mailing lists. I am wondering because I would like to\nsee the subject index added the digest versions of the following lists:\n\npgsql-sql\npgsql-general\npgsql-admin\n\nThanks for your time\n---------\nChris Williams\nSterling Software\nRome, New York\nPhone: (315) 336-0500\nEmail: [email protected]\n\n",
"msg_date": "Tue, 12 Jan 1999 08:36:32 -0500",
"msg_from": "\"Chris Williams\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Who is the maintainer of ALL the postgres Mailing lists"
},
{
"msg_contents": "\nDone...\n\n\nOn Tue, 12 Jan 1999, Chris Williams wrote:\n\n> I was wondering who maintains the postgres mailing lists. I am wondering because I would like to\n> see the subject index added the digest versions of the following lists:\n> \n> pgsql-sql\n> pgsql-general\n> pgsql-admin\n> \n> Thanks for your time\n> ---------\n> Chris Williams\n> Sterling Software\n> Rome, New York\n> Phone: (315) 336-0500\n> Email: [email protected]\n> \n> \n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Tue, 12 Jan 1999 10:09:06 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Who is the maintainer of ALL the postgres Mailing lists"
}
] |
[
{
"msg_contents": "I have just encountered some applications that really need\ntransactions and so have been perusing the transaction statements and\nthe lock man page. Thinking of the possibility of deadlocks if two\nprocesses try to acquire locks in opposite order suggested a solution.\n\nCouldn't the parser syntax be expanded to\n\n\t LOCK [TABLE] table1 [, table2 [, table3 [...]]]\n\nin which case locks on the entire group of tables could be obtained\natomically. If one fails, the process should release locks on all the\nrest, wait a bit, and retry. This should prevent infinite deadlocks\nsince all locks (not just the most recent one of several independent\nlocks) would be released at some point, allowing other processes to\nassert theirs.\n\nI say all this based only on the man page and my naive understanding\nof how the locking code actually works. Perhaps the man page doesn't\nreally reflect the actual deadlock problem, though.\n\nCheers,\nBrook\n",
"msg_date": "Tue, 12 Jan 1999 08:13:43 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "lock deadlocks"
},
{
"msg_contents": "> I have just encountered some applications that really need\n> transactions and so have been perusing the transaction statements and\n> the lock man page. Thinking of the possibility of deadlocks if two\n> processes try to acquire locks in opposite order suggested a solution.\n> \n> Couldn't the parser syntax be expanded to\n> \n> \t LOCK [TABLE] table1 [, table2 [, table3 [...]]]\n> \n> in which case locks on the entire group of tables could be obtained\n> atomically. If one fails, the process should release locks on all the\n> rest, wait a bit, and retry. This should prevent infinite deadlocks\n> since all locks (not just the most recent one of several independent\n> locks) would be released at some point, allowing other processes to\n> assert theirs.\n\nYou give a nice extension of the LOCK statement, that is quite valid,\n_and_ can not be simulated with multiple lock statements.\n\nComplex kernel locking systems, mostly multi-cpu kernels, have to do\nsimilar things. You want an 'all or nothing' lock statement. I will\nadd this to the TODO list.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 12 Jan 1999 12:25:09 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] lock deadlocks"
}
] |
[
{
"msg_contents": "[Sorry if this sent twice. I never saw my copy the first time.]\n\nDoes this seem right?\n\ndruid=> SELECT COUNT(*) FROM acctrans;\ncount\n-----\n 0\n(1 row)\n\ndruid=> SELECT client_id, SUM(tramount) FROM acctrans GROUP BY client_id;\nclient_id|sum\n---------+---\n | \n(1 row)\n\nIf there are no rows in the table then shouldn't the result be no rows\nwhen GROUP BY is used? Further, What about this?\n\ndruid=> SELECT SUM(tramount) FROM acctrans;\nsum\n---\n \n(1 row)\n\nShouldn't that be 0.00?\n\nWhat will the NUMERIC or DECIMAL types do in these situations? It\nlooks like INTEGER has the same behaviour as MONEY (which tramount is.)\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Tue, 12 Jan 1999 10:38:51 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "SUM() and GROUP BY"
}
] |
[
{
"msg_contents": "Does this seem right?\n\ndruid=> SELECT COUNT(*) FROM acctrans;\ncount\n-----\n 0\n(1 row)\n\ndruid=> SELECT client_id, SUM(tramount) FROM acctrans GROUP BY client_id;\nclient_id|sum\n---------+---\n | \n(1 row)\n\nIf there are no rows in the table then shouldn't the result be no rows\nwhen GROUP BY is used? Further, What about this?\n\ndruid=> SELECT SUM(tramount) FROM acctrans;\nsum\n---\n \n(1 row)\n\nShouldn't that be 0.00?\n\nWhat will the NUMERIC or DECIMAL types do in these situations? It\nlooks like INTEGER has the same behaviour as MONEY (which tramount is.)\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Tue, 12 Jan 1999 21:43:36 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "SUM() and GROUP BY"
},
{
"msg_contents": "> Does this seem right?\n> druid=> SELECT client_id, SUM(tramount)\n> FROM acctrans GROUP BY client_id;\n> client_id|sum\n> ---------+---\n> |\n> (1 row)\n> If there are no rows in the table then shouldn't the result be no rows\n> when GROUP BY is used? Further, What about this?\n\nNot sure. Someone may want to try this query on another DB. I know the\nanswer to the next one though...\n\n> druid=> SELECT SUM(tramount) FROM acctrans;\n> sum\n> ---\n> \n> (1 row)\n> Shouldn't that be 0.00?\n\nNo. It is returning NULL, because NULL means \"don't know\". It doesn't\nmean \"nothing\" or \"zero\". That is certainly the correct behavior if the\ntable were populated with all NULLs in that column. And by extension, it\nis the correct result if there are no rows at all, since \"don't know\"\nfor a bunch should give the same result as \"don't know\" for a few or for\nnone.\n\n> What will the NUMERIC or DECIMAL types do in these situations? It\n> looks like INTEGER has the same behaviour as MONEY (which tramount \n> is.)\n\nThey will all behave the same.\n\n - Tom\n",
"msg_date": "Wed, 13 Jan 1999 06:18:42 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY"
},
{
"msg_contents": "D'Arcy J.M. Cain wrote:\n\n>\n> Does this seem right?\n>\n> druid=> SELECT COUNT(*) FROM acctrans;\n> count\n> -----\n> 0\n> (1 row)\n>\n> druid=> SELECT client_id, SUM(tramount) FROM acctrans GROUP BY client_id;\n> client_id|sum\n> ---------+---\n> |\n> (1 row)\n>\n> If there are no rows in the table then shouldn't the result be no rows\n> when GROUP BY is used? Further, What about this?\n>\n> druid=> SELECT SUM(tramount) FROM acctrans;\n> sum\n> ---\n>\n> (1 row)\n>\n> Shouldn't that be 0.00?\n>\n> What will the NUMERIC or DECIMAL types do in these situations? It\n> looks like INTEGER has the same behaviour as MONEY (which tramount is.)\n\n NUMERIC and DECIMAL will behave exactly as above, because\n it's the (irritating) correct behaviour. It is handled in the\n generic grouping and aggregate code (in fact none of the\n aggregate functions will ever be called if there isn't\n anything to count/average/sum).\n\n To get a zero count, you need a subselect in the targetlist\n (not implemented yet). Currently the only way to simulate a\n subselect in the targetlist is to put the count() into an SQL\n function that takes the arguments you need to build the\n qualification and returns the counted number.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Wed, 13 Jan 1999 12:20:51 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY"
},
{
"msg_contents": "\"Thomas G. Lockhart\" <[email protected]> writes:\n>> druid=> SELECT SUM(tramount) FROM acctrans;\n>> sum\n>> ---\n>> \n>> (1 row)\n>> Shouldn't that be 0.00?\n\n> No. It is returning NULL, because NULL means \"don't know\". It doesn't\n> mean \"nothing\" or \"zero\". That is certainly the correct behavior if the\n> table were populated with all NULLs in that column. And by extension, it\n> is the correct result if there are no rows at all, since \"don't know\"\n> for a bunch should give the same result as \"don't know\" for a few or for\n> none.\n\nI disagree ... the sum of zero items has traditionally been defined as\nzero by any mathematician you care to ask. No logical problems are\nintroduced by doing so, and it avoids an unpleasant special case that\napplications would otherwise be forced to deal with. (Example: if\nD'Arcy's tramount column has been declared NOT NULL, then it seems to me\nthat his code is entitled to expect to get a non-NULL result from SUM().\nHe should not have to cope with a NULL just because the table is empty.)\n\nNow, if the sum is taken over a set of rows that includes at least *one*\nNULL, then I agree that SUM should return NULL: you \"don't know\" what\nthe sum ought to be if there are some inputs that you \"don't know\".\nBut we do know what the sum of no items is: it's zero. There are no\nunknown inputs and therefore no reason to claim we don't know the sum.\n\nIOW, D'Arcy's right. This is a bug.\n\nYou do have to take it on a case-by-case basis, though. For example\nAVG() of no items should (and does) return NULL, because the average\nof no items is not defined (since 0/0 is indeterminate).\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 13 Jan 1999 10:12:01 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY "
},
{
"msg_contents": "Thus spake Tom Lane\n> \"Thomas G. Lockhart\" <[email protected]> writes:\n> I disagree ... the sum of zero items has traditionally been defined as\n> zero by any mathematician you care to ask. No logical problems are\n> introduced by doing so, and it avoids an unpleasant special case that\n> applications would otherwise be forced to deal with. (Example: if\n> D'Arcy's tramount column has been declared NOT NULL, then it seems to me\n> that his code is entitled to expect to get a non-NULL result from SUM().\n> He should not have to cope with a NULL just because the table is empty.)\n\nOr, if I add a \"WHERE tramount IS NOT NULL\" clause to force it to add\nnon-null items, right? It should depend on the values in the selected\nset, not the values in the full table.\n\n> You do have to take it on a case-by-case basis, though. For example\n> AVG() of no items should (and does) return NULL, because the average\n> of no items is not defined (since 0/0 is indeterminate).\n\nWhat about GROUP BY though.\n\nSELECT client_id, AVG(tramount) FROM acctrans\n WHERE client_id = 'NOBODY' GROUP BY client_id;\n\nIt's not that I want a special value returned, I want no rows returned.\nCurrently it returns 1 row even thought 'NOBODY' doesn't exist in the\ndatabase.\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Wed, 13 Jan 1999 11:13:56 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY"
},
{
"msg_contents": "\"D'Arcy\" \"J.M.\" Cain <[email protected]> writes:\n> Or, if I add a \"WHERE tramount IS NOT NULL\" clause to force it to add\n> non-null items, right? It should depend on the values in the selected\n> set, not the values in the full table.\n\nSure. No items is no items, no matter how you selected them.\n\n> What about GROUP BY though.\n>\n> SELECT client_id, AVG(tramount) FROM acctrans\n> WHERE client_id = 'NOBODY' GROUP BY client_id;\n>\n> It's not that I want a special value returned, I want no rows returned.\n> Currently it returns 1 row even thought 'NOBODY' doesn't exist in the\n> database.\n\nHmm. I tried:\n\nplay=> create table empty (i int4);\n\nCASE 1:\n\nplay=> select i from empty;\ni\n-\n(0 rows)\n\nNo rows returned, certainly correct.\n\nCASE 2:\n\nplay=> select avg(i) from empty;\navg\n---\n\n(1 row)\n\nOne row containing the null aggregate-function result, also OK.\n\nCASE 3:\n\nplay=> select i from empty group by i;\ni\n-\n(0 rows)\n\nNo groups, no rows returned, OK.\n\nCASE 4:\n\nplay=> select avg(i) from empty group by i;\navg\n---\n\n(1 row)\n\nI suppose you could argue that this is consistent with cases 2 and 3,\nin a weird way, but it's certainly not the way I'd expect it to work.\nIf there are no groups created by GROUP BY, then AVG should never be\ninvoked at all, therefore there should be no rows returned. This is not\nthe same as case 2, wherein AVG was invoked once over the whole table\n(which just happened to be empty, causing AVG to return NULL as it\nshould for an empty input set). Here, AVG should be invoked once for\neach group created by GROUP BY. If no groups, there should be no\nresults, not an artificially created row with a null result.\n\nThe point is perhaps more obvious if you consider\n\nplay=> select i,avg(i) from empty group by i;\ni|avg\n-+---\n |\n(1 row)\n\nHere, the system has made up a NULL value for i out of thin air.\nIt will do this even if the i column is declared NOT NULL!\nI say this behavior is completely unsupportable. (If the table\nactually *contains* some NULL entries, then returning a row with\nNULL i is correct --- and indeed the system handles that case properly.\nGROUP BY's behavior for an empty table is erroneous because it makes\nit look like the table contains NULL i values when it doesn't.)\n\nSo, again I agree with D'Arcy. But I'm not the one who might have\nto try to fix this...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 13 Jan 1999 12:38:51 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY "
},
{
"msg_contents": "> I disagree ... the sum of zero items has traditionally been defined as\n> zero by any mathematician you care to ask. No logical problems are\n> introduced by doing so, and it avoids an unpleasant special case that\n> applications would otherwise be forced to deal with. (Example: if\n> D'Arcy's tramount column has been declared NOT NULL, then it seems to me\n> that his code is entitled to expect to get a non-NULL result from SUM().\n> He should not have to cope with a NULL just because the table is empty.)\n\nInformix returns NULL for sum. It returns a zero only for count().\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 13 Jan 1999 12:43:07 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY"
},
{
"msg_contents": "> I suppose you could argue that this is consistent with cases 2 and 3,\n> in a weird way, but it's certainly not the way I'd expect it to work.\n> If there are no groups created by GROUP BY, then AVG should never be\n> invoked at all, therefore there should be no rows returned.\n\nAgreed.\n\n> So, again I agree with D'Arcy.\n\nI'm missing something. Is there another issue for GROUP BY for which we\ndon't have a consensus? An aggregate on an entire column can return\nNULL, and aggregates on columns with GROUP BY columns of NULL should\nnot.\n\n> But I'm not the one who might have\n> to try to fix this...\n\nAnd why not? :)\n\n - Tom\n",
"msg_date": "Thu, 14 Jan 1999 06:11:02 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY"
},
{
"msg_contents": "Thus spake Thomas G. Lockhart\n> > So, again I agree with D'Arcy.\n> \n> I'm missing something. Is there another issue for GROUP BY for which we\n> don't have a consensus? An aggregate on an entire column can return\n> NULL, and aggregates on columns with GROUP BY columns of NULL should\n> not.\n\nWell, I still think that an aggregate on a whole column when no rows\nare selected (None exist or WHERE clause deselects everything) should\nreturn 0 rows, not NULL but the above would be better than what happens\nnow.\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Thu, 14 Jan 1999 05:38:23 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] SUM() and GROUP BY"
}
] |
[
{
"msg_contents": "\n\t>druid=> SELECT client_id, SUM(tramount) FROM acctrans GROUP BY\n>client_id;\n\t>client_id|sum\n\t>---------+---\n\t> | \n\t>(1 row)\n\t>\n\t>If there are no rows in the table then shouldn't the result be no\nrows\n\t>when GROUP BY is used? Further, What about this?\n\nYes, should return no rows. seems like a fix is needed here :-(\n\nAndreas\n\n",
"msg_date": "Wed, 13 Jan 1999 09:07:08 +0100",
"msg_from": "Zeugswetter Andreas IZ5 <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] SUM() and GROUP BY"
}
] |
[
{
"msg_contents": "As it stands in 6.4.2 pg_dump does not dump sequences created by the\nSERIAL datatype when the -t tablename option is used. The following\npatch fixes that (with a couple of cosmetic cleanups) so that whenever\nthe -t option is used, the appropriate *_id_seq sequence is also\ndumped if it exists.\n\nThe second patch just fixes a bunch of cosmetic details concerning\nalignment and changes nothing substantive in the code.\n\nCheers,\nBrook\n\n===========================================================================\n\n--- bin/pg_dump/pg_dump.c.orig\tWed Jan 13 09:37:40 1999\n+++ bin/pg_dump/pg_dump.c\tWed Jan 13 10:08:35 1999\n@@ -2593,17 +2592,26 @@\n \tint\t\t\ti,\n \t\t\t\tj,\n \t\t\t\tk;\n-\tchar\t\tq[MAXQUERYLEN];\n-\tchar\t **parentRels;\t\t/* list of names of parent relations */\n+\tchar\t\t\tq[MAXQUERYLEN];\n+\tchar\t\t\t*serialSeq = NULL;\t\t/* implicit sequence name created by SERIAL datatype */\n+\tconst char\t\t*serialSeqSuffix = \"_id_seq\";\t/* suffix for implicit SERIAL sequences */\n+\tchar\t\t\t**parentRels;\t\t\t/* list of names of parent relations */\n \tint\t\t\tnumParents;\n-\tint\t\t\tactual_atts;\t/* number of attrs in this CREATE statment */\n+\tint\t\t\tactual_atts;\t\t\t/* number of attrs in this CREATE statment */\n \n \t/* First - dump SEQUENCEs */\n+\tif (tablename)\n+\t {\n+\t serialSeq = malloc (strlen (tablename) + strlen (serialSeqSuffix) + 1);\n+\t strcpy (serialSeq, tablename);\n+\t strcat (serialSeq, serialSeqSuffix);\n+\t }\n \tfor (i = 0; i < numTables; i++)\n \t{\n \t\tif (!(tblinfo[i].sequence))\n \t\t\tcontinue;\n-\t\tif (!tablename || (!strcmp(tblinfo[i].relname, tablename)))\n+\t\tif (!tablename || (!strcmp(tblinfo[i].relname, tablename))\n+\t\t || (serialSeq && !strcmp(tblinfo[i].relname,serialSeq)))\n \t\t{\n \t\t\tbecomeUser(fout, tblinfo[i].usename);\n \t\t\tdumpSequence(fout, tblinfo[i]);\n@@ -2611,6 +2619,8 @@\n \t\t\t\tdumpACL(fout, tblinfo[i]);\n \t\t}\n \t}\n+\tif (tablename)\n+\t free (serialSeq);\n \n \tfor (i = 0; i < numTables; i++)\n \t{\n\n===========================================================================\n\n--- bin/pg_dump/pg_dump.c.orig\tWed Jan 13 09:37:40 1999\n+++ bin/pg_dump/pg_dump.c\tWed Jan 13 10:08:35 1999\n@@ -56,7 +56,7 @@\n #include <stdio.h>\n #include <string.h>\n #include <ctype.h>\n-#include <sys/param.h>\t\t\t/* for MAXHOSTNAMELEN on most */\n+#include <sys/param.h>\t\t\t\t/* for MAXHOSTNAMELEN on most */\n #ifdef solaris_sparc\n #include <netdb.h>\t\t\t\t/* for MAXHOSTNAMELEN on some */\n #endif\n@@ -96,25 +96,25 @@\n static char *GetPrivileges(const char *s);\n static void becomeUser(FILE *fout, const char *username);\n \n-extern char *optarg;\n+extern char\t*optarg;\n extern int\toptind,\n-\t\t\topterr;\n+\t\topterr;\n \n /* global decls */\n bool\t\tg_verbose;\t\t\t/* User wants verbose narration of our\n-\t\t\t\t\t\t\t\t * activities. */\n-int\t\t\tg_last_builtin_oid; /* value of the last builtin oid */\n-FILE\t *g_fout;\t\t\t\t/* the script file */\n-PGconn\t *g_conn;\t\t\t\t/* the database connection */\n-\n-bool\t\tforce_quotes;\t\t/* User wants to suppress double-quotes */\n-int\t\t\tdumpData;\t\t\t/* dump data using proper insert strings */\n-int\t\t\tattrNames;\t\t\t/* put attr names into insert strings */\n-int\t\t\tschemaOnly;\n-int\t\t\tdataOnly;\n-int\t\t\taclsOption;\n+\t\t\t\t\t\t * activities. */\n+int\t\tg_last_builtin_oid;\t\t/* value of the last builtin oid */\n+FILE\t\t*g_fout;\t\t\t/* the script file */\n+PGconn\t\t*g_conn;\t\t\t/* the database connection */\n+\n+bool\t\tforce_quotes;\t\t\t/* User wants to suppress double-quotes */\n+int\t\tdumpData;\t\t\t/* dump data using proper insert strings */\n+int\t\tattrNames;\t\t\t/* put attr names into insert strings */\n+int\t\tschemaOnly;\n+int\t\tdataOnly;\n+int\t\taclsOption;\n \n-char\t\tg_opaque_type[10];\t/* name for the opaque type */\n+char\t\tg_opaque_type[10];\t\t/* name for the opaque type */\n \n /* placeholders for the delimiters for comments */\n char\t\tg_comment_start[10];\n@@ -179,8 +179,8 @@\n static bool\n isViewRule(char *relname)\n {\n-\tPGresult *res;\n-\tint\t\t\tntups;\n+\tPGresult\t*res;\n+\tint\t\tntups;\n \tchar\t\tquery[MAXQUERYLEN];\n \n \tres = PQexec(g_conn, \"begin\");\n@@ -315,13 +315,13 @@\n \t\t\t\t\t const TableInfo tblinfo, bool oids)\n {\n \n-\tPGresult *res;\n+\tPGresult\t*res;\n \tchar\t\tquery[255];\n-\tint\t\t\tactual_atts;\t/* number of attrs in this a table */\n+\tint\t\tactual_atts;\t/* number of attrs in this a table */\n \tchar\t\texpandbuf[COPYBUFSIZ];\n \tchar\t\tq[MAXQUERYLEN];\n-\tint\t\t\ttuple;\n-\tint\t\t\tfield;\n+\tint\t\ttuple;\n+\tint\t\tfield;\n \n \tsprintf(query, \"SELECT * FROM %s\", fmtId(classname, force_quotes));\n \tres = PQexec(g_conn, query);\n@@ -421,7 +421,7 @@\n {\n \n \tint\t\t\ti;\n-\tchar\t *all_only;\n+\tchar\t\t\t*all_only;\n \n \tif (onlytable == NULL)\n \t\tall_only = \"all\";\n@@ -482,12 +482,11 @@\n prompt_for_password(char *username, char *password)\n {\n \tchar\t\tbuf[512];\n-\tint\t\t\tlength;\n+\tint\t\tlength;\n \n #ifdef HAVE_TERMIOS_H\n-\tstruct termios t_orig,\n-\t\t\t\tt;\n-\n+\tstruct termios\tt_orig,\n+\t\t\tt;\n #endif\n \n \tprintf(\"Username: \");\n@@ -535,21 +534,21 @@\n int\n main(int argc, char **argv)\n {\n-\tint\t\t\tc;\n-\tconst char *progname;\n-\tconst char *filename = NULL;\n-\tconst char *dbname = NULL;\n-\tconst char *pghost = NULL;\n-\tconst char *pgport = NULL;\n-\tchar\t *tablename = NULL;\n-\tint\t\t\toids = 0;\n-\tTableInfo *tblinfo;\n-\tint\t\t\tnumTables;\n+\tint\t\tc;\n+\tconst char\t*progname;\n+\tconst char\t*filename = NULL;\n+\tconst char\t*dbname = NULL;\n+\tconst char\t*pghost = NULL;\n+\tconst char\t*pgport = NULL;\n+\tchar\t\t*tablename = NULL;\n+\tint\t\toids = 0;\n+\tTableInfo\t*tblinfo;\n+\tint\t\tnumTables;\n \tchar\t\tconnect_string[512] = \"\";\n \tchar\t\ttmp_string[128];\n \tchar\t\tusername[100];\n \tchar\t\tpassword[100];\n-\tint\t\t\tuse_password = 0;\n+\tint\t\tuse_password = 0;\n \n \tg_verbose = false;\n \tforce_quotes = true;\n@@ -727,11 +726,11 @@\n TypeInfo *\n getTypes(int *numTypes)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n-\tTypeInfo *tinfo;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n+\tTypeInfo\t\t*tinfo;\n \n \tint\t\t\ti_oid;\n \tint\t\t\ti_typowner;\n@@ -856,12 +855,12 @@\n OprInfo *\n getOperators(int *numOprs)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n+\tchar\t\t\tquery[MAXQUERYLEN];\n \n-\tOprInfo *oprinfo;\n+\tOprInfo\t\t\t*oprinfo;\n \n \tint\t\t\ti_oid;\n \tint\t\t\ti_oprname;\n@@ -1199,11 +1198,11 @@\n AggInfo *\n getAggregates(int *numAggs)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n-\tAggInfo *agginfo;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n+\tAggInfo\t\t\t*agginfo;\n \n \tint\t\t\ti_oid;\n \tint\t\t\ti_aggname;\n@@ -1293,11 +1292,11 @@\n FuncInfo *\n getFuncs(int *numFuncs)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n-\tFuncInfo *finfo;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n+\tFuncInfo\t\t*finfo;\n \n \tint\t\t\ti_oid;\n \tint\t\t\ti_proname;\n@@ -1393,11 +1392,11 @@\n TableInfo *\n getTables(int *numTables, FuncInfo *finfo, int numFuncs)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n-\tTableInfo *tblinfo;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n+\tTableInfo\t\t*tblinfo;\n \n \tint\t\t\ti_oid;\n \tint\t\t\ti_relname;\n@@ -1726,11 +1725,11 @@\n InhInfo *\n getInherits(int *numInherits)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n-\tInhInfo *inhinfo;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n+\tInhInfo\t\t\t*inhinfo;\n \n \tint\t\t\ti_inhrel;\n \tint\t\t\ti_inhparent;\n@@ -1791,13 +1790,13 @@\n {\n \tint\t\t\ti,\n \t\t\t\tj;\n-\tchar\t\tq[MAXQUERYLEN];\n+\tchar\t\t\tq[MAXQUERYLEN];\n \tint\t\t\ti_attname;\n \tint\t\t\ti_typname;\n \tint\t\t\ti_atttypmod;\n \tint\t\t\ti_attnotnull;\n \tint\t\t\ti_atthasdef;\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \n \tfor (i = 0; i < numTables; i++)\n@@ -1902,10 +1901,10 @@\n getIndices(int *numIndices)\n {\n \tint\t\t\ti;\n-\tchar\t\tquery[MAXQUERYLEN];\n-\tPGresult *res;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n-\tIndInfo *indinfo;\n+\tIndInfo\t\t\t*indinfo;\n \n \tint\t\t\ti_indexrelname;\n \tint\t\t\ti_indrelname;\n@@ -1994,7 +1993,7 @@\n \t\t TypeInfo *tinfo, int numTypes)\n {\n \tint\t\t\ti;\n-\tchar\t\tq[MAXQUERYLEN];\n+\tchar\t\t\tq[MAXQUERYLEN];\n \tint\t\t\tfuncInd;\n \n \tfor (i = 0; i < numTypes; i++)\n@@ -2167,7 +2166,7 @@\n \t\t\tTypeInfo *tinfo, int numTypes)\n {\n \tchar\t\tq[MAXQUERYLEN];\n-\tint\t\t\tj;\n+\tint\t\tj;\n \tchar\t\t*func_def;\n \tchar\t\tfunc_lang[NAMEDATALEN + 1];\n \n@@ -2268,7 +2267,7 @@\n dumpOprs(FILE *fout, OprInfo *oprinfo, int numOperators,\n \t\t TypeInfo *tinfo, int numTypes)\n {\n-\tint\t\t\ti;\n+\tint\t\ti;\n \tchar\t\tq[MAXQUERYLEN];\n \tchar\t\tleftarg[MAXQUERYLEN];\n \tchar\t\trightarg[MAXQUERYLEN];\n@@ -2375,14 +2374,14 @@\n dumpAggs(FILE *fout, AggInfo *agginfo, int numAggs,\n \t\t TypeInfo *tinfo, int numTypes)\n {\n-\tint\t\t\ti;\n+\tint\t\ti;\n \tchar\t\tq[MAXQUERYLEN];\n \tchar\t\tsfunc1[MAXQUERYLEN];\n \tchar\t\tsfunc2[MAXQUERYLEN];\n \tchar\t\tbasetype[MAXQUERYLEN];\n \tchar\t\tfinalfunc[MAXQUERYLEN];\n \tchar\t\tcomma1[2],\n-\t\t\t\tcomma2[2];\n+\t\t\tcomma2[2];\n \n \tfor (i = 0; i < numAggs; i++)\n \t{\n@@ -2514,9 +2513,9 @@\n {\n \tconst char *acls = tbinfo.relacl;\n \tchar\t *aclbuf,\n-\t\t\t *tok,\n-\t\t\t *eqpos,\n-\t\t\t *priv;\n+\t\t *tok,\n+\t\t *eqpos,\n+\t\t *priv;\n \n \tif (strlen(acls) == 0)\n \t\treturn;\t\t\t\t\t/* table has default permissions */\n@@ -2726,15 +2736,15 @@\n \tint\t\t\ti,\n \t\t\t\tk;\n \tint\t\t\ttableInd;\n-\tchar\t\tattlist[1000];\n-\tchar\t *classname[INDEX_MAX_KEYS];\n-\tchar\t *funcname;\t\t/* the name of the function to comput the\n-\t\t\t\t\t\t\t\t * index key from */\n+\tchar\t\t\tattlist[1000];\n+\tchar\t\t\t*classname[INDEX_MAX_KEYS];\n+\tchar\t\t\t*funcname;\t\t/* the name of the function to comput the\n+\t\t\t\t\t\t\t * index key from */\n \tint\t\t\tindkey,\n \t\t\t\tindclass;\n \tint\t\t\tnclass;\n \n-\tchar\t\tq[MAXQUERYLEN],\n+\tchar\t\t\tq[MAXQUERYLEN],\n \t\t\t\tid1[MAXQUERYLEN],\n \t\t\t\tid2[MAXQUERYLEN];\n \tPGresult *res;\n@@ -2887,7 +2897,7 @@\n \t\t\t\tk;\n \tint\t\t\tm,\n \t\t\t\tn;\n-\tchar\t **outVals = NULL; /* values to copy out */\n+\tchar\t\t\t**outVals = NULL; /* values to copy out */\n \n \tn = PQntuples(res);\n \tm = PQnfields(res);\n@@ -2940,7 +2950,7 @@\n static void\n setMaxOid(FILE *fout)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tOid\t\t\tmax_oid;\n \n \tres = PQexec(g_conn, \"CREATE TABLE pgdump_oid (dummy int4)\");\n@@ -2993,7 +3003,7 @@\n static int\n findLastBuiltinOid(void)\n {\n-\tPGresult *res;\n+\tPGresult\t\t*res;\n \tint\t\t\tntups;\n \tint\t\t\tlast_oid;\n \n@@ -3025,9 +3035,9 @@\n static char *\n checkForQuote(const char *s)\n {\n-\tchar\t *r;\n+\tchar\t\t*r;\n \tchar\t\tc;\n-\tchar\t *result;\n+\tchar\t\t*result;\n \n \tint\t\t\tj = 0;\n \n@@ -3056,16 +3066,16 @@\n static void\n dumpSequence(FILE *fout, TableInfo tbinfo)\n {\n-\tPGresult *res;\n-\tint4\t\tlast,\n+\tPGresult\t\t*res;\n+\tint4\t\t\tlast,\n \t\t\t\tincby,\n \t\t\t\tmaxv,\n \t\t\t\tminv,\n \t\t\t\tcache;\n-\tchar\t\tcycled,\n+\tchar\t\t\tcycled,\n \t\t\t\tcalled,\n-\t\t\t *t;\n-\tchar\t\tquery[MAXQUERYLEN];\n+\t\t\t\t*t;\n+\tchar\t\t\tquery[MAXQUERYLEN];\n \n \tsprintf(query,\n \t\t\t\"SELECT sequence_name, last_value, increment_by, max_value, \"\n",
"msg_date": "Wed, 13 Jan 1999 10:24:19 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "fix pg_dump to dump sequences created by SERIAL datatype"
},
{
"msg_contents": "Applied.\n\n> As it stands in 6.4.2 pg_dump does not dump sequences created by the\n> SERIAL datatype when the -t tablename option is used. The following\n> patch fixes that (with a couple of cosmetic cleanups) so that whenever\n> the -t option is used, the appropriate *_id_seq sequence is also\n> dumped if it exists.\n> \n> The second patch just fixes a bunch of cosmetic details concerning\n> alignment and changes nothing substantive in the code.\n> \n> Cheers,\n> Brook\n> \n> ===========================================================================\n> \n> --- bin/pg_dump/pg_dump.c.orig\tWed Jan 13 09:37:40 1999\n> +++ bin/pg_dump/pg_dump.c\tWed Jan 13 10:08:35 1999\n> @@ -2593,17 +2592,26 @@\n> \tint\t\t\ti,\n> \t\t\t\tj,\n> \t\t\t\tk;\n> -\tchar\t\tq[MAXQUERYLEN];\n> -\tchar\t **parentRels;\t\t/* list of names of parent relations */\n> +\tchar\t\t\tq[MAXQUERYLEN];\n> +\tchar\t\t\t*serialSeq = NULL;\t\t/* implicit sequence name created by SERIAL datatype */\n> +\tconst char\t\t*serialSeqSuffix = \"_id_seq\";\t/* suffix for implicit SERIAL sequences */\n> +\tchar\t\t\t**parentRels;\t\t\t/* list of names of parent relations */\n> \tint\t\t\tnumParents;\n> -\tint\t\t\tactual_atts;\t/* number of attrs in this CREATE statment */\n> +\tint\t\t\tactual_atts;\t\t\t/* number of attrs in this CREATE statment */\n> \n> \t/* First - dump SEQUENCEs */\n> +\tif (tablename)\n> +\t {\n> +\t serialSeq = malloc (strlen (tablename) + strlen (serialSeqSuffix) + 1);\n> +\t strcpy (serialSeq, tablename);\n> +\t strcat (serialSeq, serialSeqSuffix);\n> +\t }\n> \tfor (i = 0; i < numTables; i++)\n> \t{\n> \t\tif (!(tblinfo[i].sequence))\n> \t\t\tcontinue;\n> -\t\tif (!tablename || (!strcmp(tblinfo[i].relname, tablename)))\n> +\t\tif (!tablename || (!strcmp(tblinfo[i].relname, tablename))\n> +\t\t || (serialSeq && !strcmp(tblinfo[i].relname,serialSeq)))\n> \t\t{\n> \t\t\tbecomeUser(fout, tblinfo[i].usename);\n> \t\t\tdumpSequence(fout, tblinfo[i]);\n> @@ -2611,6 +2619,8 @@\n> \t\t\t\tdumpACL(fout, tblinfo[i]);\n> \t\t}\n> \t}\n> +\tif (tablename)\n> +\t free (serialSeq);\n> \n> \tfor (i = 0; i < numTables; i++)\n> \t{\n> \n> ===========================================================================\n> \n> --- bin/pg_dump/pg_dump.c.orig\tWed Jan 13 09:37:40 1999\n> +++ bin/pg_dump/pg_dump.c\tWed Jan 13 10:08:35 1999\n> @@ -56,7 +56,7 @@\n> #include <stdio.h>\n> #include <string.h>\n> #include <ctype.h>\n> -#include <sys/param.h>\t\t\t/* for MAXHOSTNAMELEN on most */\n> +#include <sys/param.h>\t\t\t\t/* for MAXHOSTNAMELEN on most */\n> #ifdef solaris_sparc\n> #include <netdb.h>\t\t\t\t/* for MAXHOSTNAMELEN on some */\n> #endif\n> @@ -96,25 +96,25 @@\n> static char *GetPrivileges(const char *s);\n> static void becomeUser(FILE *fout, const char *username);\n> \n> -extern char *optarg;\n> +extern char\t*optarg;\n> extern int\toptind,\n> -\t\t\topterr;\n> +\t\topterr;\n> \n> /* global decls */\n> bool\t\tg_verbose;\t\t\t/* User wants verbose narration of our\n> -\t\t\t\t\t\t\t\t * activities. */\n> -int\t\t\tg_last_builtin_oid; /* value of the last builtin oid */\n> -FILE\t *g_fout;\t\t\t\t/* the script file */\n> -PGconn\t *g_conn;\t\t\t\t/* the database connection */\n> -\n> -bool\t\tforce_quotes;\t\t/* User wants to suppress double-quotes */\n> -int\t\t\tdumpData;\t\t\t/* dump data using proper insert strings */\n> -int\t\t\tattrNames;\t\t\t/* put attr names into insert strings */\n> -int\t\t\tschemaOnly;\n> -int\t\t\tdataOnly;\n> -int\t\t\taclsOption;\n> +\t\t\t\t\t\t * activities. */\n> +int\t\tg_last_builtin_oid;\t\t/* value of the last builtin oid */\n> +FILE\t\t*g_fout;\t\t\t/* the script file */\n> +PGconn\t\t*g_conn;\t\t\t/* the database connection */\n> +\n> +bool\t\tforce_quotes;\t\t\t/* User wants to suppress double-quotes */\n> +int\t\tdumpData;\t\t\t/* dump data using proper insert strings */\n> +int\t\tattrNames;\t\t\t/* put attr names into insert strings */\n> +int\t\tschemaOnly;\n> +int\t\tdataOnly;\n> +int\t\taclsOption;\n> \n> -char\t\tg_opaque_type[10];\t/* name for the opaque type */\n> +char\t\tg_opaque_type[10];\t\t/* name for the opaque type */\n> \n> /* placeholders for the delimiters for comments */\n> char\t\tg_comment_start[10];\n> @@ -179,8 +179,8 @@\n> static bool\n> isViewRule(char *relname)\n> {\n> -\tPGresult *res;\n> -\tint\t\t\tntups;\n> +\tPGresult\t*res;\n> +\tint\t\tntups;\n> \tchar\t\tquery[MAXQUERYLEN];\n> \n> \tres = PQexec(g_conn, \"begin\");\n> @@ -315,13 +315,13 @@\n> \t\t\t\t\t const TableInfo tblinfo, bool oids)\n> {\n> \n> -\tPGresult *res;\n> +\tPGresult\t*res;\n> \tchar\t\tquery[255];\n> -\tint\t\t\tactual_atts;\t/* number of attrs in this a table */\n> +\tint\t\tactual_atts;\t/* number of attrs in this a table */\n> \tchar\t\texpandbuf[COPYBUFSIZ];\n> \tchar\t\tq[MAXQUERYLEN];\n> -\tint\t\t\ttuple;\n> -\tint\t\t\tfield;\n> +\tint\t\ttuple;\n> +\tint\t\tfield;\n> \n> \tsprintf(query, \"SELECT * FROM %s\", fmtId(classname, force_quotes));\n> \tres = PQexec(g_conn, query);\n> @@ -421,7 +421,7 @@\n> {\n> \n> \tint\t\t\ti;\n> -\tchar\t *all_only;\n> +\tchar\t\t\t*all_only;\n> \n> \tif (onlytable == NULL)\n> \t\tall_only = \"all\";\n> @@ -482,12 +482,11 @@\n> prompt_for_password(char *username, char *password)\n> {\n> \tchar\t\tbuf[512];\n> -\tint\t\t\tlength;\n> +\tint\t\tlength;\n> \n> #ifdef HAVE_TERMIOS_H\n> -\tstruct termios t_orig,\n> -\t\t\t\tt;\n> -\n> +\tstruct termios\tt_orig,\n> +\t\t\tt;\n> #endif\n> \n> \tprintf(\"Username: \");\n> @@ -535,21 +534,21 @@\n> int\n> main(int argc, char **argv)\n> {\n> -\tint\t\t\tc;\n> -\tconst char *progname;\n> -\tconst char *filename = NULL;\n> -\tconst char *dbname = NULL;\n> -\tconst char *pghost = NULL;\n> -\tconst char *pgport = NULL;\n> -\tchar\t *tablename = NULL;\n> -\tint\t\t\toids = 0;\n> -\tTableInfo *tblinfo;\n> -\tint\t\t\tnumTables;\n> +\tint\t\tc;\n> +\tconst char\t*progname;\n> +\tconst char\t*filename = NULL;\n> +\tconst char\t*dbname = NULL;\n> +\tconst char\t*pghost = NULL;\n> +\tconst char\t*pgport = NULL;\n> +\tchar\t\t*tablename = NULL;\n> +\tint\t\toids = 0;\n> +\tTableInfo\t*tblinfo;\n> +\tint\t\tnumTables;\n> \tchar\t\tconnect_string[512] = \"\";\n> \tchar\t\ttmp_string[128];\n> \tchar\t\tusername[100];\n> \tchar\t\tpassword[100];\n> -\tint\t\t\tuse_password = 0;\n> +\tint\t\tuse_password = 0;\n> \n> \tg_verbose = false;\n> \tforce_quotes = true;\n> @@ -727,11 +726,11 @@\n> TypeInfo *\n> getTypes(int *numTypes)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> -\tTypeInfo *tinfo;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> +\tTypeInfo\t\t*tinfo;\n> \n> \tint\t\t\ti_oid;\n> \tint\t\t\ti_typowner;\n> @@ -856,12 +855,12 @@\n> OprInfo *\n> getOperators(int *numOprs)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> \n> -\tOprInfo *oprinfo;\n> +\tOprInfo\t\t\t*oprinfo;\n> \n> \tint\t\t\ti_oid;\n> \tint\t\t\ti_oprname;\n> @@ -1199,11 +1198,11 @@\n> AggInfo *\n> getAggregates(int *numAggs)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> -\tAggInfo *agginfo;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> +\tAggInfo\t\t\t*agginfo;\n> \n> \tint\t\t\ti_oid;\n> \tint\t\t\ti_aggname;\n> @@ -1293,11 +1292,11 @@\n> FuncInfo *\n> getFuncs(int *numFuncs)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> -\tFuncInfo *finfo;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> +\tFuncInfo\t\t*finfo;\n> \n> \tint\t\t\ti_oid;\n> \tint\t\t\ti_proname;\n> @@ -1393,11 +1392,11 @@\n> TableInfo *\n> getTables(int *numTables, FuncInfo *finfo, int numFuncs)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> -\tTableInfo *tblinfo;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> +\tTableInfo\t\t*tblinfo;\n> \n> \tint\t\t\ti_oid;\n> \tint\t\t\ti_relname;\n> @@ -1726,11 +1725,11 @@\n> InhInfo *\n> getInherits(int *numInherits)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> -\tInhInfo *inhinfo;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> +\tInhInfo\t\t\t*inhinfo;\n> \n> \tint\t\t\ti_inhrel;\n> \tint\t\t\ti_inhparent;\n> @@ -1791,13 +1790,13 @@\n> {\n> \tint\t\t\ti,\n> \t\t\t\tj;\n> -\tchar\t\tq[MAXQUERYLEN];\n> +\tchar\t\t\tq[MAXQUERYLEN];\n> \tint\t\t\ti_attname;\n> \tint\t\t\ti_typname;\n> \tint\t\t\ti_atttypmod;\n> \tint\t\t\ti_attnotnull;\n> \tint\t\t\ti_atthasdef;\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \n> \tfor (i = 0; i < numTables; i++)\n> @@ -1902,10 +1901,10 @@\n> getIndices(int *numIndices)\n> {\n> \tint\t\t\ti;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> -\tPGresult *res;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> -\tIndInfo *indinfo;\n> +\tIndInfo\t\t\t*indinfo;\n> \n> \tint\t\t\ti_indexrelname;\n> \tint\t\t\ti_indrelname;\n> @@ -1994,7 +1993,7 @@\n> \t\t TypeInfo *tinfo, int numTypes)\n> {\n> \tint\t\t\ti;\n> -\tchar\t\tq[MAXQUERYLEN];\n> +\tchar\t\t\tq[MAXQUERYLEN];\n> \tint\t\t\tfuncInd;\n> \n> \tfor (i = 0; i < numTypes; i++)\n> @@ -2167,7 +2166,7 @@\n> \t\t\tTypeInfo *tinfo, int numTypes)\n> {\n> \tchar\t\tq[MAXQUERYLEN];\n> -\tint\t\t\tj;\n> +\tint\t\tj;\n> \tchar\t\t*func_def;\n> \tchar\t\tfunc_lang[NAMEDATALEN + 1];\n> \n> @@ -2268,7 +2267,7 @@\n> dumpOprs(FILE *fout, OprInfo *oprinfo, int numOperators,\n> \t\t TypeInfo *tinfo, int numTypes)\n> {\n> -\tint\t\t\ti;\n> +\tint\t\ti;\n> \tchar\t\tq[MAXQUERYLEN];\n> \tchar\t\tleftarg[MAXQUERYLEN];\n> \tchar\t\trightarg[MAXQUERYLEN];\n> @@ -2375,14 +2374,14 @@\n> dumpAggs(FILE *fout, AggInfo *agginfo, int numAggs,\n> \t\t TypeInfo *tinfo, int numTypes)\n> {\n> -\tint\t\t\ti;\n> +\tint\t\ti;\n> \tchar\t\tq[MAXQUERYLEN];\n> \tchar\t\tsfunc1[MAXQUERYLEN];\n> \tchar\t\tsfunc2[MAXQUERYLEN];\n> \tchar\t\tbasetype[MAXQUERYLEN];\n> \tchar\t\tfinalfunc[MAXQUERYLEN];\n> \tchar\t\tcomma1[2],\n> -\t\t\t\tcomma2[2];\n> +\t\t\tcomma2[2];\n> \n> \tfor (i = 0; i < numAggs; i++)\n> \t{\n> @@ -2514,9 +2513,9 @@\n> {\n> \tconst char *acls = tbinfo.relacl;\n> \tchar\t *aclbuf,\n> -\t\t\t *tok,\n> -\t\t\t *eqpos,\n> -\t\t\t *priv;\n> +\t\t *tok,\n> +\t\t *eqpos,\n> +\t\t *priv;\n> \n> \tif (strlen(acls) == 0)\n> \t\treturn;\t\t\t\t\t/* table has default permissions */\n> @@ -2726,15 +2736,15 @@\n> \tint\t\t\ti,\n> \t\t\t\tk;\n> \tint\t\t\ttableInd;\n> -\tchar\t\tattlist[1000];\n> -\tchar\t *classname[INDEX_MAX_KEYS];\n> -\tchar\t *funcname;\t\t/* the name of the function to comput the\n> -\t\t\t\t\t\t\t\t * index key from */\n> +\tchar\t\t\tattlist[1000];\n> +\tchar\t\t\t*classname[INDEX_MAX_KEYS];\n> +\tchar\t\t\t*funcname;\t\t/* the name of the function to comput the\n> +\t\t\t\t\t\t\t * index key from */\n> \tint\t\t\tindkey,\n> \t\t\t\tindclass;\n> \tint\t\t\tnclass;\n> \n> -\tchar\t\tq[MAXQUERYLEN],\n> +\tchar\t\t\tq[MAXQUERYLEN],\n> \t\t\t\tid1[MAXQUERYLEN],\n> \t\t\t\tid2[MAXQUERYLEN];\n> \tPGresult *res;\n> @@ -2887,7 +2897,7 @@\n> \t\t\t\tk;\n> \tint\t\t\tm,\n> \t\t\t\tn;\n> -\tchar\t **outVals = NULL; /* values to copy out */\n> +\tchar\t\t\t**outVals = NULL; /* values to copy out */\n> \n> \tn = PQntuples(res);\n> \tm = PQnfields(res);\n> @@ -2940,7 +2950,7 @@\n> static void\n> setMaxOid(FILE *fout)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tOid\t\t\tmax_oid;\n> \n> \tres = PQexec(g_conn, \"CREATE TABLE pgdump_oid (dummy int4)\");\n> @@ -2993,7 +3003,7 @@\n> static int\n> findLastBuiltinOid(void)\n> {\n> -\tPGresult *res;\n> +\tPGresult\t\t*res;\n> \tint\t\t\tntups;\n> \tint\t\t\tlast_oid;\n> \n> @@ -3025,9 +3035,9 @@\n> static char *\n> checkForQuote(const char *s)\n> {\n> -\tchar\t *r;\n> +\tchar\t\t*r;\n> \tchar\t\tc;\n> -\tchar\t *result;\n> +\tchar\t\t*result;\n> \n> \tint\t\t\tj = 0;\n> \n> @@ -3056,16 +3066,16 @@\n> static void\n> dumpSequence(FILE *fout, TableInfo tbinfo)\n> {\n> -\tPGresult *res;\n> -\tint4\t\tlast,\n> +\tPGresult\t\t*res;\n> +\tint4\t\t\tlast,\n> \t\t\t\tincby,\n> \t\t\t\tmaxv,\n> \t\t\t\tminv,\n> \t\t\t\tcache;\n> -\tchar\t\tcycled,\n> +\tchar\t\t\tcycled,\n> \t\t\t\tcalled,\n> -\t\t\t *t;\n> -\tchar\t\tquery[MAXQUERYLEN];\n> +\t\t\t\t*t;\n> +\tchar\t\t\tquery[MAXQUERYLEN];\n> \n> \tsprintf(query,\n> \t\t\t\"SELECT sequence_name, last_value, increment_by, max_value, \"\n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 17 Jan 1999 23:59:55 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] fix pg_dump to dump sequences created by SERIAL\n datatype"
}
] |
[
{
"msg_contents": "In light of the recent inquiry about postgresql RPMs, it might make\nsense to have on our web page some references to packaged up versions\n(or other means of easily installing postgresql) for specific\noperating systems. As an effort to start that, here are some relevant\nURLs pointing to the NetBSD package system for postgresql (and some\nmore general ones). Note that even though the package page is\nretrievable via ftp it is a browsable page.\n\n\t NetBSD PostgreSQL package\tftp://ftp.netbsd.org/pub/NetBSD/packages/pkgsrc/databases/postgresql/README.html\n\t NetBSD home page\t\thttp://www.netbsd.org\n\t NetBSD packages information\thttp://www.netbsd.org/Documentation/software/packages.html\n\nPerhaps a page that went something like the following would be useful.\nIt could be filled in with relevant entries as we become aware of\nthem, and might point people to relatively easy installation pathways\nif they need it.\n\nCheers,\nBrook\n\n===========================================================================\nSeveral packages for PostgreSQL have been created to ease the process\nof installing the system under specific operating systems. See the\nfollowing documents for more information.\n\nOperating system\tPackage name\n---------------- ------------\nNetBSD <ref URL>\tdatabases/postgresql <ref URL>\n...\n\n\n",
"msg_date": "Wed, 13 Jan 1999 10:51:38 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "references to packaged versions of PostgreSQL"
},
{
"msg_contents": "Brook Milligan <[email protected]> writes:\n\n> In light of the recent inquiry about postgresql RPMs, it might make\n> sense to have on our web page some references to packaged up versions\n> (or other means of easily installing postgresql) for specific\n> operating systems. As an effort to start that, here are some relevant\n...\n> Perhaps a page that went something like the following would be useful.\n...\n\nLooks great! I'll have a go at it tonight.\nObjections, comments, anyone?\n---\nHal / [email protected] / [email protected]\n",
"msg_date": "13 Jan 1999 12:02:48 -0600",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] references to packaged versions of PostgreSQL"
},
{
"msg_contents": "Brook Milligan wrote:\n >In light of the recent inquiry about postgresql RPMs, it might make\n >sense to have on our web page some references to packaged up versions\n >(or other means of easily installing postgresql) for specific\n >operating systems.\n\nPostgreSQL for Debian GNU/Linux:\n\nWeb page - http://www.debian.org\nFTP - ftp://ftp.debian.org and mirrors worldwide\n\n\nThe PostgreSQL source package (postgresql) produces several binary\npackages:\n\n Section Package name [comments]\n ------- ------------ ----------\n misc postgresql\n libs libpgsql [libpq.so.1]\n libs libpgsql2 [libpq.so.2]\n libs libpgtcl\n libs libpgperl\n libs ecpg\n libs libpgjava [not available yet for 6.4.2]\n devel postgresql-dev\n doc postgresql-doc\n\n [the following packages are only in Debian's unstable release]\n libs postgresql-pl\n libs python-postgresql [does not exist yet]\n misc postgresql-contrib\n misc postgresql-test\n\nThe current Debian stable release (hamm) contains release 6.3.2 of\nPostgreSQL, as will the currently frozen next release (slink); the\nunstable release (potato) contains 6.4.2.\n\nReleases are found in the dists/ directory in the ftp archive.\n\n\n-- \nOliver Elphick [email protected]\nIsle of Wight http://www.lfix.co.uk/oliver\n PGP key from public servers; key ID 32B8FAA1\n ========================================\n \"Wherefore let him that thinketh he standeth take heed \n lest he fall.\" I Corinthians 10:12 \n\n\n",
"msg_date": "Wed, 13 Jan 1999 21:09:27 +0000",
"msg_from": "\"Oliver Elphick\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] references to packaged versions of PostgreSQL "
}
] |
[
{
"msg_contents": "\nIt seems that pg_dump (in 6.4.2) still has problem with quoting \\ and ' when they appears together.\nThe following example illustrates this.\n\n\nterekhov=> create table test (i int, t text);\nCREATE\nterekhov=> insert into test values (1,'');\nINSERT 19066 1\nterekhov=> insert into test values (2,'abcd');\nINSERT 19067 1\nterekhov=> insert into test values (3,'\\\\');\nINSERT 19068 1\nterekhov=> insert into test values (4,'\\'');\nINSERT 19069 1\nterekhov=> insert into test values (5,'\\\\\\'');\nINSERT 19070 1\nterekhov=> select * from test;\ni|t \n-+----\n1| \n2|abcd\n3|\\ \n4|' \n5|\\' \n(5 rows)\n\nterekhov=> \\q\nterekhov(575)~>pg_dump -d -D -f ter.db terekhov\nterekhov(576)~>cat ter.db\nCREATE TABLE \"test\" (\n\t\"i\" int4,\n\t\"t\" text);\nINSERT INTO \"test\" (\"i\",\"t\") values (1,'');\nINSERT INTO \"test\" (\"i\",\"t\") values (2,'abcd');\nINSERT INTO \"test\" (\"i\",\"t\") values (3,'\\');\nINSERT INTO \"test\" (\"i\",\"t\") values (4,'''');\nINSERT INTO \"test\" (\"i\",\"t\") values (5,'\\''');\n\n\t\t Wrong quoting here ----^\n\nterekhov(577)~>psql -h luc0134 -p 5434 \nWelcome to the POSTGRESQL interactive sql monitor:\n Please read the file COPYRIGHT for copyright terms of POSTGRESQL\n\n type \\? for help on slash commands\n type \\q to quit\n type \\g or terminate with semicolon to execute query\n You are currently connected to the database: terekhov\n\nterekhov=> alter table test rename to test0;\nRENAME\nterekhov=> \\q\nterekhov(578)~>psql -h luc0134 -p 5434 -e <ter.db\nCREATE TABLE \"test\" (\n\t\"i\" int4,\n\t\"t\" text);\nQUERY: CREATE TABLE \"test\" (\n\t\"i\" int4,\n\t\"t\" text);\nCREATE\nINSERT INTO \"test\" (\"i\",\"t\") values (1,'');\nQUERY: INSERT INTO \"test\" (\"i\",\"t\") values (1,'');\nINSERT 19082 1\nINSERT INTO \"test\" (\"i\",\"t\") values (2,'abcd');\nQUERY: INSERT INTO \"test\" (\"i\",\"t\") values (2,'abcd');\nINSERT 19083 1\nINSERT INTO \"test\" (\"i\",\"t\") values (3,'\\');\nINSERT INTO \"test\" (\"i\",\"t\") values (4,'''');\nINSERT INTO \"test\" (\"i\",\"t\") values (5,'\\''');\n \\? -- help\n \\a -- toggle field-alignment (currently on)\n \\C [<captn>] -- set html3 caption (currently '')\n \\connect <dbname|-> <user> -- connect to new database (currently 'terekhov')\n \\copy table {from | to} <fname>\n \\d [<table>] -- list tables and indices, columns in <table>, or * for all\n \\da -- list aggregates\n \\dd [<object>]- list comment for table, field, type, function, or operator.\n \\df -- list functions\n \\di -- list only indices\n \\do -- list operators\n \\ds -- list only sequences\n \\dS -- list system tables and indexes\n \\dt -- list only tables\n \\dT -- list types\n \\e [<fname>] -- edit the current query buffer or <fname>\n \\E [<fname>] -- edit the current query buffer or <fname>, and execute\n \\f [<sep>] -- change field separater (currently '|')\n \\g [<fname>] [|<cmd>] -- send query to backend [and results in <fname> or pipe]\n \\h [<cmd>] -- help on syntax of sql commands, * for all commands\n \\H -- toggle html3 output (currently off)\n \\i <fname> -- read and execute queries from filename\n \\l -- list all databases\n \\m -- toggle monitor-like table display (currently off)\n \\o [<fname>] [|<cmd>] -- send all query results to stdout, <fname>, or pipe\n \\p -- print the current query buffer\n \\q -- quit\n \\r -- reset(clear) the query buffer\n \\s [<fname>] -- print history or save it in <fname>\n \\t -- toggle table headings and row count (currently on)\n \\T [<html>] -- set html3.0 <table ...> options (currently '')\n \\x -- toggle expanded output (currently off)\n \\w <fname> -- output current buffer to a file\n \\z -- list current grant/revoke permissions\n \\! [<cmd>] -- shell escape or command\nEOF\nterekhov(579)~>\n\n\n\nRegards,\nMikhail Terekhov\n",
"msg_date": "Wed, 13 Jan 1999 14:49:03 -0500 (EST)",
"msg_from": "Mikhail Terekhov <[email protected]>",
"msg_from_op": true,
"msg_subject": "pg_dump again"
}
] |
[
{
"msg_contents": "> -----Original Message-----\n> Generic triggers in C that are argument driven would be\n> possible. But the drawback is that those triggers have to be\n> very smart to use saved SPI plans (one for every different\n> argument set). And it must be row level triggers, so for an\n> update to a 2 meg row table they will be fired 2 million\n> times and run their queries inside - will take some time.\n> \n> More painful in the 2 meg row situation is that trigger\n> invocation has to be delayed until COMMIT if the constraint\n> is deferred. I think we cannot remember 2 million OLD plus 2\n> million NEW tuples if one tuple can have up to 8K (will be\n> 32GB to remember plus overhead), so we need to remember at\n> least the CTID's of OLD and NEW and refetch them for the\n> trigger invocation. OUTCH - the OLD ones are at the head and\n> all the NEW ones are at the end of the tables file!\nAgreed scrap the triggers idea.\n\n> > Maybe let's go with the rewrite system, because it works, and is\n> > flexible and strangely designed for this type of problem. \n> Similar to\n> > how we use the rewrite system for views.\n> \n> And the other changes I've planned for the rewrite system\n> will improve this much more.\n> \n> 1. Change pg_rewrite.ev_attr into an int28. This would be\n> useful for ON UPDATE rules so the rewrite system can\n> easily check if a rule has to be applied or not. If none\n> of named attributes gets something different assigned\n> than it's own OLD value, they aren't updated so the rule\n> could never result in an action and can be omitted\n> completely.\nWould we add a system table for constraints? I believe that the\nDEFERRABLE option means we need special information about all\nconstraints. (ie. is_deferrable boolean, is_deferred boolean).\n\n> 2. Create cross reference catalog that lists all relations\n> used in a rule (rangetable). If we have a DELETE CASCADE\n> constraint, the rule is triggered on the key table and\n> the action is a DELETE from the referencing table. If now\n> the referencing table is dropped, the rule get's\n> corrupted because the resulting querytree isn't plannable\n> any longer (the relations in the rules rangetable are\n> identified by the OID in pg_class, not by relname). You\n> can see the effect if you create a view and drop one of\n> the base tables.\n*Bleck* We can't place cascading or restriction information into the\ntable definition. Don't know if you had thought about it, but you could\nonly create them in reverse order (child<-parent), very messy when you\nhave many 'child' tables to cascade.\n\n> Well, we need to define what to do if a table is dropped\n> that occurs in the crossref. First of all, the rules have\n> to be dropped too, but in the case of a view rule, maybe\n> the whole view too?\nI agree that we should drop rules defined on the dropped table, but not\nthose that simply access the dropped table, including views. If I need\nto drop a table which has a view defined for it simply to rearrange\ncolumn order, I'd like not to have to recreate the view as well and/or\nany RI-constraints that reference that table as well.\n\n> And in the case where a key table to which another one\n> has a CHECK reference is dropped? The rule action will\n> allways abort, so it isn't useful any more. But I\n> wouldn't like to silently drop it, because someone might\n> want to drop and recreate the key table and this would\n> silently result in that all the constraints have been\n> lost.\nThis is what I mean about views and RI-constraints.\n\n> Maybe we should change the rulesystem at all so that the\n> rangetable entries in the rule actions etc. are updated\n> with a lookup from pg_class at rewrite time. Must be done\n> carefully because someone might drop a table and recreate\n> it with a different schema corrupting the parsetree of\n> the rule actions though.\nCould we allow a 'recompile' of the rules? Which would update the rule\nactions and report which rules have been corrupted. And possibly a way\nto simple drop all invalid rules. We could simply mark the rules that\nneed to be recompiled on a DROP or ALTER TABLE (to allow for dropping a\ncolumn, or changing a column type) from the crossref. \n\n> 3. Allow an unlimited number of rules on a relation.\n> Currently there is a hard coded limit on the number of\n> rules the relation can hold in it's slots.\nI agree with this one without reference, especially if we actually\nimplement RI/constraints using rewrite.\n\n> > I am basically asking for a reason _not_ to use the rewrite \n> system for\n> > this. I can't think of one myself.\n> \n> It might interfere with the new MVCC code. The rule actions\n> must see exactly the OLD tuples that where used in the\n> original statements. Not only those in the updated table\n> itself, think of an INSERT...SELECT or an UPDATE where the\n> TLE or qual expressions are values from other tables.\nShouldn't be too hard to accomplish. We simply need to be sure that the\nvalues are either cached or re-cacheable/re-producible (if there are\nmemory concerns) until the end of the transaction. So, either:\n 1. cache all the values\n 2. store the part of the parsetree needed to recreate the values and\nhave MVCC mark any needed rows off limits to vacuum\nI know that there are many other subtleties that I'm not addressing such\nas indexes, but the above will get you the answer.\nFWIW, RI and CONSTRAINTS would be affected by this.\n\n> Not a real reason, just something to have in mind and maybe\n> switching silently to another MVCC isolation level if\n> constraint rules get applied, so all tables read from now on\n> will get a read lock applied and cannot get updated\n> concurrently until COMMIT.\nNot needed, see the above. We just need to mark the rows with the tid,\nor some-such, because dropped/updated rows will hang around until a\nvacuum.\n\n> And it's a problem I've came across just writing this note\n> where MVCC already could have broken rewrite rule system\n> semantics.\nMaybe, but I know you'll straighten it all out. I have faith in you,\nJan.\n\n> Jan\n\t-DEJ\n\nP.S. I love the discussion that my simple little request has sparked.\n",
"msg_date": "Wed, 13 Jan 1999 13:50:01 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] CONSTRAINTS..."
