threads
listlengths 1
2.99k
|
|---|
[
{
"msg_contents": "\n>> This might lead the large object implementation to confuse\n>> large object relations with other relations.\n>>\n>> According to me this is a bug. I'm I right?\n>\n>Yes, and no. LargeObjects are supposed to run within a transaction (if you\n>don't then some fun things happen), and (someone correct me if I'm wrong)\n>if newoid() is called from within the transaction, it is safe?\n>\nI see no evidence in the code that suggests that it is safe in transactions.\nThe GetNewObjectIdBlock() function which generates the OID blocks _does_\nacquire a spinlock before it generates a new block of oids so usually all\nwill be well.\nBut sometimes ((a chance of <usercount>/32) when there <usercount> active\nusers\nfor the same db) the newoid might have a quite different value than\nfileoid+1.\n\nAgain I see no evidence in the code that it is safe in transactions. I only\nsee evidence that it will _usually_ work.\n\nyes, but currently it is very hard to produce this behavior, since we still only have table locks.\nYou would need more than 32 lob tables, accessed concurrently (not sure on that) ?\nThis area has to be specially watched when page, or row locks get implemented.\n\nActually I wonder how it could be efficiently made safe within transactions\ngiven\nthat the oids generated are guaranteed to be unique within an\n_entire_ postgres installation. This would seem to imply that, effectively,\nonly one transaction would be possible at the same time in an entire\npostgresql database.\n\nI think this is why a lot of us (hands up) want to reduce the use of oid's in user tables,\nuser tables would only have oid's iff the table is created with 'with oid'.\nPer default normal user tables would not have oid's. I strongly support this as a strategy.\n\nMy current strategy to solve this problem involves the use of a new\nsystem catalog which I call pg_large_object. This catalog contains\ninformation about each large object in the system.\n\nhmmm ... another bottleneck ? one table many users ?\n\nCurrently the information maintained is:\n- identification of heap and index relations used by the large_object\n- the size of the large object\n- information about the type of the large object.\nI still need to figure out how to create a new _unique_ index on a system\ncatalog using information in the indexing.h file.\n\nI would propose a strategy, where the large object is referenced by a physical position (ctid)\nand is stored in one file per lob column. You have to always remember, that filesystems\nonly behave well if they have less than xx members per directory xx usually beeing between 1000 - 25000\nMore members per directory will get file stat times of 20 ms and more, not to forget about\nthe many open files. While it is hard to have 20000+ tables it is easy to have millions of rows,\ndefinitely too much for one directory file (this is not OS specific).\nI would also suggest to hard link large objects to an owning row. Meaning, if the row is deleted\nthe lob is also deleted. I would not make this a trigger issue at the user, or type programmer level,\nbut handle it generically in the backend. Writing a lob type is hard enough to not make it \neven more complex.\n\nGiven an oid this table allow us to determine if it is a valid large object.\nI think this is necesary (to be able to maintain referential integrity) if\nwe're ever\ngoing to have large object type.\n\nSimilarly I have defined a table pg_tuple which allows one to\ndetermine if a given oid is a valid tuple.\n\nplease remember, that a la long not all user tuples can have oids. This would always be\na major performance problem.\n\nThis together with some other minor changes allows some cool\nobject oriented features for postgresql.\n\nYes, definitely. I don't know how to resolve my inner conflict on the two seemingly contrary issues,\nperformance versus OO features. \n\nFancy the idea of persistent Java object which live in postgresql databases?\n\nAnyway if it all works as expected I'll submit some patches.\n\nThanks,\nMaurice\n\nAndreas\n\n\n\n",
"msg_date": "Mon, 9 Mar 1998 12:43:29 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] newoid in invapi.c"
}
] |
[
{
"msg_contents": "Hi all,\n\nat work I´m using RDB 6.2. I can create triggers with the following\ncommand:\n\nAFTER UPDATE OF\n SIM_ICCV_L,\n SIM_ICCNR_L,\n START_DATUM_D,\n STAT_SEIT_D,\n PREISKLASSE_T\nON ent_TEILNEHMER REFERENCING OLD AS update_old NEW AS update_new\n ( UPDATE ent_TEILNEHMER t\n SET t.update_l = update_old.update_l + 1,\n t.insert_datum_d = cast(current_timestamp as date)\n WHERE ent_TEILNEHMER.dbkey = t.dbkey )\nFOR EACH ROW;\n\nor:\n\nAFTER INSERT ON ent_TEILNEHMER\n (UPDATE ent_TEILNEHMER t\n SET t.update_l = 1,\n t.insert_datum_d = cast(current_timestamp as date)\n WHERE ent_TEILNEHMER.dbkey = t.dbkey)\nFOR EACH ROW;\n\nWith postgresql 6.2 or 6.3 I must use a user defined function to get the\nsame result. \n\nWill there be some enhancements to triggers to use the above syntax ? I\nthink, RDB is more user friendly ( at least for trigger defintions)...\n\nBye\nAndreas\n",
"msg_date": "Mon, 09 Mar 1998 15:11:53 +0100",
"msg_from": "Andreas Joppich <[email protected]>",
"msg_from_op": true,
"msg_subject": "create trigger question"
},
{
"msg_contents": "\nAndreas Joppich wrote:\n>\n> Hi all,\n>\n> at work I=B4m using RDB 6.2. I can create triggers with the following\n> command:\n>\n> AFTER UPDATE OF\n> SIM_ICCV_L,\n> SIM_ICCNR_L,\n> START_DATUM_D,\n> STAT_SEIT_D,\n> PREISKLASSE_T\n> ON ent_TEILNEHMER REFERENCING OLD AS update_old NEW AS update_new\n> ( UPDATE ent_TEILNEHMER t\n> SET t.update_l =3D update_old.update_l + 1,\n> t.insert_datum_d =3D cast(current_timestamp as date)\n> WHERE ent_TEILNEHMER.dbkey =3D t.dbkey )\n> FOR EACH ROW;\n>\n> or:\n>\n> AFTER INSERT ON ent_TEILNEHMER\n> (UPDATE ent_TEILNEHMER t\n> SET t.update_l =3D 1,\n> t.insert_datum_d =3D cast(current_timestamp as date)\n> WHERE ent_TEILNEHMER.dbkey =3D t.dbkey)\n> FOR EACH ROW;\n>\n> With postgresql 6.2 or 6.3 I must use a user defined function to get the\n> same result. =\n>\n>\n> Will there be some enhancements to triggers to use the above syntax ? I\n> think, RDB is more user friendly ( at least for trigger defintions)...\n\n Now that 6.3 is out, I'm beginning with PL/pgSQL. It will not\n cover the above exactly, but it will allow an SQL like\n procedural language to be used to create the trigger\n procedures so you don't have to write the triggers in 'C'. In\n the meantime you might want to take a look into src/pl/tcl\n where you'll find Tcl as a loadable PL that can do it for\n you.\n\n If PL/pgSQL get's ready someday, I'll take a look at the\n parser if we can enhance it to accept the above syntax and\n automagically create an appropriate trigger procedure.\n\n\nUntil later, Jan\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, 9 Mar 1998 16:08:55 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] create trigger question"
}
] |
[
{
"msg_contents": "> On Mon, 9 Mar 1998, Andrew Martin wrote:\n> \n> > > \tThen, using what was left, I did:\n> > > \n> > > psql -e < db.dump \n> > > \n> > > \tAgain. Failed misearbly, with the following coming from the 'copy\n> > > to <relname> from stdin;' section:\n> > > \n> > > 344984 johnb xgSldZdYEgIWo clio.trends.ca n \\N\n> > \n> > I got something kind-of similar with a core dump... This was 'cos of a \n> > column name which is now a reserved word.\n> \n> \tAck, I fear you are correct...the third field above is 'password',\n> which became a reserved word in v6.3...oh man, is this upgrade ever going\n> to hurt...my 'db.dump' file is 84Meg...vi just loves it :)\n> \n> \nI do think it would be a good idea if the pg_dumpall from the new release\nlooked for these now-reserved column names and warned the user. You could\nthen at least know where problems will occur. In upgrades from d/b versions\nwhich support ALTER TABLE xxx RENAME COLUMN yyy TO zzz, you would actually\nbe able to do something about it before re-dumping...\n\nAndrew\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) [email protected] (Home) [email protected]\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n",
"msg_date": "Mon, 9 Mar 1998 14:26:49 GMT",
"msg_from": "Andrew Martin <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] How to...?"