},
{
"msg_contents": "Pardon.\n\n First of all I strongly recommend that someone reads the\n section about the rewrite rule system in the programmers\n manual. From some statements in this whole discussion I'm in\n doubt if anyone really knows how the rule system in detail\n works and what it actually does with a query.\n\n And please - there where substantial changes in the rule\n system for v6.4. So anyone who THINKS he knows, update your\n knowlege.\n\n I really didn't wanted to step on someones feet, but detailed\n knowlege of the rule system is required here.\n\nDeJuan Jackson wrote:\n\n>\n> Agreed scrap the triggers idea.\n>\n> > [...]\n> >\n> Would we add a system table for constraints? I believe that the\n> DEFERRABLE option means we need special information about all\n> constraints. (ie. is_deferrable boolean, is_deferred boolean).\n\n No, we don't need another system catalog for that. The only\n thing that must be assured is that a rule which is deferrable\n is a NON-INSTEAD one with exactly one RAISE statement as\n action. Can be checked at rule create time at the same place\n the checks for view rules are currently done.\n\n I thought to add is_deferrable and is_initdeferred to\n pg_rewrite itself to hold the information about DEFERRABLE\n and INITIAL DEFERRED. If a constraint actually has to be\n deferred (added to global deferred querytree list) or not is\n a per transaction runtime information and isn't useful in the\n catalog.\n\n>\n> > 2. Create cross reference catalog that lists all relations\n> > used in a rule (rangetable). If we have a DELETE CASCADE\n> > constraint, the rule is triggered on the key table and\n> > the action is a DELETE from the referencing table. If now\n> > the referencing table is dropped, the rule get's\n> > corrupted because the resulting querytree isn't plannable\n> > any longer (the relations in the rules rangetable are\n> > identified by the OID in pg_class, not by relname). You\n> > can see the effect if you create a view and drop one of\n> > the base tables.\n> *Bleck* We can't place cascading or restriction information into the\n> table definition. Don't know if you had thought about it, but you could\n> only create them in reverse order (child<-parent), very messy when you\n> have many 'child' tables to cascade.\n\n I don't want to put it into the table definition. Anything\n that will be required for checks and cascaded deletes are\n just rules going into pg_rewrite.\n\n OTOH as I read Oracles SQL Language Quick Reference I can do\n a\n\n ALTER TABLE tab1 MODIFY key1 integer\n CONSTRAINT ref_tab2 REFERENCES ... ON DELETE CASCADE;\n\n I think I can create the tables in any order and later define\n all the (maybe circular) constraints.\n\n>\n> > Well, we need to define what to do if a table is dropped\n> > that occurs in the crossref. First of all, the rules have\n> > to be dropped too, but in the case of a view rule, maybe\n> > the whole view too?\n> I agree that we should drop rules defined on the dropped table, but not\n> those that simply access the dropped table, including views. If I need\n> to drop a table which has a view defined for it simply to rearrange\n> column order, I'd like not to have to recreate the view as well and/or\n> any RI-constraints that reference that table as well.\n>\n> [...]\n>\n> Could we allow a 'recompile' of the rules? Which would update the rule\n> actions and report which rules have been corrupted. And possibly a way\n> to simple drop all invalid rules. We could simply mark the rules that\n> need to be recompiled on a DROP or ALTER TABLE (to allow for dropping a\n> column, or changing a column type) from the crossref.\n\n Nice idea. Would require the crossref catalog, one more\n catalog to store the rules definition text and a compile\n status field in pg_rewrite.\n\n If a relation used in a rule is dropped or altered, the\n status is set to RECOMPILE. So the rule system can check the\n rule locks and try to recompile those rules from the source\n before applying them.\n\n Yeah - I really like it! This way only those rules defined on\n the dropped table have to be dropped too (as it is now). And\n as long as another table used in the rule isn't\n recreated/altered with a suitable schema, the action that\n fires the rule cannot be performed on that table (the\n recompile would fail and abort the transaction).\n\n There are some subtle points about the rule caching\n performed. Actually the relation descriptor contains rule\n locks where in memory parsetrees are placed at heap_open(),\n but it must be possible.\n\n> > It might interfere with the new MVCC code. The rule actions\n> > must see exactly the OLD tuples that where used in the\n> > original statements. Not only those in the updated table\n> > itself, think of an INSERT...SELECT or an UPDATE where the\n> > TLE or qual expressions are values from other tables.\n> Shouldn't be too hard to accomplish. We simply need to be sure that the\n> values are either cached or re-cacheable/re-producible (if there are\n> memory concerns) until the end of the transaction. So, either:\n> 1. cache all the values\n> 2. store the part of the parsetree needed to recreate the values and\n> have MVCC mark any needed rows off limits to vacuum\n> I know that there are many other subtleties that I'm not addressing such\n> as indexes, but the above will get you the answer.\n> FWIW, RI and CONSTRAINTS would be affected by this.\n\n We cannot cache all the values without the danger running out\n of memory.\n\n Those parts of the parsetree needed to recreate the values\n are actually incorporated into the parsetrees of the rule\n actions. BUT THESE ARE PARSETREES! They result in another\n execution plan and when they get executed they perform their\n own table scans!\n\n>\n> > Not a real reason, just something to have in mind and maybe\n> > switching silently to another MVCC isolation level if\n> > constraint rules get applied, so all tables read from now on\n> > will get a read lock applied and cannot get updated\n> > concurrently until COMMIT.\n> Not needed, see the above. We just need to mark the rows with the tid,\n> or some-such, because dropped/updated rows will hang around until a\n> vacuum.\n\n As I understood MVCC, it is possible that two identical scans\n in one and the same transaction, can result in different\n rows.\n\n Maybe I misunderstood something. Isn't changing the isolation\n level exactly that what can force MVCC to do locking/marking\n with the result that the two identical scans will produce the\n same rows?\n\n Vadim - please clearify!\n\n But uh - oh! Now it's me who forgot how in detail the rule\n system works :-)\n\n The deferred constraint queries performed at transaction end\n will do their own scans. So tuples removed/changed during\n this transaction could not be rescanned as they where\n (command counter is incremented!). And the incremented\n command counter is required to see that finally all circular\n constraints are satisfied.\n\n This one in detail now. Let's have the following setup:\n\n CREATE TABLE t1 (key1 int4);\n CREATE TABLE t2 (ref1 int4);\n CREATE TABLE t3 (a int4, b int4);\n\n The check constraint for ref1->key1 is the rule\n\n CREATE DEFERRABLE RULE _CU_t2_ref1 AS ON UPDATE TO t2 DO\n RAISE 'key \"%\" not in t1', new.ref1\n WHERE NOT EXISTS (SELECT key1 FROM t1 WHERE key1 = new.ref1);\n\n Now we execute\n\n BEGIN;\n SET CONSTRAINT ALL DEFERRED;\n UPDATE t2 SET ref1 = t3.b WHERE ref1 = t3.a;\n DELETE FROM t3;\n INSERT INTO t1 VALUES (4);\n COMMIT;\n\n The rewritten parsetree for the rule action would look like\n one for the query\n\n RAISE 'key \"%\" not in t1', t3.b\n FROM t2, t3\n WHERE t2.ref1 = t3.a\n AND NOT EXISTS (SELECT key1 FROM t1 WHERE key1 = t3.b);\n\n This parsetree is deferred until the end of the transaction,\n the UPDATE will already have set cmax in the tuples of t2\n that where updated. So this RAISE cannot work as expected! To\n do so, the scans of t2 and t3 must be done against the\n command counter of the UPDATE, but the scan of t1 against the\n final command counter at COMMIT.\n\n This mess is why I changed the rule system for v6.4 to\n perform the rule actions BEFORE the original query itself.\n They cannot work AFTER (what would happen for deferred ones).\n\n\n> P.S. I love the discussion that my simple little request has sparked.\n\n Me too!\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Thu, 14 Jan 1999 11:38:15 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CONSTRAINTS..."
}
] |
[
{
"msg_contents": "Hi,\n\nIt seems that pg_dump (in 6.4.2) still has a problem when quoting\n\\ and ' when these characters appears together.\nConsider the following example:\n\n\nterekhov=> create table test (i int, t text);\nCREATE\nterekhov=> insert into test values (1,'');\nINSERT 19066 1\nterekhov=> insert into test values (2,'abcd');\nINSERT 19067 1\nterekhov=> insert into test values (3,'\\\\');\nINSERT 19068 1\nterekhov=> insert into test values (4,'\\'');\nINSERT 19069 1\nterekhov=> insert into test values (5,'\\\\\\'');\nINSERT 19070 1\nterekhov=> select * from test;\ni|t \n-+----\n1| \n2|abcd\n3|\\ \n4|' \n5|\\' \n(5 rows)\n\nterekhov=> \\q\nterekhov(575)~>pg_dump -d -D -f ter.db terekhov\nterekhov(576)~>cat ter.db\nCREATE TABLE \"test\" (\n\t\"i\" int4,\n\t\"t\" text);\nINSERT INTO \"test\" (\"i\",\"t\") values (1,'');\nINSERT INTO \"test\" (\"i\",\"t\") values (2,'abcd');\nINSERT INTO \"test\" (\"i\",\"t\") values (3,'\\');\nINSERT INTO \"test\" (\"i\",\"t\") values (4,'''');\nINSERT INTO \"test\" (\"i\",\"t\") values (5,'\\''');\n ^\n wrong quoting here -----|\n\nterekhov(577)~>psql \nWelcome to the POSTGRESQL interactive sql monitor:\n Please read the file COPYRIGHT for copyright terms of POSTGRESQL\n\n type \\? for help on slash commands\n type \\q to quit\n type \\g or terminate with semicolon to execute query\n You are currently connected to the database: terekhov\n\nterekhov=> alter table test rename to test0;\nRENAME\nterekhov=> \\q\nterekhov(578)~>psql -e <ter.db\nCREATE TABLE \"test\" (\n\t\"i\" int4,\n\t\"t\" text);\nQUERY: CREATE TABLE \"test\" (\n\t\"i\" int4,\n\t\"t\" text);\nCREATE\nINSERT INTO \"test\" (\"i\",\"t\") values (1,'');\nQUERY: INSERT INTO \"test\" (\"i\",\"t\") values (1,'');\nINSERT 19082 1\nINSERT INTO \"test\" (\"i\",\"t\") values (2,'abcd');\nQUERY: INSERT INTO \"test\" (\"i\",\"t\") values (2,'abcd');\nINSERT 19083 1\nINSERT INTO \"test\" (\"i\",\"t\") values (3,'\\');\nINSERT INTO \"test\" (\"i\",\"t\") values (4,'''');\nINSERT INTO \"test\" (\"i\",\"t\") values (5,'\\''');\n \\? -- help\n \\a -- toggle field-alignment (currently on)\n \\C [<captn>] -- set html3 caption (currently '')\n \\connect <dbname|-> <user> -- connect to new database (currently 'terekhov')\n \\copy table {from | to} <fname>\n \\d [<table>] -- list tables and indices, columns in <table>, or * for all\n \\da -- list aggregates\n \\dd [<object>]- list comment for table, field, type, function, or operator.\n \\df -- list functions\n \\di -- list only indices\n \\do -- list operators\n \\ds -- list only sequences\n \\dS -- list system tables and indexes\n \\dt -- list only tables\n \\dT -- list types\n \\e [<fname>] -- edit the current query buffer or <fname>\n \\E [<fname>] -- edit the current query buffer or <fname>, and execute\n \\f [<sep>] -- change field separater (currently '|')\n \\g [<fname>] [|<cmd>] -- send query to backend [and results in <fname> or pipe]\n \\h [<cmd>] -- help on syntax of sql commands, * for all commands\n \\H -- toggle html3 output (currently off)\n \\i <fname> -- read and execute queries from filename\n \\l -- list all databases\n \\m -- toggle monitor-like table display (currently off)\n \\o [<fname>] [|<cmd>] -- send all query results to stdout, <fname>, or pipe\n \\p -- print the current query buffer\n \\q -- quit\n \\r -- reset(clear) the query buffer\n \\s [<fname>] -- print history or save it in <fname>\n \\t -- toggle table headings and row count (currently on)\n \\T [<html>] -- set html3.0 <table ...> options (currently '')\n \\x -- toggle expanded output (currently off)\n \\w <fname> -- output current buffer to a file\n \\z -- list current grant/revoke permissions\n \\! [<cmd>] -- shell escape or command\nEOF\n\nRegards,\nMikhail Terekhov\n",
"msg_date": "Wed, 13 Jan 1999 15:47:39 -0500",
"msg_from": "Mikhail Terekhov <[email protected]>",
"msg_from_op": true,
"msg_subject": "pg_dump again"
},
{
"msg_contents": "Mikhail Terekhov <[email protected]> writes:\n> It seems that pg_dump (in 6.4.2) still has a problem when quoting\n> \\ and ' when these characters appears together.\n> Consider the following example:\n> ... snip ...\n> terekhov(575)~>pg_dump -d -D -f ter.db terekhov\n\nAh, I see the problem: pg_dump's -d/-D option doesn't do quoting of\nspecial characters properly --- in fact, the only thing it does do\nis double-up single quote marks. My testing was without that option ---\nwithout -d or -D, pg_dump uses COPY, which quotes everything properly.\nThis is probably why most people weren't seeing a problem.\n\nI'll put this on my to-do list for 6.5, but in the meantime, I'd suggest\nnot using -d or -D... a dump script with one INSERT per table row is\nhorribly slow compared to COPY anyway...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 13 Jan 1999 19:22:46 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] pg_dump again "
},
{
"msg_contents": "Tom, is this fixed?\n\n\n\n> Mikhail Terekhov <[email protected]> writes:\n> > It seems that pg_dump (in 6.4.2) still has a problem when quoting\n> > \\ and ' when these characters appears together.\n> > Consider the following example:\n> > ... snip ...\n> > terekhov(575)~>pg_dump -d -D -f ter.db terekhov\n> \n> Ah, I see the problem: pg_dump's -d/-D option doesn't do quoting of\n> special characters properly --- in fact, the only thing it does do\n> is double-up single quote marks. My testing was without that option ---\n> without -d or -D, pg_dump uses COPY, which quotes everything properly.\n> This is probably why most people weren't seeing a problem.\n> \n> I'll put this on my to-do list for 6.5, but in the meantime, I'd suggest\n> not using -d or -D... a dump script with one INSERT per table row is\n> horribly slow compared to COPY anyway...\n> \n> \t\t\tregards, tom lane\n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 15 Mar 1999 09:25:27 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] pg_dump again"
},
{
"msg_contents": "Bruce Momjian <[email protected]> writes:\n> Tom, is this fixed?\n>> Mikhail Terekhov <[email protected]> writes:\n>>>> It seems that pg_dump (in 6.4.2) still has a problem when quoting\n>>>> \\ and ' when these characters appears together.\n\nYes, I fixed that.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 15 Mar 1999 10:23:44 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] pg_dump again "
}
] |
[
{
"msg_contents": "Sorry, I don't know the answer to this, so I am copying this to the\nPostgreSQL Hackers list: can anyone help, please?\n\nDaniel Lundin wrote:\n >\n >Hi Oliver,\n >\n >I've run into trouble when building PostgreSQL 6.4.2 from the source\n >debian package you've created (postgresql_6.4.2-1.dsc).\n >I managed to get it compiled, but when running initdb it freezes after a\n >while and bails out with a 'FATAL: s_lock(202ba370) at spin.c:114, stuck\n >spinlock. Aborting.' error message.\n >\n >I saw somsone posting about a smiliar problem on the postgresql mailing\n >list archive, but quite some while ago. Do you have any idea what to do\n >and if there are any patches/workarounds for this (or if it's a known\n >user-fault, me that is).\n >\n >Machine config:\n >hardware: Samsung AlphaPC 164LX 21164 based hardware\n > uname: Linux gollum 2.2.0-pre4 #3 alpha unknown\n > system: Debian GNU/Linux 2.1\n >compiler: egcs-2.91.60\n > ld 2.9.1\n > \n >Regards,\n >Daniel Lundin , MediaCenter\n >\n >_________________________________________________________________ /\\__ \n > \\/ \n > Daniel Lundin \n > MediaCenter, UNIX and BeOS Developer \n > http://www.umc.se/~daniel/\n >\n\n-- \nOliver Elphick [email protected]\nIsle of Wight http://www.lfix.co.uk/oliver\n PGP key from public servers; key ID 32B8FAA1\n ========================================\n \"Wherefore let him that thinketh he standeth take heed \n lest he fall.\" I Corinthians 10:12 \n\n\n",
"msg_date": "Wed, 13 Jan 1999 21:09:21 +0000",
"msg_from": "\"Oliver Elphick\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: PostgreSQL 6.4.2 build problems on Linux/Alpha "
}
] |
[
{
"msg_contents": "-----BEGIN PGP SIGNED MESSAGE-----\n\nI recently had a query fail with the following error:\n\n keystone=> select members.id, members.element, slips.sid, slips.problem \n keystone-> from slips, members \n keystone-> where members.element = slips.sid\n keystone-> and members.id = 'quickslips';\n ERROR: There is no operator '=' for types 'varchar' and 'int4'\n\t You will either have to retype this query using an explicit cast,\n\t or you will have to define the operator using CREATE OPERATOR\n\ncasting the varchar column to int4 works, but if I try to cast the\nint4 to varchar I get another error:\n\n keystone=> select members.id, members.element, slips.sid, slips.problem \n keystone-> from slips, members \n keystone-> where members.element = varchar(slips.sid)\n keystone-> and members.id = 'quickslips';\n ERROR: parser: parse error at or near \"slips\"\n\nI would expect that the int4 --> varchar conversion would be the\n\"safer\" one to do. By \"safer\" I mean that an int4 value should map to\na distinct varchar value whereas all non-numeric varchar values\ngenerate an error in pg_atoi....\n\nroland\n- -- \n\t\t PGP Key ID: 66 BC 3B CD\nRoland B. Roberts, PhD Custom Software Solutions\[email protected] 101 West 15th St #4NN\[email protected] New York, NY 10011\n\n-----BEGIN PGP SIGNATURE-----\nVersion: 2.6.2\nComment: Processed by Mailcrypt 3.4, an Emacs/PGP interface\n\niQCVAwUBNp1N4uoW38lmvDvNAQG/PAQAr1eU8jG6lsfFACxb732mFA/rnxKkKIED\nbn0t3AbAZMMyWe1GajQ5iEXz3ZrZmQGgGS05WMtLyo6HF/Zxvdj235qSJFVsucC3\nnQR/knEO+we3/PUCiqKEPPZNkufCxhyFjFEctws1wcp5XDnrO3EiYTBC5vTUXNF1\n1nW+k38StyU=\n=gWkW\n-----END PGP SIGNATURE-----\n",
"msg_date": "13 Jan 1999 20:52:36 -0500",
"msg_from": "Roland Roberts <[email protected]>",
"msg_from_op": true,
"msg_subject": "Postgres 6.3.2 varchar <--> int4 conversions"
}
] |
[
{
"msg_contents": "subscribe pgsql-hackers\n\n\n",
"msg_date": "Thu, 14 Jan 1999 16:02:27 +0100",
"msg_from": "Stefan Kapfhammer <[email protected]>",
"msg_from_op": true,
"msg_subject": "None"
}
] |
[
{
"msg_contents": "I can help in maintaining the RPMs for postgresql. Let me know.\n\nal dev\n\nYou wrote:\n\n> I was wondering if any one is responsible for making\n> the RPM for postgresql? \"rpm\" could be added as a\n> make target and the sources required to build the\n> rpm could be added to the CVS repository. In this\n> way those people who only want binaries could be\n> better supported.\n\nThat would be great. The RedHat folks are swamped with maintaining\n*lots* of packages, and I'm sure it would help if we could do the\nmaintenance of the rpm for them.\n\nNobody else has stepped forward for that yet, though it has been a need\nfor a while. You're welcome to it ;)\n\nPerhaps once the rpm build is codified, it would be easier for others to\n\nhelp. I've got the rpm book, but have not waded through enough to figure\n\nout how to build a from-scratch installation.\n\n - Tom\n\n\n\n",
"msg_date": "Thu, 14 Jan 1999 20:06:39 +0000",
"msg_from": "aldev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] RPM maintainer?"
}
] |
[
{
"msg_contents": "Okay, after a long fight with modula3 I know have a working version\nof cvsup for Solaris 7 SPARC. Know, what I would like to do is to use\ncvsup to update my local cvs repoistory. How would I go about doing this?\n\nThanks,\n\nMatt\n----------\nMatthew C. Aycock\nOperating Systems Analyst/Admin, Senior\nDept Math/CS\nEmory University, Atlanta, GA \nInternet: [email protected] \t\t\n\n\n",
"msg_date": "Thu, 14 Jan 1999 15:36:17 -0500 (EST)",
"msg_from": "\"Matthew C. Aycock\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "CVS...."
},
{
"msg_contents": "> Okay, after a long fight with modula3 I know have a working version\n> of cvsup for Solaris 7 SPARC. Know, what I would like to do is to use\n> cvsup to update my local cvs repoistory. How would I go about doing \n> this?\n\nLook at doc/src/sgml/cvs.sgml (no product yet since it it's a jumble\nstill). I can send you a copy if you don't have a recent tarball.\n\n - Tom\n",
"msg_date": "Fri, 15 Jan 1999 02:01:27 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CVS...."
}
] |
[
{
"msg_contents": "OK, this is a complete novice question.\nDo we or do we not have temp tables (and/or unnamed tables)? \nIf we don't why not?\nI can see them making quite a few places of current development a lot\neasier and flexible. I do realize TANSTAAFL, but as I grok it could be\nused for multiple tuple return from functions, the\ndeferrable-constraints problems, implementing the SQL92 join syntax, and\nsubselects in from clauses and select lists.\n\nJust more of those ideas,\n\t-DEJ\n",
"msg_date": "Thu, 14 Jan 1999 16:00:42 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Novice Question"
},
{
"msg_contents": "> OK, this is a complete novice question.\n> Do we or do we not have temp tables (and/or unnamed tables)? \n> If we don't why not?\n> I can see them making quite a few places of current development a lot\n> easier and flexible. I do realize TANSTAAFL, but as I grok it could be\n> used for multiple tuple return from functions, the\n> deferrable-constraints problems, implementing the SQL92 join syntax, and\n> subselects in from clauses and select lists.\n\nIt is on the TODO list, and is not a hard job at all.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Thu, 14 Jan 1999 17:56:53 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Novice Question"
}
] |
[
{
"msg_contents": "> > OK, this is a complete novice question.\n> > Do we or do we not have temp tables (and/or unnamed tables)? \n> > If we don't why not?\n> > I can see them making quite a few places of current \n> development a lot\n> > easier and flexible. I do realize TANSTAAFL, but as I grok \n> it could be\n> > used for multiple tuple return from functions, the\n> > deferrable-constraints problems, implementing the SQL92 \n> join syntax, and\n> > subselects in from clauses and select lists.\n> \n> It is on the TODO list, and is not a hard job at all.\n\nWell, in that case...\nI am a visionary when not well informed, so before I get more knowledge\nthan my optimism can stomach, let me share my vision(s).\n\nREWRITE/RULE SYSTEM:\n If we could easily keep the targetlist result in a temporary table\n(hey it's already in memory, pipe it out to disk async if we run out of\nram) we could easily recreate any statepoint of a transaction. All we\nneed is a way of determining which targetlist we need before we need\nthem. I think that the crossref table could greatly assist here. (Now\nI read Vadim's message about the above and I think this is already\nfigured out, but I was lost at about line 3 of his reply.) \n\nThis would also allow application of certain currently restricted cases,\nsuch as view on select union and select distinct. \"How?\" you may ask,\nbecause the rule could be executed as is and the new query applied to\nthe temp table results (more costly than the current method but it will\nget the job done). I'll let Jan take the idea from there, but I think\nyou get the idea.\n\nSQL92 JOIN syntax:\n I get into everything don't I, Tom. \"What is my temp-table view for\nJOINS?\" you ask. Well, I'm glad you did.\nIf unnammed-temp-tables were treated as normal tables within a\nselect/JOIN clause easily implemented JOINS. This also takes multiple\npasses, but if I'm not mistake something like his is already done for\njoins anyway. Example of my idea:\n (t1 LEFT OUTER JOIN t2 ON t1.i=t2.i) a1\n\twould generate a temporary table named/aliased a1 that had the\nsame results as (I think this will work):\n\tselect t1.*, t2.* from t1, t2 WHERE t1.i=t2.i\n UNION ALL \n select t1.*, NULL, NULL, ... FROM t1 WHERE NOT EXISTS (select 1\nFROM t2 where t1.i=t2.i);\nwhich is then taken and used in the next join clause. This also make it\neasy to include subselects in the from clause, just make the results a\ntemp-table (I know that this doesn't give you much extra power, but I'm\na sucker for easily realized flexibility). So, Tom what do you think?\n\nTUPLES RETURNED FROM FUNCTIONS:\n I believe this is another area under Jan's domain, but I could be\nwrong. The idea would be to treat these function as a temp-table.\n\"Why?\" (I know you're thinking it) because then we could have these\nfunctions in our from clause. Alias the function and you can access\nit's return members just as a table (ie select * from my_function();).\n\nOh well, I have more vision, but I'll let you guys either diminish or\naugment these before I share any more. SO, be gentle with me :).\n\n\t-DEJ\n",
"msg_date": "Thu, 14 Jan 1999 17:48:14 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Novice Question"
}
] |
[
{
"msg_contents": "Here is a problem report from a HP-UX/11.0/cc user.\n\nsrc/makefiles/Makefile.hpux seems to lack:\n\n%.sl: %.o\n $(LD) -b -o $@ $<\n\nThis makes regression test failed. Please note that in 6.4\nMakefile.hpux has above.\n\nComments?\n--\nTatsuo Ishii\n\n",
"msg_date": "Fri, 15 Jan 1999 11:42:07 +0900",
"msg_from": "Tatsuo Ishii <[email protected]>",
"msg_from_op": true,
"msg_subject": "6.4.2 HP-UX makefile problem?"