},
{
"msg_contents": "> I do think it would be a good idea if the pg_dumpall from the new release\n> looked for these now-reserved column names and warned the user. You could\n> then at least know where problems will occur. In upgrades from d/b versions\n> which support ALTER TABLE xxx RENAME COLUMN yyy TO zzz, you would actually\n> be able to do something about it before re-dumping...\n\npg_dumpall is just a shell script. You would have to do it in pg_dump. \nI hesistate to meddle with that code unless I have good reason for it. \nWe could have a separate function that scans pg_attribute looking for\nbad column names.\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",
"msg_date": "Mon, 9 Mar 1998 10:14:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] How to...?"
}
] |
[
{
"msg_contents": "-----Original Message-----\nFrom: Zeugswetter Andreas <[email protected]>\nTo: 'Maurice Gittens' <[email protected]>; '[email protected]'\n<[email protected]>\nDate: maandag 9 maart 1998 18:58\nSubject: AW: [HACKERS] newoid in invapi.c\n\n\n>\n>>> This might lead the large object implementation to confuse\n>>> large object relations with other relations.\n>>>\n>>> According to me this is a bug. I'm I right?\n>>\n>>Yes, and no. LargeObjects are supposed to run within a transaction (if you\n>>don't then some fun things happen), and (someone correct me if I'm wrong)\n>>if newoid() is called from within the transaction, it is safe?\n>>\n>I see no evidence in the code that suggests that it is safe in\ntransactions.\n>The GetNewObjectIdBlock() function which generates the OID blocks _does_\n>acquire a spinlock before it generates a new block of oids so usually all\n>will be well.\n>But sometimes ((a chance of <usercount>/32) when there <usercount> active\n>users\n>for the same db) the newoid might have a quite different value than\n>fileoid+1.\n>\n>Again I see no evidence in the code that it is safe in transactions. I only\n>see evidence that it will _usually_ work.\n>\n>yes, but currently it is very hard to produce this behavior, since we still\nonly have table locks.\nI think this may not be true since I said <usercount> in the above and I\nshould have\nsaid <connection_count>. Multiple (persistent) connections from http daemons\netc. seem likely to be common.\n\n>You would need more than 32 lob tables, accessed concurrently (not sure on\nthat) ?\n\nI don't expect this to be true.\n>This area has to be specially watched when page, or row locks get\nimplemented.\n\n>\n>Actually I wonder how it could be efficiently made safe within transactions\n>given\n>that the oids generated are guaranteed to be unique within an\n>_entire_ postgres installation. This would seem to imply that, effectively,\n>only one transaction would be possible at the same time in an entire\n>postgresql database.\n>\n>I think this is why a lot of us (hands up) want to reduce the use of oid's\nin user tables,\n>user tables would only have oid's iff the table is created with 'with oid'.\n>Per default normal user tables would not have oid's. I strongly support\nthis as a strategy.\n\nAccording to me it is impossible to support general sematics\nfor object orientation without support for the notion of identity.\nSo maybe we'll get rid of the oid and introduce some other \"thing\" with\nalso gives us identity for instances of classes.\nBut IMO no identity equates to no OO. And currently identity is only\nprovided\nby oids, so I would vote to keep them in user tables too.\n\nI expect it wouldn't be trivial to optionally remove oids (this would be\nalmost a rewrite of the system as far as I have seen.)\n\nOne thing we may be able to insure is that the oids be unique in a single\ndatabase instead of in a complete postgresql installation.\nI think this implies that there would be no more sharing of system\ncatalogs between databases. So pg_variable (or whatever it;s called)\nwould be a local database and would pull in all the other global\ndatabases as well. It wouldn't be easily possible to share databases\nbetween users any more.\nThis would be similar to using different postgresql installations for\ndifferent databases.\n\nI must admit that (apart from possible performance issues) I really\nlike the oids.\n>\n>My current strategy to solve this problem involves the use of a new\n>system catalog which I call pg_large_object. This catalog contains\n>information about each large object in the system.\n>\n>hmmm ... another bottleneck ? one table many users ?\nIn general I believe this to be true. But how much will it cost in practice?\n(and we can still be clever can't we?).\n\n>\n>Currently the information maintained is:\n>- identification of heap and index relations used by the large_object\n>- the size of the large object\n>- information about the type of the large object.\n>I still need to figure out how to create a new _unique_ index on a system\n>catalog using information in the indexing.h file.\n>\n>I would propose a strategy, where the large object is referenced by a\nphysical position (ctid)\n>and is stored in one file per lob column. You have to always remember, that\nfilesystems\n>only behave well if they have less than xx members per directory xx usually\nbeeing between 1000 - 25000\n>More members per directory will get file stat times of 20 ms and more, not\nto forget about\n>the many open files. While it is hard to have 20000+ tables it is easy to\nhave millions of rows,\n>definitely too much for one directory file (this is not OS specific).\n>I would also suggest to hard link large objects to an owning row. Meaning,\nif the row is deleted\n>the lob is also deleted. I would not make this a trigger issue at the user,\nor type programmer level,\n>but handle it generically in the backend. Writing a lob type is hard enough\nto not make it\n>even more complex.\nI agree with you where you say that a new file for each large object is not\n\"the right thing\". My experience using the large objects also confirms what\nyou\nare saying. I believe that large objects are broken by design and\nthat some rethinking is needed to do it the right way.\n\nProperly done I expect this all to be integrated with some type of PL\nlanguage\nanyway. There has been talk about this and I hope this new language will\nhave\nOO features.\n>\n>Given an oid this table allow us to determine if it is a valid large\nobject.\n>I think this is necesary (to be able to maintain referential integrity) if\n>we're ever\n>going to have large object type.\n>\n>Similarly I have defined a table pg_tuple which allows one to\n>determine if a given oid is a valid tuple.\n>\n>please remember, that a la long not all user tuples can have oids. This\nwould always be\n>a major performance problem.\nI do not think that this is true because an oid is a so called \"system\nattribute\"\nand all heap tuples in postgresql carry these attributes if the user knows\nit or not.\n\nOn the other hand consider the benefits. IMO this would be the foundation of\na\npersistent (possible OO) language which could be integrated in postgresql.\nTo do this we need object identity.\n>\n>This together with some other minor changes allows some cool\n>object oriented features for postgresql.