},
{
"msg_contents": "Tatsuo Ishii <[email protected]> writes:\n> Here is a problem report from a HP-UX/11.0/cc user.\n> src/makefiles/Makefile.hpux seems to lack:\n> %.sl: %.o\n> $(LD) -b -o $@ $<\n> This makes regression test failed. Please note that in 6.4\n> Makefile.hpux has above.\n\nI removed that rule from Makefile.hpux because I believed it to be\ndead code --- src/Makefile.shlib is now responsible for causing\nshared libs to be built correctly for each platform.\n\nI have not seen any problems building on HPUX 9 or 10. Can you\nprovide more info about the failure on HPUX 11?\n\n\t\t\tregards, tom lane\n",
"msg_date": "Fri, 15 Jan 1999 02:19:36 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4.2 HP-UX makefile problem? "
},
{
"msg_contents": "I wrote:\n> Tatsuo Ishii <[email protected]> writes:\n>> Here is a problem report from a HP-UX/11.0/cc user.\n>> src/makefiles/Makefile.hpux seems to lack:\n>> %.sl: %.o\n>> \t$(LD) -b -o $@ $<\n>> This makes regression test failed. Please note that in 6.4\n>> Makefile.hpux has above.\n\n> I removed that rule from Makefile.hpux because I believed it to be\n> dead code --- src/Makefile.shlib is now responsible for causing\n> shared libs to be built correctly for each platform.\n> I have not seen any problems building on HPUX 9 or 10. Can you\n> provide more info about the failure on HPUX 11?\n\nSigh. Tatsuo is quite right: I screwed up. The 6.4.2 fileset\n*builds* OK on HPUX, but it will not run the regression test.\n\nThe regression test includes building a shared library (two of 'em\nin fact) ... and those makefiles do not use Makefile.shlib.\nThey rely on the port makefile to supply a default build rule for\nshared libraries. So that rule is not quite as unused as I thought.\n\nI don't know how I missed this --- I must not have tried to run the\nregression tests after that set of changes, figuring that it was\npurely a build-time issue and if the system built then things were OK.\nMea maxima culpa.\n\nFor the moment I will re-insert the .sl rule into Makefile.hpux in\nboth current and REL6_4 trees. A better fix would be to make the\nvarious makefiles in src/test and contrib use Makefile.shlib ...\nbut currently I'm thinking that where we really want to go is to\nstart using GNU libtool to control shared library construction,\nso as to not have to maintain all this platform-specific build\nknowledge ourselves. So I'd rather spend time on that.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sat, 16 Jan 1999 19:14:59 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] 6.4.2 HP-UX makefile problem? "
},
{
"msg_contents": "> I wrote:\n> > Tatsuo Ishii <[email protected]> writes:\n> >> Here is a problem report from a HP-UX/11.0/cc user.\n> >> src/makefiles/Makefile.hpux seems to lack:\n> >> %.sl: %.o\n> >> \t$(LD) -b -o $@ $<\n> >> This makes regression test failed. Please note that in 6.4\n> >> Makefile.hpux has above.\n> \n> > I removed that rule from Makefile.hpux because I believed it to be\n> > dead code --- src/Makefile.shlib is now responsible for causing\n> > shared libs to be built correctly for each platform.\n> > I have not seen any problems building on HPUX 9 or 10. Can you\n> > provide more info about the failure on HPUX 11?\n> \n> Sigh. Tatsuo is quite right: I screwed up. The 6.4.2 fileset\n> *builds* OK on HPUX, but it will not run the regression test.\n> \n> The regression test includes building a shared library (two of 'em\n> in fact) ... and those makefiles do not use Makefile.shlib.\n> They rely on the port makefile to supply a default build rule for\n> shared libraries. So that rule is not quite as unused as I thought.\n> \n> I don't know how I missed this --- I must not have tried to run the\n> regression tests after that set of changes, figuring that it was\n> purely a build-time issue and if the system built then things were OK.\n> Mea maxima culpa.\n> \n> For the moment I will re-insert the .sl rule into Makefile.hpux in\n> both current and REL6_4 trees. A better fix would be to make the\n> various makefiles in src/test and contrib use Makefile.shlib ...\n\nThank you for taking care of that. BTW, the fix was actually suggested\nby Kawakami, not by me (I don't have access to HP-UX at all). Please\nremind this if you are going to write down in CVS log or somewhere.\n---\nTatsuo Ishii\n",
"msg_date": "Sun, 17 Jan 1999 23:20:25 +0900",
"msg_from": "Tatsuo Ishii <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] 6.4.2 HP-UX makefile problem? "
}
] |
[
{
"msg_contents": "I'm writing up the full set of allowed date/time formats, and thought\nI'd include the recognized time zones.\n\nI want to confirm that AHST, Alaska-Hawaii Std Time, should have the\nsame value as Australia Eastern Std Time, which is what the conversion\ntable claims. Seems to me that it is on the wrong side of the date line.\nDoes someone know what the correct value is?\n\n - Tom\n",
"msg_date": "Fri, 15 Jan 1999 05:36:04 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Time zones"
},
{
"msg_contents": "Australian Eastern Std Time is GMT-10. Alaska and Hawaii are\ndefinitely on the other side of the dateline. They should not be the\nsame.\n\nOn Fri, 15 Jan 1999, Thomas G. Lockhart wrote:\n\n> I'm writing up the full set of allowed date/time formats, and thought\n> I'd include the recognized time zones.\n> \n> I want to confirm that AHST, Alaska-Hawaii Std Time, should have the\n> same value as Australia Eastern Std Time, which is what the conversion\n> table claims. Seems to me that it is on the wrong side of the date line.\n> Does someone know what the correct value is?\n> \n> - Tom\n> \n> \n\n+----------------------+---+\n| Ross Johnson | | E-Mail: [email protected]\n| Info Sciences and Eng|___|\n| University of Canberra | FAX: +61 6 2015227\n| PO Box 1 |\n| Belconnen ACT 2616 | WWW: http://willow.canberra.edu.au/~rpj/\n| AUSTRALIA |\n+--------------------------+\n\n\n",
"msg_date": "Fri, 15 Jan 1999 17:39:17 +1100 (EST)",
"msg_from": "Ross Johnson <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Time zones"
},
{
"msg_contents": "Ross Johnson <[email protected]> writes:\n> Australian Eastern Std Time is GMT-10.\n\nGMT+10, surely, unless the sun started rising in the west\nwhen I wasn't looking. Hawaii is in the right place to\nbe GMT-10 ... maybe that's the source of the confusion?\n\n>> I want to confirm that AHST, Alaska-Hawaii Std Time, should have the\n>> same value as Australia Eastern Std Time, which is what the conversion\n>> table claims.\n\nThe most recent atlas I have says that eastern Alaska is GMT-9,\nwestern Alaska and Hawaii are GMT-10, and the westernmost Aleutians\nare GMT-11. No info about daylight savings habits. A ten-year-old\natlas claims that all of mainland Alaska is GMT-9. They may have\nchanged things up there recently ... I think we need a native\nguide ...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Fri, 15 Jan 1999 02:10:21 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Time zones "
},
{
"msg_contents": "Tom Lane <[email protected]> writes:\n\n> The most recent atlas I have says that eastern Alaska is GMT-9,\n> western Alaska and Hawaii are GMT-10, and the westernmost Aleutians\n> are GMT-11. No info about daylight savings habits. A ten-year-old\n> atlas claims that all of mainland Alaska is GMT-9. They may have\n> changed things up there recently ... I think we need a native\n> guide ...\n\n...or just Olson. Here are the relevant bits I find in the zoneinfo\nlibrary. The GMTOFF column is the basic (non-DST) offset, RULES is\n\"-\" if there's no DST, the constant \"1:00\" for the three week period\nof the 1933 Hawaiian DST experiment, and \"US\" to indicate that DST is\nused, and follows the US rules I append at the bottom of the list. In\nthis case, the FORMAT column has a \"%s\" that's replaced by the\nrelevant letter (\"S\", \"W\" or \"D\") from the LETTERS column of the rule.\n\nSo, getting back to what someone (Thomas Lockhart?) said:\n\n> I want to confirm that AHST, Alaska-Hawaii Std Time, should have the\n> same value as Australia Eastern Std Time, which is what the\n> conversion table claims.\n\nI think not. AH[SD]T used to cover Hawaii and parts of Alaska, but\nnowadays it's just parts of Alaska. Eastern Australia is another four\nhours west, over on the other side of the date line, at +10:00 instead\nof -10:00.\n\n\n# Zone\tNAME\t\tGMTOFF\tRULES\tFORMAT\t[UNTIL]\nZone Pacific/Honolulu\t-10:31:26 -\tLMT\t1900 Jan 1 12:00\n\t\t\t-10:30\t-\tHST\t1933 Apr 30 2:00\n\t\t\t-10:30\t1:00\tHDT\t1933 May 21 2:00\n\t\t\t-10:30\tUS\tH%sT\t1947 Jun 8 2:00\n\t\t\t-10:00\t-\tHST\n\n# Zone\tNAME\t\tGMTOFF\tRULES\tFORMAT\t[UNTIL]\nZone America/Juneau\t -8:57:41 -\tLMT\t1900 Aug 20 12:00\n\t\t\t -8:00\t-\tPST\t1942\n\t\t\t -8:00\tUS\tP%sT\t1946\n\t\t\t -8:00\t-\tPST\t1969\n\t\t\t -8:00\tUS\tP%sT\t1983 Oct 30 2:00\n\t\t\t -9:00\tUS\tAK%sT\nZone America/Yakutat\t -9:18:55 -\tLMT\t1900 Aug 20 12:00\n\t\t\t -9:00\t-\tYST\t1942\n\t\t\t -9:00\tUS\tY%sT\t1946\n\t\t\t -9:00\t-\tYST\t1969\n\t\t\t -9:00\tUS\tY%sT\t1983 Oct 30 2:00\n\t\t\t -9:00\tUS\tAK%sT\nZone America/Anchorage\t -9:59:36 -\tLMT\t1900 Aug 20 12:00\n\t\t\t-10:00\t-\tCAT\t1942\n\t\t\t-10:00\tUS\tCAT/CAWT 1946\n\t\t\t-10:00\t-\tCAT\t1967 Apr\n\t\t\t-10:00\t-\tAHST\t1969\n\t\t\t-10:00\tUS\tAH%sT\t1983 Oct 30 2:00\n\t\t\t -9:00\tUS\tAK%sT\nZone America/Nome\t-11:01:38 -\tLMT\t1900 Aug 20 12:00\n\t\t\t-11:00\t-\tNST\t1942\n\t\t\t-11:00\tUS\tN%sT\t1946\n\t\t\t-11:00\t-\tNST\t1967 Apr\n\t\t\t-11:00\t-\tBST\t1969\n\t\t\t-11:00\tUS\tB%sT\t1983 Oct 30 2:00\n\t\t\t -9:00\tUS\tAK%sT\nZone America/Adak\t-11:46:38 -\tLMT\t1900 Aug 20 12:00\n\t\t\t-11:00\t-\tNST\t1942\n\t\t\t-11:00\tUS\tN%sT\t1946\n\t\t\t-11:00\t-\tNST\t1967 Apr\n\t\t\t-11:00\t-\tBST\t1969\n\t\t\t-11:00\tUS\tB%sT\t1983 Oct 30 2:00\n\t\t\t-10:00\tUS\tHA%sT\n\n# Rule\tNAME\tFROM\tTO\tTYPE\tIN\tON\tAT\tSAVE\tLETTER/S\nRule\tUS\t1918\t1919\t-\tMar\tlastSun\t2:00\t1:00\tW # War\nRule\tUS\t1918\t1919\t-\tOct\tlastSun\t2:00\t0\tS\nRule\tUS\t1942\tonly\t-\tFeb\t9\t2:00\t1:00\tW # War\nRule\tUS\t1945\tonly\t-\tSep\t30\t2:00\t0\tS\nRule\tUS\t1967\tmax\t-\tOct\tlastSun\t2:00\t0\tS\nRule\tUS\t1967\t1973\t-\tApr\tlastSun\t2:00\t1:00\tD\nRule\tUS\t1974\tonly\t-\tJan\t6\t2:00\t1:00\tD\nRule\tUS\t1975\tonly\t-\tFeb\t23\t2:00\t1:00\tD\nRule\tUS\t1976\t1986\t-\tApr\tlastSun\t2:00\t1:00\tD\nRule\tUS\t1987\tmax\t-\tApr\tSun>=1\t2:00\t1:00\tD\n\n-tih\n-- \nPopularity is the hallmark of mediocrity. --Niles Crane, \"Frasier\"\n",
"msg_date": "15 Jan 1999 09:31:18 +0100",
"msg_from": "Tom Ivar Helbekkmo <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Time zones"
},
{
"msg_contents": "On Fri, 15 Jan 1999, Tom Lane wrote:\n\n> Ross Johnson <[email protected]> writes:\n> > Australian Eastern Std Time is GMT-10.\n> \n> GMT+10, surely, unless the sun started rising in the west\n> when I wasn't looking. Hawaii is in the right place to\n> be GMT-10 ... maybe that's the source of the confusion?\n\nYes. Sorry. Glad I was able to illustrate the confusion for everyone\n:-)\n\n+----------------------+---+\n| Ross Johnson | | E-Mail: [email protected]\n| Info Sciences and Eng|___|\n| University of Canberra | FAX: +61 6 2015227\n| PO Box 1 |\n| Belconnen ACT 2616 | WWW: http://willow.canberra.edu.au/~rpj/\n| AUSTRALIA |\n+--------------------------+\n\n\n",
"msg_date": "Fri, 15 Jan 1999 22:28:38 +1100 (EST)",
"msg_from": "Ross Johnson <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Time zones "
},
{
"msg_contents": "> > Hawaii is in the right place to\n> > be GMT-10 ... maybe that's the source of the confusion?\n\nGreat. I'll assume it was a sign flip, and fix it unless I hear\notherwise.\n\nThanks.\n\n - Tom\n",
"msg_date": "Fri, 15 Jan 1999 14:52:57 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Time zones"
}
] |
[
{
"msg_contents": "Sorry to intrude onto the hackers list, but I've got a problem that I think\nmight be worth looking at.\n\nI've got two databases, one a humongous DB ported from Solid to Postgres,\nthe other using just a few tables and indices from the first. The\ndatabases are called apx3 and apx5, I created apx5 by pg_dumping two tables\nfrom apx3 and running `psql -f apx3.out apx5`.\n\nBoth sets of tables also contain identical sets of data.\n\nBut when I run the following join query on the tables in apx3:\n\n EXPLAIN SELECT a.headline, b.article_type\n from articles a, article_types b\n where a.art_type_id = b.art_type_id;\n\n... I get the following output:\n\n Nested Loop (cost=3.20 size=2 width=28)\n -> Seq Scan on articles a (cost=1.07 size=2 width=14)\n -> Seq Scan on article_types b (cost=1.07 size=2 width=14)\n\nWhen I run it against apx5, it produces:\n\n Merge Join (cost=0.00 size=1 width=28)\n -> Seq Scan (cost=0.00 size=0 width=0)\n -> Sort (cost=0.00 size=0 width=0)\n -> Seq Scan on articles a (cost=0.00 size=0 width=14)\n -> Seq Scan (cost=0.00 size=0 width=0)\n -> Sort (cost=0.00 size=0 width=0)\n -> Seq Scan on article_types b (cost=0.00 size=0 width=14)\n\n\nLike I said, the two sets of tables (articles and article_types) are\nidentical. The difference between the two databases is that apx3 contains\na large number of other tables/views.\n\npg_dumping the two tables from apx3 and apx5 produces identical output:\n\nCREATE TABLE \"articles\" (\"article_id\" \"int4\" NOT NULL, \"subject1_id\"\n\"int2\", \"subject2_id\" \"int2\", \"subject3_id\" \"int2\", \"headline\"\nvarchar(256), \"subhead\" varchar(256), \"print_page\" varchar(64), \"byline\"\nvarchar(128), \"country_id\" \"int2\", \"dateline_id\" \"int2\", \"volume\" \"int2\",\n\"issue\" \"int2\", \"publ_type\" varchar(32), \"publ_date\" \"date\", \"art_src_id\"\n\"int2\", \"art_type_id\" \"int2\", \"contributor1\" varchar(128), \"contributor2\"\nvarchar(128), \"contributor3\" varchar(128), \"editor\" varchar(32),\n\"managing_editor\" varchar(32), \"copy_editor\" varchar(32),\n\"production_person\" varchar(32), \"entered_date\" \"date\", \"entered_time\"\n\"time\", \"entered_by\" varchar(50), \"split600\" \"int2\", \"wets_front\" \"int2\",\n\"wets_priority\" \"int2\", \"published\" \"int2\", \"article_text\" \"text\",\n\"wait_for_publish\" \"int2\");\nCREATE UNIQUE INDEX \"articles_article_id_key\" on \"articles\" using btree (\n\"article_id\" \"int4_ops\" );\nCREATE INDEX \"art_publdate_ind\" on \"articles\" using btree ( \"publ_date\"\n\"date_ops\" );\nCREATE INDEX \"art_wfi_ind\" on \"articles\" using btree ( \"wets_front\"\n\"int2_ops\" );\nCREATE INDEX \"art_date_id_ind\" on \"articles\" using btree ( \"dateline_id\"\n\"int2_ops\" );\nCREATE INDEX \"articles_art_type_ind\" on \"articles\" using btree (\n\"art_type_id\" \"int2_ops\" );\nCREATE TABLE \"article_types\" (\"art_type_id\" \"int2\" NOT NULL, \"article_type\"\nvarchar(63), \"listed\" \"int2\", \"priority\" \"int2\");\nCREATE UNIQUE INDEX \"article_types_pkey\" on \"article_types\" using btree (\n\"art_type_id\" \"int2_ops\" );\n\nMy uneducated guess is that there's some other factor in apx3 that's\naffecting the optimizer's ability to generate the most desirable query plan\n(i.e., the efficient Merge Join rather than the expensive Nested Loop of\nSequential Scans).\n\nThe ugly part of this problem is that apx3 is -- at least to my mind --\nquite monstrous. It was built by dumping the Solid DB, then quickly\nediting the output file and loading that into Postgres as apx2, then (when\nthis strange behavior started to appear) dumping apx2 out of Postgres and\nreloading it as apx3. The resulting database is far from perfect: There\nare cases where \"primary key\" declarations inside CREATE TABLE statements\nwere followed by redundant CREATE INDEX statements, resulting in multiple\nindices being created on single columns. There are about 80 tables/views\nand rougly as many indices.\n\nI'd be happy to send dumps of the two databases if anyone is interested in\ninvestigating this problem. I'd happily do it myself, but beyond\nindividually dumping tables from apx3, loading them into apx5 and waiting\nfor the join query to fail, I'm not sure how to proceed. Or if someone\nwants to give me some instructions on how to debug this sort of thing...?\n\n(BTW, I'm not on the hackers list, so please cc me on any replies.)\n\nThanks,\nCharlie\n\n-------------------------------------------\nPREVIOUS E-MAIL ABOUT THIS PROBLEM FOLLOWS:\n-------------------------------------------\n\nX-POP3-Rcpt: clhmail@yompy\nTo: Charles Hornberger <[email protected]>\nSubject: Re: very slow response time on large, multi-table view \nDate: Thu, 14 Jan 1999 20:02:39 -0500\nFrom: Tom Lane <[email protected]>\n\n> As a matter of fact, when I created new (albeit smaller) tables to\n> perform the same joins, I started getting some more logical results\n> from optimizer:\n\nThat's really weird. I wonder if the system's info about those original\ntables might've gotten corrupted somehow? It might be worth dumping and\nreloading the database just to see... (btw, you did do an initdb and\nreload when you upgraded to 6.4 from prior version, I hope...)\n\n> (BTW, what's the difference between a Merge Join and a Hash Join?)\n\nMerge Join works by scanning two pre-sorted tables in sync, like a\ntraditional list-merge sort. The particular plan you show did the\nsort the hard way --- I'd kind of expect it to read the tables in\norder using an index, if there is a btree index available.\n\nHash Join works by making a hashtable describing all the entries in\none table, then probing the hashtable for each record in the other\ntable.\n\nI'm not sure about which one is faster, but I expect either one would\nbeat nested loop handily...\n\nIf you don't figure out what's going on, I'd suggest posting the new\ninfo to the hackers list --- this could be a real bug in the optimizer.\n\n\t\t\tregards, tom lane\n\n---------------------------------------------------------------------\n\nTo: Tom Lane <[email protected]>\nSubject: Re: very slow response time on large, multi-table view \nDate: Thu, 14 Jan 1999 19:09:39 -0900\nFrom: Charles Hornberger <[email protected]>\n\nAt 08:02 PM 1/14/99 -0500, you wrote:\n>> As a matter of fact, when I created new (albeit smaller) tables to\n>> perform the same joins, I started getting some more logical results\n>> from optimizer:\n>\n>That's really weird. I wonder if the system's info about those original\n>tables might've gotten corrupted somehow? It might be worth dumping and\n>reloading the database just to see... (btw, you did do an initdb and\n>reload when you upgraded to 6.4 from prior version, I hope...)\n\nActually, this database was created under 6.4. There's one oddity I can\nthink of that might've messed up the tables: When I dumped the Solid DB,\nit included a bunch of \"primary key\" declarations, as well as explicit\nCREATE INDEX statements on some of the columns that were already declared\nas primary keys. Perhaps the existence of multiple indices on single\ncolumns caused the problem? I've since deleted the redundant indices; now\nthere's one per indexed column. (As I mentioned in my first post, though\nperhaps not clearly, I dumped a Solid DB, did a quick & dirty edit of all\nthe CREATE TABLE & CREATE VIEW statements to make them compatible with\nPostgres, then reloaded the thing into Postgres.)\n\nIn any case, I'm going to dump & reload the DB now, (And about initdb, I\nsuppose that the guy who did the upgrade did do that, but I'll ask to make\nsure.) But since *some* of the tables in this database work fine, and\nothers don't, I'd *guess* that it has to do with the table structures. Of\ncourse, that's one of the most uneducated guesses I've ever made....\n\nIt just occurred to me that there's NO reason for me to be writing this\ne-mail until AFTER I do the dump & reload, so ....\n\nOK, I've dumped and reloaded the original database (apx2) as apx3. And\nnothing seems to have been fixed.\n\napx3=> explain select * from test_view;\nNOTICE: QUERY PLAN:\n\nNested Loop (cost=3.20 size=2 width=28)\n -> Seq Scan on articles a (cost=1.07 size=2 width=14)\n -> Seq Scan on article_types at (cost=1.07 size=2 width=14)\n\nEXPLAIN\napx3=> explain select * from int2_test;\nNOTICE: QUERY PLAN:\n\nMerge Join (cost=0.00 size=1 width=28)\n -> Seq Scan (cost=0.00 size=0 width=0)\n -> Sort (cost=0.00 size=0 width=0)\n -> Seq Scan on int2_articles a (cost=0.00 size=0 width=14)\n -> Seq Scan (cost=0.00 size=0 width=0)\n -> Sort (cost=0.00 size=0 width=0)\n -> Seq Scan on int2_types at (cost=0.00 size=0 width=14)\n\nHere are the CREATE RULE statements executed when I created apx3 from\napx2's dump file:\n\nCREATE RULE \"_RETint2_test\" AS ON SELECT TO \"int2_test\" DO INSTEAD SELECT\n\"a\".\"headline\", \"at\".\"article_type\" FROM \"int2_articles\" \"a\", \"int2_types\"\n\"at\" WHERE \"a\".\"art_type_id\" = \"at\".\"art_type_id\";\nCREATE\nCREATE RULE \"_RETtest_view\" AS ON SELECT TO \"test_view\" DO INSTEAD SELECT\n\"a\".\"headline\", \"at\".\"article_type\" FROM \"articles\" \"a\", \"article_types\"\n\"at\" WHERE \"a\".\"art_type_id\" = \"at\".\"art_type_id\";\nCREATE\n\n\n>> (BTW, what's the difference between a Merge Join and a Hash Join?)\n>\n>Merge Join works by scanning two pre-sorted tables in sync, like a\n>traditional list-merge sort. The particular plan you show did the\n>sort the hard way --- I'd kind of expect it to read the tables in\n>order using an index, if there is a btree index available.\n>\n>Hash Join works by making a hashtable describing all the entries in\n>one table, then probing the hashtable for each record in the other\n>table.\n>\n>I'm not sure about which one is faster, but I expect either one would\n>beat nested loop handily...\n>\n>If you don't figure out what's going on, I'd suggest posting the new\n>info to the hackers list --- this could be a real bug in the optimizer.\n\nWell, what do you think? Should I post all this stuff to the hackers list? \n\nThanks,\nCharlie\n\n\n",
"msg_date": "Thu, 14 Jan 1999 22:26:08 -0800",
"msg_from": "Charles Hornberger <[email protected]>",
"msg_from_op": true,
"msg_subject": "identical joins on identical tables produce different output"
}
] |
[
{
"msg_contents": "I had to find this kind of thing out once and found a navy site at\n\nhttp://tycho.usno.navy.mil/tzones.html\n\nthat has a lot of places(well most of the big ones at least :)\n\nIt has the following entry\n\nUSA Hawaii: -10 hours \nUSA Alaska: -9 hours (Local summer -8 hours)\nAustralia New South Wales: +10 hours (Local summer +11 hours)\n\nHope this helps :)\n\nCheers,\nCiaran\n\nBTW Hi Ross... Im an ex student of the uni of Canberra. Hows it going ? :)\n\n>Australian Eastern Std Time is GMT-10. Alaska and Hawaii are\n>definitely on the other side of the dateline. They should not be the\n>same.\n>\n>On Fri, 15 Jan 1999, Thomas G. Lockhart wrote:\n>\n>> I'm writing up the full set of allowed date/time formats, and thought\n>> I'd include the recognized time zones.\n>> \n>> I want to confirm that AHST, Alaska-Hawaii Std Time, should have the\n>> same value as Australia Eastern Std Time, which is what the conversion\n>> table claims. Seems to me that it is on the wrong side of the date line.\n>> Does someone know what the correct value is?\n\n\n",
"msg_date": "Fri, 15 Jan 1999 18:25:46 +1100",
"msg_from": "\"Ciaran Dunn\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Time zones"
}
] |
[
{
"msg_contents": "I downloaded latest 6.4.2 and tried compiling on linux 2.0.34 (redhat\n5.1) with latest\nflex version 2.5.4\nBut the make complains - you have buggy version of flex!! Help.\n\nmake[1]: Entering directory\n`/home/src/redhat/BUILD/postgresql-6.4.2/src'\nmake -C lextest all\nmake[2]: Entering directory\n`/home/src/redhat/BUILD/postgresql-6.4.2/src/lextest'\nflex scan.l\ngcc -c lex.yy.c\ngcc -c lextest.c\ngcc -o lextest lex.yy.o lextest.o\nmake[2]: Leaving directory\n`/home/src/redhat/BUILD/postgresql-6.4.2/src/lextest'\n\nYou have a buggy version of flex. Read INSTALL and search for flex for\na fix.\n\nmake[1]: *** [lexverify] Error 1\nmake[1]: Leaving directory `/home/src/redhat/BUILD/postgresql-6.4.2/src'\n\nmake: *** [all] Error 2\n\n\n",
"msg_date": "Fri, 15 Jan 1999 18:07:10 +0000",
"msg_from": "aldev <[email protected]>",
"msg_from_op": true,
"msg_subject": "Compiling 6.4.2: Flex problem"
}
] |
[
{
"msg_contents": "Hello\n\nAssume I have some range-like(start,end) type, how can I enable \nrtree indexing for it? It's all what I need is to be able quickly\nfind appropriate range where given number belongs to.\n\nI've read through documentation but found some info only on btree\nstrategy. Which operation on custom type must be defined to get\nrtree index?\n\nThanks in advance!\n\n-- \nIlya Varlashkin\n",
"msg_date": "Fri, 15 Jan 1999 21:32:04 +0300 (MSK)",
"msg_from": "Ilya Varlashkin <[email protected]>",
"msg_from_op": true,
"msg_subject": "user-defined type and rtree index"
}
] |
[
{
"msg_contents": "Thomas,\n\nI looked at this and it tells you how to get the sources via\nCVSup or cvs, but not how to keep them in sync so to speek.\nI would like the ability to make my own branches and so forth\nto play with certain aspects of the database. However, I would\nlike the ability to fold in patches in a big group without having\nto apply each patch by hand.\n\nAny other ideas?\n\nMatt\n>> Okay, after a long fight with modula3 I know have a working version\n>> of cvsup for Solaris 7 SPARC. Know, what I would like to do is to use\n>> cvsup to update my local cvs repoistory. How would I go about doing \n>> this?\n>\n>Look at doc/src/sgml/cvs.sgml (no product yet since it it's a jumble\n>still). I can send you a copy if you don't have a recent tarball.\n>\n> - Tom\n\n----------\nMatthew C. Aycock\nOperating Systems Analyst/Admin, Senior\nDept Math/CS\nEmory University, Atlanta, GA \nInternet: [email protected] \t\t\n\n\n",
"msg_date": "Fri, 15 Jan 1999 14:09:26 -0500 (EST)",
"msg_from": "\"Matthew C. Aycock\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] CVS...."
},
{
"msg_contents": "On Fri, 15 Jan 1999, Matthew C. Aycock wrote:\n\n> Thomas,\n> \n> I looked at this and it tells you how to get the sources via\n> CVSup or cvs, but not how to keep them in sync so to speek.\n> I would like the ability to make my own branches and so forth\n> to play with certain aspects of the database. However, I would\n> like the ability to fold in patches in a big group without having\n> to apply each patch by hand.\n> \n> Any other ideas?\n\nI don't believe that you can...my understanding of CVSup is that you can\n\"mirror\" the remote site 's cvs repository with it, and checkout as you\nwant, but if you check something back in again, it will be gone the next\ntime an update comes down...\n\nHere are my settings for pulling down the CVS repository for\nFreeBSD...I've never tried to do a branch on it, so my above paragraph\nisn't based on tryin git...\n\n# Defaults that apply to all the collections\n*default host=cvsup2.FreeBSD.org\n*default base=/usr\n*default prefix=/usr/local/freebsd-cvsroot\n*default release=cvs \n#*default tag=.\n*default delete use-rel-suffix\n\n\nNoice thta the 'tag' line is commented out...\n\nThere is a pretty good description of CVSup at:\n\n\thttp://www.freebsd.org/handbook/cvsup.html\n\nAnd John Polstra tends to be very responsive to questions about his\nsoftware :)\n\n\n\n > > Matt\n> >> Okay, after a long fight with modula3 I know have a working version\n> >> of cvsup for Solaris 7 SPARC. Know, what I would like to do is to use\n> >> cvsup to update my local cvs repoistory. How would I go about doing \n> >> this?\n> >\n> >Look at doc/src/sgml/cvs.sgml (no product yet since it it's a jumble\n> >still). I can send you a copy if you don't have a recent tarball.\n> >\n> > - Tom\n> \n> ----------\n> Matthew C. Aycock\n> Operating Systems Analyst/Admin, Senior\n> Dept Math/CS\n> Emory University, Atlanta, GA \n> Internet: [email protected] \t\t\n> \n> \n> \n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Fri, 15 Jan 1999 15:46:23 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CVS...."
},
{
"msg_contents": "> > I looked at this and it tells you how to get the sources via\n> > CVSup or cvs, but not how to keep them in sync so to speek.\n> > I would like the ability to make my own branches and so forth\n> > to play with certain aspects of the database. However, I would\n> > like the ability to fold in patches in a big group without having\n> > to apply each patch by hand.\n> > Any other ideas?\n\nThe way I work is to use CVSup to check out versions locally on my\nmachine. Then, when I change a file I save a version as filename.orig\nand these changed files tend to accumulate. When I'm ready to submit a\npatch, I run a little utility I wrote which traverses the development\ntree and generates a patch file for each file/file.orig pair, putting\nall of the files in a separate patch directory. I've got another utility\nwhich can apply this set of patch files to the cvs source tree at\npostgresql.org. Afterwards, I'll clean up by removing the .orig files\nand doing a CVSup refresh then a cvs update on my tree.\n\nIf I'm doing a longer-term project, I'll freeze my tree and check out a\nnew one when I need to test with a current development tree.\n\nBruce has a utility he included in the distribution which generates a\nsingle patch file for all file/file.orig pairs (afaik). Look in the\nsrc/tools directory. I don't use that because I quite often have more\nthan one set of patches going at a time so will delete the patch files\nwhich aren't relevant before submitting or committing them.\n\n> And John Polstra tends to be very responsive to questions about his\n> software :)\n\nVery much so. There is a new rpm distribution of Modula-3 and John has\ngenerated a new version of CVSup which is tailored to Linux (just\nslightly different from his main FreeBSD version). I've been sitting on\nthe new code, intending to generate new static binaries for glibc2 and\nfor libc5. Will post on postgresql.org when they are ready.\n\n - Tom\n",
"msg_date": "Sat, 16 Jan 1999 02:21:57 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CVS...."
}
] |
[
{
"msg_contents": "pg_dump won't drop stuff before trying to create it; this makes\ndropping a single table (say) and recreating it difficult to automate\nsince it is subject to error if all the stuff doesn't get dropped\nproperly. The following patch causes pg_dump to emit DROP ...\nstatements prior to emitting CREATE ... statements.\n\nI don't have experience with things like aggregates, languages, types,\nfunctions, ... so I'm not quite sure I got the DROP ... syntax correct\nin all instances. I tried to follow the man pages and the fields put\ninto the CREATE ... statements, so it should be ok. Help testing this\nwould be great. (Do our regression tests test pg_dump?)\n\nAt this point, I don't understand how the CREATE TRIGGER statements\nare constructed; they don't follow the same style as anything else in\npg_dump. As a result, I'm not clear on how to implement the DROP\nTRIGGER statements. Help appreciated here.\n\nFinally, this patch should go in after the earlier one I submitted to\ncreate SERIAL sequencies.\n\nCheers,\nBrook\n\n===========================================================================\n\n--- bin/pg_dump/pg_dump.c.orig\tFri Jan 15 12:26:34 1999\n+++ bin/pg_dump/pg_dump.c\tFri Jan 15 13:02:34 1999\n@@ -1630,6 +1630,13 @@\n \t\t\t\t\texit_nicely(g_conn);\n \t\t\t\t}\n \t\t\t\ttgfunc = finfo[findx].proname;\n+\n+#if 0\t\t\t\t\n+\t\t\t\t/* XXX - how to emit this DROP TRIGGER? */\n+\t\t\t\tsprintf(query, \"DROP TRIGGER %s;\\n\", fmtId(PQgetvalue(res2, i2, i_tgname), force_quotes));\n+\t\t\t\tfputs(query, fout);\n+#endif\n+\n \t\t\t\tsprintf(query, \"CREATE TRIGGER %s \", fmtId(PQgetvalue(res2, i2, i_tgname), force_quotes));\n \t\t\t\t/* Trigger type */\n \t\t\t\tfindx = 0;\n@@ -2026,6 +2033,9 @@\n \n \t\tbecomeUser(fout, tinfo[i].usename);\n \n+\t\tsprintf(q, \"DROP TYPE %s;\\n\", fmtId(tinfo[i].typname, force_quotes));\n+\t\tfputs(q, fout);\n+\n \t\tsprintf(q,\n \t\t\t\t\"CREATE TYPE %s \"\n \t\t\t\t\"( internallength = %s, externallength = %s, input = %s, \"\n@@ -2122,6 +2132,8 @@\n \t\tlanname = checkForQuote(PQgetvalue(res, i, i_lanname));\n \t\tlancompiler = checkForQuote(PQgetvalue(res, i, i_lancompiler));\n \n+\t\tfprintf(fout, \"DROP PROCEDURAL LANGUAGE '%s';\\n\", lanname);\n+\n \t\tfprintf(fout, \"CREATE %sPROCEDURAL LANGUAGE '%s' \"\n \t\t\t\"HANDLER %s LANCOMPILER '%s';\\n\",\n \t\t\t(PQgetvalue(res, i, i_lanpltrusted)[0] == 't') ? \"TRUSTED \" : \"\",\n@@ -2237,6 +2249,20 @@\n \t\tPQclear(res);\n \t}\n \n+\tsprintf(q, \"DROP FUNCTION %s (\", fmtId(finfo[i].proname, force_quotes));\n+\tfor (j = 0; j < finfo[i].nargs; j++)\n+\t{\n+\t\tchar\t *typname;\n+\n+\t\ttypname = findTypeByOid(tinfo, numTypes, finfo[i].argtypes[j]);\n+\t\tsprintf(q, \"%s%s%s\",\n+\t\t\t\tq,\n+\t\t\t\t(j > 0) ? \",\" : \"\",\n+\t\t\t\tfmtId(typname, false));\n+\t}\n+\tsprintf (q, \"%s);\\n\", q);\n+\tfputs(q, fout);\n+\n \tsprintf(q, \"CREATE FUNCTION %s (\", fmtId(finfo[i].proname, force_quotes));\n \tfor (j = 0; j < finfo[i].nargs; j++)\n \t{\n@@ -2347,6 +2373,11 @@\n \n \t\tbecomeUser(fout, oprinfo[i].usename);\n \n+\t\tsprintf(q, \"DROP OPERATOR %s (%s, %s);\\n\", oprinfo[i].oprname, \n+\t\t\tfmtId(findTypeByOid(tinfo, numTypes, oprinfo[i].oprleft), false),\n+\t\t\tfmtId(findTypeByOid(tinfo, numTypes, oprinfo[i].oprright), false));\n+\t\tfputs(q, fout);\n+\n \t\tsprintf(q,\n \t\t\t\t\"CREATE OPERATOR %s \"\n \t\t\t\t\"(PROCEDURE = %s %s %s %s %s %s %s %s %s);\\n \",\n@@ -2442,6 +2473,10 @@\n \n \t\tbecomeUser(fout, agginfo[i].usename);\n \n+\t\tsprintf(q, \"DROP AGGREGATE %s %s;\\n\", agginfo[i].aggname,\n+\t\t\tfmtId(findTypeByOid(tinfo, numTypes, agginfo[i].aggbasetype), false));\n+\t\tfputs(q, fout);\n+\n \t\tsprintf(q, \"CREATE AGGREGATE %s ( %s %s%s %s%s %s );\\n\",\n \t\t\t\tagginfo[i].aggname,\n \t\t\t\tbasetype,\n@@ -2641,6 +2676,9 @@\n \n \t\t\tbecomeUser(fout, tblinfo[i].usename);\n \n+\t\t\tsprintf(q, \"DROP TABLE %s;\\n\", fmtId(tblinfo[i].relname, force_quotes));\n+\t\t\tfputs(q, fout);\n+\n \t\t\tsprintf(q, \"CREATE TABLE %s (\\n\\t\", fmtId(tblinfo[i].relname, force_quotes));\n \t\t\tactual_atts = 0;\n \t\t\tfor (j = 0; j < tblinfo[i].numatts; j++)\n@@ -2857,6 +2895,10 @@\n \n \t\t\tstrcpy(id1, fmtId(indinfo[i].indexrelname, force_quotes));\n \t\t\tstrcpy(id2, fmtId(indinfo[i].indrelname, force_quotes));\n+\n+\t\t\tsprintf(q, \"DROP INDEX %s;\\n\", id1);\n+\t\t\tfputs(q, fout);\n+\n \t\t\tfprintf(fout, \"CREATE %s INDEX %s on %s using %s (\",\n \t\t\t (strcmp(indinfo[i].indisunique, \"t\") == 0) ? \"UNIQUE\" : \"\",\n \t\t\t\t\tid1,\n@@ -3116,6 +3158,9 @@\n \tcalled = *t;\n \n \tPQclear(res);\n+\n+\tsprintf(query, \"DROP SEQUENCE %s;\\n\", fmtId(tbinfo.relname, force_quotes));\n+\tfputs(query, fout);\n \n \tsprintf(query,\n \t\t\t\"CREATE SEQUENCE %s start %d increment %d maxvalue %d \"\n",
"msg_date": "Fri, 15 Jan 1999 14:08:46 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "drop before create in pg_dump"
},
{
"msg_contents": "Brook Milligan <[email protected]> writes:\n> pg_dump won't drop stuff before trying to create it; this makes\n> dropping a single table (say) and recreating it difficult to automate\n> since it is subject to error if all the stuff doesn't get dropped\n> properly. The following patch causes pg_dump to emit DROP ...\n> statements prior to emitting CREATE ... statements.\n\nI think that ought to be driven by a switch to pg_dump ...\nmuch of the time, I would *want* pg_dump's script to fail\nif there's already an existing item of the given name.\n\nI see your concern, I just don't think it's the only scenario.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Fri, 15 Jan 1999 18:44:32 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] drop before create in pg_dump "
},
{
"msg_contents": "-----BEGIN PGP SIGNED MESSAGE-----\n\n>>>>> \"bm\" == Brook Milligan <[email protected]> writes:\n\n bm> The following patch causes pg_dump to emit DROP ... statements\n bm> prior to emitting CREATE ... statements.\n\nI'm not sure I want pg_dump output to automatically drop tables. I'd\ncertainly prefer having to ask for this behavior....\n\nroland\n- -- \n\t\t PGP Key ID: 66 BC 3B CD\nRoland B. Roberts, PhD Custom Software Solutions\[email protected] 101 West 15th St #4NN\[email protected] New York, NY 10011\n\n-----BEGIN PGP SIGNATURE-----\nVersion: 2.6.2\nComment: Processed by Mailcrypt 3.4, an Emacs/PGP interface\n\niQCVAwUBNp/5duoW38lmvDvNAQGJ4QP9EmXQWyUtgJi+5EkykZVU5IW9IMj/YcFa\nyGXF6eGP1SfLsW7rQPHgIljPvbT9B6WobYpGrluqA00XoW0/Wsm4TmjXuGQxRiI0\nVxveN127wFsKrA0oJb8g9IfziyR2oeUba9oPNK3Yg2sYAXvd8/schYf6Ub2KUH5Q\nFS7ylJV1awY=\n=y2FK\n-----END PGP SIGNATURE-----\n",
"msg_date": "15 Jan 1999 21:29:12 -0500",
"msg_from": "Roland Roberts <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] drop before create in pg_dump"
},
{
"msg_contents": "> Brook Milligan <[email protected]> writes:\n> > pg_dump won't drop stuff before trying to create it; this makes\n> > dropping a single table (say) and recreating it difficult to automate\n> > since it is subject to error if all the stuff doesn't get dropped\n> > properly. The following patch causes pg_dump to emit DROP ...\n> > statements prior to emitting CREATE ... statements.\n> \n> I think that ought to be driven by a switch to pg_dump ...\n> much of the time, I would *want* pg_dump's script to fail\n> if there's already an existing item of the given name.\n> \n> I see your concern, I just don't think it's the only scenario.\n\nI agree. I don't like it removing stuff before creating it. pg_dump\nalready does a lot. But I do agree it is difficult to automate\ndestroy/create.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Fri, 15 Jan 1999 22:50:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] drop before create in pg_dump"
},
{
"msg_contents": " Brook Milligan <[email protected]> writes:\n > pg_dump won't drop stuff before trying to create it; this makes\n > dropping a single table (say) and recreating it difficult to automate\n > since it is subject to error if all the stuff doesn't get dropped\n > properly. The following patch causes pg_dump to emit DROP ...\n > statements prior to emitting CREATE ... statements.\n\n I think that ought to be driven by a switch to pg_dump ...\n much of the time, I would *want* pg_dump's script to fail\n if there's already an existing item of the given name.\n\nA switch is fine. Is there any concensus as to what it should be,\nsince the two that come to mind (-D and -d) are both used?\n\nCheers,\nBrook\n",
"msg_date": "Sun, 17 Jan 1999 15:11:47 -0700 (MST)",
"msg_from": "Brook Milligan <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] drop before create in pg_dump"
},
{
"msg_contents": "> Brook Milligan <[email protected]> writes:\n> > pg_dump won't drop stuff before trying to create it; this makes\n> > dropping a single table (say) and recreating it difficult to automate\n> > since it is subject to error if all the stuff doesn't get dropped\n> > properly. The following patch causes pg_dump to emit DROP ...\n> > statements prior to emitting CREATE ... statements.\n> \n> I think that ought to be driven by a switch to pg_dump ...\n> much of the time, I would *want* pg_dump's script to fail\n> if there's already an existing item of the given name.\n> \n> A switch is fine. Is there any concensus as to what it should be,\n> since the two that come to mind (-D and -d) are both used?\n\n-c for clean.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 17 Jan 1999 23:51:14 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] drop before create in pg_dump"
}
] |
[
{
"msg_contents": "\nWith this morning's conversations about the time zones in Alaska and\nHawaii, I asked a friend of mine about Hawaii (he just moved back to \nTexas after 20+ years there).\n\n-----------------\nHawaii Standard I guess is the Time Zone. They do NOT observe Daylight\nSavings..even tho' Alaska, in the same Time Zone does.\n\nCurrently, Hawaii is 5 hours behind EST, 4 hours behind CST, 3hours behind\nMST, and 2hours behind Califrisco or PST. During Daylight Savings Time add\none hour to those numbers. If I remember correctly, Hawaii is 10 hours\nbehind GMT..but you need to check that.\n\nI'd do it for you in my Map Control Panel on my Mac...but I'm about to rush\nout the door.\n\nHope this helps....\n\nBTW Honolulu, ( and all the Islands in the Hawaiian chain, are exactly 12\nhours opposite Durban, South Africa. Don't ask why I know that.\n-----------------\n\nHe seems to indicate there is only one Alaska time zone, I didn't ask\nthat directly and my fone book doesn't cover west of the Pacific zone.\nAccording to my cell phone, *MOST* of Alaska is in the same tz but the\nwestern most cost is in the next one over - however it doesn't give the\noption of selecting it as a valid time zone.\n\nVince.\n-- \n==========================================================================\nVince Vielhaber -- KA8CSH email: [email protected] flame-mail: /dev/null\n # include <std/disclaimers.h> TEAM-OS2\n Online Searchable Campground Listings http://www.camping-usa.com\n \"There is no outfit less entitled to lecture me about bloat\n than the federal government\" -- Tony Snow\n==========================================================================\n\n\n",
"msg_date": "Fri, 15 Jan 1999 19:39:14 -0500 (EST)",
"msg_from": "Vince Vielhaber <[email protected]>",
"msg_from_op": true,
"msg_subject": "Time Zones"
},
{
"msg_contents": "> With this morning's conversations about the time zones in Alaska and\n> Hawaii, I asked a friend of mine about Hawaii (he just moved back to\n> Texas after 20+ years there).\n\nThanks Vince. More than I ever wanted to know; now I suppose I'll have\nto go back and check that everything else matches up too? ;)\n\n - Tom\n",
"msg_date": "Sat, 16 Jan 1999 02:42:47 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Time Zones"
},
{
"msg_contents": "Another question on time zones:\n\nDoes \"Iran Time\", abbreviated \"IT\", actually fall +3:40 away from UTC?\nOr do I have another typo to fix??\n\n - Tom\n",
"msg_date": "Wed, 20 Jan 1999 02:43:19 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Time Zones"
},
{
"msg_contents": "\nOn 20-Jan-99 Thomas G. Lockhart wrote:\n> Another question on time zones:\n> \n> Does \"Iran Time\", abbreviated \"IT\", actually fall +3:40 away from UTC?\n> Or do I have another typo to fix??\n> \n> - Tom\n> \n\nThis is from: http://www.tradenz.govt.nz/intelligence/profiles/iran.html\n\nTime Zones - Iran\n\n3 1/2 hours ahead of GMT \n8 1/2 hours behind NZ \n9 hours behind NZ daylight savings.\n\nI saw a couple of references to 30 mins, so far none for 40 mins.\n\nVince.\n-- \n==========================================================================\nVince Vielhaber -- KA8CSH email: [email protected] flame-mail: /dev/null\n # include <std/disclaimers.h> TEAM-OS2\n Online Searchable Campground Listings http://www.camping-usa.com\n \"There is no outfit less entitled to lecture me about bloat\n than the federal government\" -- Tony Snow\n==========================================================================\n\n\n",
"msg_date": "Tue, 19 Jan 1999 21:59:51 -0500 (EST)",
"msg_from": "Vince Vielhaber <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Time Zones"
}
] |
[
{
"msg_contents": "Please back out this recently added patch... it doesn't compile:\n\n-- log entry --\nFrom: Magnus Hagander <[email protected]>\n\nHere's another patch for the libpq backend areas. This patch removes all\nusage of \"FILE *\" on the communications channel. It also cleans up the\ncomments and headers in the pqcomm.c file - a lot of things were either\nmissing or incorrect. Finally, it removes a couple of unused functions\n(leftovers from the time of shared code between the libpq backend and\nfrontend).\n----\n\nFiles modified:\npgsql/src/backend/libpq/pqcomm.c\npgsql/src/backend/libpq/pqcomprim.c\npgsql/src/include/libpq/libpq.h\npgsql/src/include/libpq/pqcomm.h\n\nTaral -- maintenant de retour\n",
"msg_date": "Fri, 15 Jan 1999 22:40:53 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": true,
"msg_subject": "Repository no longer compiles"
}
] |
[
{
"msg_contents": "Apologies if that repository problem was already mentioned... for some reason I\nwas no longer subscribed to this list.\n\nTaral\n",
"msg_date": "Fri, 15 Jan 1999 22:47:35 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": true,
"msg_subject": "Unsubscribed?"