\n>\n>Yes, definitely. I don't know how to resolve my inner conflict on the two\nseemingly contrary issues,\n>performance versus OO features.\nGood OO designs are very clean. Consider the implementation of postgresql.\nDon't you think it is we'll designed? I see the design and implementation of\npostgresql as evidence that OO _can_ be implemented using C as well.\nI expect C++ would have allowed for an even cleaner implementation not\nin the least because of those virtual functions.\n\nI like postgresql for it's OO features and I hope to be able to enhance it\nin this\narea. Right now there are a number of fundamental OO operations that are\nnot implemented fully.\n\nFor instance:\n1. Mapping from an oid to it's most derived class.\n2. It seems that triggers are not inherited by subclasses. This would allow\nfor\npolymorphism in postgresql. I expect that this feature by it's self\n(incombination\nwith (1)) would make the use of inheritance more abundant.\n3. Allowing for abstract classes (so that I define relations which\nneed not exists as files on disk but are intended as base classes for\ninheritance).\nFor instance in the current implementation all heap tuples have the same\nset of system attributes. This could be made obvious to the OO literate by\nintroducing an abstact class like pg_tuple (or pg_object, etc).\nThis class could then be extended by concrete classes or other abstract\nclasses.\n\nHmm... does CREATE ABSTRACT [CLASS|TABLE] ... look good?\n\n4. A unified namespace for object id's;\nI expect this to allow for some pretty neat features too.\n\n5. One must be able determine if an oid represents a heap tuple object or\nnot.\n\nI have an implementation of (1,5) and I'm trying to implement (4), because\nI found no immediately apparant trivial implementation for (2) and (3).\n\nRegards from Maurice.\n\n\n\n",
"msg_date": "Mon, 9 Mar 1998 16:36:07 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] newoid in invapi.c"
}
] |
[
{
"msg_contents": "Bruce, \n\nCould we not just create a symbolic link from libpq.so -> Current version. \nThen only those who need to use\na different version from the default would have to make any changes.\n\nMatt\n\n\n>\n> At 15:17 -0000 on 5/3/98, The Hermit Hacker wrote:\n>\n>\n> > You should technically be able to run it on a different port,\n> > *but* you might have problems with the shared libraries, where trying \nto > > run v6.3 is seeing v6.1's shared libraries, and won't work...\n>\n> May I make a suggestion? In future versions, include the version number \nin > the names of the libraries.\n\nThen everyone has to update all their Makefiles after an upgrade.\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",
"msg_date": "Mon, 9 Mar 1998 11:15:05 -0500",
"msg_from": "Matt Aycock/IntelliQuest<@pipe.pcpipeline.com>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Testing Postgresql v6.3"
},
{
"msg_contents": "> \n> Bruce, \n> \n> Could we not just create a symbolic link from libpq.so -> Current version. \n> Then only those who need to use\n> a different version from the default would have to make any changes.\n\n*.so is a platform-specific library extension.\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",
"msg_date": "Mon, 9 Mar 1998 11:39:57 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Testing Postgresql v6.3"
},
{
"msg_contents": "On Mon, 9 Mar 1998 [email protected] wrote:\n\n> Bruce, \n> \n> Could we not just create a symbolic link from libpq.so -> Current version. \n> Then only those who need to use\n> a different version from the default would have to make any changes.\n> \n> Matt\n> \n> \n> >\n> > At 15:17 -0000 on 5/3/98, The Hermit Hacker wrote:\n> >\n> >\n> > > You should technically be able to run it on a different port,\n> > > *but* you might have problems with the shared libraries, where trying \n> to > > run v6.3 is seeing v6.1's shared libraries, and won't work...\n> >\n> > May I make a suggestion? In future versions, include the version number \n> in > the names of the libraries.\n> \n> Then everyone has to update all their Makefiles after an upgrade.\n\nHow about setting a so-name? I don't know much about libraries, but AFAIK \nyou can at least give dyn. libs a so-name, ie. libtermcap.so.2.0.8, where \n2.0.8 is the so-name (2 is major, and there are some minor numbers etc.). \nlibraries with the same major so-number are binary compatible (ie. don't \nrequire a relink of your programs).\n\nThere got to be some ppl around here that know more about this stuff then \nI do....\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Mon, 9 Mar 1998 21:27:05 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Testing Postgresql v6.3"
}
] |
[
{
"msg_contents": "> So maybe we'll get rid of the oid and introduce some other \"thing\" with\n> also gives us identity for instances of classes.\n> But IMO no identity equates to no OO. And currently identity is only\n> provided\n> by oids, so I would vote to keep them in user tables too.\n\nIn relational speak a tuple is always identified by it's primary key, which also guarantees\nfast access. The where oid = <value> is only fast if the user defines an index on oid.\nThe extensive use of oid's is also a nightmare for all those that want to reorganize tables.\nThere is simply very much that speaks against the use of oid's a la long.\n\nIllustra defines a \"reference\" maybe we should dig into that ?\ncreate table person (\n\tname char (16),\n\tmother ref(person),\n\tfather ref(person)\n)\n\nA unique pointer to a row for me is always:\ndbid + tableid + fileid + primary key (or even rowid)\n\nI personally like the idea of a physical address as an alterntive to oid. The problem\nwith this is that physical position changes over time. As the past has shown the\nsame problem is also present for oid's. The problem could maybe be solved\nwith a physical position tracking system, that gets reset at vacuum time. Or maybe \nthe existing logic for indexes could be reused in a somewhat modified manner.\n\nAndreas\n\n",
"msg_date": "Mon, 9 Mar 1998 18:19:16 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] newoid in invapi.c"
}
] |
[
{
"msg_contents": "> > I do think it would be a good idea if the pg_dumpall from the new release\n> > looked for these now-reserved column names and warned the user. You could\n> > then at least know where problems will occur. In upgrades from d/b versions\n> > which support ALTER TABLE xxx RENAME COLUMN yyy TO zzz, you would actually\n> > be able to do something about it before re-dumping...\n> \n> pg_dumpall is just a shell script. You would have to do it in pg_dump. \n> I hesistate to meddle with that code unless I have good reason for it. \nHaving psql core dump seems like a pretty good reason to me :-)\n\n> We could have a separate function that scans pg_attribute looking for\n> bad column names.\nWhatever... This could be run as the first stage of pg_dumpall to say\n\"Hey these column names are no longer allowed...\"\n\n\n> Bruce Momjian | 830 Blythe Avenue\n\nAndrew\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) [email protected] (Home) [email protected]\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n",
"msg_date": "Mon, 9 Mar 1998 17:39:20 GMT",
"msg_from": "Andrew Martin <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] How to...?"