}
] |
[
{
"msg_contents": "Seems at some point someone decided not to be machine-independent in\nbackend/utils/adt/float.c\n\nRedhat 5.2 system with gcc 2.8.1 doesn't define NAN unless _GNU_SOURCE is\ndefined first. Although include/utils/dt.h *does* check for NAN (and HUGE_VAL)\nbefore using them, the float.c doesn't.\n\nWhat happened?\n\nTaral\n",
"msg_date": "Fri, 15 Jan 1999 23:27:41 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": true,
"msg_subject": "backend/utils/adt/float.c uses non-existent NAN value"
},
{
"msg_contents": "> Seems at some point someone decided not to be machine-independent in\n> backend/utils/adt/float.c\n> \n> Redhat 5.2 system with gcc 2.8.1 doesn't define NAN unless _GNU_SOURCE is\n> defined first. Although include/utils/dt.h *does* check for NAN (and HUGE_VAL)\n> before using them, the float.c doesn't.\n> \n> What happened?\n\nI don't see NAN used in any file except numberic.c. I don't see it in\nfloat.c, at least in 6.5.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sat, 16 Jan 1999 00:40:20 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] backend/utils/adt/float.c uses non-existent NAN value"
},
{
"msg_contents": ">\n> Seems at some point someone decided not to be machine-independent in\n> backend/utils/adt/float.c\n>\n> Redhat 5.2 system with gcc 2.8.1 doesn't define NAN unless _GNU_SOURCE is\n> defined first. Although include/utils/dt.h *does* check for NAN (and HUGE_VAL)\n> before using them, the float.c doesn't.\n>\n> What happened?\n\n Don't know exactly - but\n\n the new NUMERIC data type started recently a discussion about\n NAN, because I tried to be able to convert NAN from/to\n float's, and that started some trouble. I'm not sure how this\n went into adt/float.c. At least I think we should fix all the\n NAN handling for v6.5.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Mon, 18 Jan 1999 09:36:10 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] backend/utils/adt/float.c uses non-existent NAN value"
},
{
"msg_contents": "> >\n> > Seems at some point someone decided not to be machine-independent in\n> > backend/utils/adt/float.c\n> >\n> > Redhat 5.2 system with gcc 2.8.1 doesn't define NAN unless _GNU_SOURCE is\n> > defined first. Although include/utils/dt.h *does* check for NAN (and HUGE_VAL)\n> > before using them, the float.c doesn't.\n> >\n> > What happened?\n> \n> Don't know exactly - but\n> \n> the new NUMERIC data type started recently a discussion about\n> NAN, because I tried to be able to convert NAN from/to\n> float's, and that started some trouble. I'm not sure how this\n> went into adt/float.c. At least I think we should fix all the\n> NAN handling for v6.5.\n> \n\nI fixed that. Added the NAN define from numeric.c.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 07:27:48 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] backend/utils/adt/float.c uses non-existent NAN value"
},
{
"msg_contents": "> > Seems at some point someone decided not to be machine-independent in\n> > backend/utils/adt/float.c\n\nThat was me.\n\n> > Redhat 5.2 system with gcc 2.8.1 doesn't define NAN unless \n> > _GNU_SOURCE is defined first. Although include/utils/dt.h *does* \n> > check for NAN (and HUGE_VAL) before using them, the float.c doesn't.\n> > What happened?\n> the new NUMERIC data type started recently a discussion about\n> NAN, because I tried to be able to convert NAN from/to\n> float's, and that started some trouble. I'm not sure how this\n> went into adt/float.c. At least I think we should fix all the\n> NAN handling for v6.5.\n\nJan started with NAN handling for numeric, and I'd already done the same\nthing for the date/time types quite a while ago. We also had a\ndiscussion recently that we should be handling 'NaN' and 'Infinity' on\nfloat8 input values so that dump/reload is more robust and symmetric, so\nI added that in ('-Infinity' still needs to be done).\n\nI added in the new code, knowing that there was some fixup and\nconsolidation which needs to be done. Unless someone has worked out\nanother scheme, I'd suggest using the date/time code in dt.h as an\nexample to work from. We might want to use PG_NAN and PG_INFINITY, which\nwe can then define in one central place.\n\nIf someone want to work this out, I'd be happy to use the result. Or I\ncan do something directly. I'd like a solution with the equivalent of\nNAN and HUGE_VAL (and -HUGE_VAL). Jan? Taral?\n\n - Tom\n",
"msg_date": "Mon, 18 Jan 1999 14:33:17 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] backend/utils/adt/float.c uses non-existent NAN value"
},
{
"msg_contents": "On Mon, 18 Jan 1999, you wrote:\n>If someone want to work this out, I'd be happy to use the result. Or I\n>can do something directly. I'd like a solution with the equivalent of\n>NAN and HUGE_VAL (and -HUGE_VAL). Jan? Taral?\n\nDoesn't dt.h have some replacements for NAN and HUGE_VAL? Something like that?\nCouldn't we use those?\n\nAha... DT_INVALID is the generic NAN, and DT_NOBEGIN/DT_NOEND are the generic\nHUGE_VAL and -HUGE_VAL...\n\nTaral\n",
"msg_date": "Mon, 18 Jan 1999 13:26:08 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] backend/utils/adt/float.c uses non-existent NAN value"
},
{
"msg_contents": "unsubscribe\n\n",
"msg_date": "Tue, 19 Jan 1999 09:41:29 +0700",
"msg_from": "Puchong <[email protected]>",
"msg_from_op": false,
"msg_subject": "(no subject)"
}
] |
[
{
"msg_contents": "On Fri, 15 Jan 1999, you wrote:\n>I don't see NAN used in any file except numberic.c. I don't see it in\n>float.c, at least in 6.5.\n\nThis is from the latest repository checkout...\n\nTara\n",
"msg_date": "Sat, 16 Jan 1999 00:11:11 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": true,
"msg_subject": "Fwd: Re: [HACKERS] backend/utils/adt/float.c uses non-existent NAN\n\tvalue"
}
] |
[
{
"msg_contents": "Hello.\n\nI was hoping someone could shed some light on the \nfollowing problem. I am currently working on an \nintranet web site (who isn't?) and am using\nPostgreSQL 6.4 as the database. On a daily basis, \nthe database needs to integrate several megs of data\ninto tables. Since the webserver needs to be up \n24 hours a day, 7 days a week, this integration process\nneeds to occur as quickly as possible.\n\nCurrently, the code which integrates the data which\nis ftp'd to the server from a mainframe, uses\nthe DELETE statement to empty existing tables. It does\nDROP indexes before importing and\nrecreates them after the import step. The\ndelete takes a significant amount of time more than\na DROP/CREATE TABLE pair, but makes it easier to change\nthe table constraints, attributes, etc without changing\ncode, and, in the future, the tables in question\nwill be accessed from multiple users with varying\npermissions. As a result, the code to DROP/CREATE\nthe tables would also have to reissue all the GRANTs\nnecessary to maintain the current state of user\npermissions.\n\nI was hoping to increase the speed of the operation\nby adding support to PostgreSQL for the Oracle\nTRUNCATE command. With version 7, Oracle introduced\nthe TRUNCATE command to quickly eliminate all the\nrows in a table. According to the Oracle docs, \nthe TRUNCATE command possesses the following qualities:\n\n1. It is a DDL statement and therefore cannot be \n rolled back\n2. It does not fire a table's DELETE triggers\n\nThis is where I am in what is my first attempt to \nmodify PostgreSQL in any meaningful way:\n\n1. I've added the grammer to gram.y, keywords.c, \n parse.h, pgindent.\n2. I've added the code to be called in \n tcop/utility.c\n3. I've added a RelationTruncate() to \n relcache.c (although it should probably be in \n creatinh.c)\n\nThe RelationTruncate() command calls heap_openr()\nto open the relation specified in the TRUNCATE\nTABLE statement, calls heap_truncate(), and\nthen calls TruncateBuffers().\n\nHere's where I have problem #1:\n\nheap_truncate() does the following:\nA. elog's an error if the relation is either a System\n relation or a Temporary relation\nB. Calls ReleaseRelationBuffers() on the relation\nC. Calls smgrtruncate() to truncate the relation to\n 0 blocks in size\nD. Calls smgrclose() to actually close the file\n descriptor\nE. Calls smgrcommit() to force the truncate to take\n place\nF. Calls heap_close() to close the relation\nG. And finally calls RelationForgetRelation() to\n eliminate the relation from the relation cache.\n\nOkay. When I first implemented it, I basically \nmirrored the DROP TABLE code until the point of\nheap_truncate() where I truncated the relation to 0\nbytes. This appeared to work. Obviously I still\nwould have to implement a RelationTruncateIndexes()\nto truncate all the indexes as well, but I haven't\ngotten that far yet.\n\nHere's what happens:\n\n> psql emptoris\nemptoris=> CREATE TABLE employees (\nemployee VARCHAR(10) not null\n);\nCREATE\nemptoris=> INSERT INTO employees values ('hello');\nINSERT 17481 1\nemptoris=> SELECT * FROM employees;\nemployee\n--------\nhello \n(1 row)\n\nemptoris=> TRUNCATE TABLE employees;\nTRUNCATE\nemptoris=> SELECT * FROM employees;\nemployee\n--------\n(0 rows)\n\n/* It took me about two hours of coding to get to */\n/* that point. And I've spent two days trying to */\n/* solve what happens next: */\n\nemptoris=> INSERT INTO employees VALUES ('one');\nINSERT 17505 1\nemptoris=> SELECT * from employees;\nemployee\n--------\nhello \none \n(2 rows)\n\nThis is where I get lost! Obviously something is \nsitting around somewhere either in the relation\ncache (which I explicitly cleared using \nRelationFogetRelation()) or in the Shared Memory\nSegment's Buffer Cache. After the TRUNCATE, the\n'employees' file on the filesystem is truncated to \n0 bytes in size. If I close the front-end and \nreissue the INSERT/SELECT the truncated data remains\nas above. However, if I close the backend after \nthe TRUNCATE, and restart it, the data is correctly\ntruncated to an empty table. I can't find where this\nin-memory copy of the table is to delete. I've\ntried marking all buffers in the buffer cache whose\nBufferDesc's have the same dbase and relid as the\nrelation being truncated as having been deleted:\n\nTruncateBuffers(Relation rel) {\n\n BufferDesc *buf;\n int i;\n\n SpinAcquire(BufMgrLock);\n for (i = 1; i <= NBuffers; i++) {\n buf = &BufferDescriptors[i - 1];\n if ((buf->tag.relId.dbId == MyDatabaseId) &&\n\t(buf->tag.relId.relId == \n RelationGetRelid(rel))) {\n buf->tag.relId.relId = InvalidOid;\n buf->flags |= BM_DELETED; \n }\n }\n SpinRelease(BufMgrLock);\n\nAs previously stated, I've also blown away the \nrelation cache by calling RelationForgetRelation().\nAnd the only reason I called smgrclose() was in a \nfailed attempt to force a flushing of wherever this\nmagical data is stored. I've also tried various\nCommitTransactions() calls, etc. All with no \nsuccess. If I only knew WHERE this data is being\ncached, despite my room temp. I.Q., I feel I could\nfigure out how to correctly purge it.\n\nI don't know where this data is continuing to live\nto purge it from the backend. I even took the\nliberty of perusing Stefan's (sp?) Master's Thesis.\n\nAny help would be appreciated. Even if its to let\nme know to not bother trying to modify a ferrari with\na sledge hammer.\n\nMarcus Mascari ([email protected])\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n_________________________________________________________\nDO YOU YAHOO!?\nGet your free @yahoo.com address at http://mail.yahoo.com\n\n",
"msg_date": "Sat, 16 Jan 1999 02:13:33 -0800 (PST)",
"msg_from": "Marcus Mascari <[email protected]>",
"msg_from_op": true,
"msg_subject": "Help with cache flushing -- trying to implement TRUNCATE"
}
] |
[
{
"msg_contents": "Comparing the current-CVS FAQ files with those in the REL6_4 branch\nis depressing --- almost every one is messed up in one branch or the\nother.\n\nThe main FAQ is newer in the 6.4 branch than in the current branch; the\n6.4 version seems to have numerous small fixes that are not in the main\nbranch. FAQ_CVS is also newer in 6.4; it's been de-HTMLized in 6.4\n(good) but not in current. FAQ_HPUX, FAQ_Irix, and FAQ_Linux are\ncompletely trashed in the 6.4 branch --- all three have been overwritten\nwith some generic document.\n\nI'd check in some fixes, but a look at the CVS logs makes me think that\npart of the problem is too many cooks stirring the broth already.\nWhoever is primary maintainer of these files probably should take\nresponsibility for getting the branches back in sync.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sat, 16 Jan 1999 17:52:12 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "doc/FAQ* files are hosed"
},
{
"msg_contents": "> Comparing the current-CVS FAQ files with those in the REL6_4 branch\n> is depressing --- almost every one is messed up in one branch or the\n> other.\n> \n> The main FAQ is newer in the 6.4 branch than in the current branch; the\n> 6.4 version seems to have numerous small fixes that are not in the main\n> branch. FAQ_CVS is also newer in 6.4; it's been de-HTMLized in 6.4\n> (good) but not in current. FAQ_HPUX, FAQ_Irix, and FAQ_Linux are\n> completely trashed in the 6.4 branch --- all three have been overwritten\n> with some generic document.\n\nI must have done that during the 6.4.2 release process. I decided to\ngrab them from the web site, but it looks like I got the wrong pages\nsomehow.\n\nThey get over-written with ever release, so I have another chance to try\nit for 6.5.\n\n\nThis does not apply to the FAQ_CVS. That's someone else.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sat, 16 Jan 1999 20:43:59 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] doc/FAQ* files are hosed"
},
{
"msg_contents": "Bruce Momjian <[email protected]> writes:\n>> Comparing the current-CVS FAQ files with those in the REL6_4 branch\n>> is depressing --- almost every one is messed up in one branch or the\n>> other.\n\n> They get over-written with ever release, so I have another chance to try\n> it for 6.5.\n\nOh? Aren't the CVS versions considered the master copies? If not,\nwhere *are* the master copies?\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 17 Jan 1999 15:00:18 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] doc/FAQ* files are hosed "
},
{
"msg_contents": "\nHello -\n\nI've been lurking on this list for a little while. I have just done a\nlittle tinkering with Postgres 6.4.2, comparing it to Oracle and mySQL on\nLinux. It would appear that just doing a lot of inserts (ie 40000) in a\nloop is enormously slower in Postgres in two ways that I tried it.\nOne - using a loop in Java that makes a JDBC call to insert.\nTwo- using plpgsql as a comparision to PL/SQL.\n\nPerhaps these are bad techniques and a C based proc would do better?\n\nI really like the idea of an open source DB and am impressed with a lot I\nsee about PostgreSQL - however the speed difference is pretty bad -\n4.5 minutes versus about 20 seconds.\n\nIs that just the way it goes? Are there options that would make it faster\nthat are not the defaults?\n\nAny reaction would be appreciated.\n\nJohn Holland\n\n",
"msg_date": "Sun, 17 Jan 1999 21:08:28 -0500 (EST)",
"msg_from": "John Holland <[email protected]>",
"msg_from_op": false,
"msg_subject": "Postgres Speed or lack thereof"
},
{
"msg_contents": "John Holland wrote:\n> \n> Hello -\n> \n> I've been lurking on this list for a little while. I have just done a\n> little tinkering with Postgres 6.4.2, comparing it to Oracle and mySQL on\n> Linux. It would appear that just doing a lot of inserts (ie 40000) in a\n> loop is enormously slower in Postgres in two ways that I tried it.\n> One - using a loop in Java that makes a JDBC call to insert.\n> Two- using plpgsql as a comparision to PL/SQL.\n> \n> Perhaps these are bad techniques and a C based proc would do better?\n> \n> I really like the idea of an open source DB and am impressed with a lot I\n> see about PostgreSQL - however the speed difference is pretty bad -\n> 4.5 minutes versus about 20 seconds.\n> \n> Is that just the way it goes? Are there options that would make it faster\n> that are not the defaults?\n\nOracle uses chained transaction mode (i.e. - all queries run in\nsingle transaction untill explicit COMMIT/ABORT), MySQL hasn't\ntransaction at all...\n\nPostgreSQL uses un-chained transaction mode: each query runs\nin own transaction - 40000 transaction commits !!!\n\nTry to use BEGIN/END to run all inserts in single transaction\nand please let us know results.\n\nVadim\n",
"msg_date": "Mon, 18 Jan 1999 09:29:45 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> Bruce Momjian <[email protected]> writes:\n> >> Comparing the current-CVS FAQ files with those in the REL6_4 branch\n> >> is depressing --- almost every one is messed up in one branch or the\n> >> other.\n> \n> > They get over-written with ever release, so I have another chance to try\n> > it for 6.5.\n> \n> Oh? Aren't the CVS versions considered the master copies? If not,\n> where *are* the master copies?\n\nThe master copies are on the web site. In fact, people send e-mail to a\nspecial account here, and they get transfered to the web site\nautomatically.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sun, 17 Jan 1999 22:03:41 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] doc/FAQ* files are hosed"
},
{
"msg_contents": "\nOn Sun, 17 Jan 1999, Tom Lane wrote:\n\n> I tried this myself and found that wrapping BEGIN/END around a series of\n> INSERT statements didn't make much difference at all.\n\n Using what API? Some APIs control autocommit for you, so execing\n\"BEGIN\" and \"END\" commands may not do anything. You might want to check\nyour API for a way to set autocommit off. In fact, I would suggest that\nyou never use BEGIN and END except via cli.\n\nTom\n\n",
"msg_date": "Sun, 17 Jan 1999 19:14:05 -0800 (PST)",
"msg_from": "Tom <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "\nOn Sun, 17 Jan 1999, Tom Lane wrote:\n\n> Tom <[email protected]> writes:\n> > On Sun, 17 Jan 1999, Tom Lane wrote:\n> >> I tried this myself and found that wrapping BEGIN/END around a series of\n> >> INSERT statements didn't make much difference at all.\n> \n> > Using what API?\n> \n> Sorry, I neglected to specify that it was psql (being driven by\n> hand-trimmed pg_dump scripts).\n\n This also wouldn't you give you any benefit from \"prepare\" that would\nspeed up repeated executions of the same statement. PostgreSQL does allow\nstatements to be prepared right? If it doesn't, chances are COPY has a\nway of doing it. It saves a lot of parsing and planning time.\n\n> > Some APIs control autocommit for you, so execing\n> > \"BEGIN\" and \"END\" commands may not do anything.\n> \n> That's not the case for psql. If it were, I would have measured no\n> difference in speed, rather than a small difference.\n> \n> \t\t\tregards, tom lane\n\nTom\n\n",
"msg_date": "Sun, 17 Jan 1999 19:23:15 -0800 (PST)",
"msg_from": "Tom <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Vadim Mikheev <[email protected]> writes:\n> John Holland wrote:\n>> I've been lurking on this list for a little while. I have just done a\n>> little tinkering with Postgres 6.4.2, comparing it to Oracle and mySQL on\n>> Linux. It would appear that just doing a lot of inserts (ie 40000) in a\n>> loop is enormously slower in Postgres in two ways that I tried it.\n>> One - using a loop in Java that makes a JDBC call to insert.\n>> Two- using plpgsql as a comparision to PL/SQL.\n>> \n>> I really like the idea of an open source DB and am impressed with a lot I\n>> see about PostgreSQL - however the speed difference is pretty bad -\n>> 4.5 minutes versus about 20 seconds.\n\n> Try to use BEGIN/END to run all inserts in single transaction\n> and please let us know results.\n\nI tried this myself and found that wrapping BEGIN/END around a series of\nINSERT statements didn't make much difference at all.\n\nOn an HP 715 (75MHz RISC box, not very heavy iron by current standards),\nI inserted about 13000 records into an initially-empty table having 38\ncolumns (just because it's data I had handy...). I timed it at:\n\nIndividual INSERT statements (as generated by pg_dump -d):\n\t33 inserts/sec\nSame with BEGIN/END wrapped around 'em:\n\t34 inserts/sec\nSingle COPY statement (as generated by pg_dump without -d):\n\t1400 inserts/sec\n\nThis was for a simple table with no indexes. In reality, this table\ntype has four b-tree indexes on different columns in our normal usage.\nThe COPY speed dropped to 325 inserts/sec when I had the indexes in\nplace. I didn't bother trying the INSERTs that way.\n\nThe conventional wisdom is that you should use COPY for bulk loading,\nand this result supports it...\n\nJohn didn't say what hardware he's using, so I don't know how comparable\nmy result is to his 150 inserts/sec --- that might have been on a table\nwith many fewer columns, or maybe his machine is just faster.\n\nAs for where the time is going: \"top\" showed that the time for the\nINSERT ops was almost all going into backend CPU time. My guess is\nthat most of the cost is in parsing and planning the INSERT statements.\nPre-planned statement skeletons might help, but I think the real answer\nwill be to find a way to avoid constructing and parsing SQL statements\nfor simple operations like INSERT. (One reason I'm interested in the\nidea of a CORBA interface is that it might help here.)\n\nMy guess is that Oracle and mySQL have APIs that allow the construction\nof an SQL INSERT command to be bypassed, and that's why they show up\nbetter on this operation.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 17 Jan 1999 22:34:39 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom <[email protected]> writes:\n> On Sun, 17 Jan 1999, Tom Lane wrote:\n>> I tried this myself and found that wrapping BEGIN/END around a series of\n>> INSERT statements didn't make much difference at all.\n\n> Using what API?\n\nSorry, I neglected to specify that it was psql (being driven by\nhand-trimmed pg_dump scripts).\n\n> Some APIs control autocommit for you, so execing\n> \"BEGIN\" and \"END\" commands may not do anything.\n\nThat's not the case for psql. If it were, I would have measured no\ndifference in speed, rather than a small difference.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 17 Jan 1999 23:21:45 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom Lane wrote:\n> \n> > Try to use BEGIN/END to run all inserts in single transaction\n> > and please let us know results.\n> \n> I tried this myself and found that wrapping BEGIN/END around a series of\n> INSERT statements didn't make much difference at all.\n> \n> On an HP 715 (75MHz RISC box, not very heavy iron by current standards),\n> I inserted about 13000 records into an initially-empty table having 38\n ^^\n> columns (just because it's data I had handy...). I timed it at:\n ^^^^^^^\nSo much.\n\n> Individual INSERT statements (as generated by pg_dump -d):\n> 33 inserts/sec\n> Same with BEGIN/END wrapped around 'em:\n> 34 inserts/sec\n> Single COPY statement (as generated by pg_dump without -d):\n> 1400 inserts/sec\n> \n...\n> \n> John didn't say what hardware he's using, so I don't know how comparable\n> my result is to his 150 inserts/sec --- that might have been on a table\n> with many fewer columns, or maybe his machine is just faster.\n> \n> As for where the time is going: \"top\" showed that the time for the\n> INSERT ops was almost all going into backend CPU time. My guess is\n> that most of the cost is in parsing and planning the INSERT statements.\n> Pre-planned statement skeletons might help, but I think the real answer\n> will be to find a way to avoid constructing and parsing SQL statements\n> for simple operations like INSERT. (One reason I'm interested in the\n> idea of a CORBA interface is that it might help here.)\n> \n> My guess is that Oracle and mySQL have APIs that allow the construction\n> of an SQL INSERT command to be bypassed, and that's why they show up\n> better on this operation.\n\nFor the table with single int4 column & 2000 INSERTs I have:\n\nBEGIN/END:\t\t3.5sec\t\t~600insert/sec\nWithout:\t 151sec\t\t~13insert/sec\n\n(IDE disk!)\n\nI guess that you don't use -F flag..?\n\nVadim\n",
"msg_date": "Mon, 18 Jan 1999 11:54:44 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "I wrote:\n> As for where the time is going: \"top\" showed that the time for the\n> INSERT ops was almost all going into backend CPU time. My guess is\n> that most of the cost is in parsing and planning the INSERT\n> statements.\n\nHaving an idle hour this evening, I thought it'd be interesting to build\na backend with profiling enabled, so as to confirm or deny the above\nguess. It seems that indeed a lot of time is being wasted, but where\nit's being wasted might surprise you!\n\nI profiled a backend built from this morning's current CVS sources,\nbeing driven by a single psql run reading a series of exactly 5000\nINSERT commands. The first few are:\n\nINSERT INTO \"fehistory_1\" values (64,14,252,'D','-','-','s',1,4027,NULL,'S',1,4088,NULL,32,'Thu Jan 07 15:15:00 1999 EST','Fri Jan 08 08:30:38 1999 EST',4027,32,NULL,NULL,46,NULL);\nINSERT INTO \"fehistory_1\" values (65,14,252,'P','-','-','S',1,4086,NULL,NULL,NULL,NULL,NULL,32,'Fri Jan 08 16:00:00 1999 EST',NULL,NULL,NULL,NULL,NULL,45,NULL);\nINSERT INTO \"fehistory_1\" values (66,14,135,'P','-','-','S',1,13619,NULL,'s',1,12967,NULL,100,'Fri Jan 08 02:00:00 1999 EST',NULL,NULL,NULL,NULL,NULL,44,NULL);\n\nand it goes on for a long while in that vein :-). Total runtime was\n278.09 CPU seconds in the backend, 307 seconds elapsed realtime. The\ntable being inserted into has no indexes and was initially empty.\n\nThe profiler is GNU gprof, which provides both flat and structured\nprofile data. The top few entries in the flat profile are:\n\n % cumulative self self total \n time seconds seconds calls ms/call ms/call name \n 34.96 97.21 97.21 _mcount\n 7.18 117.19 19.98 8047816 0.00 0.00 tree_cut\n 6.56 135.43 18.24 962187 0.02 0.02 recv\n 3.55 145.31 9.88 8234278 0.00 0.00 tree_insert\n 3.49 155.02 9.71 4292947 0.00 0.01 malloc\n 3.31 164.22 9.20 1965056 0.00 0.01 SearchSysCache\n 3.24 173.23 9.01 4291775 0.00 0.01 free\n 2.44 180.01 6.78 6171667 0.00 0.00 tree_concatenate\n 2.17 186.04 6.03 8234069 0.00 0.00 tree_delete\n 1.07 189.02 2.98 10250 0.29 0.29 write\n 1.05 191.94 2.92 _psort_cmp\n 0.95 194.57 2.63 2983301 0.00 0.00 newNode\n 0.92 197.14 2.57 4251762 0.00 0.00 OrderedElemPop\n 0.88 199.60 2.46 2746172 0.00 0.00 comphash\n 0.82 201.88 2.28 4252495 0.00 0.01 AllocSetAlloc\n 0.77 204.02 2.14 1965056 0.00 0.00 SearchSysCacheTuple\n 0.73 206.05 2.03 265000 0.01 0.01 yylex\n 0.69 207.96 1.91 $$remU\n 0.68 209.86 1.90 4252500 0.00 0.00 OrderedElemPushHead\n 0.62 211.58 1.72 4247473 0.00 0.00 palloc\n 0.60 213.26 1.68 25004 0.07 0.07 send\n 0.56 214.81 1.55 1965056 0.00 0.00 CatalogCacheComputeHashIndex\n 0.55 216.33 1.52 5000 0.30 1.07 yyparse\n 0.54 217.82 1.49 4251762 0.00 0.01 AllocSetFree\n 0.53 219.29 1.47 4252494 0.00 0.00 MemoryContextAlloc\n 0.51 220.71 1.42 4252500 0.00 0.00 OrderedElemPushInto\n 0.49 222.07 1.36 3430844 0.00 0.00 strncmp\n 0.48 223.41 1.34 4205327 0.00 0.00 OrderedSetGetHead\n 0.47 224.73 1.32 elog\n 0.45 225.99 1.26 1964976 0.00 0.00 DLRemove\n 0.43 227.19 1.20 2510857 0.00 0.00 strcmp\n 0.43 228.38 1.19 4200327 0.00 0.00 OrderedElemGetBase\n 0.42 229.56 1.18 4245740 0.00 0.01 PortalHeapMemoryAlloc\n 0.38 230.62 1.06 4252500 0.00 0.00 OrderedElemPush\n 0.37 231.66 1.04 4205327 0.00 0.00 AllocSetGetFirst\n 0.36 232.66 1.00 1965145 0.00 0.00 DLAddHead\n\n_mcount is part of the profiling apparatus and can be ignored (although\nthe fact that it's so high up implies that we're executing an awful lot\nof short routine calls, because mcount runs once per entry and exit of\nprofilable routines). tree_cut, tree_insert, tree_concatenate, and\ntree_delete are subroutines of malloc/free. Altogether, malloc/free\nand friends accounted for 61.39 seconds out of the 180.88 non-overhead\nCPU seconds in this run.\n\nIn other words, we're spending a third of our time mallocing and freeing\nmemory. A tad high, what?\n\nActually, it's worse than that, because AllocSetAlloc,\nPortalHeapMemoryAlloc, AllocSetFree, and all of the OrderedElemXXX\nroutines represent our own bookkeeping layer atop malloc/free.\nThat's another 18.66 seconds spent in these top routines, which means\nthat we are real close to expending half the backend's runtime on\nmemory bookkeeping. This needs work.\n\nThe other thing that jumps out here is the unreasonably high position of\nrecv(), which is called 962187 times. The script being read by psql was\nonly 957186 characters. Evidently we're invoking a kernel recv() call\nonce per character read from the frontend. I suspect this is an\ninefficiency introduced by Magnus Hagander's recent rewrite of backend\nlibpq (see, I told you there was a reason for using stdio ;-)). We're\ngonna have to do something about that, though it's not as critical as\nthe memory-allocation issue. It also appears that send() is now being\ninvoked multiple times per backend reply, which is going to create\ninefficiencies outside the backend (ie, multiple packets per reply).\nOn a test case with a lot of SELECTs that would show up more than it\ndoes here.\n\n\nMoving on to the dynamic profile, the most interesting items are:\n\nindex % time self children called name\n-----------------------------------------------\n 0.05 118.06 1/1 DoBackend [7]\n[8] 65.3 0.05 118.06 1 PostgresMain [8]\n 0.04 57.64 5000/5000 pg_exec_query [9]\n 0.01 39.83 5000/5000 CommitTransactionCommand [11]\n 0.02 18.86 5001/5001 ReadCommand [27]\n 0.01 1.18 5000/5000 StartTransactionCommand [118]\n 0.01 0.36 5001/5001 ReadyForQuery [190]\n 0.04 0.00 5001/5001 EnableNotifyInterrupt [385]\n 0.04 0.00 5000/5000 strspn [388]\n 0.00 0.01 1/1 InitPostgres [495]\n 0.00 0.00 5000/548573 strlen [172]\n 0.00 0.00 5001/93349 memset [366]\n 0.00 0.00 1/1 read_pg_options [601]\n 0.00 0.00 2/2 pq_putint [637]\n 0.00 0.00 1/15002 pq_putnchar [123]\n 0.00 0.00 2/10013 getenv [241]\n 0.00 0.00 1/1 SetPgUserName [683]\n 0.00 0.00 1/1 pq_init [687]\n 0.00 0.00 1/1 proc_exit [704]\n 0.00 0.00 1/1 pq_close [721]\n 0.00 0.00 1/5002 getpgusername [521]\n 0.00 0.00 1/2 FindExec [722]\n 0.00 0.00 5/8 getopt [728]\n 0.00 0.00 5001/5001 DisableNotifyInterrupt [761]\n 0.00 0.00 9/34 pqsignal [806]\n 0.00 0.00 2/5 atoi [837]\n 0.00 0.00 1/2 SetProcessingMode [868]\n 0.00 0.00 1/10002 TransactionFlushEnabled [749]\n 0.00 0.00 1/1 sigsetjmp [934]\n\n(\"self\" is the amount of time (CPU sec) spent directly in the named function,\nwhile \"children\" is the amount of time spent in its callees, pro-rated\nby the number of calls. For example, strlen was actually measured to\nuse 0.47 sec in the whole run, but since PostgresMain called it less\nthan 1% of all the times it was called, PostgresMain gets blamed for\nless than 0.01 sec of that total.)\n\nThe interesting thing here is the relative times indicated for\npg_exec_query, CommitTransactionCommand, and ReadCommand. We already\nfound out why ReadCommand is so slow (recv() per character) ... but why\nis CommitTransactionCommand so high, when I already discovered that\ndoing only one commit doesn't help much? Patience, we'll get there.\n\n-----------------------------------------------\n 0.03 57.61 5000/5000 pg_exec_query [9]\n[10] 31.9 0.03 57.61 5000 pg_exec_query_dest [10]\n 0.09 39.12 5000/5000 pg_parse_and_plan [13]\n 0.00 17.37 5000/5000 ProcessQuery [36]\n 0.02 0.60 5000/5000 SetQuerySnapshot [159]\n 0.01 0.34 5000/5000 CommandCounterIncrement [196]\n 0.02 0.05 10000/4291775 free [22]\n-----------------------------------------------\n 0.09 39.12 5000/5000 pg_exec_query_dest [10]\n[13] 21.7 0.09 39.12 5000 pg_parse_and_plan [13]\n 0.02 19.28 5000/5000 parser [26]\n 0.03 17.63 5000/5000 planner [34]\n 0.00 1.98 5000/5000 QueryRewrite [90]\n 0.02 0.05 10000/4292947 malloc [21]\n 0.02 0.05 10000/4291775 free [22]\n 0.00 0.02 5000/497779 lappend [91]\n 0.01 0.00 5000/5000 IsAbortedTransactionBlockState [529]\n 0.00 0.00 5000/72779 length [342]\n\nIn other words, parsing, planning, and executing an INSERT each take\nabout the same amount of time.\n\n-----------------------------------------------\n 0.01 39.83 5000/5000 PostgresMain [8]\n[11] 22.0 0.01 39.83 5000 CommitTransactionCommand [11]\n 0.02 39.81 5000/5000 CommitTransaction [12]\n-----------------------------------------------\n 0.02 39.81 5000/5000 CommitTransactionCommand [11]\n[12] 22.0 0.02 39.81 5000 CommitTransaction [12]\n 0.00 38.12 5000/5000 AtCommit_Memory [16]\n 0.01 1.10 5000/5000 RecordTransactionCommit [122]\n 0.00 0.30 5000/5000 AtCommit_Locks [211]\n 0.02 0.10 5000/5000 AtEOXact_portals [286]\n 0.03 0.07 5000/5000 DestroyTempRels [305]\n 0.03 0.00 5000/5000 _lo_commit [425]\n 0.01 0.00 5000/5000 AtCommit_Notify [522]\n 0.01 0.00 5000/5000 CloseSequences [523]\n 0.01 0.00 5000/5000 RelationPurgeLocalRelation [531]\n 0.00 0.00 5000/10000 AtCommit_Cache [750]\n-----------------------------------------------\n 0.00 38.12 5000/5000 CommitTransaction [12]\n[16] 21.1 0.00 38.12 5000 AtCommit_Memory [16]\n 0.01 38.11 5000/5000 EndPortalAllocMode [17]\n 0.00 0.00 5000/10552 MemoryContextSwitchTo [747]\n-----------------------------------------------\n 0.01 38.11 5000/5000 AtCommit_Memory [16]\n[17] 21.1 0.01 38.11 5000 EndPortalAllocMode [17]\n 0.54 37.55 5000/5000 AllocSetReset [18]\n 0.01 0.00 5000/10000 PortalHeapMemoryGetVariableMemory [449]\n 0.01 0.00 5000/5000 FixedStackPop [528]\n 0.00 0.00 5000/51434 MemoryContextFree [367]\n-----------------------------------------------\n 0.54 37.55 5000/5000 EndPortalAllocMode [17]\n[18] 21.1 0.54 37.55 5000 AllocSetReset [18]\n 1.47 32.51 4200327/4251762 AllocSetFree [20]\n 1.04 2.53 4205327/4205327 AllocSetGetFirst [64]\n\nIn other words, essentially *all* of the CPU time spent in\nCommitTransaction is spent freeing memory. That's probably why\nganging the transactions doesn't help --- it's the same number of\nmemory blocks getting allocated and freed.\n\n 0.02 19.28 5000/5000 pg_parse_and_plan [13]\n[26] 10.7 0.02 19.28 5000 parser [26]\n 0.02 13.88 5000/5000 parse_analyze [41]\n 1.52 3.81 5000/5000 yyparse [54]\n 0.01 0.01 5000/5000 init_io [474]\n 0.00 0.01 5000/420496 pstrdup [124]\n 0.01 0.00 5000/5000 parser_init [533]\n\nThomas might know why parse_analyze is taking so much time compared to\nthe rest of the parsing machinery...\n\n\nI won't bother the list with the rest of the profile, although I'll be\nhappy to send it to anyone who wants to see all of it. Our immediate\nperformance problems seem pretty clear, however: time to rethink memory\nmanagement.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 18 Jan 1999 00:31:43 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "On Sun, Jan 17, 1999 at 09:08:28PM -0500, John Holland wrote:\n> I've been lurking on this list for a little while. I have just done a\n> little tinkering with Postgres 6.4.2, comparing it to Oracle and mySQL on\n> Linux. It would appear that just doing a lot of inserts (ie 40000) in a\n> loop is enormously slower in Postgres in two ways that I tried it.\n> One - using a loop in Java that makes a JDBC call to insert.\n> Two- using plpgsql as a comparision to PL/SQL.\n\nDid you use the '-F' option? I tried somethings similar from C and found\nsimilar results when calling fsync() after each write. Once I used -F it was\nway faster and comparable to Oracle.\n\n> Perhaps these are bad techniques and a C based proc would do better?\n\nFeel free to try src/interfaces/ecpg/test/perftest.c\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Mon, 18 Jan 1999 08:48:36 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "On Sun, Jan 17, 1999 at 07:23:15PM -0800, Tom wrote:\n> This also wouldn't you give you any benefit from \"prepare\" that would\n> speed up repeated executions of the same statement. PostgreSQL does allow\n> statements to be prepared right? If it doesn't, chances are COPY has a\n\nSince when? I thought we're still looking for someone to implement it, i.e.\nat least the PREPARE statement.\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Mon, 18 Jan 1999 08:49:24 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Tom Lane wrote:\n> \n> The other thing that jumps out here is the unreasonably high position of\n> recv(), which is called 962187 times. The script being read by psql was\n> only 957186 characters. Evidently we're invoking a kernel recv() call\n> once per character read from the frontend. I suspect this is an\n> inefficiency introduced by Magnus Hagander's recent rewrite of backend\n> libpq (see, I told you there was a reason for using stdio ;-)).\n\nAt least part of the problem is also the fe-be protocol itself, maybe \nnot in the backend receiving side, but certainly while front-end is \nreceiving.\n\nThe main problem is that you must very often wait for a specific end \ncharacter (instead of sending first the length and then reading the \nrequired amount in one chunk), and it requires some trickery to do it\nefficiently without reading each character separately.\n\nI once did the fe-be protocol in python (an interpreted scripting \nlanguage, see http://www.python.org), it was for v6.2. I was quite \namazed by the baroqueness of the protocol - it uses a mix three \ntechniques - 1. wait for EOT char, 2. receive a pascal style string and\n3.\nget a record consisting of a mix of 1 and 2.\n\n> We're\n> gonna have to do something about that, though it's not as critical as\n> the memory-allocation issue. It also appears that send() is now being\n> invoked multiple times per backend reply, which is going to create\n> inefficiencies outside the backend (ie, multiple packets per reply).\n> On a test case with a lot of SELECTs that would show up more than it\n> does here.\n\nI am sure that we need to change the protocol sometime soon, if we \nwant to increase the performance. I have contemplated something like \nthe X-window protocol, that seems to be quite well designed.\n\nAnd of course we need some sort of CLI that can do prepared statements \nand that can use binary fields (not only the file interface).\n\nCurrently we have some of it, bu only in SPI.\n\nUnfortunately most inserts are not done using SPI :(\n\nIt may be the CORBA interface that somebody may be working on, or \nit may be something simpler.\n\n---------------------\nHannu Krosing\n",
"msg_date": "Mon, 18 Jan 1999 14:35:27 +0200",
"msg_from": "Hannu Krosing <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "\n\nI tried to do this.....\n\nWhat I've done is create a plpgpsql function that does the inserts.\nThere is a begin/end in the function of course...I also tried doing it\nwith a begin work before calling the function in psql, and it did seem\nthat this worked as it should - ie a rollback work would get rid of the \n40000 inserts.\n\nStill, it took about 3.5 minutes, and Oracle does it in about 30 seconds.\n\nfactor of 7.....\n\nOn the plus side for Postgres, trying to find some queries against the\ndata for other speed tests, I found that it seemed to respond quickly -\nnot much thinking to do with one table of integers I guess. Actually I \ncouldn't come up with a query that would take very long - need another\ntable I suppose.\n\nI found that UPDATE TEST SET A = B -1 took only 20 seconds for 40000\nrecords.\n\nThe real plus is that Oracle couldn't do that transaction because it was\nrunning out of rollback segment space - something I have fixed in the past\nbut I recall being a pain to fix - need to make bigger rollback datafiles\nor whatever.\n\n\nAny suggested speed tests would be appreciated. \n\n> \n> Try to use BEGIN/END to run all inserts in single transaction\n> and please let us know results.\n> \n> Vadim\n> \n\n",
"msg_date": "Mon, 18 Jan 1999 10:31:34 -0500 (EST)",
"msg_from": "John Holland <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Hannu Krosing <[email protected]> writes:\n> I am sure that we need to change the protocol sometime soon, if we \n> want to increase the performance.\n\nThe immediate problem is just that we've given back performance that we\nhad a week ago. stdio-based libpq didn't have this problem, because it\ndealt with the kernel a bufferload at a time not a character at a time.\n\nAlso, while the fe/be protocol is certainly a bit crufty, I'm not\nconvinced that a well-designed frontend is going to suffer any real\nperformance problem from parsing the protocol. The trick again is to\nread into a buffer and parse from the buffer. The last time I profiled\nfrontend libpq, it turned out to be spending all its time in memory\nallocation not parsing. (Which I fixed, btw.)\n\n> And of course we need some sort of CLI that can do prepared statements \n> and that can use binary fields (not only the file interface).\n\n> Currently we have some of it, bu only in SPI.\n\nSPI? I missed something ... what's that?\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 18 Jan 1999 11:47:32 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom Lane wrote:\n> \n> Also, while the fe/be protocol is certainly a bit crufty, I'm not\n> convinced that a well-designed frontend is going to suffer any real\n> performance problem from parsing the protocol. The trick again is to\n> read into a buffer and parse from the buffer.\n\nThat's also what I did, but still better protocol would have helped\n\n> > And of course we need some sort of CLI that can do prepared statements\n> > and that can use binary fields (not only the file interface).\n> \n> > Currently we have some of it, bu only in SPI.\n> \n> SPI? I missed something ... what's that?\n\nI think it stands for Server Programming Interface (?). Vadim (?) added \nit as the interface for C and other programming language functions \nto execute queries inside the backend. AFAIK, it follows the standard \nprepare-bind-execute-fetch model for its queries.\n\nIt is quite well documented in the standard PG docs.\n\n-------------------\nHannu Krosing\n",
"msg_date": "Tue, 19 Jan 1999 01:15:06 +0200",
"msg_from": "Hannu Krosing <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Vadim Mikheev <[email protected]> writes:\n> For the table with single int4 column & 2000 INSERTs I have:\n\n> BEGIN/END:\t\t3.5sec\t\t~600insert/sec\n> Without:\t 151sec\t\t~13insert/sec\n\n> (IDE disk!)\n\n> I guess that you don't use -F flag..?\n\n*Do* use -F, you meant of course. You're right, I do. With an\nfsync after every transaction, I'm sure my test would've been much\nslower.\n\nHowever, with a UPS and an OS that only crashes about once every\nother year, I feel pretty secure using -F ...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 18 Jan 1999 21:03:08 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Michael Meskes wrote:\n> \n> On Sun, Jan 17, 1999 at 07:23:15PM -0800, Tom wrote:\n> > This also wouldn't you give you any benefit from \"prepare\" that would\n> > speed up repeated executions of the same statement. PostgreSQL does allow\n> > statements to be prepared right? If it doesn't, chances are COPY has a\n> \n> Since when? I thought we're still looking for someone to implement it, i.e.\n> at least the PREPARE statement.\n\n1. PG can handle prepared queries.\n Parser & Planner can parse/plan (i.e. prepare) query with\n parameters once, so such prepared query can be executed \n many times by executor without parser/planner invocation.\n2. We have to implement some interface to use this ability.\n\nVadim\n",
"msg_date": "Tue, 19 Jan 1999 09:15:01 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "On Mon, 18 Jan 1999, Vadim Mikheev wrote:\n>1. PG can handle prepared queries.\n> Parser & Planner can parse/plan (i.e. prepare) query with\n> parameters once, so such prepared query can be executed \n> many times by executor without parser/planner invocation.\n>2. We have to implement some interface to use this ability.\n\nYes, I need prepared statements for CORBA, too.\n\nTaral\n",
"msg_date": "Mon, 18 Jan 1999 20:26:40 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": false,
"msg_subject": "Prepared statements (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "Tom Lane wrote:\n> \n> Vadim Mikheev <[email protected]> writes:\n> > For the table with single int4 column & 2000 INSERTs I have:\n> \n> > BEGIN/END: 3.5sec ~600insert/sec\n> > Without: 151sec ~13insert/sec\n> \n> > (IDE disk!)\n> \n> > I guess that you don't use -F flag..?\n> \n> *Do* use -F, you meant of course. You're right, I do. With an\n\nActually no -:)\n\n> fsync after every transaction, I'm sure my test would've been much\n> slower.\n\nJust use BEGIN/END to run all queries in single transaction -:)\n\n> However, with a UPS and an OS that only crashes about once every\n> other year, I feel pretty secure using -F ...\n\nHowever, it's easy to crash Postgres itself and lose\ncommitted transactions -:(\n\nVadim\n",
"msg_date": "Tue, 19 Jan 1999 09:31:57 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Vadim Mikheev <[email protected]> writes:\n> Tom Lane wrote:\n>> However, with a UPS and an OS that only crashes about once every\n>> other year, I feel pretty secure using -F ...\n\n> However, it's easy to crash Postgres itself and lose\n> committed transactions -:(\n\nSurely not? The docs say (and a quick look at the code confirms)\nthat -F suppresses calls to fsync(2). It does not suppress writes.\nThus, a commit will still write the data out to kernel disk buffers.\nAll that fsync does is force the kernel to execute immediate disk\nwrites for those buffers. If I don't fsync, and the backend crashes,\nthe modified file data is still in kernel disk buffers and the kernel\nis still responsible for seeing that those dirty buffers get written\nout eventually.\n\n(Of course, errors inside Postgres might cause it to write bogus\ndata, but -F isn't going to help or hurt for that.)\n\nNot using -F means that you don't trust your OS, your hardware,\nand/or your power supply. It has nothing to do with whether you\ntrust Postgres.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Tue, 19 Jan 1999 00:25:48 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom Lane wrote:\n> \n> Vadim Mikheev <[email protected]> writes:\n> > Tom Lane wrote:\n> >> However, with a UPS and an OS that only crashes about once every\n> >> other year, I feel pretty secure using -F ...\n> \n> > However, it's easy to crash Postgres itself and lose\n> > committed transactions -:(\n> \n> Surely not? The docs say (and a quick look at the code confirms)\n> that -F suppresses calls to fsync(2). It does not suppress writes.\n> Thus, a commit will still write the data out to kernel disk buffers.\n> All that fsync does is force the kernel to execute immediate disk\n> writes for those buffers. If I don't fsync, and the backend crashes,\n> the modified file data is still in kernel disk buffers and the kernel\n> is still responsible for seeing that those dirty buffers get written\n> out eventually.\n\nYou're right.\n\nVadim\n",
"msg_date": "Tue, 19 Jan 1999 12:37:19 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> Not using -F means that you don't trust your OS, your hardware,\n> and/or your power supply. It has nothing to do with whether you\n> trust Postgres.\n\nI couldn't have said it better.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 19 Jan 1999 01:03:16 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "On Tue, Jan 19, 1999 at 09:15:01AM +0700, Vadim Mikheev wrote:\n> 1. PG can handle prepared queries.\n> Parser & Planner can parse/plan (i.e. prepare) query with\n> parameters once, so such prepared query can be executed \n> many times by executor without parser/planner invocation.\n> 2. We have to implement some interface to use this ability.\n\nI see. Thanks.\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Tue, 19 Jan 1999 09:34:26 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "At 11:29 19/01/99 -0500, you wrote:\n>can you explain the -F flag? when is it passed to what? Can this be done\n>in plpgsql? how?\n\nOr passed by libpq in a C program that needs to insert (as fast as\npossible) a lot of new rows? \n>\n>\n>john\n>On\n>Mon, 18 Jan 1999, Tom Lane wrote:\n>\n>> Vadim Mikheev <[email protected]> writes:\n>> > For the table with single int4 column & 2000 INSERTs I have:\n>> \n>> > BEGIN/END:\t\t3.5sec\t\t~600insert/sec\n>> > Without:\t 151sec\t\t~13insert/sec\n>> \n>> > (IDE disk!)\n>> \n>> > I guess that you don't use -F flag..?\n>> \n>> *Do* use -F, you meant of course. You're right, I do. With an\n>> fsync after every transaction, I'm sure my test would've been much\n>> slower.\n>> \n>> However, with a UPS and an OS that only crashes about once every\n>> other year, I feel pretty secure using -F ...\n>> \n>> \t\t\tregards, tom lane\n>> \n>\n>\n>\n>\n------------------------------------------------------------------\nEng. Roberto Jo�o Lopes Garcia E-mail: [email protected]\nF. 55 11 848 9906 FAX 55 11 848 9955\n\nMHA Engenharia Ltda\nE-mail: [email protected] WWW: http://www.mha.com.br\n\nAv Maria Coelho Aguiar, 215 Bloco D 2 Andar\nCentro Empresarial de Sao Paulo\nSao Paulo - BRASIL - 05805 000\n-------------------------------------------------------------------\n\n",
"msg_date": "Tue, 19 Jan 1999 13:43:43 -0200",
"msg_from": "Roberto Joao Lopes Garcia <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "can you explain the -F flag? when is it passed to what? Can this be done\nin plpgsql? how?\n\n\njohn\nOn\nMon, 18 Jan 1999, Tom Lane wrote:\n\n> Vadim Mikheev <[email protected]> writes:\n> > For the table with single int4 column & 2000 INSERTs I have:\n> \n> > BEGIN/END:\t\t3.5sec\t\t~600insert/sec\n> > Without:\t 151sec\t\t~13insert/sec\n> \n> > (IDE disk!)\n> \n> > I guess that you don't use -F flag..?\n> \n> *Do* use -F, you meant of course. You're right, I do. With an\n> fsync after every transaction, I'm sure my test would've been much\n> slower.\n> \n> However, with a UPS and an OS that only crashes about once every\n> other year, I feel pretty secure using -F ...\n> \n> \t\t\tregards, tom lane\n> \n\n",
"msg_date": "Tue, 19 Jan 1999 11:29:35 -0500 (EST)",
"msg_from": "John Holland <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "On Mon, 18 Jan 1999, Taral wrote:\n\n> On Mon, 18 Jan 1999, Vadim Mikheev wrote:\n> >1. PG can handle prepared queries.\n> > Parser & Planner can parse/plan (i.e. prepare) query with\n> > parameters once, so such prepared query can be executed \n> > many times by executor without parser/planner invocation.\n> >2. We have to implement some interface to use this ability.\n> \n> Yes, I need prepared statements for CORBA, too.\n\nThis would improve JDBC's PreparedStatement as well.\n\n-- \n Peter T Mount [email protected]\n Main Homepage: http://www.retep.org.uk\nPostgreSQL JDBC Faq: http://www.retep.org.uk/postgres\n Java PDF Generator: http://www.retep.org.uk/pdf\n\n",
"msg_date": "Tue, 19 Jan 1999 21:05:49 +0000 (GMT)",
"msg_from": "Peter T Mount <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Prepared statements (was: Postgres Speed or lack\n\tthereof)"
},
{
"msg_contents": "John Holland <[email protected]> writes:\n> can you explain the -F flag? when is it passed to what?\n\n-F is a command-line flag passed to the backend at backend startup.\nSince backends are normally started by the postmaster, what you\nreally do in practice is to start the postmaster with \"-o -F\".\nFor example my postmaster start script looks like\n\nnohup postmaster -i -o \"-F\" >server.log 2>&1 </dev/null &\n\nWhat the -F switch actually does is to disable calls to fsync(2),\nthereby allowing modified file blocks to hang around in kernel\nmemory for a little while (up to 30 seconds in most Unixes)\nrather than being force-written to disk as soon as each transaction\ncommits. If the same block of the database file gets modified\nagain within that time window (very likely under a repeated-update\nload), you just saved a disk write. On the other hand, if your OS\ncrashes or your power goes out in those 30 sec, you just lost a\ndatabase update that you thought you had committed.\n\nI'm not sure I believe the argument that omitting -F buys very much\nsafety, even if you do not trust your power company. Murphy's law\nsays that a power flicker will happen in the middle of committing\na transaction, not during the 30-second-max window between when you\ncould've had the data flushed to disk if only you'd used fsync()\nand when the swapper process will fsync it on its own. And in that\ncase you have a corrupted database anyway. So my theory is you use\na reliable OS, and get yourself a UPS if your power company isn't\nreliable (lord knows mine ain't), and back up your database as often\nas you can. -F buys enough speed that it's worth the small extra risk.\n\nThere are competent experts with conflicting opinions, however ;-)\n\nSee doc/README.fsync for more about -F and the implications of\nusing it.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Wed, 20 Jan 1999 00:57:20 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "> John Holland <[email protected]> writes:\n> > can you explain the -F flag? when is it passed to what?\n> \n> -F is a command-line flag passed to the backend at backend startup.\n> Since backends are normally started by the postmaster, what you\n> really do in practice is to start the postmaster with \"-o -F\".\n> For example my postmaster start script looks like\n> \n> nohup postmaster -i -o \"-F\" >server.log 2>&1 </dev/null &\n> \n> What the -F switch actually does is to disable calls to fsync(2),\n> thereby allowing modified file blocks to hang around in kernel\n> memory for a little while (up to 30 seconds in most Unixes)\n> rather than being force-written to disk as soon as each transaction\n> commits. If the same block of the database file gets modified\n> again within that time window (very likely under a repeated-update\n> load), you just saved a disk write. On the other hand, if your OS\n> crashes or your power goes out in those 30 sec, you just lost a\n> database update that you thought you had committed.\n> \n> I'm not sure I believe the argument that omitting -F buys very much\n> safety, even if you do not trust your power company. Murphy's law\n> says that a power flicker will happen in the middle of committing\n> a transaction, not during the 30-second-max window between when you\n> could've had the data flushed to disk if only you'd used fsync()\n> and when the swapper process will fsync it on its own. And in that\n> case you have a corrupted database anyway. So my theory is you use\n> a reliable OS, and get yourself a UPS if your power company isn't\n> reliable (lord knows mine ain't), and back up your database as often\n> as you can. -F buys enough speed that it's worth the small extra risk.\n> \n> There are competent experts with conflicting opinions, however ;-)\n> \n> See doc/README.fsync for more about -F and the implications of\n> using it.\n\nWell said, and I think we have to address this shortcoming. Vadim has\nbeen reluctant to turn off fsync by default. I have been trying to come\nup with some soft-fsync on my own but haven't hit on anything he agrees\nwith yet.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 20 Jan 1999 01:22:55 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": ">For example my postmaster start script looks like>\n>\n>nohup postmaster -i -o \"-F\" >server.log 2>&1 </dev/null &\n>\n\nI trayed the -F option and the results is bellow. I use a C program (libPQ)\nthat read from 3 dbf files and past it to 3 tables in postgreSQL 4.0.0.\nEche row in dbf file generate a SQL insert command. I�m not using begin\ntransaction and commit to make all insert in just one transaction.\n\nTest without -F option:\n\ntable 1: 697 rows = 4' 39\"\ntable 2: 22760 rows = 1h 9' 37\"\ntable 3: 22758 rows = 43' 44\"\nindex/vacuun/grant all tables 12' 58\" \nTotal time: 2h 10' 58\"\n\nTest with -F option (dbf file was updated and grow up a litle):\n\ntable 1: 698 rows = 1' 52\"\ntable 2: 22798 rows = 15' 38\"\ntable 3: 22796 rows = 7' 42\"\nindex/vacuun/grant all tables 11' 39\" \nTotal time: 36' 51\" ~= 72% reducion in time \n\nit will allow me to update postgreSQL tables from the dbf 2 times a day\ninsted of one at night!\n\nThank you who point me the -F option. I will tray with begin transaction\nand commit to make all inserts in just one transaction\n\nRoberto\n\n\n\n\n------------------------------------------------------------------\nEng. Roberto Jo�o Lopes Garcia E-mail: [email protected]\nF. 55 11 848 9906 FAX 55 11 848 9955\n\nMHA Engenharia Ltda\nE-mail: [email protected] WWW: http://www.mha.com.br\n\nAv Maria Coelho Aguiar, 215 Bloco D 2 Andar\nCentro Empresarial de Sao Paulo\nSao Paulo - BRASIL - 05805 000\n-------------------------------------------------------------------\n\n",
"msg_date": "Thu, 21 Jan 1999 09:14:38 -0200",
"msg_from": "Roberto Joao Lopes Garcia <[email protected]>",
"msg_from_op": false,
"msg_subject": "Performance test with -F (Was Postgres Speed or lack thereof)"
},
{
"msg_contents": "On Wed, 20 Jan 1999, Tom Lane wrote:\n\n> What the -F switch actually does is to disable calls to fsync(2),\n> thereby allowing modified file blocks to hang around in kernel\n> memory for a little while (up to 30 seconds in most Unixes)\n> rather than being force-written to disk as soon as each transaction\n> commits. If the same block of the database file gets modified\n> again within that time window (very likely under a repeated-update\n> load), you just saved a disk write. \n\nYour effeciencies from running asynchronously run far beyond that.\nBy delaying writes, you allow the kernel to order them and pass them in\nlarge chunks to the drive, getting very good effeciency. Compare ten\naggregated writes (the asynchronous case) to ten individual writes\n(the synchronous case): in the former case, the disk arm needs to move\nover the platter once, and you pay the cost of initiating a write once,\nwhereas in the later case the disk arm needs to do ten seeks, and you\npay the cost of initiating ten different disk operations, servicing ten\ninterrupts instead of one, etc. You save a lot more than just optimizing\nout cancelling writes.\n\n> I'm not sure I believe the argument that omitting -F buys very much\n> safety, even if you do not trust your power company. Murphy's law\n> says that a power flicker will happen in the middle of committing\n> a transaction, not during the 30-second-max window between when you\n> could've had the data flushed to disk if only you'd used fsync()\n> and when the swapper process will fsync it on its own. And in that\n> case you have a corrupted database anyway.\n\nThat depends on how you process your transactions, I guess. It is\npossible to perform transactions safely, such that no interruption can\nviolate your db's integrity. I have not read and understood enough\nof psql's transaction engine to know how safe it is in the face of\ninterruption, but there's no inherent reason that it need be unsafe.\n\nThe real way to solve this problem is to build a file system which is\nmore suitable for use in this way than a traditional unix fs. You're\nworking around the coarse-grained-ness of fsync() no matter what you do,\nand there is no good solution until you move away from fsync() and towards\nsomething like fsync(block) or fsync(transaction).\n\n--\nTodd Graham Lewis 32�49'N,83�36'W (800) 719-4664, x2804\n******Linux****** MindSpring Enterprises [email protected]\n\n\"Those who write the code make the rules.\" -- Jamie Zawinski\n\n",
"msg_date": "Fri, 22 Jan 1999 13:24:16 -0500 (EST)",
"msg_from": "Todd Graham Lewis <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Todd Graham Lewis wrote:\n> \n> On Wed, 20 Jan 1999, Tom Lane wrote:\n> \n> > What the -F switch actually does is to disable calls to fsync(2),\n> > thereby allowing modified file blocks to hang around in kernel\n\nCould you use an mmapped file and flush it according to some algorythm? \nHard on memory but fast.\n\n-- \nJohn Ryan\nCEO Ryan Net Works Visit us at www.cybertrace.com\[email protected] (703) 998-1751 (w)\n3515-A S. Stafford St. (703) 820-6304 (f)\nArlington, VA 22206-1191 (703) 626-4130 (c)\n\n-----BEGIN PGP PUBLIC KEY BLOCK----- Version: 2.6.2\nmQBtAzO3iTYAAAEDAOvDARDRxluZdieT11m6ukP/niHI3qnSL9DKmRrLil5M+Cdp\n3w4fV2/HvoBhsUvJE0uLwtbFCCBUP5YkGWNM0oU+XxQHP2+yUY8CDEHLncZ6KHV5\nKLlfQKDHSQgZ+58PcQAFE7QjSm9obiBSeWFuIDxqb2huQHd3dy5jeWJlcnRyYWNlLmNvbT4==Wl7V\n-----END PGP PUBLIC KEY BLOCK-----\n",
"msg_date": "Fri, 22 Jan 1999 16:11:34 -0800",
"msg_from": "John Ryan <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Tom Lane wrote:\n> \n> Having an idle hour this evening, I thought it'd be interesting to build\n> a backend with profiling enabled, so as to confirm or deny the above\n> guess. It seems that indeed a lot of time is being wasted, but where\n> it's being wasted might surprise you!\n...\n> In other words, we're spending a third of our time mallocing and freeing\n> memory. A tad high, what?\n> \n> Actually, it's worse than that, because AllocSetAlloc,\n> PortalHeapMemoryAlloc, AllocSetFree, and all of the OrderedElemXXX\n> routines represent our own bookkeeping layer atop malloc/free.\n> That's another 18.66 seconds spent in these top routines, which means\n> that we are real close to expending half the backend's runtime on\n> memory bookkeeping. This needs work.\n>\n\nYes, it's suprizing!\n\nI added some debug code to palloc/pfree and it shows that for\nINSERT:\n\n1. 80% of allocations are made for <= 32 bytes.\n2. pfree is used for 25% of them only (others are freed\n after statement/transaction is done).\n\nNote that our mmgr adds 16 bytes to each allocation\n(+ some bytes in malloc) - a great overhead, yes?\n\nI added code to allocate a few big (16K-64K) blocks\nof memory for these small allocations to speed up \npalloc by skiping AllocSetAlloc/malloc. New code\ndon't free allocated memory (to make bookkeeping fast)\nbut keeping in mind 2. above and memory overhead\nit seems as appropriate thing to do. These code also\nspeed up freeing when statement/transaction is done,\nbecause of only a few blocks have to be freed now.\n\nI did 5000 INSERTS (into tables with 3 ints and 33 ints)\nwith BEGIN/END, -F and -B 512 (I run postgres directly,\nwithout postmaster). User times:\n\n\t\t\told\t\tnew\n-----------------------------------------\ntable with 3 ints\t 9.7 sec\t 7.6 sec\ntable with 33 ints\t59.5 sec\t39.9 sec\n\nSo, new code 20%-30% faster. Process sizes are the same.\n\nTom, could you run new code under profiling?\n\nThere are still some things to do:\n\n1. SELECT/UPDATE/DELETE often palloc/pfree tuples\n (sizes are > 32 bytes), but pfree now requires \n additional lookup to see is memory allocated by \n AllocSetAlloc or new code. We can avoid this.\n\n2. Index scans palloc/pfree IndexResult for each\n tuple returned by index. This one was annoying me\n for long time. IndexResult should be part of\n IndexScan structure...\n\n3. psort uses leftist structure (16 bytes) when\n disk is used for sorting. Obviously, big block\n allocation should be used by lselect code itself.\n\n4. Actually, new mode shouldn't be used by Executor\n in some cases.\n\nI'll address this in a few days...\n\nBTW, look at memutils.h: new code is like \"tunable\"\nmode described there.\n\n> \n> In other words, essentially *all* of the CPU time spent in\n> CommitTransaction is spent freeing memory. That's probably why\n> ganging the transactions doesn't help --- it's the same number of\n> memory blocks getting allocated and freed.\n\nIt shows that we should get rid of system malloc/free and do\nall things in mmgr itself - this would allow us much faster\nfree memory contexts at statement/transaction end.\n\nVadim\n",
"msg_date": "Sun, 24 Jan 1999 12:29:51 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Vadim Mikheev <[email protected]> writes:\n> Tom Lane wrote:\n>> In other words, we're spending a third of our time mallocing and freeing\n>> memory. A tad high, what?\n\n> I added some debug code to palloc/pfree and it shows that for\n> INSERT:\n> 1. 80% of allocations are made for <= 32 bytes.\n> 2. pfree is used for 25% of them only (others are freed\n> after statement/transaction is done).\n\nI had suspected the former (because most of the allocs are\nparser nodes) and knew the latter from profile call counts.\n\n> Note that our mmgr adds 16 bytes to each allocation\n> (+ some bytes in malloc) - a great overhead, yes?\n\nYouch ... for a list-node-sized request, that's a lot of overhead.\n\nI did some more profiling this weekend, after Magnus Hagander and I\nrepaired libpq's little problem with calling recv() once per input byte.\nUnsurprisingly, memory allocation is still 50% of backend CPU time in\nmy test consisting of a long series of INSERT statements. It's less\nthan that, but still an interesting number, in two other test scenarios\nI set up:\n\n(1) SELECT the whole contents of a 6500-row, 38-column table; repeat 5 times.\n Memory allocation takes 8.74 of 54.45 backend CPU sec, or 16%.\n\n(2) Load a database from a pg_dump script. Script size is 1.3MB.\n This is of course mostly CREATE TABLE, COPY from stdin, CREATE INDEX.\n Memory allocation takes 11.08 of 38.25 sec = 29%.\n\nI believe that the reason memory alloc is such a large percentage of the\nINSERT scenario is that this is the only one where lots of SQL parsing\nis going on. In the SELECT scenario, most of the palloc/pfree traffic\nis temporary field value representations inside printtup (more about\nthat below). In the database load scenario, it seems that btree index\nbuilding is the biggest hit; the heaviest callers of palloc are:\n\n 0.00 0.01 10642/438019 _bt_searchr [86]\n 0.00 0.01 13531/438019 heap_formtuple [149]\n 0.00 0.01 13656/438019 CopyFrom [24]\n 0.00 0.01 14196/438019 textin [247]\n 0.01 0.01 18275/438019 datetime_in [42]\n 0.01 0.03 32595/438019 bpcharin [182]\n 0.01 0.03 38882/438019 index_formtuple [79]\n 0.01 0.03 39028/438019 _bt_formitem [212]\n 0.07 0.16 204564/438019 _bt_setsortkey [78]\n[88] 1.3 0.14 0.34 438019 palloc [88]\n\n> I added code to allocate a few big (16K-64K) blocks of memory for\n> these small allocations to speed up palloc by skiping\n> AllocSetAlloc/malloc. New code don't free allocated memory (to make\n> bookkeeping fast) but keeping in mind 2. above and memory overhead it\n> seems as appropriate thing to do. These code also speed up freeing\n> when statement/transaction is done, because of only a few blocks have\n> to be freed now.\n\nThis is pretty much what I had in mind, but I think that it is not OK\nto leave everything to be freed at statement end. In particular I've\njust been looking at printtup(), the guts of SELECT output, and I see\nthat each field value is palloc'd (by a type-specific output routine)\nand then pfree'd by printtup after emitting it. If you make the pfrees\nall no-ops then this means the temporary space needed for SELECT is\nat least equal to the total volume of output data ... not good for\nhuge tables. (COPY in and out also seem to palloc/pfree each field\nvalue that goes past.)\n\nHowever, I think we could get around this by making more use of memory\ncontexts. The space allocated inside printtup could be palloc'd from a\n\"temporary\" context that gets flushed every so often (say, after every\nfew hundred tuples of output). All the pallocs done inside the parser\ncould go to a \"statement\" context that is flushed at the end of each\nstatement. It'd probably also be a good idea to have two kinds of\nmemory contexts, the current kind where each block is individually\nfree-able and the new kind where we only keep track of how to free the\nwhole context's contents at once. That way, any particular routine that\nwas going to allocate a *large* chunk of memory would have a way to free\nthat chunk explicitly when done with it (but would have to remember\nwhich context it got it from in order to do the free). For the typical\nlist-node-sized request, we wouldn't bother with the extra bookkeeping.\n\nI think that palloc() could be defined as always giving you the bulk-\nallocated kind of chunk. pfree() would become a no-op and eventually\ncould be removed entirely. Anyone who wants the other kind of chunk\nwould call a different pair of routines, where the controlling memory\ncontext has to be named at both alloc and free time. (This approach\ngets rid of your concern about the cost of finding which kind of context\na block has been allocated in inside of pfree; we expect the caller\nto know that.)\n\nGetting this right is probably going to take some thought and work,\nbut it looks worthwhile from a performance standpoint. Maybe for 6.6?\n\n> It shows that we should get rid of system malloc/free and do\n> all things in mmgr itself - this would allow us much faster\n> free memory contexts at statement/transaction end.\n\nI don't think we can or should stop using malloc(), but we can\nask it for large blocks and do our own allocations inside those\nblocks --- was that what you meant?\n\n\t\t\tregards, tom lane\n",
"msg_date": "Sun, 24 Jan 1999 14:54:52 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom Lane wrote:\n> I don't think we can or should stop using malloc(), but we can\n> ask it for large blocks and do our own allocations inside those\n> blocks --- was that what you meant?\n\nI will just jump in with a small idea.\nThe Gnome crowd is investigating using alloca instead of malloc/free\nwhere applicable. It is a huge save in CPU cycles where it can be used.\n\n\tregards,\n-- \n-----------------\nG�ran Thyni\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Sun, 24 Jan 1999 23:09:57 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "\nBUGS\n The alloca() function is machine dependent; its use is discouraged.\n\n\nOn Sun, 24 Jan 1999, Goran Thyni wrote:\n\n> Tom Lane wrote:\n> > I don't think we can or should stop using malloc(), but we can\n> > ask it for large blocks and do our own allocations inside those\n> > blocks --- was that what you meant?\n> \n> I will just jump in with a small idea.\n> The Gnome crowd is investigating using alloca instead of malloc/free\n> where applicable. It is a huge save in CPU cycles where it can be used.\n> \n> \tregards,\n> -- \n> -----------------\n> G�ran Thyni\n> This is Penguin Country. On a quiet night you can hear Windows NT\n> reboot!\n> \n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Sun, 24 Jan 1999 18:26:45 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "The Hermit Hacker wrote:\n> BUGS\n> The alloca() function is machine dependent; its use is discouraged.\n\nGain a big potential speed boost, loose some portability.\nI think most modern unices has a good alloca, DOS and Mac\ndon't, but who's porting the server there?\n\nAny unices out there missing alloca?\n\n\tmvh,\n-- \n-----------------\nG�ran Thyni\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Sun, 24 Jan 1999 23:59:29 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "Tom Lane wrote:\n> \n> > Note that our mmgr adds 16 bytes to each allocation\n> > (+ some bytes in malloc) - a great overhead, yes?\n> \n> Youch ... for a list-node-sized request, that's a lot of overhead.\n\nAnd lists are very often used by planner and - never pfreed.\n\n> Getting this right is probably going to take some thought and work,\n> but it looks worthwhile from a performance standpoint. Maybe for 6.6?\n\nYes - I have to return to MVCC stuff...\nSo, I consider my exercizes with mmgr as vacation from MVCC -:)\n\n> > It shows that we should get rid of system malloc/free and do\n> > all things in mmgr itself - this would allow us much faster\n> > free memory contexts at statement/transaction end.\n> \n> I don't think we can or should stop using malloc(), but we can\n> ask it for large blocks and do our own allocations inside those\n> blocks --- was that what you meant?\n\nNo. We could ask brk() for large blocks.\nThe problem is where to handle dynamic allocations.\nAs long as they are handled by malloc we can't put\nthem in proper blocks of current memory context.\nBut having our own handling malloc would become useless.\n\nVadim\n",
"msg_date": "Mon, 25 Jan 1999 10:01:39 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "\nOn 25-Jan-99 Vadim Mikheev wrote:\n> \n> Yes - I have to return to MVCC stuff...\n> So, I consider my exercizes with mmgr as vacation from MVCC -:)\n\nDumb question. What is MVCC? It almost sounds like a M$ compiler.\n\nVince.\n-- \n==========================================================================\nVince Vielhaber -- KA8CSH email: [email protected] flame-mail: /dev/null\n # include <std/disclaimers.h> TEAM-OS2\n Online Searchable Campground Listings http://www.camping-usa.com\n \"There is no outfit less entitled to lecture me about bloat\n than the federal government\" -- Tony Snow\n==========================================================================\n\n\n",
"msg_date": "Sun, 24 Jan 1999 23:02:32 -0500 (EST)",
"msg_from": "Vince Vielhaber <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "On Sun, 24 Jan 1999, Goran Thyni wrote:\n\n> The Hermit Hacker wrote:\n> > BUGS\n> > The alloca() function is machine dependent; its use is discouraged.\n> \n> Gain a big potential speed boost, loose some portability.\n> I think most modern unices has a good alloca, DOS and Mac\n> don't, but who's porting the server there?\n> \n> Any unices out there missing alloca?\n\nMake you a deal...\n\nYou build the code such that \"#ifdef HAVE_ALLOCA\" is true, so that those\nplatforms that either don't support it, or have broken suport for it,\naren't affect, and you are most welcome to work on it...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 25 Jan 1999 01:38:54 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "> On Sun, 24 Jan 1999, Goran Thyni wrote:\n> \n> > The Hermit Hacker wrote:\n> > > BUGS\n> > > The alloca() function is machine dependent; its use is discouraged.\n> > \n> > Gain a big potential speed boost, loose some portability.\n> > I think most modern unices has a good alloca, DOS and Mac\n> > don't, but who's porting the server there?\n> > \n> > Any unices out there missing alloca?\n> \n> Make you a deal...\n> \n> You build the code such that \"#ifdef HAVE_ALLOCA\" is true, so that those\n> platforms that either don't support it, or have broken suport for it,\n> aren't affect, and you are most welcome to work on it...\n\nAs far as I know, alloca is only useful for memory that is used by the\ncurrent function or its children. I don't think we have many cases\nwhere we could use that.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 25 Jan 1999 01:10:49 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": ">>>> The Hermit Hacker wrote:\n>> You build the code such that \"#ifdef HAVE_ALLOCA\" is true, so that those\n>> platforms that either don't support it, or have broken suport for it,\n>> aren't affect, and you are most welcome to work on it...\n\nalloca is treated as a builtin function by gcc, so it should work on\nany platform if you use gcc. If your compiler is not gcc then alloca\nis *very* iffy --- as far as I know, it is not specified by any\nrecognized standard. So I concur with Marc's opinion: a speedup on\nalloca-supporting platforms is nice, but it is not acceptable to break\nthe code on machines without it.\n\nBruce Momjian <[email protected]> writes:\n> As far as I know, alloca is only useful for memory that is used by the\n> current function or its children. I don't think we have many cases\n> where we could use that.\n\nRight, storage allocated by alloca is part of the calling function's\nstack frame --- it goes away when that function exits, no sooner and\nno later.\n\nI have noticed some places in Postgres where routines palloc storage\nthat they pfree before exiting. alloca would be a fine substitute\nfor that usage. But as far as I've seen in profiling, the primary\ncallers of palloc are allocating storage that they will *not* free,\nbut indeed will return to their caller --- the type-specific conversion\nroutines all work that way, for example. So I'm doubtful that adding\nalloca to our tool arsenal would be worth the portability headaches\nit would cost.\n\nI have been toying with the notion of nested memory contexts, which\nwould be created at the start of certain major routines and deleted\nwhen they exit. palloc would allocate storage from the most closely\nnested context, and thus would mean \"allocate storage that will live\nthrough the current major operation, whatever it is\". This is still\na very half-baked notion and I can't say that it will work, but I\nthought I'd throw it into the thread...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 25 Jan 1999 01:46:22 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: alloca (was: Postgres Speed or lack thereof) "
},
{
"msg_contents": "Vadim Mikheev <[email protected]> writes:\n> Tom Lane wrote:\n>> I don't think we can or should stop using malloc(), but we can\n>> ask it for large blocks and do our own allocations inside those\n>> blocks --- was that what you meant?\n\n> No. We could ask brk() for large blocks.\n\nI think that would be a bad idea. brk() is a Unix-ism; I doubt it's\nsupported on Win NT, for example. malloc() is a lot more portable.\n\nAnother potential portability issue is whether malloc() will coexist\nwith calling brk() ourselves. (It *ought* to, but I can believe that\nthe feature might be broken on some platforms, since it's so seldom\nexercised...) We can't stop all uses of malloc(), because parts of the\nC library use it --- stdio, qsort, putenv all do on my machine.\n\nIf we're going to grab large chunks and keep them, then any small\ninefficiency in doing the grabbing isn't really worth worrying about;\nso I don't see the need to bypass malloc() for that.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 25 Jan 1999 10:24:25 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom Lane wrote:\n> \n> Vadim Mikheev <[email protected]> writes:\n> > Tom Lane wrote:\n> >> I don't think we can or should stop using malloc(), but we can\n> >> ask it for large blocks and do our own allocations inside those\n> >> blocks --- was that what you meant?\n> \n> > No. We could ask brk() for large blocks.\n> \n> I think that would be a bad idea. brk() is a Unix-ism; I doubt it's\n> supported on Win NT, for example. malloc() is a lot more portable.\n> \n> Another potential portability issue is whether malloc() will coexist\n> with calling brk() ourselves. (It *ought* to, but I can believe that\n> the feature might be broken on some platforms, since it's so seldom\n> exercised...) We can't stop all uses of malloc(), because parts of the\n> C library use it --- stdio, qsort, putenv all do on my machine.\n> \n> If we're going to grab large chunks and keep them, then any small\n> inefficiency in doing the grabbing isn't really worth worrying about;\n> so I don't see the need to bypass malloc() for that.\n\nOk, I agreed.\n\nVadim\n",
"msg_date": "Mon, 25 Jan 1999 22:27:12 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> Right, storage allocated by alloca is part of the calling function's\n> stack frame --- it goes away when that function exits, no sooner and\n> no later.\n\nAnd as I'm sure many already know, there are limits on stack frame size\non many architectures, and limits on total stack size on every platform.\nThe available size is not in the same league as malloc()-style heap\nallocation.\n\n - Tom\n",
"msg_date": "Mon, 25 Jan 1999 16:23:25 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "The Hermit Hacker wrote:\n> Make you a deal...\n> \n> You build the code such that \"#ifdef HAVE_ALLOCA\" is true, so that those\n> platforms that either don't support it, or have broken suport for it,\n> aren't affect, and you are most welcome to work on it...\n\nYeah right, I should keep my mouth shut. :-)\n\nIt is a big job, starting getting with getting profiling working on \nmy box, analyzing lots of code and then come up with some optimations.\n\nWe can start with the patch below to detect alloca() in autoconf.\n\n\tbest regards,\n-- \n-----------------\nG�ran Thyni\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!",
"msg_date": "Tue, 26 Jan 1999 17:39:36 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "Bruce Momjian wrote:\n> As far as I know, alloca is only useful for memory that is used by the\n> current function or its children. I don't think we have many cases\n> where we could use that.\n\nPerhaps not in so many obvious places, but\nwith some restructuring perhaps.\nI plan to look into this too, time permitting, don't hold your breath.\n\n\tregards,\n-- \n-----------------\nG�ran Thyni\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n\n\n",
"msg_date": "Tue, 26 Jan 1999 17:43:07 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "\n\n\nBefore this patch is applied, havae we determined wheteher or not its even\na useful thing to pursue? I thought the general conscious was that, for\nus, it wasn't...\n\n On Tue, 26 Jan 1999, Goran Thyni wrote:\n\n> The Hermit Hacker wrote:\n> > Make you a deal...\n> > \n> > You build the code such that \"#ifdef HAVE_ALLOCA\" is true, so that those\n> > platforms that either don't support it, or have broken suport for it,\n> > aren't affect, and you are most welcome to work on it...\n> \n> Yeah right, I should keep my mouth shut. :-)\n> \n> It is a big job, starting getting with getting profiling working on \n> my box, analyzing lots of code and then come up with some optimations.\n> \n> We can start with the patch below to detect alloca() in autoconf.\n> \n> \tbest regards,\n> -- \n> -----------------\n> G�ran Thyni\n> This is Penguin Country. On a quiet night you can hear Windows NT\n> reboot!\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Tue, 26 Jan 1999 14:07:52 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "The Hermit Hacker <[email protected]> writes:\n> Before this patch is applied, havae we determined wheteher or not its even\n> a useful thing to pursue? I thought the general conscious was that, for\n> us, it wasn't...\n\nIf we can demonstrate through profiling that there's someplace where\nusing alloca offers a worthwhile speedup, then let's go for it.\n\nBut, until we find such a hotspot, I'd be inclined not to spend\nconfigure cycles on testing for alloca...\n\n\t\t\tregards, tom lane\n",
"msg_date": "Tue, 26 Jan 1999 14:00:54 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: alloca (was: Postgres Speed or lack thereof) "
},
{
"msg_contents": "The Hermit Hacker wrote:\n> Before this patch is applied, havae we determined wheteher or not its even\n> a useful thing to pursue? I thought the general conscious was that, for\n> us, it wasn't...\n\nI would have to investigate further before making any final judgement.\nBut having the test in configure/config.h can't hurt should me or\nsomeone\nelse find a worthwhile speedup using alloca, can it?\n\nIf you want something more substantial you have to wait,\nI just posted the patch to make it available to other\neager investigators.\n\n\thappy hacking,\n-- \n-----------------\nG�ran Thyni\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Wed, 27 Jan 1999 00:35:34 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "On Wed, 27 Jan 1999, Goran Thyni wrote:\n\n> The Hermit Hacker wrote:\n> > Before this patch is applied, havae we determined wheteher or not its even\n> > a useful thing to pursue? I thought the general conscious was that, for\n> > us, it wasn't...\n> \n> I would have to investigate further before making any final judgement.\n> But having the test in configure/config.h can't hurt should me or\n> someone\n> else find a worthwhile speedup using alloca, can it?\n> \n> If you want something more substantial you have to wait,\n> I just posted the patch to make it available to other\n> eager investigators.\n\nLet's go with the wait scenario...one thing I've always hated with other\npackages and configure is watching them run forever checking for things\nthat are never used, but checking \"just cause it can\" *roll eyes*\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Tue, 26 Jan 1999 19:46:35 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: alloca (was: Postgres Speed or lack thereof)"
},
{
"msg_contents": "> \n> Tom Lane wrote:\n> > \n> > > Note that our mmgr adds 16 bytes to each allocation\n> > > (+ some bytes in malloc) - a great overhead, yes?\n> > \n> > Youch ... for a list-node-sized request, that's a lot of overhead.\n> \n> And lists are very often used by planner and - never pfreed.\n> \n> > Getting this right is probably going to take some thought and work,\n> > but it looks worthwhile from a performance standpoint. Maybe for 6.6?\n> \n> Yes - I have to return to MVCC stuff...\n> So, I consider my exercizes with mmgr as vacation from MVCC -:)\n> \n> > > It shows that we should get rid of system malloc/free and do\n> > > all things in mmgr itself - this would allow us much faster\n> > > free memory contexts at statement/transaction end.\n> > \n> > I don't think we can or should stop using malloc(), but we can\n> > ask it for large blocks and do our own allocations inside those\n> > blocks --- was that what you meant?\n> \n> No. We could ask brk() for large blocks.\n> The problem is where to handle dynamic allocations.\n> As long as they are handled by malloc we can't put\n> them in proper blocks of current memory context.\n> But having our own handling malloc would become useless.\n> \n> Vadim\n\nWe could use 4 methods for dynamic allocation:\n\n1)\tmalloc/free - for persistent storage allocation\n\n2)\tpalloc/pfree - for storage belonging to some context and\n\twhich we can keep track of and free explicitly\n\n3)\tfast_palloc - for storage which impossible, too difficult or too\n\texpensive to keep track of. This storage should be allocated with\n\tfast and simple inline code from bigger chunks allocated with palloc.\n\tThis storage would never freed explicitly, so that code could be\n\tsimple and fast, but the big chunks would be freed automatically at\n\tthe end of the transaction.\n\n4)\tfast_talloc - we could introduce a `tuple' context handled like\n\tfast_palloc for storage used only while processing one tuple.\n\tThis storage could be fast allocated from few big chunks allocated\n\twith palloc and freed explicitly after the tuple has been processed.\n\tThis could avoid the overhead of many malloc/palloc while reducing\n\tthe overall memory usage for transaction which process many rows.\n\tThe total cost per tuple could be one palloc and one pfree.\n\tWe could also simply reuse the chunks for every tuple and pfree them \n\tonly at the end of the transaction. This would cost one palloc/pfree\n\tper transaction.\n\nThis would require revising the code and changing palloc/pfree with the new\nfunctions where appropriate, but this could be done gradually because the\nold palloc/pfree are always safe.\n\n-- \nMassimo Dal Zotto\n\n+----------------------------------------------------------------------+\n| Massimo Dal Zotto email: [email protected] |\n| Via Marconi, 141 phone: ++39-0461534251 |\n| 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n| Italy pgp: finger [email protected] |\n+----------------------------------------------------------------------+\n",
"msg_date": "Thu, 28 Jan 1999 10:38:10 +0100 (MET)",
"msg_from": "Massimo Dal Zotto <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Massimo Dal Zotto <[email protected]> writes:\n> We could use 4 methods for dynamic allocation:\n\nThis sounds pretty close to what I was thinking, too. We already have\nthe notion of \"memory contexts\" which can have different alloc/free\nimplementations, so there's already a structure to put these different\nalloc methods into.\n\nHowever, we probably don't really want to touch each individual palloc()\ncall in the system to change it to a MemoryContextAlloc(some-context, size)\ncall. Not only would that be horribly tedious, but an awful lot of\nthese calls are made for the purpose of creating an object that the\nrequesting routine intends to pass back to its caller. In that\nsituation, the requesting routine doesn't really *know* what the\nlifetime of the storage needs to be --- it's the caller that determines\nwhat's going to happen.\n\nI think what we want is to define palloc() as allocating\nnot-individually-freeable storage from the current context, which is set\nby the most closely nested main control routine (parser, planner,\nexecutor, etc). That should handle the majority of the existing code\nand thus minimize the number of places that have to be touched.\npfree() goes away, eventually, and can be made a no-op meanwhile.\n\nThis assumes that the number of places that really *want* to free\nstorage earlier than end of context is much smaller than the number\nof places that don't want to be bothered --- therefore, it makes more\nsense to redefine palloc() as the not-separately-freeable kind and\nchange the places that do want to control their storage lifetime,\ninstead of changing all the other places.\n\n> 4)\tfast_talloc - we could introduce a `tuple' context handled like\n> \tfast_palloc for storage used only while processing one tuple.\n\nRight, that could be the most closely nested context for operations that\niterate over tuples. A further idea I had is to define allocations\nfrom that context as being only good for one tuple, but to not bother to\nphysically clear the context after every single tuple --- do it every\nten or hundred or so tuples, instead. The amount of storage wasted\nwon't be very large, and it cuts the per-tuple time cost even more.\n(Of course, if resetting a context to empty only takes a few\ninstructions, you might as well just do it every time. I'm not sure\noffhand if that will be true or not.)\n\n\t\t\tregards, tom lane\n",
"msg_date": "Thu, 28 Jan 1999 14:39:15 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "I have found a bug and tried to trace the source but now I need some\npointers.\n\nENV: Pgsql 6.4.2 (and current) compiled with locale (using Swedish\nlocale)\n\nBUG:\nA small table: CREATE TABLE x(txt text);\n\ngoran=> SELECT * FROM x;\ntxt \n------\nabc \n��� \nGr�n\nGunnar\nGUNNAR\nG�ran \ng�ran \nG�RAN \n(8 rows)\n\nbut:\n\ngoran=> select * from x WHERE txt LIKE 'G%';\ntxt \n------\nGr�n\nGunnar\nGUNNAR\n(3 rows)\n\nthe same goes for: \"select * from x WHERE txt ~ '^G'\" which is the same\nas above LIKE-stmt in the backend.\nOtherwise regex works correct.\n\nTo sum up:\nCase sensitive regex anchored with '^' and where the first char\nfollowing\nthe match is a non-ascii char gives misses.\n\nI have tracked it down to the following, 3 functions are called to test\nthe expression:\n1. textregexeq OK for 5 instances\n2. text_ge('G','G') also OK\n3. text_lt('G','G\\0xFF') this is not correct!\n\nCase 3 work not work with strcoll(), in varstr_cmp().\nIf I change it to strcoll() to strncmp() it works as expected,\nbut it probably breaks sorting etc big time.\n\nAny suggestions how to proceed?\n\n\tTIA,\n-- \n---------------------------------------------\nG�ran Thyni, JMS Bildbasen, Kiruna\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Thu, 28 Jan 1999 22:44:33 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "tough locale bug"
},
{
"msg_contents": "I have found a bug and tried to trace the source but now I need some\npointers.\n\nENV: Pgsql 6.4.2 (and current) compiled with locale (using Swedish\nlocale)\n\nBUG:\nA small table: CREATE TABLE x(txt text);\n\ngoran=> SELECT * FROM x;\ntxt \n------\nabc \n��� \nGr�n\nGunnar\nGUNNAR\nG�ran \ng�ran \nG�RAN \n(8 rows)\n\nbut:\n\ngoran=> select * from x WHERE txt LIKE 'G%';\ntxt \n------\nGr�n\nGunnar\nGUNNAR\n(3 rows)\n\nthe same goes for: \"select * from x WHERE txt ~ '^G'\" which is the same\nas above LIKE-stmt in the backend.\nOtherwise regex works correct.\n\nTo sum up:\nCase sensitive regex anchored with '^' and where the first char\nfollowing\nthe match is a non-ascii char gives misses.\n\nI have tracked it down to the following, 3 functions are called to test\nthe expression:\n1. textregexeq OK for 5 instances\n2. text_ge('G','G') also OK\n3. text_lt('G','G\\0xFF') this is not correct!\n\nCase 3 work not work with strcoll(), in varstr_cmp().\nIf I change it to strcoll() to strncmp() it works as expected,\nbut it probably breaks sorting etc big time.\n\nAny suggestions how to proceed?\n\n\tTIA,\n-- \n---------------------------------------------\nG�ran Thyni, JMS Bildbasen, Kiruna\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Thu, 28 Jan 1999 22:47:12 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "tough locale bug"
},
{
"msg_contents": "Tom Lane wrote:\n\n> However, we probably don't really want to touch each individual palloc()\n> call in the system to change it to a MemoryContextAlloc(some-context, size)\n> call. Not only would that be horribly tedious, but an awful lot of\n> these calls are made for the purpose of creating an object that the\n> requesting routine intends to pass back to its caller. In that\n> situation, the requesting routine doesn't really *know* what the\n> lifetime of the storage needs to be --- it's the caller that determines\n> what's going to happen.\n\n There are about 900 calls of palloc() in the backend code. It\n is much less than I expected (we have over 200,000 lines of\n code).\n\n So I vote for doing it ONCE HIGH QUALITY, instead of half way\n but easy to do.\n\n And I don't think you could avoid at least looking at each\n individual palloc(). You defined nesting levels, where at\n their end the allocated memory get's free'd. But there are\n many places where deeper nested functions put things into\n structures that are held in upper levels and thus live\n longer. These upper levels continue to use the things the\n callee's placed in there.\n\n\nJan\n\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Fri, 29 Jan 1999 15:15:19 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "[email protected] (Jan Wieck) writes:\n> Tom Lane wrote:\n>> However, we probably don't really want to touch each individual palloc()\n>> call in the system ...\n\n> There are about 900 calls of palloc() in the backend code. It\n> is much less than I expected (we have over 200,000 lines of\n> code).\n\nMuch less than I thought, also. And (grep|wc...) over 300 of these\ncalls are in the utils/adt code, ie, they are for passing back the\nresults of type conversion functions. We'll only need to think through\nhow that should work once, and then it's just a mechanical edit to all\nthe ADT files.\n\n> So I vote for doing it ONCE HIGH QUALITY, instead of half way\n> but easy to do.\n\nOK, I agree.\n\n> And I don't think you could avoid at least looking at each\n> individual palloc(). You defined nesting levels, where at\n> their end the allocated memory get's free'd. But there are\n> many places where deeper nested functions put things into\n> structures that are held in upper levels and thus live\n> longer. These upper levels continue to use the things the\n> callee's placed in there.\n\nWhat about my prior point that the bottom-level function may not know\nhow long the caller needs the storage? Will we have to add a \"memory\ncontext to put result in\" parameter to a lot of routines? Ugh, but\nmaybe it's the only way.\n\nI've noticed that everyone else contributing to this thread has been\nthinking in terms of inventing multiple allocation functions with\ndifferent names, ie a routine might have to call \"palloc\" or\n\"fast_palloc\" or \"tmp_palloc\" or whatever depending on what behavior it\nwants. I really think we are better off to stick to the structure that\nwe already have (cf. include/nodes/memnodes.h and include/utils/mcxt.h),\nin which there's one set of interface routines that take a \"context\"\nparameter, and the context determines the semantics of allocation.\n(The interface routines probably ought to be macros, not out-of-line\ncode, but that's a minor optimization.) This will make it easier to\nhandle cases where one routine has to tell another one how to allocate\nits result space: you pass a MemoryContext parameter.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Fri, 29 Jan 1999 11:21:15 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Tom Lane wrote:\n\n>\n> [email protected] (Jan Wieck) writes:\n> > Tom Lane wrote:\n> >> However, we probably don't really want to touch each individual palloc()\n> >> call in the system ...\n>\n> > There are about 900 calls of palloc() in the backend code. It\n> > is much less than I expected (we have over 200,000 lines of\n> > code).\n>\n> Much less than I thought, also. And (grep|wc...) over 300 of these\n> calls are in the utils/adt code, ie, they are for passing back the\n> results of type conversion functions. We'll only need to think through\n> how that should work once, and then it's just a mechanical edit to all\n> the ADT files.\n\n Exactly these 300 ones (except for 20 or 30 of them) are\n those that the caller needs :-)\n\n> What about my prior point that the bottom-level function may not know\n> how long the caller needs the storage? Will we have to add a \"memory\n> context to put result in\" parameter to a lot of routines? Ugh, but\n> maybe it's the only way.\n\n I've browsed a little throug the code (well, the candidates I\n thought are good choices for optimization). The problem there\n is, that the order of allocation hasn't anything to do with\n the lifecycle of the objects. So if we want to handle\n allocations different depending on the object lifecycle, the\n allocators (mainly their callers) would have to switch often\n between different contexts.\n\n It wouldn't make the code much better readable if between\n most of the statements some MemoryContextStumbleOver()\n appears. And many memory contexts means on the other hand\n many concurrently read/written pages. Thus I would expect it\n to affect the caching of the CPU.\n\n>\n> I've noticed that everyone else contributing to this thread has been\n> thinking in terms of inventing multiple allocation functions with\n> different names, ie a routine might have to call \"palloc\" or\n> \"fast_palloc\" or \"tmp_palloc\" or whatever depending on what behavior it\n> wants.\n\n So I.\n\n> I really think we are better off to stick to the structure that\n> we already have (cf. include/nodes/memnodes.h and include/utils/mcxt.h),\n> in which there's one set of interface routines that take a \"context\"\n> parameter, and the context determines the semantics of allocation.\n> (The interface routines probably ought to be macros, not out-of-line\n> code, but that's a minor optimization.) This will make it easier to\n> handle cases where one routine has to tell another one how to allocate\n> its result space: you pass a MemoryContext parameter.\n\n I came to the same conclusion. So I continued with the\n approach of bigger chunks handled in palloc(). What I have\n now is something, that gains about 10% speedup at the\n regression test, while memory consumption (visibly watched\n with top(1)) seems not to raise compared against old version.\n\n Since the bigger blocks are built on top of the existing\n memory context model, it would not conflict with any enhanced\n usage we could make with it.\n\n I include a patch at the end - please comment.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\ndiff -cr src.orig/backend/utils/mmgr/mcxt.c src/backend/utils/mmgr/mcxt.c\n*** src.orig/backend/utils/mmgr/mcxt.c\tThu Jan 28 16:47:39 1999\n--- src/backend/utils/mmgr/mcxt.c\tFri Jan 29 20:15:41 1999\n***************\n*** 106,111 ****\n--- 106,112 ----\n static struct GlobalMemory TopGlobalMemoryData = {\n \tT_GlobalMemory,\t\t\t\t/* NodeTag\t\t\t\ttag\t\t */\n \t&GlobalContextMethodsData,\t/* ContextMethods\t\tmethod\t */\n+ \tNULL,\t\t\t\t\t\t/* char* smallchunk_block\t*/\n \t{{0}},\t\t\t\t\t\t/* uninitialized OrderedSetData allocSetD */\n \t\"TopGlobal\",\t\t\t\t/* char* name\t */\n \t{0}\t\t\t\t\t\t\t/* uninitialized OrderedElemData elemD */\ndiff -cr src.orig/backend/utils/mmgr/palloc.c src/backend/utils/mmgr/palloc.c\n*** src.orig/backend/utils/mmgr/palloc.c\tThu Jan 28 16:47:39 1999\n--- src/backend/utils/mmgr/palloc.c\tFri Jan 29 21:57:06 1999\n***************\n*** 24,29 ****\n--- 24,57 ----\n \n #include \"utils/palloc.h\"\n \n+ \n+ typedef struct PallocBlock {\n+ \tMemoryContext\tmcxt;\n+ \tint\t\t\t\trefcount;\n+ \tchar\t\t *unused_chunk;\n+ \tchar\t\t *freeptr;\n+ \tchar\t\t *endptr;\n+ } PallocBlock;\n+ \n+ typedef struct PallocChunk {\n+ \tPallocBlock\t\t*blk;\n+ \tSize\t\t\tsize;\n+ } PallocChunk;\n+ \n+ \n+ #define PALLOC_BLOCKSIZE\t\t\\\n+ \t\t\t(8192 - sizeof(OrderedElemData) - sizeof(Size))\n+ #define PALLOC_CHUNK_LIMIT\t\t512\n+ #define PALLOC_SIZE_ALIGN(s)\t((s + 15) & ~15)\n+ \n+ #define PALLOC_CHUNK_BLKPTR(c)\t(((PallocChunk *)(c))[-1].blk)\n+ #define PALLOC_CHUNK_SIZE(c)\t(((PallocChunk *)(c))[-1].size)\n+ \n+ #define PALLOC_CHUNK_HDRSZ\t\tMAXALIGN(sizeof(PallocChunk))\n+ #define PALLOC_BLOCK_HDRSZ\t\tMAXALIGN(sizeof(PallocBlock))\n+ \n+ #define PALLOC_FREESPACE(b)\t\t((b)->endptr - (b)->freeptr)\n+ \n /* ----------------------------------------------------------------\n *\t\tUser library functions\n * ----------------------------------------------------------------\n***************\n*** 66,72 ****\n #ifdef PALLOC_IS_MALLOC\n \treturn malloc(size);\n #else\n! \treturn MemoryContextAlloc(CurrentMemoryContext, size);\n #endif\t /* PALLOC_IS_MALLOC */\n }\n \n--- 94,168 ----\n #ifdef PALLOC_IS_MALLOC\n \treturn malloc(size);\n #else\n! \tPallocBlock\t\t*block;\n! \tchar\t\t\t*chunk;\n! \tSize\t\t\tasize = PALLOC_SIZE_ALIGN(size);\n! \n! \tif (asize >= PALLOC_CHUNK_LIMIT)\n! \t{\n! \t\tchunk = (char *)MemoryContextAlloc(CurrentMemoryContext, size + \n! \t\t\t\t\t\t\t\t\t\t\t\t\t\tPALLOC_CHUNK_HDRSZ);\n! \t\tchunk += PALLOC_CHUNK_HDRSZ;\n! \t\tPALLOC_CHUNK_BLKPTR(chunk) = NULL;\n! \t\treturn (void *)chunk;\n! \t}\n! \n! \tblock = (PallocBlock *)(CurrentMemoryContext->smallchunk_block);\n! \n! \tif (block != NULL && PALLOC_FREESPACE(block) < asize)\n! \t{\n! \t\tchar\t*prev = NULL;\n! \t\tSize\tchunk_size;\n! \n! \t\tchunk = block->unused_chunk;\n! \t\twhile (chunk)\n! \t\t{\n! \t\t\tchunk_size = PALLOC_CHUNK_SIZE(chunk);\n! \t\t\tif (asize == chunk_size)\n! \t\t\t{\n! \t\t\t\tif (prev == NULL)\n! \t\t\t\t\tblock->unused_chunk = (char *)PALLOC_CHUNK_BLKPTR(chunk);\n! \t\t\t\telse\n! \t\t\t\t\tPALLOC_CHUNK_BLKPTR(prev) = PALLOC_CHUNK_BLKPTR(chunk);\n! \t\t\t\t\t\t\t\t\t\n! \t\t\t\tblock->refcount++;\n! \t\t\t\tPALLOC_CHUNK_BLKPTR(chunk) = block;\n! \t\t\t\t/*\n! \t\t\t\tPALLOC_CHUNK_SIZE(chunk) = size;\n! \t\t\t\t*/\n! \t\t\t\treturn (void *)chunk;\n! \t\t\t}\n! \t\t\tprev = chunk;\n! \t\t\tchunk = (char *)PALLOC_CHUNK_BLKPTR(chunk);\n! \t\t}\n! \n! \t\tblock = NULL;\n! \t}\n! \n! \tif (block == NULL)\n! \t{\n! \t\tblock = (PallocBlock *)MemoryContextAlloc(CurrentMemoryContext, \n! \t\t\t\t\t\t\t\t\tPALLOC_BLOCKSIZE);\n! \t\tblock->mcxt = CurrentMemoryContext;\n! \t\tblock->unused_chunk = NULL;\n! \t\tblock->refcount = 0;\n! \t\tblock->freeptr = ((char *)block) + PALLOC_BLOCK_HDRSZ + \n! \t\t\t\t\t\t\t\t\t\t\tPALLOC_CHUNK_HDRSZ;\n! \t\tblock->endptr = ((char *)block) + PALLOC_BLOCKSIZE;\n! \n! \t\tCurrentMemoryContext->smallchunk_block = (void *)block;\n! \t}\n! \n! \tchunk = block->freeptr;\n! \tblock->freeptr += PALLOC_CHUNK_HDRSZ + asize;\n! \tblock->refcount++;\n! \tPALLOC_CHUNK_BLKPTR(chunk) = block;\n! \tPALLOC_CHUNK_SIZE(chunk) = asize;\n! \n! \tif (block->freeptr >= block->endptr)\n! \t\tblock->mcxt->smallchunk_block = NULL;\n! \n! \treturn (void *)chunk;\n #endif\t /* PALLOC_IS_MALLOC */\n }\n \n***************\n*** 76,82 ****\n #ifdef PALLOC_IS_MALLOC\n \tfree(pointer);\n #else\n! \tMemoryContextFree(CurrentMemoryContext, pointer);\n #endif\t /* PALLOC_IS_MALLOC */\n }\n \n--- 172,202 ----\n #ifdef PALLOC_IS_MALLOC\n \tfree(pointer);\n #else\n! \tPallocBlock\t\t*block = PALLOC_CHUNK_BLKPTR(pointer);\n! \n! \tif (block == NULL)\n! \t{\n! \t\tMemoryContextFree(CurrentMemoryContext, (char *)pointer - PALLOC_CHUNK_HDRSZ);\n! \t\treturn;\n! \t}\n! \n! \tPALLOC_CHUNK_BLKPTR(pointer) = (PallocBlock *)(block->unused_chunk);\n! \tblock->unused_chunk = (char *)pointer;\n! \n! \tblock->refcount--;\n! \tif (block->refcount == 0)\n! \t{\n! \t\tif (block == (PallocBlock *)(block->mcxt->smallchunk_block))\n! \t\t{\n! \t\t\tblock->freeptr = ((char *)block) + PALLOC_BLOCK_HDRSZ +\n! \t\t\t\t\t\t\t\t\t\t\t\tPALLOC_CHUNK_HDRSZ;\n! \t\t\tblock->unused_chunk = NULL;\n! \t\t}\n! \t\telse\n! \t\t{\n! \t\t\tMemoryContextFree(block->mcxt, (void *)block);\n! \t\t}\n! \t}\n #endif\t /* PALLOC_IS_MALLOC */\n }\n \n***************\n*** 100,106 ****\n #ifdef PALLOC_IS_MALLOC\n \treturn realloc(pointer, size);\n #else\n! \treturn MemoryContextRealloc(CurrentMemoryContext, pointer, size);\n #endif\n }\n \n--- 220,248 ----\n #ifdef PALLOC_IS_MALLOC\n \treturn realloc(pointer, size);\n #else\n! \tPallocBlock\t\t*block = PALLOC_CHUNK_BLKPTR(pointer);\n! \tchar \t\t\t*new;\n! \tSize\t\t\ttocopy;\n! \n! \tif (block == NULL)\n! \t{\n! \t\tnew = (char *)MemoryContextRealloc(CurrentMemoryContext, \n! \t\t\t\t\t\t\t\t(char *)pointer - PALLOC_CHUNK_HDRSZ,\n! \t\t\t\t\t\t\t\tsize + PALLOC_CHUNK_HDRSZ);\n! \t\tnew += PALLOC_CHUNK_HDRSZ;\n! \t\tPALLOC_CHUNK_BLKPTR(new) = NULL;\n! \t\treturn (void *)new;\n! \t}\n! \telse\n! \t{\n! \t\tnew = palloc(size);\n! \n! \t\ttocopy = PALLOC_CHUNK_SIZE(pointer) > size ? \n! \t\t\t\t\t\t\tsize : PALLOC_CHUNK_SIZE(pointer);\n! \t\tmemcpy(new, pointer, tocopy);\n! \t\tpfree(pointer);\n! \t\treturn (void *)new;\n! \t}\n #endif\n }\n \n***************\n*** 117,119 ****\n--- 259,263 ----\n \n \treturn nstr;\n }\n+ \n+ \ndiff -cr src.orig/backend/utils/mmgr/portalmem.c src/backend/utils/mmgr/portalmem.c\n*** src.orig/backend/utils/mmgr/portalmem.c\tThu Jan 28 16:47:39 1999\n--- src/backend/utils/mmgr/portalmem.c\tFri Jan 29 20:15:41 1999\n***************\n*** 390,395 ****\n--- 390,396 ----\n \tNodeSetTag((Node *) &portal->variable, T_PortalVariableMemory);\n \tAllocSetInit(&portal->variable.setData, DynamicAllocMode, (Size) 0);\n \tportal->variable.method = &PortalVariableContextMethodsData;\n+ \tportal->variable.smallchunk_block = NULL;\n \n \t/*\n \t * initialize portal heap context\n***************\n*** 399,404 ****\n--- 400,406 ----\n \tFixedStackInit(&portal->heap.stackData,\n \t\t\t\t offsetof(HeapMemoryBlockData, itemData));\n \tportal->heap.method = &PortalHeapContextMethodsData;\n+ \tportal->heap.smallchunk_block = NULL;\n \n \t/*\n \t * set bogus portal name\n***************\n*** 756,761 ****\n--- 758,764 ----\n \tNodeSetTag((Node *) &portal->variable, T_PortalVariableMemory);\n \tAllocSetInit(&portal->variable.setData, DynamicAllocMode, (Size) 0);\n \tportal->variable.method = &PortalVariableContextMethodsData;\n+ \tportal->variable.smallchunk_block = NULL;\n \n \t/* initialize portal heap context */\n \tNodeSetTag((Node *) &portal->heap, T_PortalHeapMemory);\n***************\n*** 763,768 ****\n--- 766,772 ----\n \tFixedStackInit(&portal->heap.stackData,\n \t\t\t\t offsetof(HeapMemoryBlockData, itemData));\n \tportal->heap.method = &PortalHeapContextMethodsData;\n+ \tportal->heap.smallchunk_block = NULL;\n \n \t/* initialize portal name */\n \tlength = 1 + strlen(name);\n***************\n*** 918,923 ****\n--- 922,928 ----\n \n \t/* free current mode */\n \tAllocSetReset(&HEAPMEMBLOCK(context)->setData);\n+ \tcontext->smallchunk_block = NULL;\n \tMemoryContextFree((MemoryContext) PortalHeapMemoryGetVariableMemory(context),\n \t\t\t\t\t context->block);\n \ndiff -cr src.orig/include/nodes/memnodes.h src/include/nodes/memnodes.h\n*** src.orig/include/nodes/memnodes.h\tThu Jan 28 16:47:39 1999\n--- src/include/nodes/memnodes.h\tFri Jan 29 20:15:41 1999\n***************\n*** 60,65 ****\n--- 60,66 ----\n {\n \tNodeTag\t\ttype;\n \tMemoryContextMethods method;\n+ \tvoid\t\t*smallchunk_block;\n }\t\t *MemoryContext;\n \n /* think about doing this right some time but we'll have explicit fields\n***************\n*** 68,73 ****\n--- 69,75 ----\n {\n \tNodeTag\t\ttype;\n \tMemoryContextMethods method;\n+ \tvoid\t\t*smallchunk_block;\n \tAllocSetData setData;\n \tchar\t *name;\n \tOrderedElemData elemData;\n***************\n*** 79,84 ****\n--- 81,87 ----\n {\n \tNodeTag\t\ttype;\n \tMemoryContextMethods method;\n+ \tvoid\t\t*smallchunk_block;\n \tAllocSetData setData;\n }\t\t *PortalVariableMemory;\n \n***************\n*** 86,91 ****\n--- 89,95 ----\n {\n \tNodeTag\t\ttype;\n \tMemoryContextMethods method;\n+ \tvoid\t\t*smallchunk_block;\n \tPointer\t\tblock;\n \tFixedStackData stackData;\n }\t\t *PortalHeapMemory;\n",