},
{
"msg_contents": "> \n> > > I do think it would be a good idea if the pg_dumpall from the new release\n> > > looked for these now-reserved column names and warned the user. You could\n> > > then at least know where problems will occur. In upgrades from d/b versions\n> > > which support ALTER TABLE xxx RENAME COLUMN yyy TO zzz, you would actually\n> > > be able to do something about it before re-dumping...\n> > \n> > pg_dumpall is just a shell script. You would have to do it in pg_dump. \n> > I hesistate to meddle with that code unless I have good reason for it. \n> Having psql core dump seems like a pretty good reason to me :-)\n\nTough to argue with this.\n\n> \n> > We could have a separate function that scans pg_attribute looking for\n> > bad column names.\n> Whatever... This could be run as the first stage of pg_dumpall to say\n> \"Hey these column names are no longer allowed...\"\n> \n> \n> > Bruce Momjian | 830 Blythe Avenue\n> \n> Andrew\n> \n> ----------------------------------------------------------------------------\n> Dr. Andrew C.R. Martin University College London\n> EMAIL: (Work) [email protected] (Home) [email protected]\n> URL: http://www.biochem.ucl.ac.uk/~martin\n> Tel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\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",
"msg_date": "Mon, 9 Mar 1998 13:04:50 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] How to...?"
}
] |
[
{
"msg_contents": "\nRedirected to 'the proper list' - [email protected]\n\nOn Mon, 9 Mar 1998, Bruce Momjian wrote:\n\n> > \n> > WARNING!\n> > \n> > Postgres 6.3 has MAJOR trouble with btree-indexed text fields.\n> > \n> > Performance levels are *10x* worse than the same indexed fields in \"varchar\"\n> > format!\n> > \n> > Be EXTREMELY careful - I got bit in the ass by this this morning, and it was\n> > very fortunate that I figured out what was going on.\n> > \n> > The reason I changed this over was that I had dumped the table and it came\n> > out of the pg_dump program with a negative size. So I figured I'd change it\n> > to TEXT and that would resolve the problem. BIG mistake.\n> > \n> > Be on guard for this folks.\n> > \n> > Developers, you might want to look into this - there's no good reason for\n> > this kind of behavior, is there?\n> \n> No good reason at all. As far as I know, text and varchar() behave\n> identically in the backend, except for the input functions which limit\n> the length of varchar.\n\n\tKarl...just curious, but what does an 'explain' show for the two\ndifferent situations? 10x worse almost sounds like the indices aren't\neven being used, don't they?\n\n\n",
"msg_date": "Mon, 9 Mar 1998 13:02:26 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] DANGER WILL ROBINSON!"
},
{
"msg_contents": "On Mon, Mar 09, 1998 at 01:02:26PM -0500, The Hermit Hacker wrote:\n> \n> Redirected to 'the proper list' - [email protected]\n> \n> On Mon, 9 Mar 1998, Bruce Momjian wrote:\n> \n> > > \n> > > WARNING!\n> > > \n> > > Postgres 6.3 has MAJOR trouble with btree-indexed text fields.\n> > > \n> > > Performance levels are *10x* worse than the same indexed fields in \"varchar\"\n> > > format!\n> > > \n> > > Be EXTREMELY careful - I got bit in the ass by this this morning, and it was\n> > > very fortunate that I figured out what was going on.\n> > > \n> > > The reason I changed this over was that I had dumped the table and it came\n> > > out of the pg_dump program with a negative size. So I figured I'd change it\n> > > to TEXT and that would resolve the problem. BIG mistake.\n> > > \n> > > Be on guard for this folks.\n> > > \n> > > Developers, you might want to look into this - there's no good reason for\n> > > this kind of behavior, is there?\n> > \n> > No good reason at all. As far as I know, text and varchar() behave\n> > identically in the backend, except for the input functions which limit\n> > the length of varchar.\n> \n> \tKarl...just curious, but what does an 'explain' show for the two\n> different situations? 10x worse almost sounds like the indices aren't\n> even being used, don't they?\n\nExplain claims the indices are being used.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 12:17:42 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] DANGER WILL ROBINSON!"
}
] |
[
{
"msg_contents": "-----Original Message-----\nFrom: Zeugswetter Andreas <[email protected]>\nTo: '[email protected]' <[email protected]>\nDate: dinsdag 10 maart 1998 0:36\nSubject: Re: [HACKERS] newoid in invapi.c\n\n\n>> So maybe we'll get rid of the oid and introduce some other \"thing\" with\n>> also gives us identity for instances of classes.\n>> But IMO no identity equates to no OO. And currently identity is only\n>> provided\n>> by oids, so I would vote to keep them in user tables too.\n>\n>In relational speak a tuple is always identified by it's primary key, which\nalso guarantees\n>fast access.\nYes, in relational speak it is.\n\n>The where oid = <value> is only fast if the user defines an index on oid.\n>The extensive use of oid's is also a nightmare for all those that want to\nreorganize tables.\n>There is simply very much that speaks against the use of oid's a la long.\nYes I agree that the way oids are implemented now has problems.\nHowever I choose to see these problems as \"implementation details\".\n\n>\n>Illustra defines a \"reference\" maybe we should dig into that ?\n>create table person (\n> name char (16),\n> mother ref(person),\n> father ref(person)\n>)\n>\nOk, if such a reference is unique within a table then we've got something\nsimilar to tids if they are not they would more resemble oids.\nI don't know which is the case for illustra but recalling it's\nheritage...?!?\n\nI presume that illustra would allow me to extend the above like in the\nfollowing:\n\ncreate table teacher (\n course char (32)\n) inherits(person);\n\nIf the illustra system allows me to insert a teacher object as my\nmother then the illustra reference is not likely to be implemented as\nphysical\nreference (tid) but more likely with some logical reference (oid).\n\nI really hope the last suggestion is the case as it would much\nresemble that which I would like to see in postgresql.\n\n>A unique pointer to a row for me is always:\n>dbid + tableid + fileid + primary key (or even rowid)\n\nIn a relational system, yes. In a OO system not necesarily.\nBecause as in the example above, a lady who happens to be my mother\nmay also happen to be a teacher. The identity of \"my mother\" and the\nidentity of \"my mother the teacher\" should be the same.\nIt would be a pity is \"my mother\" would have two identities just because\nof the way my database system stores it's data.\n>\n>I personally like the idea of a physical address as an alterntive to oid.\nThe problem\n>with this is that physical position changes over time. As the past has\nshown the\n>same problem is also present for oid's. The problem could maybe be solved\n>with a physical position tracking system, that gets reset at vacuum time.\nOr maybe\n>the existing logic for indexes could be reused in a somewhat modified\nmanner.\n\nI'm not trying to say that the physical address approach to identity is\nwrong.\nI'll try to explain with an example.\n\n<EXAMPLE>\nCREATE TABLE base (f1 int4);\nCREATE TABLE derived (f2 int4) inherits(base);\n\nINSERT INTO base values(10);\nINSERT INTO base values(20,20);\n</EXAMPLE>\n\nFor the query \"SELECT <identifier>,f1 from base;\" my ideal OO system\nmight give:\n\n<QUERY OUTPUT>\n18400, 10\n18401, 20\n(2 rows)\n</QUERY OUTPUT>\n\nFor the query \"SELECT <identifier>, f1 from derived;\" my ideal OO system\nmight give:\n\n<QUERY OUTPUT>\n18401, 20\n(1 rows)\n</QUERY OUTPUT>\n\nThis of course as a result of the so-called \"is_a\" relation between\ninstances of the\nderived class and instances of the base class.\nSo the query is polymorphic because it also operates on instances of classes\nderived from the base class.\nSo to support polymorphism we need to have some form of identity which\nis also valid between tables. As a result the current tids in postgresql\nwon't work because they are only valid within one table.\n\nYou'll have noticed that my \"ideal\" system has different semantics\nthan postgresql. So as far as I concerned there is room for improvement\nin postgresql.\n\nAccording to me in the least triggers, indices and select/update/delete\nstatements should be polymorphic\n(should work for instances of base classes and instances of derived\nclasses).\n\nThanks, with regards from Maurice.\n\n\n",
"msg_date": "Mon, 9 Mar 1998 19:52:21 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] newoid in invapi.c"
}
] |
[
{
"msg_contents": "\nThe following query returns:\n\nexplain update user set inter = person.inter where person.custid = user.custid;\n\nNOTICE: QUERY PLAN:\n \n Nested Loop (cost=1095.16 size=1 width=370)\n -> Seq Scan on person (cost=1095.16 size=16035 width=8)\n -> Seq Scan on user (cost=0.00 size=0 width=362)\n \nEXPLAIN\n\nUh, this is a problem.\n\nBoth person and user have an index on \"custid\". 6.2.1 used to utilize them.\n\nYou have to do a sequential scan on user, but NOT on person to perform \nthis update.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 13:23:54 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Heh, what's this?"