"msg_date": "Fri, 29 Jan 1999 23:10:35 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> I've noticed that everyone else contributing to this thread has been\n> thinking in terms of inventing multiple allocation functions with\n> different names, ie a routine might have to call \"palloc\" or\n> \"fast_palloc\" or \"tmp_palloc\" or whatever depending on what behavior it\n> wants. I really think we are better off to stick to the structure that\n> we already have (cf. include/nodes/memnodes.h and include/utils/mcxt.h),\n> in which there's one set of interface routines that take a \"context\"\n> parameter, and the context determines the semantics of allocation.\n> (The interface routines probably ought to be macros, not out-of-line\n> code, but that's a minor optimization.) This will make it easier to\n> handle cases where one routine has to tell another one how to allocate\n> its result space: you pass a MemoryContext parameter.\n\nI agree, inline is good. I can imagine something like makeNode is going\nto be tough.\n\nI am going to give the parser as an example. It palloc's all sorts of\nthings for rangetables, strings, structures, lists, but eventually just\nexports a huge Query structure.\n\nIf we could somehow just grab copy Query structure and free all the\nother parser palloc allocations, that would be great.\n\nSame thing with the optimizer. It makes all sorts of lists, but\neventuall just exports Plan's. All the other stuff can be pfree'ed.\n\nIt is almost as though we could copy the Query structure to another\nmemory context, and just fee all palloc'ed memory.\n\nWhat we could try is to keep the various memory contexts like cache\ncontext unchanged, but allow pallocs in the standard memory context to\nbe allocated from a pool and free'ed automatically. Our standard memory\ncontext is overused and doesn't need the pfree overhead. We could make\npalloc a macro that tested for the current context, and did standard\npallocs from a free memory pool.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Sat, 30 Jan 1999 01:06:12 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Goran Thyni wrote:\n> 3. text_lt('G','G\\0xFF') this is not correct!\n> \n> Case 3 work not work with strcoll(), in varstr_cmp().\n> If I change it to strcoll() to strncmp() it works as expected,\n> but it probably breaks sorting etc big time.\n\nHere is a hack which solves the problem without breaking sorting AFAIK.\nI have read a lot of code, but has not found any cleaner way to fix\nthis.\nA cleaner solution would be to find the highest char in locale charset\nbut I found no portable way to do that, any pointers appriciated.\n\nIt is not a beauty but it works. \nPlease apply to -current since it fixes the bug, \nunless someone has a better suggestion.\n\n\tbest regards,\n-- \n-----------------\nG�ran Thyni\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!",
"msg_date": "Sun, 31 Jan 1999 16:52:39 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Patch (was: tough locale bug)"
},
{
"msg_contents": "> #ifdef USE_LOCALE\n> ! if ((int)arg2[len2 - 1] != -1)\n\nSorry, I'm not as familiar with the LOCALE code as I should be. Why are\nwe testing explicitly for -1? Is there something special about a\ncharacter with all bits set in typical locale support code?\n\nRegards.\n\n - Tom\n",
"msg_date": "Mon, 01 Feb 1999 13:59:35 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Hi!\n\nOn Mon, 1 Feb 1999, Thomas G. Lockhart wrote:\n> > #ifdef USE_LOCALE\n> > ! if ((int)arg2[len2 - 1] != -1)\n> \n> Sorry, I'm not as familiar with the LOCALE code as I should be. Why are\n> we testing explicitly for -1? Is there something special about a\n> character with all bits set in typical locale support code?\n\n Other way around - in windows-1251 it is pretty standard character.\n\nOleg.\n---- \n Oleg Broytmann National Research Surgery Centre http://sun.med.ru/~phd/\n Programmers don't die, they just GOSUB without RETURN.\n\n",
"msg_date": "Mon, 1 Feb 1999 17:10:15 +0300 (MSK)",
"msg_from": "Oleg Broytmann <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "\"Thomas G. Lockhart\" wrote:\n> \n> > #ifdef USE_LOCALE\n> > ! if ((int)arg2[len2 - 1] != -1)\n> \n> Sorry, I'm not as familiar with the LOCALE code as I should be. Why are\n> we testing explicitly for -1? Is there something special about a\n> character with all bits set in typical locale support code?\n> \n> Regards.\n> \n> - Tom\n\nIt looks a bit cryptic yes, but it is not specific to locale.\nThe planner appends a \"(char) -1\" (or alt syntax '\\xFF') to\nthe teststring in some cases of \"<=\" on text (f.ex in \"txt ~ '^G'\").\nThis works OK with strncmp which compare char by char as unsigned ints,\nbut locale uses lookup-tables where '\\FF' not is guanteed to be\ngreater or equal to any other unsigned 8-bit character.\n\nTherefore I did a fallback to strncmp (instead of locale aware strcoll)\nfor this special case.\nIt is not the perfect solution and may break in a few cases, but as it\nis now\nit breaks most usage of LIKE and ~ when locale is enabled.\n\nIf we could in a portable way to find the last usable character in the\nused charset it would be nice, but I have not found any way to do that.\n\n\tregards,\n-- \n---------------------------------------------\nG�ran Thyni, JMS Bildbasen, Kiruna\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Mon, 01 Feb 1999 17:38:52 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Oleg Broytmann wrote:\n> \n> Hi!\n> \n> On Mon, 1 Feb 1999, Thomas G. Lockhart wrote:\n> > > #ifdef USE_LOCALE\n> > > ! if ((int)arg2[len2 - 1] != -1)\n> >\n> > Sorry, I'm not as familiar with the LOCALE code as I should be. Why are\n> > we testing explicitly for -1? Is there something special about a\n> > character with all bits set in typical locale support code?\n> \n> Other way around - in windows-1251 it is pretty standard character.\n\nOK,\nand it is probably not the last character in the alphabet either. \nIn that case we need another solution I am afraid.\nAnyone got any ideas how to resolve this?\n\n\tregards,\n-- \n---------------------------------------------\nG�ran Thyni, JMS Bildbasen, Kiruna\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Mon, 01 Feb 1999 17:45:55 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Hi!\n\nOn Mon, 1 Feb 1999, Goran Thyni wrote:\n> > Other way around - in windows-1251 it is pretty standard character.\n> and it is probably not the last character in the alphabet either. \n\n In win-1251 it is last charachter in alphabet. In koi8-r it is not.\n\nOleg.\n---- \n Oleg Broytmann http://members.xoom.com/phd2/ [email protected]\n Programmers don't die, they just GOSUB without RETURN.\n\n",
"msg_date": "Tue, 2 Feb 1999 12:43:30 +0300 (MSK)",
"msg_from": "Oleg Broytmann <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Oleg Broytmann wrote:\n> \n> Hi!\n> \n> On Mon, 1 Feb 1999, Goran Thyni wrote:\n> > > Other way around - in windows-1251 it is pretty standard character.\n> > and it is probably not the last character in the alphabet either.\n> \n> In win-1251 it is last charachter in alphabet. In koi8-r it is not.\n\nI posted a new patch that corrects the problem in a way that\ndoesn't break koi8-r (and possibly others). I hope it will get\napplied before beta starts.\n\n\tmvh,\n-- \n---------------------------------------------\nG�ran Thyni, JMS Bildbasen, Kiruna\nThis is Penguin Country. On a quiet night you can hear Windows NT\nreboot!\n",
"msg_date": "Tue, 02 Feb 1999 11:47:43 +0100",
"msg_from": "Goran Thyni <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Hi!\n\nOn Tue, 2 Feb 1999, Goran Thyni wrote:\n> I posted a new patch that corrects the problem in a way that\n> doesn't break koi8-r (and possibly others). I hope it will get\n> applied before beta starts.\n\n I saw the patch. I'll try to apply it and test.\n BTW, if you are looking into locale - could you please supply locale\nregression test for you locale? Look into ...src/test/locale, copy koi8-r\ndirectory and supply your test data.\n\nOleg.\n---- \n Oleg Broytmann National Research Surgery Centre http://sun.med.ru/~phd/\n Programmers don't die, they just GOSUB without RETURN.\n\n",
"msg_date": "Tue, 2 Feb 1999 14:08:48 +0300 (MSK)",
"msg_from": "Oleg Broytmann <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "I wrote:\n\n>\n> I came to the same conclusion. So I continued with the\n> approach of bigger chunks handled in palloc(). What I have\n> now is something, that gains about 10% speedup at the\n> regression test, while memory consumption (visibly watched\n> with top(1)) seems not to raise compared against old version.\n>\n> Since the bigger blocks are built on top of the existing\n> memory context model, it would not conflict with any enhanced\n> usage we could make with it.\n>\n> I include a patch at the end - please comment.\n>\n>\n> Jan\n\n Did anyone play around with it? I've had it installed now for\n some days and it work's well so far.\n\n How close are we to v6.5 BETA? Should I apply it to CURRENT?\n\n> diff -cr src.orig/backend/utils/mmgr/mcxt.c src/backend/utils/mmgr/mcxt.c\n> *** src.orig/backend/utils/mmgr/mcxt.c Thu Jan 28 16:47:39 1999\n> --- src/backend/utils/mmgr/mcxt.c Fri Jan 29 20:15:41 1999\n> ***************\n> *** 106,111 ****\n> --- 106,112 ----\n> static struct GlobalMemory TopGlobalMemoryData = {\n> T_GlobalMemory, /* NodeTag tag */\n> &GlobalContextMethodsData, /* ContextMethods method */\n> + NULL, /* char* smallchunk_block */\n> {{0}}, /* uninitialized OrderedSetData allocSetD */\n> \"TopGlobal\", /* char* name */\n> {0} /* uninitialized OrderedElemData elemD */\n> diff -cr src.orig/backend/utils/mmgr/palloc.c src/backend/utils/mmgr/palloc.c\n> *** src.orig/backend/utils/mmgr/palloc.c Thu Jan 28 16:47:39 1999\n> --- src/backend/utils/mmgr/palloc.c Fri Jan 29 21:57:06 1999\n> ***************\n> *** 24,29 ****\n> --- 24,57 ----\n>\n> #include \"utils/palloc.h\"\n>\n> +\n> + typedef struct PallocBlock {\n> + MemoryContext mcxt;\n> + int refcount;\n> + char *unused_chunk;\n> + char *freeptr;\n> + char *endptr;\n> + } PallocBlock;\n> +\n> + typedef struct PallocChunk {\n> + PallocBlock *blk;\n> + Size size;\n> + } PallocChunk;\n> +\n> +\n> + #define PALLOC_BLOCKSIZE \\\n> + (8192 - sizeof(OrderedElemData) - sizeof(Size))\n> + #define PALLOC_CHUNK_LIMIT 512\n> + #define PALLOC_SIZE_ALIGN(s) ((s + 15) & ~15)\n> +\n> + #define PALLOC_CHUNK_BLKPTR(c) (((PallocChunk *)(c))[-1].blk)\n> + #define PALLOC_CHUNK_SIZE(c) (((PallocChunk *)(c))[-1].size)\n> +\n> + #define PALLOC_CHUNK_HDRSZ MAXALIGN(sizeof(PallocChunk))\n> + #define PALLOC_BLOCK_HDRSZ MAXALIGN(sizeof(PallocBlock))\n> +\n> + #define PALLOC_FREESPACE(b) ((b)->endptr - (b)->freeptr)\n> +\n> /* ----------------------------------------------------------------\n> * User library functions\n> * ----------------------------------------------------------------\n> ***************\n> *** 66,72 ****\n> #ifdef PALLOC_IS_MALLOC\n> return malloc(size);\n> #else\n> ! return MemoryContextAlloc(CurrentMemoryContext, size);\n> #endif /* PALLOC_IS_MALLOC */\n> }\n>\n> --- 94,168 ----\n> #ifdef PALLOC_IS_MALLOC\n> return malloc(size);\n> #else\n> ! PallocBlock *block;\n> ! char *chunk;\n> ! Size asize = PALLOC_SIZE_ALIGN(size);\n> !\n> ! if (asize >= PALLOC_CHUNK_LIMIT)\n> ! {\n> ! chunk = (char *)MemoryContextAlloc(CurrentMemoryContext, size +\n> ! PALLOC_CHUNK_HDRSZ);\n> ! chunk += PALLOC_CHUNK_HDRSZ;\n> ! PALLOC_CHUNK_BLKPTR(chunk) = NULL;\n> ! return (void *)chunk;\n> ! }\n> !\n> ! block = (PallocBlock *)(CurrentMemoryContext->smallchunk_block);\n> !\n> ! if (block != NULL && PALLOC_FREESPACE(block) < asize)\n> ! {\n> ! char *prev = NULL;\n> ! Size chunk_size;\n> !\n> ! chunk = block->unused_chunk;\n> ! while (chunk)\n> ! {\n> ! chunk_size = PALLOC_CHUNK_SIZE(chunk);\n> ! if (asize == chunk_size)\n> ! {\n> ! if (prev == NULL)\n> ! block->unused_chunk = (char *)PALLOC_CHUNK_BLKPTR(chunk);\n> ! else\n> ! PALLOC_CHUNK_BLKPTR(prev) = PALLOC_CHUNK_BLKPTR(chunk);\n> !\n> ! block->refcount++;\n> ! PALLOC_CHUNK_BLKPTR(chunk) = block;\n> ! /*\n> ! PALLOC_CHUNK_SIZE(chunk) = size;\n> ! */\n> ! return (void *)chunk;\n> ! }\n> ! prev = chunk;\n> ! chunk = (char *)PALLOC_CHUNK_BLKPTR(chunk);\n> ! }\n> !\n> ! block = NULL;\n> ! }\n> !\n> ! if (block == NULL)\n> ! {\n> ! block = (PallocBlock *)MemoryContextAlloc(CurrentMemoryContext,\n> ! PALLOC_BLOCKSIZE);\n> ! block->mcxt = CurrentMemoryContext;\n> ! block->unused_chunk = NULL;\n> ! block->refcount = 0;\n> ! block->freeptr = ((char *)block) + PALLOC_BLOCK_HDRSZ +\n> ! PALLOC_CHUNK_HDRSZ;\n> ! block->endptr = ((char *)block) + PALLOC_BLOCKSIZE;\n> !\n> ! CurrentMemoryContext->smallchunk_block = (void *)block;\n> ! }\n> !\n> ! chunk = block->freeptr;\n> ! block->freeptr += PALLOC_CHUNK_HDRSZ + asize;\n> ! block->refcount++;\n> ! PALLOC_CHUNK_BLKPTR(chunk) = block;\n> ! PALLOC_CHUNK_SIZE(chunk) = asize;\n> !\n> ! if (block->freeptr >= block->endptr)\n> ! block->mcxt->smallchunk_block = NULL;\n> !\n> ! return (void *)chunk;\n> #endif /* PALLOC_IS_MALLOC */\n> }\n>\n> ***************\n> *** 76,82 ****\n> #ifdef PALLOC_IS_MALLOC\n> free(pointer);\n> #else\n> ! MemoryContextFree(CurrentMemoryContext, pointer);\n> #endif /* PALLOC_IS_MALLOC */\n> }\n>\n> --- 172,202 ----\n> #ifdef PALLOC_IS_MALLOC\n> free(pointer);\n> #else\n> ! PallocBlock *block = PALLOC_CHUNK_BLKPTR(pointer);\n> !\n> ! if (block == NULL)\n> ! {\n> ! MemoryContextFree(CurrentMemoryContext, (char *)pointer - PALLOC_CHUNK_HDRSZ);\n> ! return;\n> ! }\n> !\n> ! PALLOC_CHUNK_BLKPTR(pointer) = (PallocBlock *)(block->unused_chunk);\n> ! block->unused_chunk = (char *)pointer;\n> !\n> ! block->refcount--;\n> ! if (block->refcount == 0)\n> ! {\n> ! if (block == (PallocBlock *)(block->mcxt->smallchunk_block))\n> ! {\n> ! block->freeptr = ((char *)block) + PALLOC_BLOCK_HDRSZ +\n> ! PALLOC_CHUNK_HDRSZ;\n> ! block->unused_chunk = NULL;\n> ! }\n> ! else\n> ! {\n> ! MemoryContextFree(block->mcxt, (void *)block);\n> ! }\n> ! }\n> #endif /* PALLOC_IS_MALLOC */\n> }\n>\n> ***************\n> *** 100,106 ****\n> #ifdef PALLOC_IS_MALLOC\n> return realloc(pointer, size);\n> #else\n> ! return MemoryContextRealloc(CurrentMemoryContext, pointer, size);\n> #endif\n> }\n>\n> --- 220,248 ----\n> #ifdef PALLOC_IS_MALLOC\n> return realloc(pointer, size);\n> #else\n> ! PallocBlock *block = PALLOC_CHUNK_BLKPTR(pointer);\n> ! char *new;\n> ! Size tocopy;\n> !\n> ! if (block == NULL)\n> ! {\n> ! new = (char *)MemoryContextRealloc(CurrentMemoryContext,\n> ! (char *)pointer - PALLOC_CHUNK_HDRSZ,\n> ! size + PALLOC_CHUNK_HDRSZ);\n> ! new += PALLOC_CHUNK_HDRSZ;\n> ! PALLOC_CHUNK_BLKPTR(new) = NULL;\n> ! return (void *)new;\n> ! }\n> ! else\n> ! {\n> ! new = palloc(size);\n> !\n> ! tocopy = PALLOC_CHUNK_SIZE(pointer) > size ?\n> ! size : PALLOC_CHUNK_SIZE(pointer);\n> ! memcpy(new, pointer, tocopy);\n> ! pfree(pointer);\n> ! return (void *)new;\n> ! }\n> #endif\n> }\n>\n> ***************\n> *** 117,119 ****\n> --- 259,263 ----\n>\n> return nstr;\n> }\n> +\n> +\n> diff -cr src.orig/backend/utils/mmgr/portalmem.c src/backend/utils/mmgr/portalmem.c\n> *** src.orig/backend/utils/mmgr/portalmem.c Thu Jan 28 16:47:39 1999\n> --- src/backend/utils/mmgr/portalmem.c Fri Jan 29 20:15:41 1999\n> ***************\n> *** 390,395 ****\n> --- 390,396 ----\n> NodeSetTag((Node *) &portal->variable, T_PortalVariableMemory);\n> AllocSetInit(&portal->variable.setData, DynamicAllocMode, (Size) 0);\n> portal->variable.method = &PortalVariableContextMethodsData;\n> + portal->variable.smallchunk_block = NULL;\n>\n> /*\n> * initialize portal heap context\n> ***************\n> *** 399,404 ****\n> --- 400,406 ----\n> FixedStackInit(&portal->heap.stackData,\n> offsetof(HeapMemoryBlockData, itemData));\n> portal->heap.method = &PortalHeapContextMethodsData;\n> + portal->heap.smallchunk_block = NULL;\n>\n> /*\n> * set bogus portal name\n> ***************\n> *** 756,761 ****\n> --- 758,764 ----\n> NodeSetTag((Node *) &portal->variable, T_PortalVariableMemory);\n> AllocSetInit(&portal->variable.setData, DynamicAllocMode, (Size) 0);\n> portal->variable.method = &PortalVariableContextMethodsData;\n> + portal->variable.smallchunk_block = NULL;\n>\n> /* initialize portal heap context */\n> NodeSetTag((Node *) &portal->heap, T_PortalHeapMemory);\n> ***************\n> *** 763,768 ****\n> --- 766,772 ----\n> FixedStackInit(&portal->heap.stackData,\n> offsetof(HeapMemoryBlockData, itemData));\n> portal->heap.method = &PortalHeapContextMethodsData;\n> + portal->heap.smallchunk_block = NULL;\n>\n> /* initialize portal name */\n> length = 1 + strlen(name);\n> ***************\n> *** 918,923 ****\n> --- 922,928 ----\n>\n> /* free current mode */\n> AllocSetReset(&HEAPMEMBLOCK(context)->setData);\n> + context->smallchunk_block = NULL;\n> MemoryContextFree((MemoryContext) PortalHeapMemoryGetVariableMemory(context),\n> context->block);\n>\n> diff -cr src.orig/include/nodes/memnodes.h src/include/nodes/memnodes.h\n> *** src.orig/include/nodes/memnodes.h Thu Jan 28 16:47:39 1999\n> --- src/include/nodes/memnodes.h Fri Jan 29 20:15:41 1999\n> ***************\n> *** 60,65 ****\n> --- 60,66 ----\n> {\n> NodeTag type;\n> MemoryContextMethods method;\n> + void *smallchunk_block;\n> } *MemoryContext;\n>\n> /* think about doing this right some time but we'll have explicit fields\n> ***************\n> *** 68,73 ****\n> --- 69,75 ----\n> {\n> NodeTag type;\n> MemoryContextMethods method;\n> + void *smallchunk_block;\n> AllocSetData setData;\n> char *name;\n> OrderedElemData elemData;\n> ***************\n> *** 79,84 ****\n> --- 81,87 ----\n> {\n> NodeTag type;\n> MemoryContextMethods method;\n> + void *smallchunk_block;\n> AllocSetData setData;\n> } *PortalVariableMemory;\n>\n> ***************\n> *** 86,91 ****\n> --- 89,95 ----\n> {\n> NodeTag type;\n> MemoryContextMethods method;\n> + void *smallchunk_block;\n> Pointer block;\n> FixedStackData stackData;\n> } *PortalHeapMemory;\n>\n>\n\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Tue, 2 Feb 1999 15:35:30 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "[email protected] (Jan Wieck) writes:\n> [ patch to handle small-block allocation specially ]\n> Did anyone play around with it? I've had it installed now for\n> some days and it work's well so far.\n> How close are we to v6.5 BETA? Should I apply it to CURRENT?\n\nWell, I think it's a band-aid until we can redesign the memory manager\nproperly, but it could be a useful band-aid ... might as well put it in.\n\nThe only thing that bothers me about the code itself is lack of\ncomments. In particular it'd be nice to mention somewhere that freeptr\nis actually one chunkheader past the start of the free space --- took\na while to puzzle that out.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Tue, 02 Feb 1999 10:27:02 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": ">\n> [email protected] (Jan Wieck) writes:\n> > [ patch to handle small-block allocation specially ]\n> > Did anyone play around with it? I've had it installed now for\n> > some days and it work's well so far.\n> > How close are we to v6.5 BETA? Should I apply it to CURRENT?\n>\n> Well, I think it's a band-aid until we can redesign the memory manager\n> properly, but it could be a useful band-aid ... might as well put it in.\n>\n> The only thing that bothers me about the code itself is lack of\n> comments. In particular it'd be nice to mention somewhere that freeptr\n> is actually one chunkheader past the start of the free space --- took\n> a while to puzzle that out.\n\n Tricky eh? The AllocSet...() functions do it the same so it\n just continues with that style.\n\n But you're right, and when moving it in I'll shurely add\n comments.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Tue, 2 Feb 1999 16:34:06 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "On Tue, 2 Feb 1999, Jan Wieck wrote:\n\n> I wrote:\n> \n> >\n> > I came to the same conclusion. So I continued with the\n> > approach of bigger chunks handled in palloc(). What I have\n> > now is something, that gains about 10% speedup at the\n> > regression test, while memory consumption (visibly watched\n> > with top(1)) seems not to raise compared against old version.\n> >\n> > Since the bigger blocks are built on top of the existing\n> > memory context model, it would not conflict with any enhanced\n> > usage we could make with it.\n> >\n> > I include a patch at the end - please comment.\n> >\n> >\n> > Jan\n> \n> Did anyone play around with it? I've had it installed now for\n> some days and it work's well so far.\n> \n> How close are we to v6.5 BETA? Should I apply it to CURRENT?\n\nBETA starts upon Vadim's word...I'd say go for it...\n\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Tue, 2 Feb 1999 11:51:08 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> > Since the bigger blocks are built on top of the existing\n> > memory context model, it would not conflict with any enhanced\n> > usage we could make with it.\n> >\n> > I include a patch at the end - please comment.\n> >\n> >\n> > Jan\n> \n> Did anyone play around with it? I've had it installed now for\n> some days and it work's well so far.\n> \n> How close are we to v6.5 BETA? Should I apply it to CURRENT?\n\n\nGood question. I am done with temp tables, so Vadim makes the call on\nwhen to start beta.\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 2 Feb 1999 11:39:18 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": ">\n> > > Since the bigger blocks are built on top of the existing\n> > > memory context model, it would not conflict with any enhanced\n> > > usage we could make with it.\n> > >\n> > > I include a patch at the end - please comment.\n> > >\n> > >\n> > > Jan\n> >\n> > Did anyone play around with it? I've had it installed now for\n> > some days and it work's well so far.\n> >\n> > How close are we to v6.5 BETA? Should I apply it to CURRENT?\n>\n>\n> Good question. I am done with temp tables, so Vadim makes the call on\n> when to start beta.\n\n Now I'm placing another LOCK request (waiting for Vadim).\n\n The problem with constraints in ExecRelCheck() I've just\n found not only affects COPY. It also occurs with\n INSERT...SELECT or big UPDATE's. Constraints with many\n affected tuples eat up backend memory quickly until out of\n swap space.\n\n This fix must go into CURRENT and REL6_4 before we release\n v6.5 or v6.4.3 either. There are actually reports from users\n not able to reload dumps made with v6.4.2.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Tue, 2 Feb 1999 17:55:47 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> > Good question. I am done with temp tables, so Vadim makes the call on\n> > when to start beta.\n> \n> Now I'm placing another LOCK request (waiting for Vadim).\n> \n> The problem with constraints in ExecRelCheck() I've just\n> found not only affects COPY. It also occurs with\n> INSERT...SELECT or big UPDATE's. Constraints with many\n> affected tuples eat up backend memory quickly until out of\n> swap space.\n> \n> This fix must go into CURRENT and REL6_4 before we release\n> v6.5 or v6.4.3 either. There are actually reports from users\n> not able to reload dumps made with v6.4.2.\n\nYou can keep applying fixes like this until near final release date. I\nhave patches in my mailbox to apply today too.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 2 Feb 1999 12:05:03 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Do you want this applied?\n\n\n[Charset iso-8859-1 unsupported, filtering to ASCII...]\n> Goran Thyni wrote:\n> > 3. text_lt('G','G\\0xFF') this is not correct!\n> > \n> > Case 3 work not work with strcoll(), in varstr_cmp().\n> > If I change it to strcoll() to strncmp() it works as expected,\n> > but it probably breaks sorting etc big time.\n> \n> Here is a hack which solves the problem without breaking sorting AFAIK.\n> I have read a lot of code, but has not found any cleaner way to fix\n> this.\n> A cleaner solution would be to find the highest char in locale charset\n> but I found no portable way to do that, any pointers appriciated.\n> \n> It is not a beauty but it works. \n> Please apply to -current since it fixes the bug, \n> unless someone has a better suggestion.\n> \n> \tbest regards,\n> -- \n> -----------------\n> G_ran Thyni\n> This is Penguin Country. On a quiet night you can hear Windows NT\n> reboot!\n\n> diff -cr cvs/pgsql/src/backend/utils/adt/varlena.c cvswork/pgsql/src/backend/utils/adt/varlena.c\n> *** cvs/pgsql/src/backend/utils/adt/varlena.c\tMon Dec 14 07:01:37 1998\n> --- cvswork/pgsql/src/backend/utils/adt/varlena.c\tSun Jan 31 16:32:52 1999\n> ***************\n> *** 496,528 ****\n> varstr_cmp(char *arg1, int len1, char *arg2, int len2)\n> {\n> \tint\t\t\tresult;\n> ! \tchar\t *a1p,\n> ! \t\t\t *a2p;\n> ! \n> #ifdef USE_LOCALE\n> ! \ta1p = (unsigned char *) palloc(len1 + 1);\n> ! \ta2p = (unsigned char *) palloc(len2 + 1);\n> ! \n> ! \tmemcpy(a1p, arg1, len1);\n> ! \t*(a1p + len1) = '\\0';\n> ! \tmemcpy(a2p, arg2, len2);\n> ! \t*(a2p + len2) = '\\0';\n> ! \n> ! \tresult = strcoll(a1p, a2p);\n> ! \n> ! \tpfree(a1p);\n> ! \tpfree(a2p);\n> ! \n> ! #else\n> ! \n> ! \ta1p = arg1;\n> ! \ta2p = arg2;\n> ! \n> ! \tresult = strncmp(a1p, a2p, Min(len1, len2));\n> ! \tif ((result == 0) && (len1 != len2))\n> ! \t\tresult = (len1 < len2) ? -1 : 1;\n> #endif\n> ! \n> \treturn result;\n> }\t/* varstr_cmp() */\n> \n> --- 496,524 ----\n> varstr_cmp(char *arg1, int len1, char *arg2, int len2)\n> {\n> \tint\t\t\tresult;\n> ! \tchar\t *a1p, *a2p;\n> #ifdef USE_LOCALE\n> ! \tif ((int)arg2[len2 - 1] != -1)\n> ! \t {\n> ! \t\ta1p = (unsigned char *) palloc(len1 + 1);\n> ! \t\ta2p = (unsigned char *) palloc(len2 + 1);\n> ! \t\tmemcpy(a1p, arg1, len1);\n> ! \t\t*(a1p + len1) = '\\0';\n> ! \t\tmemcpy(a2p, arg2, len2);\n> ! \t\t*(a2p + len2) = '\\0';\n> ! \t\tresult = strcoll(a1p, a2p);\n> ! \t\tpfree(a1p);\n> ! \t\tpfree(a2p);\n> ! \t }\n> ! \telse\n> #endif\n> ! \t {\n> ! \t\ta1p = arg1;\n> ! \t\ta2p = arg2;\n> ! \t\tresult = strncmp(a1p, a2p, Min(len1, len2));\n> ! \t\tif ((result == 0) && (len1 != len2))\n> ! \t\tresult = (len1 < len2) ? -1 : 1;\n> ! \t }\n> \treturn result;\n> }\t/* varstr_cmp() */\n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 2 Feb 1999 14:20:02 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Tom, I assume this has all been corrected, right?\n\n\n\n\n> I wrote:\n> > As for where the time is going: \"top\" showed that the time for the\n> > INSERT ops was almost all going into backend CPU time. My guess is\n> > that most of the cost is in parsing and planning the INSERT\n> > statements.\n> \n> Having an idle hour this evening, I thought it'd be interesting to build\n> a backend with profiling enabled, so as to confirm or deny the above\n> guess. It seems that indeed a lot of time is being wasted, but where\n> it's being wasted might surprise you!\n> \n> I profiled a backend built from this morning's current CVS sources,\n> being driven by a single psql run reading a series of exactly 5000\n> INSERT commands. The first few are:\n> \n> INSERT INTO \"fehistory_1\" values (64,14,252,'D','-','-','s',1,4027,NULL,'S',1,4088,NULL,32,'Thu Jan 07 15:15:00 1999 EST','Fri Jan 08 08:30:38 1999 EST',4027,32,NULL,NULL,46,NULL);\n> INSERT INTO \"fehistory_1\" values (65,14,252,'P','-','-','S',1,4086,NULL,NULL,NULL,NULL,NULL,32,'Fri Jan 08 16:00:00 1999 EST',NULL,NULL,NULL,NULL,NULL,45,NULL);\n> INSERT INTO \"fehistory_1\" values (66,14,135,'P','-','-','S',1,13619,NULL,'s',1,12967,NULL,100,'Fri Jan 08 02:00:00 1999 EST',NULL,NULL,NULL,NULL,NULL,44,NULL);\n> \n> and it goes on for a long while in that vein :-). Total runtime was\n> 278.09 CPU seconds in the backend, 307 seconds elapsed realtime. The\n> table being inserted into has no indexes and was initially empty.\n> \n> The profiler is GNU gprof, which provides both flat and structured\n> profile data. The top few entries in the flat profile are:\n> \n> % cumulative self self total \n> time seconds seconds calls ms/call ms/call name \n> 34.96 97.21 97.21 _mcount\n> 7.18 117.19 19.98 8047816 0.00 0.00 tree_cut\n> 6.56 135.43 18.24 962187 0.02 0.02 recv\n> 3.55 145.31 9.88 8234278 0.00 0.00 tree_insert\n> 3.49 155.02 9.71 4292947 0.00 0.01 malloc\n> 3.31 164.22 9.20 1965056 0.00 0.01 SearchSysCache\n> 3.24 173.23 9.01 4291775 0.00 0.01 free\n> 2.44 180.01 6.78 6171667 0.00 0.00 tree_concatenate\n> 2.17 186.04 6.03 8234069 0.00 0.00 tree_delete\n> 1.07 189.02 2.98 10250 0.29 0.29 write\n> 1.05 191.94 2.92 _psort_cmp\n> 0.95 194.57 2.63 2983301 0.00 0.00 newNode\n> 0.92 197.14 2.57 4251762 0.00 0.00 OrderedElemPop\n> 0.88 199.60 2.46 2746172 0.00 0.00 comphash\n> 0.82 201.88 2.28 4252495 0.00 0.01 AllocSetAlloc\n> 0.77 204.02 2.14 1965056 0.00 0.00 SearchSysCacheTuple\n> 0.73 206.05 2.03 265000 0.01 0.01 yylex\n> 0.69 207.96 1.91 $$remU\n> 0.68 209.86 1.90 4252500 0.00 0.00 OrderedElemPushHead\n> 0.62 211.58 1.72 4247473 0.00 0.00 palloc\n> 0.60 213.26 1.68 25004 0.07 0.07 send\n> 0.56 214.81 1.55 1965056 0.00 0.00 CatalogCacheComputeHashIndex\n> 0.55 216.33 1.52 5000 0.30 1.07 yyparse\n> 0.54 217.82 1.49 4251762 0.00 0.01 AllocSetFree\n> 0.53 219.29 1.47 4252494 0.00 0.00 MemoryContextAlloc\n> 0.51 220.71 1.42 4252500 0.00 0.00 OrderedElemPushInto\n> 0.49 222.07 1.36 3430844 0.00 0.00 strncmp\n> 0.48 223.41 1.34 4205327 0.00 0.00 OrderedSetGetHead\n> 0.47 224.73 1.32 elog\n> 0.45 225.99 1.26 1964976 0.00 0.00 DLRemove\n> 0.43 227.19 1.20 2510857 0.00 0.00 strcmp\n> 0.43 228.38 1.19 4200327 0.00 0.00 OrderedElemGetBase\n> 0.42 229.56 1.18 4245740 0.00 0.01 PortalHeapMemoryAlloc\n> 0.38 230.62 1.06 4252500 0.00 0.00 OrderedElemPush\n> 0.37 231.66 1.04 4205327 0.00 0.00 AllocSetGetFirst\n> 0.36 232.66 1.00 1965145 0.00 0.00 DLAddHead\n> \n> _mcount is part of the profiling apparatus and can be ignored (although\n> the fact that it's so high up implies that we're executing an awful lot\n> of short routine calls, because mcount runs once per entry and exit of\n> profilable routines). tree_cut, tree_insert, tree_concatenate, and\n> tree_delete are subroutines of malloc/free. Altogether, malloc/free\n> and friends accounted for 61.39 seconds out of the 180.88 non-overhead\n> CPU seconds in this run.\n> \n> In other words, we're spending a third of our time mallocing and freeing\n> memory. A tad high, what?\n> \n> Actually, it's worse than that, because AllocSetAlloc,\n> PortalHeapMemoryAlloc, AllocSetFree, and all of the OrderedElemXXX\n> routines represent our own bookkeeping layer atop malloc/free.\n> That's another 18.66 seconds spent in these top routines, which means\n> that we are real close to expending half the backend's runtime on\n> memory bookkeeping. This needs work.\n> \n> The other thing that jumps out here is the unreasonably high position of\n> recv(), which is called 962187 times. The script being read by psql was\n> only 957186 characters. Evidently we're invoking a kernel recv() call\n> once per character read from the frontend. I suspect this is an\n> inefficiency introduced by Magnus Hagander's recent rewrite of backend\n> libpq (see, I told you there was a reason for using stdio ;-)). We're\n> gonna have to do something about that, though it's not as critical as\n> the memory-allocation issue. It also appears that send() is now being\n> invoked multiple times per backend reply, which is going to create\n> inefficiencies outside the backend (ie, multiple packets per reply).\n> On a test case with a lot of SELECTs that would show up more than it\n> does here.\n> \n> \n> Moving on to the dynamic profile, the most interesting items are:\n> \n> index % time self children called name\n> -----------------------------------------------\n> 0.05 118.06 1/1 DoBackend [7]\n> [8] 65.3 0.05 118.06 1 PostgresMain [8]\n> 0.04 57.64 5000/5000 pg_exec_query [9]\n> 0.01 39.83 5000/5000 CommitTransactionCommand [11]\n> 0.02 18.86 5001/5001 ReadCommand [27]\n> 0.01 1.18 5000/5000 StartTransactionCommand [118]\n> 0.01 0.36 5001/5001 ReadyForQuery [190]\n> 0.04 0.00 5001/5001 EnableNotifyInterrupt [385]\n> 0.04 0.00 5000/5000 strspn [388]\n> 0.00 0.01 1/1 InitPostgres [495]\n> 0.00 0.00 5000/548573 strlen [172]\n> 0.00 0.00 5001/93349 memset [366]\n> 0.00 0.00 1/1 read_pg_options [601]\n> 0.00 0.00 2/2 pq_putint [637]\n> 0.00 0.00 1/15002 pq_putnchar [123]\n> 0.00 0.00 2/10013 getenv [241]\n> 0.00 0.00 1/1 SetPgUserName [683]\n> 0.00 0.00 1/1 pq_init [687]\n> 0.00 0.00 1/1 proc_exit [704]\n> 0.00 0.00 1/1 pq_close [721]\n> 0.00 0.00 1/5002 getpgusername [521]\n> 0.00 0.00 1/2 FindExec [722]\n> 0.00 0.00 5/8 getopt [728]\n> 0.00 0.00 5001/5001 DisableNotifyInterrupt [761]\n> 0.00 0.00 9/34 pqsignal [806]\n> 0.00 0.00 2/5 atoi [837]\n> 0.00 0.00 1/2 SetProcessingMode [868]\n> 0.00 0.00 1/10002 TransactionFlushEnabled [749]\n> 0.00 0.00 1/1 sigsetjmp [934]\n> \n> (\"self\" is the amount of time (CPU sec) spent directly in the named function,\n> while \"children\" is the amount of time spent in its callees, pro-rated\n> by the number of calls. For example, strlen was actually measured to\n> use 0.47 sec in the whole run, but since PostgresMain called it less\n> than 1% of all the times it was called, PostgresMain gets blamed for\n> less than 0.01 sec of that total.)\n> \n> The interesting thing here is the relative times indicated for\n> pg_exec_query, CommitTransactionCommand, and ReadCommand. We already\n> found out why ReadCommand is so slow (recv() per character) ... but why\n> is CommitTransactionCommand so high, when I already discovered that\n> doing only one commit doesn't help much? Patience, we'll get there.\n> \n> -----------------------------------------------\n> 0.03 57.61 5000/5000 pg_exec_query [9]\n> [10] 31.9 0.03 57.61 5000 pg_exec_query_dest [10]\n> 0.09 39.12 5000/5000 pg_parse_and_plan [13]\n> 0.00 17.37 5000/5000 ProcessQuery [36]\n> 0.02 0.60 5000/5000 SetQuerySnapshot [159]\n> 0.01 0.34 5000/5000 CommandCounterIncrement [196]\n> 0.02 0.05 10000/4291775 free [22]\n> -----------------------------------------------\n> 0.09 39.12 5000/5000 pg_exec_query_dest [10]\n> [13] 21.7 0.09 39.12 5000 pg_parse_and_plan [13]\n> 0.02 19.28 5000/5000 parser [26]\n> 0.03 17.63 5000/5000 planner [34]\n> 0.00 1.98 5000/5000 QueryRewrite [90]\n> 0.02 0.05 10000/4292947 malloc [21]\n> 0.02 0.05 10000/4291775 free [22]\n> 0.00 0.02 5000/497779 lappend [91]\n> 0.01 0.00 5000/5000 IsAbortedTransactionBlockState [529]\n> 0.00 0.00 5000/72779 length [342]\n> \n> In other words, parsing, planning, and executing an INSERT each take\n> about the same amount of time.\n> \n> -----------------------------------------------\n> 0.01 39.83 5000/5000 PostgresMain [8]\n> [11] 22.0 0.01 39.83 5000 CommitTransactionCommand [11]\n> 0.02 39.81 5000/5000 CommitTransaction [12]\n> -----------------------------------------------\n> 0.02 39.81 5000/5000 CommitTransactionCommand [11]\n> [12] 22.0 0.02 39.81 5000 CommitTransaction [12]\n> 0.00 38.12 5000/5000 AtCommit_Memory [16]\n> 0.01 1.10 5000/5000 RecordTransactionCommit [122]\n> 0.00 0.30 5000/5000 AtCommit_Locks [211]\n> 0.02 0.10 5000/5000 AtEOXact_portals [286]\n> 0.03 0.07 5000/5000 DestroyTempRels [305]\n> 0.03 0.00 5000/5000 _lo_commit [425]\n> 0.01 0.00 5000/5000 AtCommit_Notify [522]\n> 0.01 0.00 5000/5000 CloseSequences [523]\n> 0.01 0.00 5000/5000 RelationPurgeLocalRelation [531]\n> 0.00 0.00 5000/10000 AtCommit_Cache [750]\n> -----------------------------------------------\n> 0.00 38.12 5000/5000 CommitTransaction [12]\n> [16] 21.1 0.00 38.12 5000 AtCommit_Memory [16]\n> 0.01 38.11 5000/5000 EndPortalAllocMode [17]\n> 0.00 0.00 5000/10552 MemoryContextSwitchTo [747]\n> -----------------------------------------------\n> 0.01 38.11 5000/5000 AtCommit_Memory [16]\n> [17] 21.1 0.01 38.11 5000 EndPortalAllocMode [17]\n> 0.54 37.55 5000/5000 AllocSetReset [18]\n> 0.01 0.00 5000/10000 PortalHeapMemoryGetVariableMemory [449]\n> 0.01 0.00 5000/5000 FixedStackPop [528]\n> 0.00 0.00 5000/51434 MemoryContextFree [367]\n> -----------------------------------------------\n> 0.54 37.55 5000/5000 EndPortalAllocMode [17]\n> [18] 21.1 0.54 37.55 5000 AllocSetReset [18]\n> 1.47 32.51 4200327/4251762 AllocSetFree [20]\n> 1.04 2.53 4205327/4205327 AllocSetGetFirst [64]\n> \n> In other words, essentially *all* of the CPU time spent in\n> CommitTransaction is spent freeing memory. That's probably why\n> ganging the transactions doesn't help --- it's the same number of\n> memory blocks getting allocated and freed.\n> \n> 0.02 19.28 5000/5000 pg_parse_and_plan [13]\n> [26] 10.7 0.02 19.28 5000 parser [26]\n> 0.02 13.88 5000/5000 parse_analyze [41]\n> 1.52 3.81 5000/5000 yyparse [54]\n> 0.01 0.01 5000/5000 init_io [474]\n> 0.00 0.01 5000/420496 pstrdup [124]\n> 0.01 0.00 5000/5000 parser_init [533]\n> \n> Thomas might know why parse_analyze is taking so much time compared to\n> the rest of the parsing machinery...\n> \n> \n> I won't bother the list with the rest of the profile, although I'll be\n> happy to send it to anyone who wants to see all of it. Our immediate\n> performance problems seem pretty clear, however: time to rethink memory\n> management.\n> \n> \t\t\tregards, tom lane\n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 15 Mar 1999 09:28:15 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Can someone remind me where we left this memory context stuff?\n\n\n\n> > I've noticed that everyone else contributing to this thread has been\n> > thinking in terms of inventing multiple allocation functions with\n> > different names, ie a routine might have to call \"palloc\" or\n> > \"fast_palloc\" or \"tmp_palloc\" or whatever depending on what behavior it\n> > wants. I really think we are better off to stick to the structure that\n> > we already have (cf. include/nodes/memnodes.h and include/utils/mcxt.h),\n> > in which there's one set of interface routines that take a \"context\"\n> > parameter, and the context determines the semantics of allocation.\n> > (The interface routines probably ought to be macros, not out-of-line\n> > code, but that's a minor optimization.) This will make it easier to\n> > handle cases where one routine has to tell another one how to allocate\n> > its result space: you pass a MemoryContext parameter.\n> \n> I agree, inline is good. I can imagine something like makeNode is going\n> to be tough.\n> \n> I am going to give the parser as an example. It palloc's all sorts of\n> things for rangetables, strings, structures, lists, but eventually just\n> exports a huge Query structure.\n> \n> If we could somehow just grab copy Query structure and free all the\n> other parser palloc allocations, that would be great.\n> \n> Same thing with the optimizer. It makes all sorts of lists, but\n> eventuall just exports Plan's. All the other stuff can be pfree'ed.\n> \n> It is almost as though we could copy the Query structure to another\n> memory context, and just fee all palloc'ed memory.\n> \n> What we could try is to keep the various memory contexts like cache\n> context unchanged, but allow pallocs in the standard memory context to\n> be allocated from a pool and free'ed automatically. Our standard memory\n> context is overused and doesn't need the pfree overhead. We could make\n> palloc a macro that tested for the current context, and did standard\n> pallocs from a free memory pool.\n> \n> -- \n> Bruce Momjian | http://www.op.net/~candle\n> [email protected] | (610) 853-3000\n> + If your life is a hard drive, | 830 Blythe Avenue\n> + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 15 Mar 1999 09:43:46 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "\nAs I remember, this was fixed already, without your patch, right?\n\n\n[Charset iso-8859-1 unsupported, filtering to ASCII...]\n> Goran Thyni wrote:\n> > 3. text_lt('G','G\\0xFF') this is not correct!\n> > \n> > Case 3 work not work with strcoll(), in varstr_cmp().\n> > If I change it to strcoll() to strncmp() it works as expected,\n> > but it probably breaks sorting etc big time.\n> \n> Here is a hack which solves the problem without breaking sorting AFAIK.\n> I have read a lot of code, but has not found any cleaner way to fix\n> this.\n> A cleaner solution would be to find the highest char in locale charset\n> but I found no portable way to do that, any pointers appriciated.\n> \n> It is not a beauty but it works. \n> Please apply to -current since it fixes the bug, \n> unless someone has a better suggestion.\n> \n> \tbest regards,\n> -- \n> -----------------\n> G_ran Thyni\n> This is Penguin Country. On a quiet night you can hear Windows NT\n> reboot!\n\n> diff -cr cvs/pgsql/src/backend/utils/adt/varlena.c cvswork/pgsql/src/backend/utils/adt/varlena.c\n> *** cvs/pgsql/src/backend/utils/adt/varlena.c\tMon Dec 14 07:01:37 1998\n> --- cvswork/pgsql/src/backend/utils/adt/varlena.c\tSun Jan 31 16:32:52 1999\n> ***************\n> *** 496,528 ****\n> varstr_cmp(char *arg1, int len1, char *arg2, int len2)\n> {\n> \tint\t\t\tresult;\n> ! \tchar\t *a1p,\n> ! \t\t\t *a2p;\n> ! \n> #ifdef USE_LOCALE\n> ! \ta1p = (unsigned char *) palloc(len1 + 1);\n> ! \ta2p = (unsigned char *) palloc(len2 + 1);\n> ! \n> ! \tmemcpy(a1p, arg1, len1);\n> ! \t*(a1p + len1) = '\\0';\n> ! \tmemcpy(a2p, arg2, len2);\n> ! \t*(a2p + len2) = '\\0';\n> ! \n> ! \tresult = strcoll(a1p, a2p);\n> ! \n> ! \tpfree(a1p);\n> ! \tpfree(a2p);\n> ! \n> ! #else\n> ! \n> ! \ta1p = arg1;\n> ! \ta2p = arg2;\n> ! \n> ! \tresult = strncmp(a1p, a2p, Min(len1, len2));\n> ! \tif ((result == 0) && (len1 != len2))\n> ! \t\tresult = (len1 < len2) ? -1 : 1;\n> #endif\n> ! \n> \treturn result;\n> }\t/* varstr_cmp() */\n> \n> --- 496,524 ----\n> varstr_cmp(char *arg1, int len1, char *arg2, int len2)\n> {\n> \tint\t\t\tresult;\n> ! \tchar\t *a1p, *a2p;\n> #ifdef USE_LOCALE\n> ! \tif ((int)arg2[len2 - 1] != -1)\n> ! \t {\n> ! \t\ta1p = (unsigned char *) palloc(len1 + 1);\n> ! \t\ta2p = (unsigned char *) palloc(len2 + 1);\n> ! \t\tmemcpy(a1p, arg1, len1);\n> ! \t\t*(a1p + len1) = '\\0';\n> ! \t\tmemcpy(a2p, arg2, len2);\n> ! \t\t*(a2p + len2) = '\\0';\n> ! \t\tresult = strcoll(a1p, a2p);\n> ! \t\tpfree(a1p);\n> ! \t\tpfree(a2p);\n> ! \t }\n> ! \telse\n> #endif\n> ! \t {\n> ! \t\ta1p = arg1;\n> ! \t\ta2p = arg2;\n> ! \t\tresult = strncmp(a1p, a2p, Min(len1, len2));\n> ! \t\tif ((result == 0) && (len1 != len2))\n> ! \t\tresult = (len1 < len2) ? -1 : 1;\n> ! \t }\n> \treturn result;\n> }\t/* varstr_cmp() */\n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 15 Mar 1999 09:44:40 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Patch (was: tough locale bug)"
},
{
"msg_contents": "Bruce Momjian <[email protected]> writes:\n> Tom, I assume this has all been corrected, right?\n\nThe recv() problem is fixed, and Jan took a swipe at reducing the cost\nof palloc/pfree. I have not done a profile since Jan's changes to see\nhow successful he was.\n\nI would like to revisit memory management at some point, but it won't\nhappen for 6.5.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 15 Mar 1999 10:27:52 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "Bruce Momjian <[email protected]> writes:\n> Can someone remind me where we left this memory context stuff?\n\nJan put in some changes to allow palloc()'d space to be freed more\ncheaply. The more sweeping rewrite that I was suggesting is not done,\nbut perhaps we can reconsider it for 6.6 or sometime later. I don't\nthink we need to worry about it for 6.5.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 15 Mar 1999 10:31:04 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": ">\n> Can someone remind me where we left this memory context stuff?\n>\n\n I've changed AllocSetAlloc() and friends to manage small\n allocations in bigger blocks together where the sizes are\n round up to powers of two and free's chunks in the blocks are\n reused in place. The entire blocks get only free'd when they\n get completely free.\n\n Then I've changed palloc() etc. into macros (which started\n the discussion on mcxt.h - sorry).\n\n Overall speedup from the changes (for regression) is\n something about 10-15%.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Mon, 15 Mar 1999 22:09:25 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "> >\n> > Can someone remind me where we left this memory context stuff?\n> >\n> \n> I've changed AllocSetAlloc() and friends to manage small\n> allocations in bigger blocks together where the sizes are\n> round up to powers of two and free's chunks in the blocks are\n> reused in place. The entire blocks get only free'd when they\n> get completely free.\n> \n> Then I've changed palloc() etc. into macros (which started\n> the discussion on mcxt.h - sorry).\n> \n> Overall speedup from the changes (for regression) is\n> something about 10-15%.\n> \n\nGreat.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 15 Mar 1999 16:46:48 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "\n\nThis from January 24, 1999. Is it still relivant, and something that\nshould be on the TODO list?\n\n\n> Tom Lane wrote:\n> > \n> > Having an idle hour this evening, I thought it'd be interesting to build\n> > a backend with profiling enabled, so as to confirm or deny the above\n> > guess. It seems that indeed a lot of time is being wasted, but where\n> > it's being wasted might surprise you!\n> ...\n> > In other words, we're spending a third of our time mallocing and freeing\n> > memory. A tad high, what?\n> > \n> > Actually, it's worse than that, because AllocSetAlloc,\n> > PortalHeapMemoryAlloc, AllocSetFree, and all of the OrderedElemXXX\n> > routines represent our own bookkeeping layer atop malloc/free.\n> > That's another 18.66 seconds spent in these top routines, which means\n> > that we are real close to expending half the backend's runtime on\n> > memory bookkeeping. This needs work.\n> >\n> \n> Yes, it's suprizing!\n> \n> I added some debug code to palloc/pfree and it shows that for\n> INSERT:\n> \n> 1. 80% of allocations are made for <= 32 bytes.\n> 2. pfree is used for 25% of them only (others are freed\n> after statement/transaction is done).\n> \n> Note that our mmgr adds 16 bytes to each allocation\n> (+ some bytes in malloc) - a great overhead, yes?\n> \n> I added code to allocate a few big (16K-64K) blocks\n> of memory for these small allocations to speed up \n> palloc by skiping AllocSetAlloc/malloc. New code\n> don't free allocated memory (to make bookkeeping fast)\n> but keeping in mind 2. above and memory overhead\n> it seems as appropriate thing to do. These code also\n> speed up freeing when statement/transaction is done,\n> because of only a few blocks have to be freed now.\n> \n> I did 5000 INSERTS (into tables with 3 ints and 33 ints)\n> with BEGIN/END, -F and -B 512 (I run postgres directly,\n> without postmaster). User times:\n> \n> \t\t\told\t\tnew\n> -----------------------------------------\n> table with 3 ints\t 9.7 sec\t 7.6 sec\n> table with 33 ints\t59.5 sec\t39.9 sec\n> \n> So, new code 20%-30% faster. Process sizes are the same.\n> \n> Tom, could you run new code under profiling?\n> \n> There are still some things to do:\n> \n> 1. SELECT/UPDATE/DELETE often palloc/pfree tuples\n> (sizes are > 32 bytes), but pfree now requires \n> additional lookup to see is memory allocated by \n> AllocSetAlloc or new code. We can avoid this.\n> \n> 2. Index scans palloc/pfree IndexResult for each\n> tuple returned by index. This one was annoying me\n> for long time. IndexResult should be part of\n> IndexScan structure...\n> \n> 3. psort uses leftist structure (16 bytes) when\n> disk is used for sorting. Obviously, big block\n> allocation should be used by lselect code itself.\n> \n> 4. Actually, new mode shouldn't be used by Executor\n> in some cases.\n> \n> I'll address this in a few days...\n> \n> BTW, look at memutils.h: new code is like \"tunable\"\n> mode described there.\n> \n> > \n> > In other words, essentially *all* of the CPU time spent in\n> > CommitTransaction is spent freeing memory. That's probably why\n> > ganging the transactions doesn't help --- it's the same number of\n> > memory blocks getting allocated and freed.\n> \n> It shows that we should get rid of system malloc/free and do\n> all things in mmgr itself - this would allow us much faster\n> free memory contexts at statement/transaction end.\n> \n> Vadim\n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 6 Jul 1999 20:43:27 -0400 (EDT)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "\nI have added this to the TODO list:\n\n* improve dynamic memory allocation by introducing tuple-context memory \n allocation\n* add pooled memory allocation where allocations are freed only as a\n group \n\n\n\n> We could use 4 methods for dynamic allocation:\n> \n> 1)\tmalloc/free - for persistent storage allocation\n> \n> 2)\tpalloc/pfree - for storage belonging to some context and\n> \twhich we can keep track of and free explicitly\n> \n> 3)\tfast_palloc - for storage which impossible, too difficult or too\n> \texpensive to keep track of. This storage should be allocated with\n> \tfast and simple inline code from bigger chunks allocated with palloc.\n> \tThis storage would never freed explicitly, so that code could be\n> \tsimple and fast, but the big chunks would be freed automatically at\n> \tthe end of the transaction.\n> \n> 4)\tfast_talloc - we could introduce a `tuple' context handled like\n> \tfast_palloc for storage used only while processing one tuple.\n> \tThis storage could be fast allocated from few big chunks allocated\n> \twith palloc and freed explicitly after the tuple has been processed.\n> \tThis could avoid the overhead of many malloc/palloc while reducing\n> \tthe overall memory usage for transaction which process many rows.\n> \tThe total cost per tuple could be one palloc and one pfree.\n> \tWe could also simply reuse the chunks for every tuple and pfree them \n> \tonly at the end of the transaction. This would cost one palloc/pfree\n> \tper transaction.\n> \n> This would require revising the code and changing palloc/pfree with the new\n> functions where appropriate, but this could be done gradually because the\n> old palloc/pfree are always safe.\n> \n> -- \n> Massimo Dal Zotto\n> \n> +----------------------------------------------------------------------+\n> | Massimo Dal Zotto email: [email protected] |\n> | Via Marconi, 141 phone: ++39-0461534251 |\n> | 38057 Pergine Valsugana (TN) www: http://www.cs.unitn.it/~dz/ |\n> | Italy pgp: finger [email protected] |\n> +----------------------------------------------------------------------+\n> \n> \n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 6 Jul 1999 21:25:48 -0400 (EDT)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Bruce Momjian <[email protected]> writes:\n> This from January 24, 1999. Is it still relivant, and something that\n> should be on the TODO list?\n\nJan's rewrite of alloc sets has probably taken care of the speed issue.\nOf course we still need to look at how memory allocation is being\n*used*, in order to fix the within-statement memory leak problems.\nBut that's not what I was complaining about here.\n\n\t\t\tregards, tom lane\n\n>> Tom Lane wrote:\n>>>> \n>>>> Having an idle hour this evening, I thought it'd be interesting to build\n>>>> a backend with profiling enabled, so as to confirm or deny the above\n>>>> guess. It seems that indeed a lot of time is being wasted, but where\n>>>> it's being wasted might surprise you!\n>> ...\n>>>> In other words, we're spending a third of our time mallocing and freeing\n>>>> memory. A tad high, what?\n>>>> \n>>>> Actually, it's worse than that, because AllocSetAlloc,\n>>>> PortalHeapMemoryAlloc, AllocSetFree, and all of the OrderedElemXXX\n>>>> routines represent our own bookkeeping layer atop malloc/free.\n>>>> That's another 18.66 seconds spent in these top routines, which means\n>>>> that we are real close to expending half the backend's runtime on\n>>>> memory bookkeeping. This needs work.\n>>>> \n>> \n>> Yes, it's suprizing!\n>> \n>> I added some debug code to palloc/pfree and it shows that for\n>> INSERT:\n>> \n>> 1. 80% of allocations are made for <= 32 bytes.\n>> 2. pfree is used for 25% of them only (others are freed\n>> after statement/transaction is done).\n>> \n>> Note that our mmgr adds 16 bytes to each allocation\n>> (+ some bytes in malloc) - a great overhead, yes?\n>> \n>> I added code to allocate a few big (16K-64K) blocks\n>> of memory for these small allocations to speed up \n>> palloc by skiping AllocSetAlloc/malloc. New code\n>> don't free allocated memory (to make bookkeeping fast)\n>> but keeping in mind 2. above and memory overhead\n>> it seems as appropriate thing to do. These code also\n>> speed up freeing when statement/transaction is done,\n>> because of only a few blocks have to be freed now.\n>> \n>> I did 5000 INSERTS (into tables with 3 ints and 33 ints)\n>> with BEGIN/END, -F and -B 512 (I run postgres directly,\n>> without postmaster). User times:\n>> \n>> old\t\tnew\n>> -----------------------------------------\n>> table with 3 ints\t 9.7 sec\t 7.6 sec\n>> table with 33 ints\t59.5 sec\t39.9 sec\n>> \n>> So, new code 20%-30% faster. Process sizes are the same.\n>> \n>> Tom, could you run new code under profiling?\n>> \n>> There are still some things to do:\n>> \n>> 1. SELECT/UPDATE/DELETE often palloc/pfree tuples\n>> (sizes are > 32 bytes), but pfree now requires \n>> additional lookup to see is memory allocated by \n>> AllocSetAlloc or new code. We can avoid this.\n>> \n>> 2. Index scans palloc/pfree IndexResult for each\n>> tuple returned by index. This one was annoying me\n>> for long time. IndexResult should be part of\n>> IndexScan structure...\n>> \n>> 3. psort uses leftist structure (16 bytes) when\n>> disk is used for sorting. Obviously, big block\n>> allocation should be used by lselect code itself.\n>> \n>> 4. Actually, new mode shouldn't be used by Executor\n>> in some cases.\n>> \n>> I'll address this in a few days...\n>> \n>> BTW, look at memutils.h: new code is like \"tunable\"\n>> mode described there.\n>> \n>>>> \n>>>> In other words, essentially *all* of the CPU time spent in\n>>>> CommitTransaction is spent freeing memory. That's probably why\n>>>> ganging the transactions doesn't help --- it's the same number of\n>>>> memory blocks getting allocated and freed.\n>> \n>> It shows that we should get rid of system malloc/free and do\n>> all things in mmgr itself - this would allow us much faster\n>> free memory contexts at statement/transaction end.\n>> \n>> Vadim\n>> \n>> \n\n\n> -- \n> Bruce Momjian | http://www.op.net/~candle\n> [email protected] | (610) 853-3000\n> + If your life is a hard drive, | 830 Blythe Avenue\n> + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 07 Jul 1999 09:29:28 -0400",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
}
] |
[
{
"msg_contents": "Bruce,\n\nThis was fixed about 1 month ago by Vadim,\n\n\tKeith.\n\t\nBruce Momjian <[email protected]>\n> \n> > Hi Hackers,\n> > \n> > With the following query from the opr_sanity regression test I'm\n> > getting a backend crash.\n> > \n> > SELECT p1.oid, p1.oprcode, p2.oid, p2.oprcode\n> > FROM pg_operator AS p1, pg_operator AS p2\n> > WHERE p1.oprlsortop = p2.oid AND\n> > (p1.oprname != '=' OR\n> > p1.oprkind != 'b' OR p2.oprkind != 'b' OR\n> > p1.oprleft != p2.oprleft OR\n> > p1.oprright != p2.oprright OR\n> > p1.oprresult != 16 OR\n> > p2.oprresult != 16 OR\n> > p1.oprrsortop = 0);\n> \n> I tried the query in the current development tree, and it returned zero\n> rows and did not crash.\n> \n\n",
"msg_date": "Sun, 17 Jan 1999 12:09:28 +0000 (GMT)",
"msg_from": "Keith Parks <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Backend crash in opr_sanity regression test."
}
] |
[
{
"msg_contents": "Bruce,\n\n\tThis problem with VIEWs was fixed by Jan WiecK on or\n\taround 18th December.\n\t\nKeith.\n\nBruce Momjian <[email protected]>\n\n> \n> > Hi,\n> > \n> > I think I need some help here\n> > \n> > Taking part of the pg_views rule shows it thinks the views\n> > are NOT views.\n> > \n> > postgres=> SELECT relname AS viewname, pg_get_userbyid(relowner) AS \nviewowner, \n> > pg_get_viewdef(relname) AS definition FROM pg_class WHERE relhasrules;\n> > viewname |viewowner|definition\n> > ----------+---------+----------\n> > pg_user |postgres |Not a view\n> > pg_rules |postgres |Not a view\n> > pg_views |postgres |Not a view\n> > pg_tables |postgres |Not a view\n> > pg_indexes|postgres |Not a view\n> > (5 rows)\n> \n> Works in the current development tree.\n> \n> \n\n\n",
"msg_date": "Sun, 17 Jan 1999 12:12:36 +0000 (GMT)",
"msg_from": "Keith Parks <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] CURRENT: crash in select_view regression test..."
}
] |
[
{
"msg_contents": " really hesitate to make this criticism given all the work people are\ndoing on the system and all the great features and fixes that are\nbeing added but could we please make it a hard and fast rule that\nnothing gets added into the tree until the person adding it has at\nleast compiled the source, if not a full regression test. I have\nfinally jumped in and tried to add full primary key support but I\nkeep stumbling over simple syntax errors in files I am not working\non but that keep me from fully testing my own changes. Here's what\nI get compiling gram.y[c].\n\ngcc -I../../include -I../../backend -I/usr/local/include -O2 -pipe -Wall -Wmissing-prototypes -I.. -Wno-error -c gram.c -o gram.o\n/usr/local/share/bison.simple: In function `yyparse':\n/usr/local/share/bison.simple:327: warning: implicit declaration of function `yyerror'\n/usr/local/share/bison.simple:387: warning: implicit declaration of function `yylex'\ngram.y:2797: `forUpdate' undeclared (first use in this function)\ngram.y:2797: (Each undeclared identifier is reported only once\ngram.y:2797: for each function it appears in.)\ngram.y:2800: syntax error before `->'\ngram.y:2800: syntax error before `)'\ngram.y:2804: `n' undeclared (first use in this function)\ngram.y:2815: case label not within a switch statement\ngram.y:2834: break statement not within loop or switch\ngram.y:2835: case label not within a switch statement\ngram.y:2838: break statement not within loop or switch\ngram.y:2839: case label not within a switch statement\ngram.y:2843: break statement not within loop or switch\n...etc for many more lines.\n\nI assume that line 2800 in gram.y is missing a paren and should have the\nfollowing change.\n\n*** ../src.original/./backend/parser/gram.y Sun Jan 17 23:49:29 1999\n--- ./backend/parser/gram.y Sun Jan 17 23:59:13 1999\n***************\n*** 2797,2803 ****\n first_select>forUpdate = $3;\n $$ = (Node *)first_select;\n }\n! if ((SelectStmt *)$$)->forUpdate != NULL)\n {\n SelectStmt *n = (SelectStmt *)$1;\n \n--- 2797,2803 ----\n first_select>forUpdate = $3;\n $$ = (Node *)first_select;\n }\n! if (((SelectStmt *)$$)->forUpdate != NULL)\n {\n SelectStmt *n = (SelectStmt *)$1;\n\nBut I can't tell what should be done for the missing declaration for\nforUpdate. Can whoever added this please complete it. It appears\n4 times in the file.\n\n\nFurther, backend/libpq/pqcomprim.c is missing a header that causes the\ncompile to fail.\n\n*** ../src.original/./backend/libpq/pqcomprim.c Sun Jan 17 23:43:29 1999\n--- ./backend/libpq/pqcomprim.c Sun Jan 17 23:43:43 1999\n***************\n*** 1,5 ****\n--- 1,6 ---- \n #include <stdlib.h> \n #include <stdio.h> \n+ #include <errno.h> \n #include <sys/types.h>\n #include <sys/socket.h>\n #include <netinet/in.h>\n\nSorry if I seem a little frustrated. It's only because I am. This isn't\nthe first time this has happened while I was trying to work on the code.\nI don't think a full compile after extracting your diffs but before\nsending them in is so much to ask, is it? I'm not, after all, asking\nfor full regression testing.\n\nAnyway, I have (I think) all the changes needed to put full primary key\nsupport in. As soon as I can test them I'll send them in. Or maybe I\nshould send them in now without fully testing them. (Wry joke.)\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Mon, 18 Jan 1999 00:18:16 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "Syntax errors in current tree"
},
{
"msg_contents": "\"D'Arcy J.M. Cain\" wrote:\n> \n> really hesitate to make this criticism given all the work people are\n> doing on the system and all the great features and fixes that are\n> being added but could we please make it a hard and fast rule that\n> nothing gets added into the tree until the person adding it has at\n> least compiled the source, if not a full regression test. I have\n ^^^^^^^^^^^^^^^^^^^^^^^^^\nThis is what I always do!\nforUpdate was added by me into gram.y and parsenodes.h:SelectStmt.\nI guess that I committed both backend & include dirs\n(cvs update inside include dir doesn't show anything for\nparsenodes.h)...\n\n> finally jumped in and tried to add full primary key support but I\n> keep stumbling over simple syntax errors in files I am not working\n> on but that keep me from fully testing my own changes. Here's what\n> I get compiling gram.y[c].\n> \n> gcc -I../../include -I../../backend -I/usr/local/include -O2 -pipe -Wall -Wmissing-prototypes -I.. -Wno-error -c gram.c -o gram.o\n> /usr/local/share/bison.simple: In function `yyparse':\n> /usr/local/share/bison.simple:327: warning: implicit declaration of function `yyerror'\n> /usr/local/share/bison.simple:387: warning: implicit declaration of function `yylex'\n> gram.y:2797: `forUpdate' undeclared (first use in this function)\n\nVadim\n",
"msg_date": "Mon, 18 Jan 1999 12:41:11 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree"
},
{
"msg_contents": "> \"D'Arcy J.M. Cain\" wrote:\n> > \n> > really hesitate to make this criticism given all the work people are\n> > doing on the system and all the great features and fixes that are\n> > being added but could we please make it a hard and fast rule that\n> > nothing gets added into the tree until the person adding it has at\n> > least compiled the source, if not a full regression test. I have\n> ^^^^^^^^^^^^^^^^^^^^^^^^^\n> This is what I always do!\n> forUpdate was added by me into gram.y and parsenodes.h:SelectStmt.\n> I guess that I committed both backend & include dirs\n> (cvs update inside include dir doesn't show anything for\n> parsenodes.h)...\n\nNo, it is me. I merged the Stephan's patches into gram.y because he did\nnot have ForUpdate support in his patch. Working on it now.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 01:20:44 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree"
},
{
"msg_contents": "> really hesitate to make this criticism given all the work people are\n> doing on the system and all the great features and fixes that are\n> being added but could we please make it a hard and fast rule that\n> nothing gets added into the tree until the person adding it has at\n> least compiled the source, if not a full regression test. I have\n> finally jumped in and tried to add full primary key support but I\n> keep stumbling over simple syntax errors in files I am not working\n> on but that keep me from fully testing my own changes. Here's what\n> I get compiling gram.y[c].\n\nFixed. I assume people test the patches before submission, so I don't\nusually compile after each one. I did add some code in gram.y, and that\nwas what caused the problem.\n\nSorry. Blech (Sound of me falling on my sword. :-))\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 01:48:07 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree"
},
{
"msg_contents": "Thus spake Bruce Momjian\n> Fixed. I assume people test the patches before submission, so I don't\n\nNot an unreasonable assumption.\n\n> usually compile after each one. I did add some code in gram.y, and that\n> was what caused the problem.\n\nI don't think that the core maintainers should have to compile each\nand every patch before committing it. However, perhaps a form letter\ncan go out to each new submitter asking if their patch was tested and\nholding their first patch till they respond affirmatively. Once they\nhave a history, even of one patch, then accept their submissions as\nlong as they appear good.\n\nMaybe we need a way to track this. Anyone know of a good system for\ntracking this sort of BASic DATA. :-)\n\n> Sorry. Blech (Sound of me falling on my sword. :-))\n\nThat's OK. I just hope it was lying flat on the ground at the time. :-)\n\n-- \nD'Arcy J.M. Cain <darcy@{druid|vex}.net> | Democracy is three wolves\nhttp://www.druid.net/darcy/ | and a sheep voting on\n+1 416 424 2871 (DoD#0082) (eNTP) | what's for dinner.\n",
"msg_date": "Mon, 18 Jan 1999 07:59:48 -0500 (EST)",
"msg_from": "\"D'Arcy\" \"J.M.\" Cain <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree"
},
{
"msg_contents": "> Thus spake Bruce Momjian\n> > Fixed. I assume people test the patches before submission, so I don't\n> \n> Not an unreasonable assumption.\n> \n> > usually compile after each one. I did add some code in gram.y, and that\n> > was what caused the problem.\n> \n> I don't think that the core maintainers should have to compile each\n> and every patch before committing it. However, perhaps a form letter\n> can go out to each new submitter asking if their patch was tested and\n> holding their first patch till they respond affirmatively. Once they\n> have a history, even of one patch, then accept their submissions as\n> long as they appear good.\n> \n> Maybe we need a way to track this. Anyone know of a good system for\n> tracking this sort of BASic DATA. :-)\n\nNot really a problem. Most patch problems are either my mucking with it\nto merge it with other changes, or platform-specific problems that the\ntester would never have see. For example, Win32 used EINTR without\nincluding errno.h.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 08:26:40 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree"
},
{
"msg_contents": "Bruce Momjian <[email protected]> writes:\n> Not really a problem. Most patch problems are either my mucking with it\n> to merge it with other changes, or platform-specific problems that the\n> tester would never have see. For example, Win32 used EINTR without\n> including errno.h.\n\nAnd, by the same token, whoever does commit a patch might fail to notice\na platform-specific problem that didn't happen to show up on his\nplatform. You can't really expect things to be 100% tested before they\nhit the CVS tree (that's why we have a release cycle).\n\nMy experience so far is that the Postgres CVS sources are pretty stable;\nmost of the time they work, which can't be said for some other projects\nI'm involved in :-(. Not that we shouldn't strive to do even better,\nbut an occasional breakage is going to happen.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 18 Jan 1999 11:07:14 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree "
},
{
"msg_contents": "> Bruce Momjian <[email protected]> writes:\n> > Not really a problem. Most patch problems are either my mucking with it\n> > to merge it with other changes, or platform-specific problems that the\n> > tester would never have see. For example, Win32 used EINTR without\n> > including errno.h.\n> \n> And, by the same token, whoever does commit a patch might fail to notice\n> a platform-specific problem that didn't happen to show up on his\n> platform. You can't really expect things to be 100% tested before they\n> hit the CVS tree (that's why we have a release cycle).\n> \n> My experience so far is that the Postgres CVS sources are pretty stable;\n> most of the time they work, which can't be said for some other projects\n> I'm involved in :-(. Not that we shouldn't strive to do even better,\n> but an occasional breakage is going to happen.\n\nProblem is that I rarely have a solid block of time to apply the\npatches. (I used to.) It is done in pieces as I can get time during\nthe day. This makes for mistakes.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 11:36:48 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Syntax errors in current tree"
}
] |
[
{
"msg_contents": "I am feeling the need to jump into the code and add some of the easier\nthings that people have been asking for.\n\nHowever, I have a 7-month old baby, and just can't seem to get any free\ntime to do that. I haven't forgotten about all the things people need. \nLooks like Vadim and others have added many nice features to 6.5. Hope\nI can add some stuff for 6.6.\n\nSave some missing features for me. :-)\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 00:28:20 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "I need a PostgreSQL vacation"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> I am feeling the need to jump into the code and add some of the easier\n> things that people have been asking for.\n> \n> However, I have a 7-month old baby, and just can't seem to get any free\n> time to do that. I haven't forgotten about all the things people need.\n\n-:((\n\n> Looks like Vadim and others have added many nice features to 6.5. Hope\n> I can add some stuff for 6.6.\n> \n> Save some missing features for me. :-)\n\nSure -:)\n\nHow long you will off?\n\nVadim\n",
"msg_date": "Mon, 18 Jan 1999 14:13:30 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": ">\n> I am feeling the need to jump into the code and add some of the easier\n> things that people have been asking for.\n>\n> However, I have a 7-month old baby, and just can't seem to get any free\n> time to do that. I haven't forgotten about all the things people need.\n> Looks like Vadim and others have added many nice features to 6.5. Hope\n> I can add some stuff for 6.6.\n\n Tickle your baby - Postgres must wait (lower priority).\n\n>\n> Save some missing features for me. :-)\n\n Be sure we'll do so :-)\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Mon, 18 Jan 1999 13:19:14 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> Bruce Momjian wrote:\n> > \n> > I am feeling the need to jump into the code and add some of the easier\n> > things that people have been asking for.\n> > \n> > However, I have a 7-month old baby, and just can't seem to get any free\n> > time to do that. I haven't forgotten about all the things people need.\n> \n> -:((\n> \n> > Looks like Vadim and others have added many nice features to 6.5. Hope\n> > I can add some stuff for 6.6.\n> > \n> > Save some missing features for me. :-)\n> \n> Sure -:)\n> \n> How long you will off?\n\nI will not reduce what I have currently been doing for the past 7\nmonths. It is just that I can't get into PostgreSQL like I did a year\nago until the baby gets a little older. Not sure how much older he has\nto be before my wife let's me sneak down to the basement for hours.\n\nI already do sneak down, but for PostgreSQL, I need to do that on a\nregular basis.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 07:27:06 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "On Mon, 18 Jan 1999, Bruce Momjian wrote:\n\n> I am feeling the need to jump into the code and add some of the easier\n> things that people have been asking for.\n> \n> However, I have a 7-month old baby, and just can't seem to get any free\n\n\tWhere are the online photos being kept? :)\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 18 Jan 1999 09:46:01 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "\nOn 18-Jan-99 Bruce Momjian wrote:\n\n>> How long you will off?\n> \n> I will not reduce what I have currently been doing for the past 7\n> months. It is just that I can't get into PostgreSQL like I did a year\n> ago until the baby gets a little older. Not sure how much older he has\n> to be before my wife let's me sneak down to the basement for hours.\n> \n> I already do sneak down, but for PostgreSQL, I need to do that on a\n> regular basis.\n\nWe were in the process of moving when my youngest was born. Forseeing\nthat there would be little chance to even see a computer if it were in\nthe basement, while my wife was in the hospital I kept moving stuff to\nthe new house - the computer went into a remote corner of the living\nroom. She only complained about it for a couple of years, now she's\ngotten used to it :)\n \nVince.\n-- \n==========================================================================\nVince Vielhaber -- KA8CSH email: [email protected] flame-mail: /dev/null\n # include <std/disclaimers.h> TEAM-OS2\n Online Searchable Campground Listings http://www.camping-usa.com\n \"There is no outfit less entitled to lecture me about bloat\n than the federal government\" -- Tony Snow\n==========================================================================\n\n\n",
"msg_date": "Mon, 18 Jan 1999 10:21:19 -0500 (EST)",
"msg_from": "Vince Vielhaber <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> On Mon, 18 Jan 1999, Bruce Momjian wrote:\n> \n> > I am feeling the need to jump into the code and add some of the easier\n> > things that people have been asking for.\n> > \n> > However, I have a 7-month old baby, and just can't seem to get any free\n> \n> \tWhere are the online photos being kept? :)\n\nBottom of my web page. I need to add some more.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 10:33:44 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> \n> On 18-Jan-99 Bruce Momjian wrote:\n> \n> >> How long you will off?\n> > \n> > I will not reduce what I have currently been doing for the past 7\n> > months. It is just that I can't get into PostgreSQL like I did a year\n> > ago until the baby gets a little older. Not sure how much older he has\n> > to be before my wife let's me sneak down to the basement for hours.\n> > \n> > I already do sneak down, but for PostgreSQL, I need to do that on a\n> > regular basis.\n> \n> We were in the process of moving when my youngest was born. Forseeing\n> that there would be little chance to even see a computer if it were in\n> the basement, while my wife was in the hospital I kept moving stuff to\n> the new house - the computer went into a remote corner of the living\n> room. She only complained about it for a couple of years, now she's\n> gotten used to it :)\n\nI can't even get a _television_ into the living room. It's pretty\nformal. I do have a terminal in the kitchen and upstairs family/tv\nroom, though, all connected to the Unix machine in the basement.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 10:37:04 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> Hope I can add some stuff for 6.6.\n> Save some missing features for me. :-)\n\nThe rate I'm going, outer joins will be waiting for you ;)\n\n - Tom\n",
"msg_date": "Mon, 18 Jan 1999 17:06:38 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "On Mon, 18 Jan 1999, Bruce Momjian wrote:\n\n> > \n> > On 18-Jan-99 Bruce Momjian wrote:\n> > \n> > >> How long you will off?\n> > > \n> > > I will not reduce what I have currently been doing for the past 7\n> > > months. It is just that I can't get into PostgreSQL like I did a year\n> > > ago until the baby gets a little older. Not sure how much older he has\n> > > to be before my wife let's me sneak down to the basement for hours.\n> > > \n> > > I already do sneak down, but for PostgreSQL, I need to do that on a\n> > > regular basis.\n> > \n> > We were in the process of moving when my youngest was born. Forseeing\n> > that there would be little chance to even see a computer if it were in\n> > the basement, while my wife was in the hospital I kept moving stuff to\n> > the new house - the computer went into a remote corner of the living\n> > room. She only complained about it for a couple of years, now she's\n> > gotten used to it :)\n> \n> I can't even get a _television_ into the living room. It's pretty\n> formal. I do have a terminal in the kitchen and upstairs family/tv\n> room, though, all connected to the Unix machine in the basement.\n\n\tModern dayQuaker .. whoda guessed :)\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 18 Jan 1999 15:53:19 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> > I can't even get a _television_ into the living room. It's pretty\n> > formal. I do have a terminal in the kitchen and upstairs family/tv\n> > room, though, all connected to the Unix machine in the basement.\n> \n> \tModern dayQuaker .. whoda guessed :)\n\nSssh. Don't tell anyone. I will be shunned. :-)\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 14:56:07 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> > I can't even get a _television_ into the living room. It's pretty\n> > formal. I do have a terminal in the kitchen and upstairs family/tv\n> > room, though, all connected to the Unix machine in the basement.\n> \n> \tModern dayQuaker .. whoda guessed :)\n\nYo! Excuse me, but as a Quaker, I have to say you're confusing us with\nthe Amish. I think.\n\nSorry, back to your regularly scheduled programming...\n\n",
"msg_date": "Mon, 18 Jan 1999 16:09:20 -0500 (EST)",
"msg_from": "\"J. Michael Roberts\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "> > > I can't even get a _television_ into the living room. It's pretty\n> > > formal. I do have a terminal in the kitchen and upstairs family/tv\n> > > room, though, all connected to the Unix machine in the basement.\n> > \n> > \tModern dayQuaker .. whoda guessed :)\n> \n> Yo! Excuse me, but as a Quaker, I have to say you're confusing us with\n> the Amish. I think.\n> \n> Sorry, back to your regularly scheduled programming...\n\nSorry. I am from Pennsylvania, and we make jokes about it. They know\nthe difference.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 16:15:47 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
},
{
"msg_contents": "On Mon, 18 Jan 1999, J. Michael Roberts wrote:\n\n> > > I can't even get a _television_ into the living room. It's pretty\n> > > formal. I do have a terminal in the kitchen and upstairs family/tv\n> > > room, though, all connected to the Unix machine in the basement.\n> > \n> > \tModern dayQuaker .. whoda guessed :)\n> \n> Yo! Excuse me, but as a Quaker, I have to say you're confusing us with\n> the Amish. I think.\n\n\tD'oh...as a part Mennonite...you guys all look alike to me :)\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 18 Jan 1999 18:41:34 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] I need a PostgreSQL vacation"
}
] |
[
{
"msg_contents": "> Magnus wrote:\n> > Damn!\n> > That patch missed a few things - so now it doesn't compile any more!\n> > Here is a fix for the missing.\n> \n> I applied this patch last night --- dunno why Marc didn't do it yet,\n> since as you say the system won't build without it.\n> \n> I still had problems building the revised backend/libpq code on HP-UX.\n> I solved them by the expedient of removing the MSG_WAITALL flags from\n> the recv() calls. That flag doesn't exist on HP-UX, nor on a lot of\n> other systems I'll bet. So I went ahead and checked in that change.\n> \n> If you think MSG_WAITALL is really important for performance, you\n> could offer a patch that would do something like \"#ifdef MSG_WAITALL\".\n> My advice is not to bother though. 99% of frontend queries are going\n> to arrive in a single packet, probably, so I doubt it's worth worrying\n> about or introducing a possible cross-platform behavior \n> difference for.\n\nNope, it shuold probably be Ok to do without. Definitly not worth\nintroducing incompatibilities. Get rid of it :-)\n\n//Magnus\n",
"msg_date": "Mon, 18 Jan 1999 09:53:50 +0100",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [PATCHES] Another libpq-be patch"
}
] |
[
{
"msg_contents": "> The other thing that jumps out here is the unreasonably high \n> position of\n> recv(), which is called 962187 times. The script being read \n> by psql was\n> only 957186 characters. Evidently we're invoking a kernel recv() call\n> once per character read from the frontend. I suspect this is an\n> inefficiency introduced by Magnus Hagander's recent rewrite of backend\n> libpq (see, I told you there was a reason for using stdio ;-)). We're\n> gonna have to do something about that, though it's not as critical as\n> the memory-allocation issue. \nCould be because of that. I noticed that the backend calls pq_getchar() a\n_lot_ of times, looping for reading a single character. It did that before\ntoo. The difference was that pq_getchar() called fgetc() then, and calls\nrecv() now.\nI don't know, maybe recv() is more expensive than fgetc()? But I really\ncan't see any reason it shuold be called more often now than before.\nAn interesting fact is that pq_getchar() doesn't show up at all. Could be\nbecause it's fast, but still executed many times, right? Or it could be that\nthe 'inner loops' in pq_getchar(), pq_peekchar(), or pqGetNBytes() don't\nwork as expected. On my system (Linux 2.2), I only get one recv() call for\neach entry into these functions - as it should be - might it be different on\nyours?\n\nOk, so I give up, perhaps we need a buffer after all :-)\n\n\n//Magnus\n",
"msg_date": "Mon, 18 Jan 1999 10:07:26 +0100",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "> Could be because of that. I noticed that the backend calls pq_getchar() a\n> _lot_ of times, looping for reading a single character. It did that before\n> too. The difference was that pq_getchar() called fgetc() then, and calls\n> recv() now.\n> I don't know, maybe recv() is more expensive than fgetc()? But I really\n> can't see any reason it shuold be called more often now than before.\n> An interesting fact is that pq_getchar() doesn't show up at all. Could be\n> because it's fast, but still executed many times, right? Or it could be that\n> the 'inner loops' in pq_getchar(), pq_peekchar(), or pqGetNBytes() don't\n> work as expected. On my system (Linux 2.2), I only get one recv() call for\n> each entry into these functions - as it should be - might it be different on\n> yours?\n\nIt is very possible that fgetc() is a macro on your platform. See\n/usr/include/stdio.h. If so, it has no function call overhead.\n\nOn BSD/OS, it used to be a macro, but now with threads, it is not. They\nhave a macro version, but is under a different name.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 07:34:58 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
},
{
"msg_contents": "Magnus Hagander <[email protected]> writes:\n> I don't know, maybe recv() is more expensive than fgetc()?\n\nVastly.\n\nrecv() is a kernel call. You have the overhead of getting control into\nthe kernel, normally several times more expensive than a function call;\nof passing parameters back and forth from user space to kernel space\n(for example, the kernel will probably have to translate and range-check\nthe buffer pointer you pass it, to ensure you can't fool the kernel into\nscribbling on some other process's memory); of verifying that the\ndescriptor number you pass is open and you have permission to read it;\nand of finding the associated data buffer. Plus the scheduler may\nrun to reconsider whether to give control back to you, or switch off to\nanother user process. Etc etc etc.\n\nfgetc() is a plain C function that normally just has to fetch the next\nbyte out of a buffer that's already been read into your address space\n--- that is, when you're using stdio, you pay all the above-described\nkernel interaction overhead once per bufferload, not once per character.\n\nIf you use getc(), which is allowed to be a macro, you don't even pay\nthe function-call overhead; that form is probably less than a dozen\ninstructions, except when the buffer is empty. Judging by the profile\nnumbers, recv()'ing a single character takes close to 1400 instructions\non my system.\n\n> An interesting fact is that pq_getchar() doesn't show up at all. Could be\n> because it's fast, but still executed many times, right?\n\nRight, it doesn't run long enough to get itself into the top functions.\nIt's there though --- the dynamic profile shows:\n\n-----------------------------------------------\n 0.03 18.83 5001/5001 ReadCommand [27]\n[28] 10.4 0.03 18.83 5001 SocketBackend [28]\n 0.00 18.73 5000/5000 pq_getstr [30]\n 0.01 0.09 5001/5001 pq_getnchar [303]\n-----------------------------------------------\n 0.00 18.73 5000/5000 SocketBackend [28]\n[30] 10.4 0.00 18.73 5000 pq_getstr [30]\n 0.11 18.62 5000/5000 pqGetString [29]\n-----------------------------------------------\n 0.11 18.62 5000/5000 pq_getstr [30]\n[29] 10.4 0.11 18.62 5000 pqGetString [29]\n 0.47 18.15 957186/957186 pq_getchar [31]\n-----------------------------------------------\n 0.47 18.15 957186/957186 pqGetString [29]\n[31] 10.3 0.47 18.15 957186 pq_getchar [31]\n 18.15 0.00 957186/962187 recv [32]\n-----------------------------------------------\n 0.09 0.00 5001/962187 pqGetNBytes [315]\n 18.15 0.00 957186/962187 pq_getchar [31]\n[32] 10.1 18.24 0.00 962187 recv [32]\n-----------------------------------------------\n\nIn the old code with fgetc(), the execution time of fgetc() was probably\nnot much worse than pq_getchar --- ie, about half a second not 18\nseconds for this test sequence...\n\n\nWhat we need to do here is to re-introduce the buffering ability of\nstdio into backend libpq. If you compare the current frontend libpq,\nyou'll notice that it reads or writes the socket a bufferload at a time,\nnot a character at a time.\n\n\t\t\tregards, tom lane\n",
"msg_date": "Mon, 18 Jan 1999 11:37:12 -0500",
"msg_from": "Tom Lane <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof "
},
{
"msg_contents": "> What we need to do here is to re-introduce the buffering ability of\n> stdio into backend libpq. If you compare the current frontend libpq,\n> you'll notice that it reads or writes the socket a bufferload at a time,\n> not a character at a time.\n\nFYI, I never profiles massive inserts. I usually profiles just SELECT\nstatements on large tables.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 11:48:22 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
}
] |
[
{
"msg_contents": "Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\nwith INTERSECT and EXCEPT. Is that correct?\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Mon, 18 Jan 1999 11:40:04 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "latest parser changes"
},
{
"msg_contents": "> Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\n> with INTERSECT and EXCEPT. Is that correct?\n\nBeats me. I allowed it in the grammer while merging Stephan's stuff\ninto the code.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 12:11:38 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "Michael Meskes wrote:\n> \n> Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\n> with INTERSECT and EXCEPT. Is that correct?\n\nYes. \n\nVadim\n",
"msg_date": "Tue, 19 Jan 1999 09:56:36 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "> Michael Meskes wrote:\n> > \n> > Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\n> > with INTERSECT and EXCEPT. Is that correct?\n\nLooks like the tests you make in gram.y will still catch those.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 22:00:14 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "On Mon, Jan 18, 1999 at 12:11:38PM -0500, Bruce Momjian wrote:\n> > Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\n> > with INTERSECT and EXCEPT. Is that correct?\n> \n> Beats me. I allowed it in the grammer while merging Stephan's stuff\n> into the code.\n\nSince I cannot see the appropriate check for this, I guess we should add it\nto gram.y.\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Tue, 19 Jan 1999 09:33:56 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "On Mon, Jan 18, 1999 at 10:00:14PM -0500, Bruce Momjian wrote:\n> > > Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\n> > > with INTERSECT and EXCEPT. Is that correct?\n> \n> Looks like the tests you make in gram.y will still catch those.\n\nWhere? I just see:\n\nif (n->unionClause != NULL)\n elog(ERROR, \"SELECT FORUPDATE is not allowed with UNION clause\");\n \nbut not intersect and except. But then I haven't had to time to read the\ncode carefully.\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Tue, 19 Jan 1999 09:35:24 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "> On Mon, Jan 18, 1999 at 10:00:14PM -0500, Bruce Momjian wrote:\n> > > > Since FOR UPDATE is not allowed with UNION I suppose it is also not allowed\n> > > > with INTERSECT and EXCEPT. Is that correct?\n> > \n> > Looks like the tests you make in gram.y will still catch those.\n> \n> Where? I just see:\n> \n> if (n->unionClause != NULL)\n> elog(ERROR, \"SELECT FORUPDATE is not allowed with UNION clause\");\n> \n> but not intersect and except. But then I haven't had to time to read the\n> code carefully.\n\nI thought INTERSECT and EXCEPT used UNION as well, or is that false? I\ndon't know.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 19 Jan 1999 15:47:10 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "On Tue, Jan 19, 1999 at 03:47:10PM -0500, Bruce Momjian wrote:\n> I thought INTERSECT and EXCEPT used UNION as well, or is that false? I\n> don't know.\n\nI don't think you're right. I see three different rules for the three\nkeywords.\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Wed, 20 Jan 1999 10:01:02 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] latest parser changes"
},
{
"msg_contents": "> On Tue, Jan 19, 1999 at 03:47:10PM -0500, Bruce Momjian wrote:\n> > I thought INTERSECT and EXCEPT used UNION as well, or is that false? I\n> > don't know.\n> \n> I don't think you're right. I see three different rules for the three\n> keywords.\n\nI see. I have added the intersectClause test.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Wed, 20 Jan 1999 14:48:23 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] latest parser changes"
}
] |
[
{
"msg_contents": "Hi all,\n\nI have a question about rules. I create a rule to insert data into a\nview that works fine,\nbut when I add a row to the view, PostgreSQL replies INSERT 0 0 if the\nrow is added and it replies\nINSERT oid# 1 if the row is not added.\nI have the same behavior with RULES on UPDATE and on DELETE. \n\nHere the example:\n\n\ncreate rule \"_RIT_vista\" as on insert to vista\n where new.job='SALESMAN'\n do instead\n insert into emp (empno,ename,job) values\n(new.empno,new.ename,new.job);\nCREATE\n\ninsert into vista values ('8900','MANUEL','SALESMAN');\nINSERT 0 0 <<<<<<<<<<------------------------- Why it replies\nINSERT 0 0 if it adds the row ?\n^^^^^^^^^^^\n\nselect * from vista;\nempno|ename |job\n-----+----------+------------\n 8900|MANUEL |SALESMAN\n(1 row)\n\ninsert into vista values ('8901','JOSE','PROGRAMMER');\nINSERT 144991 1 <<<<<<<<<------------------------- Why it replies\nINSERT oid 1 if it adds no row?\n^^^^^^^^^^^^^^^\n\nselect * from vista;\nempno|ename |job\n-----+----------+------------\n 8900|MANUEL |SALESMAN\n(1 row)\n\nselect oid,* from emp;\n oid|ename |empno|job |hiredate|sal|comm|deptno|level|mgr\n------+----------+-----+------------+--------+---+----+------+-----+---\n144990|MANUEL | 8900|SALESMAN | | | | | |\n(1 row)\n\nAny ideas ?\n\n-Jose'-\n",
"msg_date": "Mon, 18 Jan 1999 14:24:41 +0100",
"msg_from": "\"Jose' Soares\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "about RULES"
},
{
"msg_contents": ">\n> Hi all,\n>\n> I have a question about rules. I create a rule to insert data into a\n> view that works fine,\n> but when I add a row to the view, PostgreSQL replies INSERT 0 0 if the\n> row is added and it replies\n> INSERT oid# 1 if the row is not added.\n> I have the same behavior with RULES on UPDATE and on DELETE.\n\n Anything is fine and behaves as (I) expected :-)\n\n I know, it's a little irritating. The reply you're seeing is\n the result of the last processed query. Your rule ...\n\n>\n> Here the example:\n>\n>\n> create rule \"_RIT_vista\" as on insert to vista\n> where new.job='SALESMAN'\n> do instead\n> insert into emp (empno,ename,job) values\n> (new.empno,new.ename,new.job);\n> CREATE\n\n ... is a conditional one (has a rule qualification WHERE\n new.job = 'SALESMAN'). In this case, the rule system split's\n your query into one that inserts into \"emp\" if rule-qual\n matched and another that inserts into \"vista\" if not. The\n query to insert into \"emp\" is executed first, the one that\n inserts into \"vista\" second.\n\n>\n> insert into vista values ('8900','MANUEL','SALESMAN');\n> INSERT 0 0 <<<<<<<<<<------------------------- Why it replies\n> INSERT 0 0 if it adds the row ?\n> ^^^^^^^^^^^\n\n 1 row inserted into \"emp\", 0 rows inserted into \"vista\".\n\n>\n> select * from vista;\n> empno|ename |job\n> -----+----------+------------\n> 8900|MANUEL |SALESMAN\n> (1 row)\n>\n> insert into vista values ('8901','JOSE','PROGRAMMER');\n> INSERT 144991 1 <<<<<<<<<------------------------- Why it replies\n> INSERT oid 1 if it adds no row?\n> ^^^^^^^^^^^^^^^\n\n 0 rows inserted into \"emp\", 1 row inserted into \"vista\"!\n\n>\n> select * from vista;\n> empno|ename |job\n> -----+----------+------------\n> 8900|MANUEL |SALESMAN\n> (1 row)\n>\n> select oid,* from emp;\n> oid|ename |empno|job |hiredate|sal|comm|deptno|level|mgr\n> ------+----------+-----+------------+--------+---+----+------+-----+---\n> 144990|MANUEL | 8900|SALESMAN | | | | | |\n> (1 row)\n>\n\n Now you wonder where Postgres left 'JOSE' - eh? He's there\n and all information is kept. But you're unable to see it,\n because on SELECT from \"vista\", the rewrite system changes\n your query to something that displays from \"emp\". But if you\n take a look into the database directory, you'll see that the\n file \"vista\" lost it's zero-size. So there is data!\n\n You can't update, delete or see him, as long as the _RETvista\n rule is active. Any query that needs to scan \"vista\" get's\n rewritten to scan from \"emp\".\n\n It's a problem of a general-purpose production rule system\n like the one Postgres has. It does not (and IMHO should not)\n check, whether the summary of all defined rules makes sense.\n Such a check would make it impossible to setup many qualified\n rules, because they have circular dependencies that cannot be\n matched when you created the first ones.\n\n In the case of a qualified instead rule, the parsetree get's\n splitted. One with the qual, one with the negated qual. If\n you only want 'SALESMAN's, you must put the qualification\n into the INSERT action of the rule as\n\n create rule \"_INSvista\" as on insert to vista\n do instead\n insert into emp select new.empno, new.ename, new.job\n where new.job='SALESMAN';\n\n This is an unqualified instead rule, where the ACTION filters\n the data. This time, the rewrite system will throw away the\n original query and only the INSERT INTO emp ... is left, but\n it will never insert anything than row's of SALESMAN. All\n others are silently thrown away.\n\n Changing from INSERT ... VALUES to INSERT ... SELECT is the\n way to make the parser happy about the WHERE clause. On\n parsetree level, there is no difference between them.\n\n Well, I expect more questions now - so come on.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Mon, 18 Jan 1999 23:37:03 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] about RULES"
},
{
"msg_contents": "> \n> In the case of a qualified instead rule, the parsetree get's\n> splitted. One with the qual, one with the negated qual. If\n> you only want 'SALESMAN's, you must put the qualification\n> into the INSERT action of the rule as\n> \n> create rule \"_INSvista\" as on insert to vista\n> do instead\n> insert into emp select new.empno, new.ename, new.job\n> where new.job='SALESMAN';\n\nI see that rules is more complex than I thougth. \n\nI tried your example Jan but it doesn't work...\n\ncreate rule \"_INSvista\" as on insert to vista\n do instead\n insert into emp select new.empno, new.ename, new.job\n where new.job='SALESMAN';\nERROR: Type of empno does not match target column ename\n\n-Jose'-\n",
"msg_date": "Tue, 19 Jan 1999 13:30:37 +0100",
"msg_from": "\"Jose' Soares\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] about RULES"
},
{
"msg_contents": "Jose' Soares ha scritto:\n\n> >\n> > In the case of a qualified instead rule, the parsetree get's\n> > splitted. One with the qual, one with the negated qual. If\n> > you only want 'SALESMAN's, you must put the qualification\n> > into the INSERT action of the rule as\n> >\n> > create rule \"_INSvista\" as on insert to vista\n> > do instead\n> > insert into emp select new.empno, new.ename, new.job\n> > where new.job='SALESMAN';\n>\n> I see that rules is more complex than I thougth.\n>\n> I tried your example Jan but it doesn't work...\n>\n> create rule \"_INSvista\" as on insert to vista\n> do instead\n> insert into emp select new.empno, new.ename, new.job\n> where new.job='SALESMAN';\n> ERROR: Type of empno does not match target column ename\n>\n> -Jose'-\n\nI had no reply to this message probably because a problem with my mail.\n\nI'm trying to make a view updatable using triggers, INSERT works fine\nbut UPDATE/DELETE\ndoesn't because the WHERE condition is evaluated FALSE all the time.\nHere an example:\n\n\ncreate table emp (\n empno int,\n ename char(10),\n job char(12),\n hiredate date,\n sal money,\n comm int,\n deptno int,\n level int,\n mgr int\n);\nCREATE\n\ninsert into emp values (8900,'MANUEL','SALESMAN',CURRENT_DATE,'$2,000');\n\nINSERT 149844 1\n\ncreate table vista\n as select empno, ename, job\n from emp\n where job='SALESMAN';\nCREATE\n\ndrop function add_vista();\ncreate function add_vista() returns opaque as '\nbegin\n raise notice ''trigger fired: % on %'',tg_when,tg_op;\n return null;\nend;\n' language 'plpgsql';\n\n\ncreate trigger t_add_vista before insert or update or delete\n on vista for each row execute procedure add_vista();\n\ndelete from vista where ename='MANUEL';\nNOTICE: trigger fired: BEFORE on DELETE\nDELETE 0\ndelete from vista ;\nNOTICE: trigger fired: BEFORE on DELETE\nDELETE 0\n\n--And now VISTA becames a view:.................\n\nCREATE RULE \"_RETvista\" AS\n ON SELECT TO \"vista\"\n DO INSTEAD\n SELECT \"empno\", \"ename\", \"job\"\n FROM \"emp\" WHERE \"job\" = 'SALESMAN'::\"bpchar\";\nCREATE\n\nselect * from vista;\n\nempno|ename |job\n-----+----------+------------\n 8900|MANUEL |SALESMAN\n(1 row)\n\n-- The where condition is all the time evaluated as FALSE...\n\ndelete from vista where ename='MANUEL'; --why this condition isn't true\n?\nDELETE 0\n\ndelete from vista ;\nNOTICE: trigger fired: BEFORE on DELETE\nDELETE 0\n\n\n--Is there a way to make views updatable ?\n\n-Jose'-\n\n\n\n",
"msg_date": "Fri, 22 Jan 1999 16:33:03 +0100",
"msg_from": "\"Jose' Soares\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] about RULES"
},
{