}
] |
[
{
"msg_contents": "\nHi again folks.\n\nWell, this problem just got a lot bigger.\n\nThe previous update that does a sequential scan where it used to (under\n6.2.1) do an index search on the outside condition just showed up in another\nprocess - and caused it to generate a 426MB core image, at which point it\nblew chunks.\n\nFolks, this is a big deal, and may be why some people are seeing problems\nwith 6.3. If someone doesn't come up with a fix for this pretty quickly\nwe're going to have to back out 6.3 here entirely.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 14:36:32 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Oh oh.... big trouble"
},
{
"msg_contents": "> \n> \n> Hi again folks.\n> \n> Well, this problem just got a lot bigger.\n> \n> The previous update that does a sequential scan where it used to (under\n> 6.2.1) do an index search on the outside condition just showed up in another\n> process - and caused it to generate a 426MB core image, at which point it\n> blew chunks.\n> \n> Folks, this is a big deal, and may be why some people are seeing problems\n> with 6.3. If someone doesn't come up with a fix for this pretty quickly\n> we're going to have to back out 6.3 here entirely.\n\nSounds like a serious problem, but honestly, this is the first problem\nof this type I have heard about. I am not saying it is not a bug, but I\nhave no other hint as to the cause.\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",
"msg_date": "Mon, 9 Mar 1998 16:58:21 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Oh oh.... big trouble"
}
] |
[
{
"msg_contents": "\nHi folks,\n\nRemember me? Remember I was bitching about hashed and indexes scans not\nbeing indexed?\n\nGuess what - it magically fixed itself.\n\nIf you want to talk about things that *bother* me, this one tops the pack.\n\nThe same query now returns an index hash query plan, which executes in a few\nseconds and requires little memory (as opposed to looped sequential scans\nrequiring 500MB on the server).\n\nThis is really, really, odd.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 15:04:59 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, 9 Mar 1998, Karl Denninger wrote:\n\n> \n> Hi folks,\n> \n> Remember me? Remember I was bitching about hashed and indexes scans not\n> being indexed?\n> \n> Guess what - it magically fixed itself.\n> \n> If you want to talk about things that *bother* me, this one tops the pack.\n> \n> The same query now returns an index hash query plan, which executes in a few\n> seconds and requires little memory (as opposed to looped sequential scans\n> requiring 500MB on the server).\n> \n> This is really, really, odd.\n\nUh, maybe a stupid question, but did you do a 'vacuum'?\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Mon, 9 Mar 1998 22:22:13 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, Mar 09, 1998 at 10:22:13PM +0100, Maarten Boekhold wrote:\n> On Mon, 9 Mar 1998, Karl Denninger wrote:\n> \n> > \n> > Hi folks,\n> > \n> > Remember me? Remember I was bitching about hashed and indexes scans not\n> > being indexed?\n> > \n> > Guess what - it magically fixed itself.\n> > \n> > If you want to talk about things that *bother* me, this one tops the pack.\n> > \n> > The same query now returns an index hash query plan, which executes in a few\n> > seconds and requires little memory (as opposed to looped sequential scans\n> > requiring 500MB on the server).\n> > \n> > This is really, really, odd.\n> \n> Uh, maybe a stupid question, but did you do a 'vacuum'?\n\nUh, yeah. Several of them, thinking that perhaps the indices were out of\ndate or otherwise damaged. Then, after some period of time where queries\nand updates were being put against the table, it started working properly (!)\n\nI have *absolutely* no idea what could have changed - I made no alterations,\ndidn't shut the server down, nothing - between the time that it was last\nreturning the wrong \"explain\" output and when it started returning the\ncorrect output.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 15:25:03 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "> > Uh, maybe a stupid question, but did you do a 'vacuum'?\n> \n> Uh, yeah. Several of them, thinking that perhaps the indices were out of\n> date or otherwise damaged. Then, after some period of time where queries\n> and updates were being put against the table, it started working properly (!)\n> \n> I have *absolutely* no idea what could have changed - I made no alterations,\n> didn't shut the server down, nothing - between the time that it was last\n> returning the wrong \"explain\" output and when it started returning the\n> correct output.\n\nMaybe the tables grew so that the optimizer decided now was to time where \nusing indices would be faster? Just trying to make some sense out of this...\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Mon, 9 Mar 1998 22:27:19 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "> \n> \n> Hi folks,\n> \n> Remember me? Remember I was bitching about hashed and indexes scans not\n> being indexed?\n> \n> Guess what - it magically fixed itself.\n> \n> If you want to talk about things that *bother* me, this one tops the pack.\n> \n> The same query now returns an index hash query plan, which executes in a few\n> seconds and requires little memory (as opposed to looped sequential scans\n> requiring 500MB on the server).\n> \n> This is really, really, odd.\n\nVacuum? Vacuum analyze? Improper index?\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",
"msg_date": "Mon, 9 Mar 1998 16:58:59 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, Mar 09, 1998 at 10:27:19PM +0100, Maarten Boekhold wrote:\n> > > Uh, maybe a stupid question, but did you do a 'vacuum'?\n> > \n> > Uh, yeah. Several of them, thinking that perhaps the indices were out of\n> > date or otherwise damaged. Then, after some period of time where queries\n> > and updates were being put against the table, it started working properly (!)\n> > \n> > I have *absolutely* no idea what could have changed - I made no alterations,\n> > didn't shut the server down, nothing - between the time that it was last\n> > returning the wrong \"explain\" output and when it started returning the\n> > correct output.\n> \n> Maybe the tables grew so that the optimizer decided now was to time where \n> using indices would be faster? Just trying to make some sense out of this...\n> \n> Maarten\n\nI don't think so. We do a lot of replaces on these tables (a shitload of\nthem in fact) but very few inserts. Vacuum typically finds about the same\nnumber of entries have been replaced as there are in the table to begin with\n(ie: the update rate is roughly 1:1 for the number of tuples in the\ndatabase, or perhaps even a bit higher).\n\nAnd, we run vacuum across the database every night through cron.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 16:20:19 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, Mar 09, 1998 at 04:58:59PM -0500, Bruce Momjian wrote:\n> > \n> > \n> > Hi folks,\n> > \n> > Remember me? Remember I was bitching about hashed and indexes scans not\n> > being indexed?\n> > \n> > Guess what - it magically fixed itself.\n> > \n> > If you want to talk about things that *bother* me, this one tops the pack.\n> > \n> > The same query now returns an index hash query plan, which executes in a few\n> > seconds and requires little memory (as opposed to looped sequential scans\n> > requiring 500MB on the server).