"msg_contents": "> > I tried your example Jan but it doesn't work...\n> >\n> > create rule \"_INSvista\" as on insert to vista\n> > do instead\n> > insert into emp select new.empno, new.ename, new.job\n> > where new.job='SALESMAN';\n> > ERROR: Type of empno does not match target column ename\n> >\n> > -Jose'-\n>\n> I had no reply to this message probably because a problem with my mail.\n\n Sorry,\n\n I just wrote it down quickly. Of cause the insert action\n must be a valid statement. In the above case I assume, the\n table 'emp' didn't had the empno first. When looking at the\n schema of emp you gave below, the rule should read:\n\n create rule \"_INSvista\" as on insert to vista\n do instead\n insert into emp (empno, ename, job)\n values (new.empno, new.ename, new.job)\n where new.job = 'SALESMAN';\n\n>\n> I'm trying to make a view updatable using triggers, INSERT works fine\n> but UPDATE/DELETE\n> doesn't because the WHERE condition is evaluated FALSE all the time.\n> Here an example:\n\n A trigger could never work for UPDATE or DELETE. A trigger is\n only fired when there is actually a row in a table to get\n updated or deleted. Views don't (or at least shouldn't)\n contain any data, so there is never a row to fire them.\n\n>\n> create table emp (\n> empno int,\n> ename char(10),\n> job char(12),\n> hiredate date,\n> sal money,\n> comm int,\n> deptno int,\n> level int,\n> mgr int\n> );\n> CREATE\n>\n> [...]\n>\n> -- The where condition is all the time evaluated as FALSE...\n>\n> delete from vista where ename='MANUEL'; --why this condition isn't true\n> ?\n> DELETE 0\n\n The rule system redirected the scan for the DELETE from\n 'vista' to a scan from 'emp' because vista is a view on emp.\n The resulting query is a scan from emp who's result tupels\n should be deleted from vista - a whole lot of nonsens and\n thus absolutely nothing happens.\n\n>\n> delete from vista ;\n> NOTICE: trigger fired: BEFORE on DELETE\n> DELETE 0\n>\n>\n> --Is there a way to make views updatable ?\n\n Read section 8 of the programmers manual to understand all\n the details of the rewrite rule system. Then do it with\n rules.\n\n\nJan\n\n--\n\n#======================================================================#\n# It's easier to get forgiveness for being wrong than for being right. #\n# Let's break this rule - forgive me. #\n#======================================== [email protected] (Jan Wieck) #\n\n\n",
"msg_date": "Fri, 22 Jan 1999 20:03:16 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] about RULES"
}
] |
[
{
"msg_contents": "Tom Lane wrote\n>Vadim Mikheev <[email protected]> writes:\n>> John Holland wrote:\n>>> I've been lurking on this list for a little while. I have just done a\n>>> little tinkering with Postgres 6.4.2, comparing it to Oracle and mySQL on\n>>> Linux. It would appear that just doing a lot of inserts (ie 40000) in a\n>>> loop is enormously slower in Postgres in two ways that I tried it.\n>>> One - using a loop in Java that makes a JDBC call to insert.\n>>> Two- using plpgsql as a comparision to PL/SQL.\n>>> \n>>> I really like the idea of an open source DB and am impressed with a lot I\n>>> see about PostgreSQL - however the speed difference is pretty bad -\n>>> 4.5 minutes versus about 20 seconds.\n>\n>> Try to use BEGIN/END to run all inserts in single transaction\n>> and please let us know results.\n>\n>I tried this myself and found that wrapping BEGIN/END around a series of\n>INSERT statements didn't make much difference at all.\n>\n>On an HP 715 (75MHz RISC box, not very heavy iron by current standards),\n>I inserted about 13000 records into an initially-empty table having 38\n>columns (just because it's data I had handy...). I timed it at:\n>\n>Individual INSERT statements (as generated by pg_dump -d):\n> 33 inserts/sec\n>Same with BEGIN/END wrapped around 'em:\n> 34 inserts/sec\n>Single COPY statement (as generated by pg_dump without -d):\n> 1400 inserts/sec\n>\n>This was for a simple table with no indexes. In reality, this table\n>type has four b-tree indexes on different columns in our normal usage.\n>The COPY speed dropped to 325 inserts/sec when I had the indexes in\n>place. I didn't bother trying the INSERTs that way.\n>\n>The conventional wisdom is that you should use COPY for bulk loading,\n>and this result supports it...\n>\n>John didn't say what hardware he's using, so I don't know how comparable\n>my result is to his 150 inserts/sec --- that might have been on a table\n>with many fewer columns, or maybe his machine is just faster.\n>\n>As for where the time is going: \"top\" showed that the time for the\n>INSERT ops was almost all going into backend CPU time. My guess is\n>that most of the cost is in parsing and planning the INSERT statements.\n>Pre-planned statement skeletons might help, but I think the real answer\n>will be to find a way to avoid constructing and parsing SQL statements\n>for simple operations like INSERT. (One reason I'm interested in the\n>idea of a CORBA interface is that it might help here.)\n>\n>My guess is that Oracle and mySQL have APIs that allow the construction\n>of an SQL INSERT command to be bypassed, and that's why they show up\n>better on this operation.\n\nOracle has something called direct path loading. It is very fast, but\nfussy. You need to drop indexes, do the direct path load, and restore\nthe indexes. If you do a direct path load with indexes, or stop \nhalfway (? was a while ago ?) the table gets stuck in a funny mode\nand you have to drop/recreate it. See the sqlload manual for\ndetails.\n\nWe used this, plus segmenting the database, plus ping-ponging between\ntwo tables[1] on one big project. Worked great except they cancelled\nthe project two weeks before we were supposed to go on-line. Still\nbitter about that one.\n\n\n-- cary\n\n[1] I.e have two tables, a and b. Load b while accesses are going to\na, then when b is all loaded and indexed, have the apps switch to looking\nup using b and load the data into a. I think we used views or something\nto do the switch.\n",
"msg_date": "Mon, 18 Jan 1999 10:32:47 -0500 (EST)",
"msg_from": "\"Cary O'Brien\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Postgres Speed or lack thereof"
}
] |
[
{
"msg_contents": "Okay, I've updated the ecpg parser yet again. Unfortunately it has one\nremaining shift/reduce conflict. But the very same conflict is in gram.y, so\nI don't dig into it very much now.\n\nAnyway, I just saw that there were minor changes made to ecpg by others. Now\nI like that but I would prefer if I was told about that. Otherwise my\nversion numbering and Changelog maintaining might break. Or simply change\nthese too. :-)\n\nAlso I had to add #include <errno.h> to backend/libpq/pqcomprim.c to be\nable to compile postgresql.\n\nPatch is attached. Since my resubscription process is still not finished\nyet, I still send them here.\n\nMichael\n\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!",
"msg_date": "Mon, 18 Jan 1999 16:49:03 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "ecpg changes"
},
{
"msg_contents": "Applied.\n\n> Okay, I've updated the ecpg parser yet again. Unfortunately it has one\n> remaining shift/reduce conflict. But the very same conflict is in gram.y, so\n> I don't dig into it very much now.\n> \n> Anyway, I just saw that there were minor changes made to ecpg by others. Now\n> I like that but I would prefer if I was told about that. Otherwise my\n> version numbering and Changelog maintaining might break. Or simply change\n> these too. :-)\n> \n> Also I had to add #include <errno.h> to backend/libpq/pqcomprim.c to be\n> able to compile postgresql.\n\nI had taken care of that last night.\n\n> \n> Patch is attached. Since my resubscription process is still not finished\n> yet, I still send them here.\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 12:16:01 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] ecpg changes"
}
] |
[
{
"msg_contents": "I tried inserting 15000 tuples each in two tables via pgsql and timed it.\nAfter insertion I also created btree indeces on both tables.\n\nSome results:\n\nFirst try was with a 6.4.1 version without -F option and all inserts in one\ntransaction:\nreal 2m11.084s\nuser 0m1.760s\nsys 0m1.680s\n\nThen I switched to 6.5 without -F:\nreal 2m35.833s\nuser 0m2.900s\nsys 0m3.910s\n\nSo some of the latest changes cost quite a lot of time. On the other hand I\nwas surprised to see that the size of the table file decreased by about 10%\nwhen switching to 6.5.\n\nThen I switched on -F:\nreal 1m11.573s\nuser 0m3.150s\nsys 0m3.760s\n\nI've seen even bigger advantages of -F.\n\nFor comparison I tried without BEGIN/END:\nreal 1m12.463s\nuser 0m3.030s\nsys 0m4.480s\n\nSo, yes it is slower than just one transaction, but not very much.\n\nFinally I tried the same script (except for changing int4 to number(4)) on\nOracle 8.0.5:\nreal 1m29.248s\nuser 0m24.460s\nsys 0m4.880s\n\nI got similar results when I tried the last time. On stuff like insertions\nwe are quite a lot faster than Oracle if we use -F. \n\nAll test were run on my AMD K6/350 with 64MB ram running Linux-2.2.0-pre7.\n\nmichael\n\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Mon, 18 Jan 1999 16:56:17 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "performance test"
},
{
"msg_contents": "Michael Meskes wrote:\n> \n> I tried inserting 15000 tuples each in two tables via pgsql and timed it.\n> After insertion I also created btree indeces on both tables.\n> \n...\n> \n> Then I switched to 6.5 without -F:\n> real 2m35.833s\n> user 0m2.900s\n> sys 0m3.910s\n> \n...\n> \n> Then I switched on -F:\n> real 1m11.573s\n> user 0m3.150s\n> sys 0m3.760s\n> \n> I've seen even bigger advantages of -F.\n\nYou've seen disadvantages of our unperfect buffer manager -:)\nWhen server need in buffer for new data comming and there is\nno unused buffers in pool (i.e. - all buffers filled with new\ndata and marked as dirty), server gets some dirty buffer,\nwrites it AND FSYNC file. So, server does many fsyncs\neven with BEGIN/END while should do _one_ fsync at COMMIT.\n\nHaving this problem fixed you wouldn't had so big difference\nbetween -F and BEGIN/END-no-F.\n\nVadim\n",
"msg_date": "Tue, 19 Jan 1999 10:27:50 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] performance test"
},
{
"msg_contents": "> You've seen disadvantages of our unperfect buffer manager -:)\n> When server need in buffer for new data comming and there is\n> no unused buffers in pool (i.e. - all buffers filled with new\n> data and marked as dirty), server gets some dirty buffer,\n> writes it AND FSYNC file. So, server does many fsyncs\n> even with BEGIN/END while should do _one_ fsync at COMMIT.\n> \n> Having this problem fixed you wouldn't had so big difference\n> between -F and BEGIN/END-no-F.\n\nIs this idea still reasonable?\n\n\n---------------------------------------------------------------------------\n\n\nHere is an archive of the pg_log discussion.\n\n---------------------------------------------------------------------------\n\nFrom: Bruce Momjian <[email protected]>\nMessage-Id: <[email protected]>\nSubject: [HACKERS] Bufferd loggins/pg_log\nTo: [email protected] (PostgreSQL-development)\nDate: Mon, 17 Nov 1997 00:42:18 -0500 (EST)\nCc: [email protected] (Vadim B. Mikheev)\nX-Mailer: ELM [version 2.4 PL25]\nMIME-Version: 1.0\nContent-Type: text/plain; charset=US-ASCII\nContent-Transfer-Encoding: 7bit\nSender: [email protected]\nPrecedence: bulk\nStatus: OR\n\nHere is my current idea for doing bufferd logging, and exists between\nthe normal fsync on every transaction and no-fsync options. I believe\nit will be very popular, because it mimicks the Unix file system\nreliability structure.\n\n---------------------------------------------------------------------------\n\nOn startup, the postmaster makes a copy of pg_log, called pg_log_live.\nEach postgres backend mmaps() this new file into its address space. A\nlock is gotten to make changes to the file. All backend use pg_log_live\nrather than pg_log. Only the postmaster write to pg_log. (I will\nsomeday remove the exec() from postmaster, so backends will get this\naddress space automatically.) \n\nThe first 512 bytes of pg_log and pg_log_live are used for log managment\ninformation. We add a new field to pg_log_live called min_xid_commit\nwhich records the lowest transaction id that any backend has committed\nsince the start of the last sync pass of the postmater. We also add\nfields to record current pg_variable oid and xid at the same time. (xid\nmay have to be moved into pg_variable so backends can fsync it (see\nbelow).)\n\nEvery 60-90 seconds, the postmaster gets a write lock on pg_log, takes\nthe minimum xid set in the start of pg_log, and resets its value. It\nrecords the current oid and xid from pg_variable. It then clears the\nlock, and starts reading from the minimum recorded xid changed to the\nend of pg_log_live, and copies it into allocated memory. It then does a\nsync (twice?), waits for completion, and then writes the pg_log_live\npartial copy it made to pg_log. We update the copies of oid and xid we\nsaved before the sync to the bottom of pg_log_live. \n\nWe can change the 60-90 seconds to be longer, but the system does it\nevery 30 seconds anyway.\n\nWhen the postmaster stops, it does this same operation before shutting\ndown, and pg_log_live is removed.\n\nWe make a copy of the current xid and oid in the front of pg_log_live,\nso that if the postmaster starts up, and pg_log_live exists, the\npostmaster adds 10,000 to xid and oid of pg_variable, so no previously\nused but unsynced values are used.\n\nWe know that the current values of pg_variable could not have been\nexceeded by 10,000, because each backend consults the pg_log copies of\nthese variable to make sure they do not exceed 10,000 from the value\nbefore the last sync. They exceed those values only by fscyn'ing every\n10,000 increments.\n\nSaid another way, if a postgres backend exceeds the pg_log last xid or\noid of pg_log, or any 10,000 multiple, it must fsync the change to\npg_variable. This way, a crash skips over any unsynced oid/xid's used,\nand this is done without having to keep fsyncing pg_variable. In most\ncases, the 10,000 will never be exceeded by a backend before the\npostmaster does a sync and increases the last xid/oid again.\n\nI think this is a very clean way to give us no-fync performance with\nfull-rollback buffered logging. The specification is clean and almost\ncomplete enough for coding.\n\nI think this gives us what we need, by having a mmap'ed() pg_log_live,\nwhich backends can use, and a postmaster-controlled pg_log, which is\nused on startup, with xid/oid controls in a crash situation to skip over\npartially committed transactions.\n\nComments?\n\n-- \nBruce Momjian\[email protected]\n\n\n---------------------------------------------------------------------------\n\nSender: [email protected]\nMessage-ID: <[email protected]>\nDate: Mon, 17 Nov 1997 14:56:05 +0700\nFrom: \"Vadim B. Mikheev\" <[email protected]>\nOrganization: ITTS (Krasnoyarsk)\nX-Mailer: Mozilla 3.01 (X11; I; FreeBSD 2.2.5-RELEASE i386)\nMIME-Version: 1.0\nTo: Bruce Momjian <[email protected]>\nCC: PostgreSQL-development <[email protected]>,\n \"Vadim B. Mikheev\" <[email protected]>\nSubject: Re: Bufferd loggins/pg_log\nReferences: <[email protected]>\nContent-Type: text/plain; charset=us-ascii\nContent-Transfer-Encoding: 7bit\nStatus: OR\n\nBruce Momjian wrote:\n> \n> On startup, the postmaster makes a copy of pg_log, called pg_log_live.\n> Each postgres backend mmaps() this new file into its address space. A\n> lock is gotten to make changes to the file. All backend use pg_log_live\n> rather than pg_log. Only the postmaster write to pg_log. (I will\n> someday remove the exec() from postmaster, so backends will get this\n> address space automatically.)\n\nWhat are advantages of mmaping entire pg_log over \"online\" pg_log\npages ?\npg_log may be very big (tens of Mb) - why we have to spend\nprocess address space for tens of Mb of mostly unused data ?\nAlso, do all systems have mmap ?\n\n> \n> Every 60-90 seconds, the postmaster gets a write lock on pg_log, takes\n> the minimum xid set in the start of pg_log, and resets its value. It\n> records the current oid and xid from pg_variable. It then clears the\n> lock, and starts reading from the minimum recorded xid changed to the\n> end of pg_log_live, and copies it into allocated memory. It then does a\n> sync (twice?), waits for completion, and then writes the pg_log_live\n ^^^^^\nman sync:\n\n The sync() function forces a write of dirty (modified) buffers in the\n ^^^^^^\n block buffer cache out to disk...\n...\n\nBUGS\n Sync() may return before the buffers are completely flushed.\n\nVadim\n\n---------------------------------------------------------------------------\n\nFrom: Bruce Momjian <[email protected]>\nMessage-Id: <[email protected]>\nSubject: [HACKERS] Re: Bufferd loggins/pg_log\nTo: [email protected] (Vadim B. Mikheev)\nDate: Mon, 17 Nov 1997 08:46:29 -0500 (EST)\nCc: [email protected] (PostgreSQL-development)\nIn-Reply-To: <[email protected]> from \"Vadim B. Mikheev\" at Nov 17, 97 02:56:05 pm\nX-Mailer: ELM [version 2.4 PL25]\nMIME-Version: 1.0\nContent-Type: text/plain; charset=US-ASCII\nContent-Transfer-Encoding: 7bit\nSender: [email protected]\nPrecedence: bulk\nStatus: OR\n\n> \n> Bruce Momjian wrote:\n> > \n> > On startup, the postmaster makes a copy of pg_log, called pg_log_live.\n> > Each postgres backend mmaps() this new file into its address space. A\n> > lock is gotten to make changes to the file. All backend use pg_log_live\n> > rather than pg_log. Only the postmaster write to pg_log. (I will\n> > someday remove the exec() from postmaster, so backends will get this\n> > address space automatically.)\n> \n> What are advantages of mmaping entire pg_log over \"online\" pg_log\n> pages ?\n> pg_log may be very big (tens of Mb) - why we have to spend\n> process address space for tens of Mb of mostly unused data ?\n> Also, do all systems have mmap ?\n\nI believe you are correct that it would be better keeping the last few\npages of pg_log in shared memory rather than using mmap().\n\nI think the important new ideas are keeping track of the oid/xid before\nsync so we can accurately add 10,000 after a crash.\n\nI am a little foggy on race condiions of growing the pg_log region while\nother backends are running, and modifying non-shared memory pages, but\nyou seem to have a handle on it.\n\nWe don't need pg_log_live if only the postmaster writes those last two\npages to pg_log, and if we keep track of a crash status somewhere else,\nperhaps at the start of pg_log.\n\n> \n> > \n> > Every 60-90 seconds, the postmaster gets a write lock on pg_log, takes\n> > the minimum xid set in the start of pg_log, and resets its value. It\n> > records the current oid and xid from pg_variable. It then clears the\n> > lock, and starts reading from the minimum recorded xid changed to the\n> > end of pg_log_live, and copies it into allocated memory. It then does a\n> > sync (twice?), waits for completion, and then writes the pg_log_live\n> ^^^^^\n> man sync:\n> \n> The sync() function forces a write of dirty (modified) buffers in the\n> ^^^^^^\n> block buffer cache out to disk...\n> ...\n> \n> BUGS\n> Sync() may return before the buffers are completely flushed.\n> \n> Vadim\n> \n\nMy BSD/OS doesn't mention this, but twice is a good idea.\n\n\n\n-- \nBruce Momjian\[email protected]\n\n\n\n\n-- \nBruce Momjian | 830 Blythe Avenue\[email protected] | Drexel Hill, Pennsylvania 19026\n + If your life is a hard drive, | (610) 353-9879(w)\n + Christ can be your backup. | (610) 853-3000(h)\n\n\n\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 19 Jan 1999 01:01:14 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] performance test"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> > You've seen disadvantages of our unperfect buffer manager -:)\n> > When server need in buffer for new data comming and there is\n> > no unused buffers in pool (i.e. - all buffers filled with new\n> > data and marked as dirty), server gets some dirty buffer,\n> > writes it AND FSYNC file. So, server does many fsyncs\n> > even with BEGIN/END while should do _one_ fsync at COMMIT.\n> >\n> > Having this problem fixed you wouldn't had so big difference\n> > between -F and BEGIN/END-no-F.\n> \n> Is this idea still reasonable?\n> \n> ---------------------------------------------------------------------------\n> \n> Here is an archive of the pg_log discussion.\n\nThanks for archive, Bruce.\nBut delayed fsync is another idea.\nbufmgr has to be fixed anyway.\n\nVadim\n",
"msg_date": "Tue, 19 Jan 1999 13:13:14 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] performance test"
},
{
"msg_contents": "On Tue, Jan 19, 1999 at 10:27:50AM +0700, Vadim Mikheev wrote:\n> You've seen disadvantages of our unperfect buffer manager -:)\n> When server need in buffer for new data comming and there is\n> no unused buffers in pool (i.e. - all buffers filled with new\n> data and marked as dirty), server gets some dirty buffer,\n> writes it AND FSYNC file. So, server does many fsyncs\n> even with BEGIN/END while should do _one_ fsync at COMMIT.\n> \n> Having this problem fixed you wouldn't had so big difference\n> between -F and BEGIN/END-no-F.\n\nI see. But without BEGIN/END a huge difference remains to be expected,\ndoesn't it?\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Tue, 19 Jan 1999 11:13:01 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] performance test"
},
{
"msg_contents": "Michael Meskes wrote:\n> \n> On Tue, Jan 19, 1999 at 10:27:50AM +0700, Vadim Mikheev wrote:\n> > You've seen disadvantages of our unperfect buffer manager -:)\n> > When server need in buffer for new data comming and there is\n> > no unused buffers in pool (i.e. - all buffers filled with new\n> > data and marked as dirty), server gets some dirty buffer,\n> > writes it AND FSYNC file. So, server does many fsyncs\n> > even with BEGIN/END while should do _one_ fsync at COMMIT.\n> >\n> > Having this problem fixed you wouldn't had so big difference\n> > between -F and BEGIN/END-no-F.\n> \n> I see. But without BEGIN/END a huge difference remains to be expected,\n> doesn't it?\n\nOf 'course! Two fsync per commit...\n\nVadim\n",
"msg_date": "Wed, 20 Jan 1999 09:04:39 +0700",
"msg_from": "Vadim Mikheev <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] performance test"
}
] |
[
{
"msg_contents": "New interface. IMHO, not as nice as pgaccess.\n\n\thttp://home.primus.baynet.de/mgeisler/kpgsql/\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 12:17:48 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "KPGsql"
},
{
"msg_contents": "\nOld interface...been around since almost the day that the first CD went\nout, and hasn't been worked on since, either...or, at least, nothing new\nhas come out :(\n\nOn Mon, 18 Jan 1999, Bruce Momjian wrote:\n\n> New interface. IMHO, not as nice as pgaccess.\n> \n> \thttp://home.primus.baynet.de/mgeisler/kpgsql/\n> \n> -- \n> Bruce Momjian | http://www.op.net/~candle\n> [email protected] | (610) 853-3000\n> + If your life is a hard drive, | 830 Blythe Avenue\n> + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n> \n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 18 Jan 1999 15:04:46 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] KPGsql"
},
{
"msg_contents": "On Mon, Jan 18, 1999 at 03:04:46PM -0400, The Hermit Hacker wrote:\n> Old interface...been around since almost the day that the first CD went\n> out, and hasn't been worked on since, either...or, at least, nothing new\n> has come out :(\n\nIs there a web site where we list URLs for tools like this?\n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Tue, 19 Jan 1999 09:33:03 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] KPGsql"
},
{
"msg_contents": "> On Mon, Jan 18, 1999 at 03:04:46PM -0400, The Hermit Hacker wrote:\n> > Old interface...been around since almost the day that the first CD went\n> > out, and hasn't been worked on since, either...or, at least, nothing new\n> > has come out :(\n> \n> Is there a web site where we list URLs for tools like this?\n\nIt is:\n\n\thttp://home.primus.baynet.de/mgeisler/kpgsql/\n\nCheck out www.xshare.com. That's where I got this info. Xshare has\nPostgreSQL, and tons of other open source software links.\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Tue, 19 Jan 1999 15:46:11 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] KPGsql"
},
{
"msg_contents": "On Tue, Jan 19, 1999 at 03:46:11PM -0500, Bruce Momjian wrote:\n> > Is there a web site where we list URLs for tools like this?\n> .. \n> Check out www.xshare.com. That's where I got this info. Xshare has\n> PostgreSQL, and tons of other open source software links.\n\nI will, but I meant a playe on www.postgresql.org where we list all tools we\nknow about that are useful in working with postgresql. \n\nMichael\n-- \nMichael Meskes | Go SF 49ers!\nTh.-Heuss-Str. 61, D-41812 Erkelenz | Go Rhein Fire!\nTel.: (+49) 2431/72651 | Use Debian GNU/Linux!\nEmail: [email protected] | Use PostgreSQL!\n",
"msg_date": "Wed, 20 Jan 1999 10:02:37 +0100",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] KPGsql"
},
{
"msg_contents": "On Wed, 20 Jan 1999, Michael Meskes wrote:\n\n> On Tue, Jan 19, 1999 at 03:46:11PM -0500, Bruce Momjian wrote:\n> > > Is there a web site where we list URLs for tools like this?\n> > .. \n> > Check out www.xshare.com. That's where I got this info. Xshare has\n> > PostgreSQL, and tons of other open source software links.\n> \n> I will, but I meant a playe on www.postgresql.org where we list all tools we\n> know about that are useful in working with postgresql. \n\nRight now, just the CD page has links from it...the 'third-party' page\nneeds to be majorly overhauled...the page should work, bu tthe categories\nare in-complete...will try and re-hit it this coming weekend...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Wed, 20 Jan 1999 15:16:14 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] KPGsql"
}
] |
[
{
"msg_contents": "> Magnus Hagander <[email protected]> writes:\n> > I don't know, maybe recv() is more expensive than fgetc()?\n> \n> Vastly.\nOk. That pretty much says it all :-)\n<snip>\n\n> What we need to do here is to re-introduce the buffering ability of\n> stdio into backend libpq. If you compare the current frontend libpq,\n> you'll notice that it reads or writes the socket a bufferload \n> at a time,\n> not a character at a time.\n\nOk. I'll take a look at that as soon as I get the time (shouldn't be long).\n\n//Magnus\n",
"msg_date": "Mon, 18 Jan 1999 19:33:05 +0100",
"msg_from": "Magnus Hagander <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Postgres Speed or lack thereof "
}
] |
[
{
"msg_contents": "Many 'wish' scripts use this trick so it does not have to hard-code the\npath into the first line:\n \n #!/bin/sh\n # the next line restarts using wish \\\n exec wish \"$0\" \"$@\"\n \nCan you add this to your script, or should I change configure so it \nlooks in tkConfig.sh for TK_PREFIX?\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 14:27:31 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "wish path in pgaccess"
}
] |
[
{
"msg_contents": "If I start pgaccess, it works. I can select tables and see the data.\n\nHowever, if I start pgaccess, and start clicking the left buttons, when\nI reach 'functions' I get the hourglass icon. The mouse is still\nactive, though.\n\nCan you reproduce that there?\n\n-- \n Bruce Momjian | http://www.op.net/~candle\n [email protected] | (610) 853-3000\n + If your life is a hard drive, | 830 Blythe Avenue\n + Christ can be your backup. | Drexel Hill, Pennsylvania 19026\n",
"msg_date": "Mon, 18 Jan 1999 14:33:33 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "pgaccess bug report"
}
] |
[
{
"msg_contents": "Is there row-level locking available yet? If not, is anyone working on it? It\nwill help CORBA support a lot...\n\nTaral\n",
"msg_date": "Mon, 18 Jan 1999 14:18:23 -0600",
"msg_from": "Taral <[email protected]>",
"msg_from_op": true,
"msg_subject": "CORBA: row-level locking available?"
},
{
"msg_contents": "On Mon, 18 Jan 1999, Taral wrote:\n\n> Is there row-level locking available yet? If not, is anyone working on it? It\n> will help CORBA support a lot...\n\nSomeoone have a backlog of email he has to catch up on? :)\n\nrow-level locking is not going to be implemented, as far as I know,\never...we have MVCC implemented instead, wich, from my understanding, is\na better way of doing pretty much the same thing...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Mon, 18 Jan 1999 16:26:33 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] CORBA: row-level locking available?"
}
] |
[
{
"msg_contents": "Najprej bi se vam rad opravi�il, ker sem za na�in obve��anja javnosti uporabil nenaro�eno po�to (angl. spam), ki ste jo nekateri prejeli tudi ve�krat. Na izbiro pa nisem imel druge mo�nosti, saj me slovenski javni mediji niso podprli. Javnim medijem sem poslal svoje prvo pismo, ki je v celoti objavljeno tudi v tem (v drugem delu pisma).\nSlovensko javnost bi �elel upozoriti na krivice, ki so zapisane v na�i ustavi in zakonih. Opozoril bi na krivico, ki doleti vse mo�ke - slu�enje voja�kega roka. Prosim, da moje pisanje vzamete skrajno resno, saj lahko skupaj to dr�avo naredimo res demokrati�no. In kaj �e se vas to ne ti�e? �e ste �e odslu�ili voja�ki rok? �e niste mo�kega spola? Potem pomislite na svoje otroke. �e pa jih �e imate in �e so dovolj stari, pa jih vpra�ajte, �e si oni �elijo slu�iti voja�ki rok. Pogovarjal sem se tudi z �e precej mo�kimi, ki so �e odslu�ili voja��ino, pa so rekli, da jim ni bilo ni� hudega. Pa ne vem, koliko so bili njihovi odgovori iskreni, saj so na vpra�anje, �e bi �li slu�iti voja��ino �e enkrat, kot iz topa vsi odgovorili: \"Ne.\".\nTo pismo je bilo poslano na pribli�no 50000 naslovov elektronske po�te slovenskih dr�avljanov. Prosim vas, da ga posredujete naprej, da bo za to izvedelo �im ve� ljudi. �e pa je v va�i mo�i, da ga objavite v javnem mediju, ali na kak�en drug na�in pomagate pri spreminjanju ustave, se vam tudi za to �e v naprej najlep�e zahvaljujem. �e bi radi stopili v stik z mano, mi lahko pi�ete na elektronski naslov [email protected].\nSre�no novo leto 1999 in upam, da bomo do novega tiso�letja �iveli �e v pravi�nej�i dr�avi.\n\nPozdravlja vas\n\t\tVoja�ki Obveznik\n\nV nadaljevanju pisma je objavljena tudi ustavna prito�ba, ki je bila poslana na dva elektronska naslova ustavnega sodi��a ([email protected], [email protected]). Ustavna prito�ba mora biti vlo�ena pisno, nikjer pa ne pi�e, da je ne morem vlo�iti s pomo�jo elektronske po�te. Javno jo objavljam zato, ker bo brez va�e podpore enostavno presli�ana.\n\n---------------------------------------------------------------------\n\nUSTAVNA PRITO�BA\n\nVlagam ustavno prito�bo, saj menim, da je v ustavo in zakone zapisana krivica, ki se izvr�uje nad vsemi mo�kimi dr�avljani.\n\nV ustavi je zapisano:\n14. �len: (enakost pred zakonom) V Sloveniji so vsakomur zagotovljene enake �lovekove pravice in temeljne svobo��ine, ne glede na narodnost, raso spol, jezik, vero, politi�no ali drugo prepri�anje, gmotno stanje, rojstvo, izobrazbo, dru�beni polo�aj ali katerokoli drugo osebno okoli��ino.\n32. �len: (svoboda gibanja) Vsakdo ima pravico, da se prosto giblje in si izbira prebivali��e, da zapusti dr�avo in se vanjo kadarkoli vrne. Ta pravica se sme omejiti z zakonom, vendar samo, �e je to potrebno, da bi se zagotovil potek kazenskega postopka, da bi se prepre�ilo �irjenje nalezljivih bolezni, se zavaroval javni red, ali �e to zahtevajo interesi obrambe dr�ave. Tujcem se na podlagi zakona lahko omeji vstop v dr�avo in �as bivanja v njej.\n53. �len: (zakonska zveza in dru�ina) Zakonska zveza temelji na enakopravnosti zakoncev. Sklene se pred pristojnim dr�avnim organom. Zakonsko zvezo in pravna razmerja v njej, v dru�ini in v zunajzakonski skupnosti ureja zakon. Dr�ava varuje dru�ino, materinstvo, o�etovstvo, otroke in mladino ter ustvarja za to varstvo potrebne razmere. \n\nV zakonu o voja�ki dol�nosti je zapisano:\n1. �len: Voja�ko dol�nost izvr�ujejo v miru in v vojni mo�ki dr�avljani pod pogoji, ki jih dolo�a ta zakon.\nPri izvr�evanju voja�ke dol�nosti v skladu s tem zakonom lahko sodelujejo tudi �enske.\n31. �len: Slu�enje voja�kega roka se na nabornikovo pro�njo odlo�i:\n- 1. odstavek, 7. to�ka: naborniku, ki po predpisih, ki urejajo varstvo dru�in oseb na obvezni voja�ki slu�bi, izpolnjuje pogoje za pridobitev statusa edinega hranilca dru�ine, �e bi z njegovim odhodom na slu�enje voja�kega roka njegova du�ina za�la v te�ak materialni polo�aj, dokler te okoli��ine trajajo, najdlje pa do konca septembra koledarskega leta, v katerem dopolnijo 27 let;\n38. �len: Nabornik, ki nasprotuje uporabi oro�ja v vseh okoli��inah, lahko uveljavlja ugovor vesti voja�ki dol�nosti in voja�ki rok slu�i brez oro�ja ali opravi nadomestno civilno slu�bo.\nNabornik lahko ugovor vesti voja�ki slu�bi uveljavlja zaradi religioznih, filozofskih ali humanitarnih razlogov. Razloge, zaradi katerih nabornik uveljavlja ugovor vesti, mora potrjevati splo�ni na�in �ivljenja in ravnanja nabornika.\nUgovor vesti voja�ki dol�nosti pod pogoji iz prvega in prej�njega odstavka lahko uveljavlja tudi voja�ki obveznik med slu�enjem ali po slu�enju voja�kega roka v skladu s tem zakonom. Dr�avljan, ki mu je priznan ugovor vesti voja�ki dol�nosti po odslu�enem voja�kem roku, je dol�an opraviti 30 dnevno usposabljanje za opravljanje nalog za��ite in re�evanja.\nProgram usposabljanja za opravljanje nalog za��ite in re�evanja iz prej�njega odstavka predpi�e minister, pristojen za varstvo pred naravnimi in drugimi nesre�ami.\n\nKot pravi 14. �len ustave, so v Sloveniji vsakomur zagotovljene enake �lovekove pravice in temeljne svobo��ine. S 1. �lenom zakona o voja�ki dol�nosti pa dr�ava dela razlike med mo�kim in �enskim spolom, saj voja�ko dol�nost izvajajo v miru in vojni mo�ki dr�avljani, skladno z zakonom pa lahko sodelujejo tudi �enske.\n38. �len zakona o voja�ki dol�nosti omogo�a tudi civilno slu�enje voja�kega roka, kar pa ni ni� drugega, kot prisilno delo, saj mora delo posameznik opravljati proti svoji volji in brez ustreznega finan�nega nadomestila. Dr�ava s tem priznava, da je prostovoljnih vojakov premalo, �e ne kar ni�.\n53. �len ustave pravi, da dr�ava varuje dru�ino, v 31. �lenu zakona o voja�ki dol�nosti pa pi�e, da se slu�enje voja�kega roka lahko odlo�i, �e je voja�ki obveznik edini hranilec dru�ine, zaradi tega pa bi njegova dru�ina za�la v te�ak materialni polo�aj, vendar najkasneje do konca septembra. In kako varuje dr�ava dru�ino po septembru?\n\nVsakemu �loveku, ki proti svoji volji slu�i voja�ki rok (kar po na�i zakonodaji ni mogo�e za �enske), so storjene naslednje krivice:\n- omejevanje svobode;\n- vsiljevanje proti volji posameznika;\n- izguba prihodka.\nTorej problema ni mogo�e re�iti na na�in, da bi uvedli obvezno slu�enje voja��ine tudi za �enske, saj bi s tem �tevilo ljudi, katerim je bila storjena krivica, le pove�ali. Zato predlagam slede�e spremembe v ustavi in zakonodaji na�e dr�ave:\n- vsak posameznik se lahko prostovoljno odlo�i za slu�enje voja�kega roka;\n- dr�ava organizira profesionalno vojsko.\n\nPredlagam pa �e naslednjo re�itev, kako financirati profesionalno vojsko. Dr�ava lahko uvede poseben davek na NEslu�enje voja�kega roka, katerega bi pla�evali vsi dr�avljani (ne glede na spol), ki (�e) niso slu�ili voja�kega roka. Ta re�itev je pravi�na, saj je dr�avljan, ki se je odlo�il za slu�enje voja�kega roka, utrpel izgubo prihodka - se pravi je svoj �as (�as pa je denar), poklonil dr�avi in tako ta davek pla�al v enem kosu. Na ta na�in bi vsi dr�avljani prispevali k obarmabi dr�ave - na tak ali druga�en na�in, imeli pa bi svobodno voljo izbire. Seveda pa tisti, ki prostovoljno odslu�ijo voja�ki rok, ostanejo v rezervni sestavi vojske in so zaradi tega do�ivljensko opro��eni pla�evanja tega davka.\nNa ta na�in bi sredstva za potrebe profesionalne vojske porazdelili na ramena vseh dr�avljanov (in to na obroke in ne v enem kosu), razen na tiste, ki se odlo�ijo za slu�enje voja�kega roka (oziroma so ga �e odslu�ili) - za te pa novi zakon ne bi predstavljal nobene razlike glede na sedanjega.\nNa ta na�in torej ne bi posamezniku ni�esar vsiljevali, njegova svoboda ne bi bila omejena (kdor bi se odlo�il za slu�enje voja��ine, bi se prostovoljno), financirali pa bi vojsko vsi dr�avljani enakomerno. Dr�ava bi imela profesionalno vojsko, ki bi bila toliko ve�ja, kolikor manj dr�avljanov bi se odlo�ilo za prostovoljno slu�enje - sredstva pa bi bila zagotovljena, saj bi se s tem avtomatsko pove�alo tudi �tevilo pla�evalcev davka na neslu�enje voja�kega roka.\n\n1.1.1999 v Sloveniji, Voja�ki Obveznik\n---------------------------------------------------------------------\n\n\nPrilagam �e svoje prvo pismo, s katerim sem �elel javnost obvestiti s pomo�jo javnih medijev, ki pa mojega pisanja niso vzeli ravno resno.\n\n\n\nTo pismo pi�em z namenom, da opozorim na krivice v na�i dr�avi, saj je �e skrajni �as, da se glede tega kaj ukrene.\nPokazal bi rad na diskriminacijo spolov, in na nasprotja zapisana v na�i ustavi.\n\nUstava pravi:\n1. �len: Slovenija je demokrati�na republika.\n14. �len: (enakost pred zakonom) V Sloveniji so vsakomur zagotovljene enake �lovekove pravice in temeljne svobo��ine, ne glede na narodnost, raso spol, jezik, vero, politi�no ali drugo prepri�anje, gmotno stanje, rojstvo, izobrazbo, dru�beni polo�aj ali katerokoli drugo osebno okoli��ino.\n32. �len: (svoboda gibanja) Vsakdo ima pravico, da se prosto giblje in si izbira prebivali��e, da zapusti dr�avo in se vanjo kadarkoli vrne. Ta pravica se sme omejiti z zakonom, vendar samo, �e je to potrebno, da bi se zagotovil potek kazenskega postopka, da bi se prepre�ilo �irjenje nalezljivih bolezni, se zavaroval javni red, ali �e to zahtevajo interesi obrambe dr�ave. Tujcem se na podlagi zakona lahko omeji vstop v dr�avo in �as bivanja v njej.\n53. �len: (zakonska zveza in dru�ina) Zakonska zveza temelji na enakopravnosti zakoncev. Sklene se pred pristojnim dr�avnim organom. Zakonsko zvezo in pravna razmerja v njej, v dru�ini in v zunajzakonski skupnosti ureja zakon. Dr�ava varuje dru�ino, materinstvo, o�etovstvo, otroke in mladino ter ustvarja za to varstvo potrebne razmere. \n\nNa vse �lene se bom skliceval v nadaljevanju tega pisma, v nasprotju pa sta si �e 14. in 32. �len, saj naj bi bili v dr�avi vsi ljudje enakopravni, nekaterim ljudem pa se omeji pravica gibanja, �e je to v interesu obrambe dr�ave. In komu se omejijo te pravice - najpogostej�i primer so vojaki, ki slu�ijo voja�ki rok. Kot pravi zakon, voja��ino morajo slu�iti mo�ki, �enske pa jo lahko. Torej si spola v na�i dr�avi nista enakopravna. Mo�kim je tako odvzeta svoboda gibanja (in �e katera druga), pri �emer je edini zlo�in, ki so ga storili le ta, da so se rodili kot mo�ki.\nSlu�enje voja��ine je:\n- omejevanje svobode;\n- vsiljevanje proti volji posameznika (kar sklepam po tem, da je slu�enje voja��ine omogo�eno �enskam, pa jih ravno ne sre�ujemo mnogo po voja�nicah);\n- izguba prihodka posameznika - gmotna �koda (53. �len ustave zavezuje dr�avo k varstvu dru�ine, ki kaj hitro lahko postane socialno ogo�ena pri izgubi morda edinega prihodka v dru�ini).\n\nPri sprejemanju ustave ni bila upo�tevana volja manj�ine - voja�kih obveznikov, ki po ve�ini no�ejo slu�iti voja�kega roka. �e dr�i 1. �len ustave in z njim demokracija, bi morali upo�tevati tudi voljo te manj�ine. Dr�ava skrbi za etni�ne manj�ine, za starostne pa ne.\n\nKer bi s tem, da za �enske uvedemo slu�enje voja��ine (konec koncev je tudi civilno slu�enje le slab izgovor dr�ave za prisilno delo njenih mo�kih prebivalcev, saj tudi �enske zmorejo opravljati veliko ve�ino teh del), pridobili enakopravnost dr�avljanov, pa zlo�in nad posameznikom ne bi bil ni� manj�i. Re�itvi za to situacijo sta torej le dve (ali kombinacija obeh):\n- vsak posameznik se lahko prostovoljno odlo�i za slu�enje voja�kega roka;\n- profesionalna vojska.\n\nNaj poudarim, da v �asu pred Napoleonom ni bilo obvezne voja��ine, torej vedno ni bilo tako in nikjer ne pi�e, da mora vedno ostati tako. Prav tako je tudi sklicevanje na druge dr�ave, kjer imajo podobna dolo�ila v ustavi, le slab izgovor, saj obstajajo tudi dr�ave s smrtno kaznijo, pa se ne zgledujemo po njih. Raje se zgledujmo po dr�avah, ki �e imajo profesionalno vojsko.\nSeveda imamo v Sloveniji tudi varuha �lovekovih pravic in zakon o varuhu �lovekovih pravic, ki pa je napisan s figo v �epu, kar je logi�no, saj mora delovati v interesu dr�ave, ne pa �loveka, kateremu je namenjen.\nGotovo pa je najve�ji krivec za zakon, ki diskriminira mo�ke, �kodo�eljnost ljudi. Zakone in ustavo so pisali po ve�ini mo�ki, ki so vsi �e odslu�ili voja�ki rok. In \"�e sem jaz moral skozi to, potem naj gredo �e drugi\".\n\nKer se v na�i dr�avi vse prepo�asi dogajajo spremembe, sem se odlo�il za boj proti krivicam v ustavi s pomo�jo:\n- obve��anja prek javnih medijev;\n- ustavne prito�be;\n- �tudentskega parlamenta;\n- �tudentskih demonstracij;\n- mno�i�nega bojkota slu�enja voja��ine.\n\nMoram priznati, da sem tudi sam voja�ki obveznik, saj se ljudje le redko lotimo re�evanja problemov, ki se nas ne ti�ejo. Morda je razlog za to prekratko bivanje posameznika na tem svetu.\n\nPismo sem poslal na naslednje naslove, saj je v rokah vseh vas mo� spreminjanja dr�ave:\n\n- Dija�ka organizacija Slovenije ([email protected])\n- �tudentska organizacija Slovenije, �tudentska vlada, �tudentski parlament (upravitelju [email protected], ki ga prosim, da to pismo posreduje predsednikom teh organizacij)\n- Ustavno sodi��e ([email protected])\n- Ministrstvo za obrambo ([email protected])\n- Vladi RS, Parlamentu RS, g. Milanu Ku�anu, g. Janezu Drnov�ku, Ministrstvu za pravosodje (upraviteljema [email protected], [email protected], katera prosim, da jim to pismo posredujeta)\n- Varuhu �lovekovih pravic (upravitelju [email protected], ki ga prosim, da to pismo posreduje g. Ivu Bizjaku)\n\n\n�ASOPISI:\n- Ve�er ([email protected])\n- Tednik Dru�ina ([email protected])\n- Slovenske Novice ([email protected])\n- Delo (upravitelju [email protected], ki ga prosim, da to pismo posreduje glavnemu uredniku)\n- Dnevnik ([email protected])\n- Mladina ([email protected])\n\nRADIJSKE POSTAJE:\n- Mar� ([email protected])\n- Radio Ognji��e ([email protected])\n- Radio �tudent (upravitelju [email protected], ki ga prosim, da to pismo posreduje glavnemu uredniku)\n- Gama MM ([email protected])\n\nTELEVIZIJSKE POSTAJE:\n- RTV Slovenija ([email protected])\n- POP TV ([email protected])\n- Gajba TV ([email protected])\n\nPOLITI�NIE STRANKE:\n- Liberalna stranka ([email protected])\n- Republikanci Slovenije ([email protected])\n- Slovenski kr��anski demokrati ([email protected])\n- Socialdemokratska stranka Slovenije ([email protected])\n- Liberalna demokracija Slovenije ([email protected])\n- Nova stranka ([email protected])\n- Demokrati Slovenije ([email protected])\n- Zdru�ena lista socialnih demokratov ([email protected])\n\nVoja�ki Obveznik\n\n",
"msg_date": "Mon, 18 Jan 1999 21:26:46 +0100 (GMT+01:00)",
"msg_from": "[email protected]",
"msg_from_op": true,
"msg_subject": "Krivice zapisane v slovenski ustavi"
}
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.