\n> > \n> > This is really, really, odd.\n> \n> Vacuum? Vacuum analyze? Improper index?\n\nNope, nope and it was working until this morning with no indicated errors.\n\nThis and the text indices are the two really, really odd things I'm seeing\nin 6.3, along with *some* operations taking a lot longer than they used to.\n\nOne of the things we do around here from libpq is an operation that looks\nlike this:\n\n\n1)\tSelect a set of records from a join on a couple of tables.\n\n2)\tSelect from a different table (using the results from the first \n\tselect), looking specifically for certain data which we then use\n\tprogrammatically to perform a set of computations.\n\nNow, the first select is just peachy. It returns ~1500 or so records.\n\nThe iterative loop on the second select used to run through the entire 1500\nrecords or so (doing a select for each) in ~20-30 seconds. Now it takes\nroughly 2-3 minutes to do the same job. I have checked with \"explain\" that\nthe indices are being used for all queries - they are.\n\nI've seen this also with a few other processes that do the same kind of\nthing, and get the same results.\n\nI'm wondering if what we're seeing here is a severe degredation of locking\nperformance. That is, could this be related to the new deadlock detection\ncode? We're doing selects/updates within several tables in these jobs, but\nall within one transaction block (begin/commit or begin/rollback).\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 16:59:59 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "> \n> Nope, nope and it was working until this morning with no indicated errors.\n> \n> This and the text indices are the two really, really odd things I'm seeing\n> in 6.3, along with *some* operations taking a lot longer than they used to.\n> \n> One of the things we do around here from libpq is an operation that looks\n> like this:\n> \n> \n> 1)\tSelect a set of records from a join on a couple of tables.\n> \n> 2)\tSelect from a different table (using the results from the first \n> \tselect), looking specifically for certain data which we then use\n> \tprogrammatically to perform a set of computations.\n> \n> Now, the first select is just peachy. It returns ~1500 or so records.\n> \n> The iterative loop on the second select used to run through the entire 1500\n> records or so (doing a select for each) in ~20-30 seconds. Now it takes\n> roughly 2-3 minutes to do the same job. I have checked with \"explain\" that\n> the indices are being used for all queries - they are.\n> \n> I've seen this also with a few other processes that do the same kind of\n> thing, and get the same results.\n> \n> I'm wondering if what we're seeing here is a severe degredation of locking\n> performance. That is, could this be related to the new deadlock detection\n> code? We're doing selects/updates within several tables in these jobs, but\n> all within one transaction block (begin/commit or begin/rollback).\n\nThe deadlock code only runs after a minute of waiting for a lock, and\nevery minute thereafter, so if you are not waiting on a lock, you don't\nrun that code.\n\nNow the queue behavior has changed. We used to give readers preference,\nbut now we do some queue management that is fairer to readers and\nwriters.\n\nSee backend/storage/lmgr/proc.c::ProcSleep for the details:\n\n /*\n * If the first entries in the waitQueue have a greater priority than\n * we have, we must be a reader, and they must be a writers, and we\n * must be here because the current holder is a writer or a reader but\n * we don't share shared locks if a writer is waiting. We put\n * ourselves after the writers. This way, we have a FIFO, but keep\n * the readers together to give them decent priority, and no one\n * starves. Because we group all readers together, a non-empty queue\n * only has a few possible configurations:\n *\n * [readers] [writers] [readers][writers] [writers][readers]\n * [writers][readers][writers]\n *\n * In a full queue, we would have a reader holding a lock, then a writer\n * gets the lock, then a bunch of readers, made up of readers who\n * could not share the first readlock because a writer was waiting,\n * and new readers arriving while the writer had the lock.\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",
"msg_date": "Mon, 9 Mar 1998 22:01:17 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, Mar 09, 1998 at 10:01:17PM -0500, Bruce Momjian wrote:\n> > 1)\tSelect a set of records from a join on a couple of tables.\n> > \n> > 2)\tSelect from a different table (using the results from the first \n> > \tselect), looking specifically for certain data which we then use\n> > \tprogrammatically to perform a set of computations.\n> > \n> > Now, the first select is just peachy. It returns ~1500 or so records.\n> > \n> > The iterative loop on the second select used to run through the entire 1500\n> > records or so (doing a select for each) in ~20-30 seconds. Now it takes\n> > roughly 2-3 minutes to do the same job. I have checked with \"explain\" that\n> > the indices are being used for all queries - they are.\n> > \n> > I've seen this also with a few other processes that do the same kind of\n> > thing, and get the same results.\n> > \n> > I'm wondering if what we're seeing here is a severe degredation of locking\n> > performance. That is, could this be related to the new deadlock detection\n> > code? We're doing selects/updates within several tables in these jobs, but\n> > all within one transaction block (begin/commit or begin/rollback).\n> \n> The deadlock code only runs after a minute of waiting for a lock, and\n> every minute thereafter, so if you are not waiting on a lock, you don't\n> run that code.\n> \n> Now the queue behavior has changed. We used to give readers preference,\n> but now we do some queue management that is fairer to readers and\n> writers.\n\nWell, the other problem that comes with this however is CPU load on the\nserver. I find that the system load goes through the ceiling - there's\nnot a lot of disk I/O going on during this time, but the CPU is damn busy,\nalmost all of it in user time.\n\nI don't think this explains all of it. Something is going on in the\ninteraction of the different processes being handled.\n\nBasically, we have the following:\n\n1)\tOne big \"select\" to get candidates.\n\n2)\tA second select on another table using values from the candidate to \n\tcheck two algorythmic values (which can't really be pushed up into \n\tthe first select as a join - if it could, that would make things \n\teasy).\n\n3)\tIf either of the two conditions are met, then we take a process\n\tdetour and do the appropriate system-level things, and update the\n\tcandidate record. If we do take either of these, we open a\n\ttransaction to make sure we either commit all changes or none of\n\tthem.\n\n4)\tIf the candidate was modified (ie: a \"begin\" was executed along\n\twith one or more \"updates\") we send a \"commit\", assuming we were\n\table to do the system-level things ok.\n\nWhile this is running, no other instance can be (we use another table with a\nsingle \"flag\" value as a means of preventing this), because if two of these\ncome along at once all hell would break loose.\n\nThis is \"set off\" by an async trigger when there could be candidates.\n\nWith 6.2.1 this worked fine - we got through the entire process in a few\nseconds, and all was well. \n\nWith 6.3 I have had to schedule this to run on 30 minute intervals, because\nits 10x+ slower than it was under 6.2.1. I'm giving serious consideration\nto a table reconstruction so I can use a join to get all of this in one\npass, which will get rid of the need to loop over the first select's \nreturned values.\n\nI've seen this same kind of behavior in a few other places as well; in\nplaces where you are doing reads and writes in a mixed environment (ie: read\nsomething, write something (typically in a different table) based on what \nyou read) performance of 6.3 has gone in the toilet. Update jobs that used \nto run in tens of seconds are requiring several minutes to complete now.\n\nAnd again, we're not seeing much disk I/O during this period - but we *ARE*\nseeing a hell of a lot of CPU activity, almost all in user mode.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 21:33:50 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "> I've seen this same kind of behavior in a few other places as well; in\n> places where you are doing reads and writes in a mixed environment (ie: read\n> something, write something (typically in a different table) based on what \n> you read) performance of 6.3 has gone in the toilet. Update jobs that used \n> to run in tens of seconds are requiring several minutes to complete now.\n> \n> And again, we're not seeing much disk I/O during this period - but we *ARE*\n> seeing a hell of a lot of CPU activity, almost all in user mode.\n\nOK, how about using postgres -t option or profiling to get the function\nusing so much cpu? This may help us tell where the problem lies. Does\nanyone else see such problems? All other reports I hear was that 6.3\nwas faster.\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",
"msg_date": "Mon, 9 Mar 1998 23:05:49 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, Mar 09, 1998 at 11:05:49PM -0500, Bruce Momjian wrote:\n> > I've seen this same kind of behavior in a few other places as well; in\n> > places where you are doing reads and writes in a mixed environment (ie: read\n> > something, write something (typically in a different table) based on what \n> > you read) performance of 6.3 has gone in the toilet. Update jobs that used \n> > to run in tens of seconds are requiring several minutes to complete now.\n> > \n> > And again, we're not seeing much disk I/O during this period - but we *ARE*\n> > seeing a hell of a lot of CPU activity, almost all in user mode.\n> \n> OK, how about using postgres -t option or profiling to get the function\n> using so much cpu? This may help us tell where the problem lies. Does\n> anyone else see such problems? All other reports I hear was that 6.3\n> was faster.\n\nDocs on this somewhere?\n\nI'll be happy to profile it if I can figure out how! :-)\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 22:15:55 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "> > was faster.\n> \n> Docs on this somewhere?\n> \n> I'll be happy to profile it if I can figure out how! :-)\n> \n\nSee the FAQ. Usually profiling is -pg.\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",
"msg_date": "Mon, 9 Mar 1998 23:21:21 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "On Mon, 9 Mar 1998, Karl Denninger wrote:\n\n> On Mon, Mar 09, 1998 at 11:05:49PM -0500, Bruce Momjian wrote:\n> > > I've seen this same kind of behavior in a few other places as well; in\n> > > places where you are doing reads and writes in a mixed environment (ie: read\n> > > something, write something (typically in a different table) based on what \n> > > you read) performance of 6.3 has gone in the toilet. Update jobs that used \n> > > to run in tens of seconds are requiring several minutes to complete now.\n> > > \n> > > And again, we're not seeing much disk I/O during this period - but we *ARE*\n> > > seeing a hell of a lot of CPU activity, almost all in user mode.\n> > \n> > OK, how about using postgres -t option or profiling to get the function\n> > using so much cpu? This may help us tell where the problem lies. Does\n> > anyone else see such problems? All other reports I hear was that 6.3\n> > was faster.\n> \n> Docs on this somewhere?\n> \n> I'll be happy to profile it if I can figure out how! :-)\n\nI have just figured this out. First, you can get some timing information \nfrom postgresql itself by including '-tpa -tpl -te' on the backend \noptions, second, you can recompile postgresql with profiling on (look in \nMakefile.global for profiling options, adjust them for your syetem). \nThen, *don't do a 'make install'* but only copy the postgres binary (in \nbackend dir) to your postgresql bin-dir, as postgres.profile. Whenever \nyou want to profile, edit your start-script to use that postgres.profile \nas a backend. Run your queries, then copy the profile-output (gmon.out \nfor gcc profiling) from the *database* directory you did your queries on \n(ie. if you connected to database 'template1', it'll be in \ndata/base/template1).\n\nThen do normal profiling stuff (foor gcc profiling, run 'gprof gmon.out \nbin/postgres.profile > prof.out' to get readable output).\n\nIf you have questions, please mail me privately.\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Tue, 10 Mar 1998 11:56:57 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
},
{
"msg_contents": "Hello,\n\nI posted a similar problem here on Saturday. Now I had a little time \nlooking at the queries we send.\n\nThis is the output of PG 6.2.1\n\n--\nboersenspiel=> explain SELECT DISTINCT spieler_nr, Trans.wpk_nr,\nstate, anzahl, buyprice, buydate, sellprice, selldate, Kurse.wpk_nr,\nname, curr, kurs, datum , Trans.oid from Trans, Kurse WHERE\nTrans.wpk_nr=Kurse.wpk_nr AND spieler_nr=3 ORDER BY Trans.wpk_nr ,\nselldate USING >; NOTICE:QUERY PLAN:\n\nUnique (cost=0.00 size=0 width=0)\n -> Sort (cost=4.10 size=0 width=0)\n -> Nested Loop (cost=4.10 size=1 width=73)\n -> Index Scan on trans (cost=2.05 size=1 width=41)\n -> Index Scan on kurse (cost=2.05 size=14305 width=32)\n\n\nNow the same query in 6.3:\n\nUnique (cost=1164.21 size=0 width=0)\n -> Sort (cost=1164.21 size=0 width=0)\n -> Hash Join (cost=1164.21 size=1 width=73)\n -> Seq Scan on kurse (cost=688.07 size=14305 width=32)\n -> Hash (cost=0.00 size=0 width=0)\n -> Index Scan on trans (cost=2.05 size=1\nwidth=41) \n\nAll indices are created (all btrees), but the index on kurse doesn't \nseem to be used. \n\n\n> Guess what - it magically fixed itself.\n> \n> If you want to talk about things that *bother* me, this one tops the pack.\n> \n> The same query now returns an index hash query plan, which executes in a few\n> seconds and requires little memory (as opposed to looped sequential scans\n> requiring 500MB on the server).\n> \n> This is really, really, odd.\n\nDito.\n\nBTW.: I tried to apply the patches from Massimo, as the only major \nproblem for us in 6.2.1p6 is the buggy deadlock code. Anybody managed \nto get it working? \n\nCiao\n\nUlrich\n\n\n\nUlrich Voss \\ \\ / /__ / ___|__ _| |\nVoCal web publishing \\ \\ / / _ \\| | / _` | |\[email protected] \\ V / (_) | |__| (_| | |\nhttp://www.vocalweb.de \\_/ \\___/ \\____\\__,_|_|\nTel: 0203-306-1560 web publishing\n",
"msg_date": "Tue, 10 Mar 1998 13:52:42 +0000",
"msg_from": "\"Boersenspielteam\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Heh, the disappearing problem!"
}
] |
[
{
"msg_contents": "> trashing postgresql doing a 'select count(*) from artist_fti'\n> \"select count(*) from artist_fti where string ~ '^lling';\"\n> \"select count(*) from artist_fti where string ~ '^tones';\"\n> and the join-statement.\n> \n> I don't have much knowledge of profiling, but it seems to me that in the \n> end this is really related to I/O. Dirty disk caches forces postgresql to \n> read everything from disk again, and this disk is not the fastest in the \n> world. Still it takes much more time than I would expect. I mean, btree \n> is designed to have few disk accesses, and then accessing the tables \n> shouldn't take too long also.\n\n[Marteen is finding word searches on a big table to take a long time. \nThis is the word slice code we talked about recently as a group. It\nrapid searches for words inside of strings. Very useful.]\n\nOK, I meant to reply to this, but forgot to. Let's take the second query:\n\n\tselect count(*) from artist_fti where string ~ '^lling';\n\nHere is the top of the profile output:\n\n---------------------------------------------------------------------------\n\nEach sample counts as 0.01 seconds.\n % cumulative self self total \n time seconds seconds calls us/call us/call name \n 10.20 0.05 0.05 18434 2.71 2.71 sstep\n 8.16 0.09 0.04 3225 12.40 12.41 RelationGetLRelId\n 6.12 0.12 0.03 5862 5.12 5.12 SpinAcquire\n 6.12 0.15 0.03 2750 10.91 10.91 ExecEvalVar\n 6.12 0.18 0.03 105 285.71 411.52 heapgettup\n 6.12 0.21 0.03 2 15000.00 54472.81 ExecAgg\n 4.08 0.23 0.02 12711 1.57 1.57 OrderedSetContains\n 4.08 0.25 0.02 5501 3.64 9.09 ExecEvalExpr\n 2.04 0.26 0.01 12711 0.79 2.36 AllocSetContains\n 2.04 0.27 0.01 7722 1.30 1.30 OrderedElemPushHead\n 2.04 0.28 0.01 7608 1.31 2.63 hash_search\n 2.04 0.29 0.01 6395 1.56 1.56 tag_hash\n 2.04 0.30 0.01 5610 1.78 6.50 PortalHeapMemoryFree\n 2.04 0.31 0.01 2293 4.36 4.36 fmgr_isbuiltin\n 2.04 0.32 0.01 1919 5.21 5.21 BufferGetBlockNumber\n 2.04 0.33 0.01 1912 5.23 9.55 ReleaseBuffer\n 2.04 0.34 0.01 1572 6.36 6.36 TransactionIdEquals\n 2.04 0.35 0.01 1571 6.37 12.73 HeapTupleSatisfiesVisibili\n\n\n---------------------------------------------------------------------------\n\nFirst 'sstep' is taking a lot of time, and that, I think, is the regex\nstuff. Author says the regex is slow and can be speeded up, so we may\nget a new release from him soon.\n\nAlso strange is the time for ExecAgg, which seems fairly high for a\nsingle function, but only 0.015 seconds. Certainly not significant in\nthe wallclock time.\n\nNow, seeing as everything else adds up to perhaps 1-2 seconds, why 30\nseconds for the query. Certainly looks like I/O is the problem.\n\nHave you tried adding -B buffers. This, I think, would help. If you\ndon't flush the cache, how long does a second identical query take? \nAlso, Vadim, when the backend is accessing a btree index, does it walk\ndown to the page it needs and read all rows off of that, or does it\ndrill down the btree to get each row. Third, remember that these are\nmade indexable by adding\n\n\t\tstring >= 'ling' and\t\n\t\tstring <= 'ling\\377'\n\nNow, if each is taken to evaluate litarally, I am guessing the backend\nis getting all >= 'ling' and then all <= 'ling\\377' and then finding the\nones that match. This is terrible for performance. I believe this is\nthe crux of the problem, and why so much I/O is going on.\n\nVadim, any way to make the optimizer realize that we are looking for a\nrange of values, and preventing it from pulling all those rows from the\nindex? Almost some index scheme that would get all >= 'ling' but stop\nwhen they get > 'ling\\377'?\n\nWith this going on, the system is spinning through the entire btree\nindex each time to get the data, even though it is a small subset.\n\nAs part of a test, would you please run:\n\n\tselect count(*) from artist_fti where string >= 'lling'\n\nand\n\n\tselect count(*) from artist_fti where string <= 'lling\\377'\n\n\nand then try:\n\n\tselect count(*) from artist_fti where string >= 'lling' and\n\t\t\t\t\t\tstring <= 'lling\\377'\n\nAlso try:\n\n\tselect count(*) from artist_fti where string ~ '^lling' and\n\t\t\t\t\tstring >= 'lling and\n\t\t\t\t\tstring <= 'lling\\377';\n\nTry these with the cache trashing code to get good numbers on the\nperformance. My guess is that the sum of the first and second execution\ntimes will equal the time for the third, and the third and fourth will\ntake the same amount of time.\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",
"msg_date": "Mon, 9 Mar 1998 21:54:26 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: your mail"
}
] |
[
{
"msg_contents": "unsubscribe\n\n\n\n\n\n\n\nunsubscribe",
"msg_date": "Tue, 10 Mar 1998 12:30:56 +0900",
"msg_from": "\"=?euc-kr?B?udqxzcXC?=\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "None"
}
] |
[
{
"msg_contents": "In article <[email protected]> you wrote:\n: Same same. They've been trying to get me to 'look at' Developer 2000. \n: They do have a SCO, and Solaris product, as well as AIX and HPUX, but no\n: BSD/OS. They also said \"no plans for it.\"\n: \n:> Personally, I use MySQL and it blows the doors off of many other\n:> databases... in one case, we had a guy asking how many SPARC servers\n:> we had running the thing (assuming Oracle or Sybase)... he was shocked\n:> to learn it was on a single P5-166 with 64Mb RAM.\n\n: I've used MySQL. It is very fast but doesn't support transactions. At\n: least it didn't when I was using it. I currently use Velocis and\n: PostgreSQL. Velocis seems slow compared to Postgres. PostgreSQL is\n: fast, stable, and flexible, but ODBC support is marginal. My problem is\n: that my customers don't get a warm fuzzy about a solution unless it has\n: a brand name attached to it. The typical response is, \"Free UNIX? Free\n: RDMBS? What, are you nuts?\" Can't seem to combat that one. Feel like\n: a lemming on the Microsoft boat heading for the cliff. :-(\n\n[CC'ed to PostgreSQL hacker list.]\n\nGee, I didn't think we were faster than Velocis. Hope you are planning\non moving to PostgreSQL 6.3. It is even better and faster.\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",
"msg_date": "Mon, 9 Mar 1998 22:45:12 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: Why no Oracle, Sybase, Informix etc."
}
] |
[
{
"msg_contents": "Hi folks.\n\nMe again.\n\nCheck this out (an excerpt from a \"pg_dump\"ed table):\n\nCREATE TABLE user (custid int4, class varchar(-5), login varchar(-5)....\n\nUh, varchar(-5) eh? And pray tell, how should I reload that table? :-)\n\nThis is kinda serious.\n\n--\n-- \nKarl Denninger ([email protected])| MCSNet - Serving Chicagoland and Wisconsin\nhttp://www.mcs.net/ | T1's from $600 monthly to FULL DS-3 Service\n\t\t\t | NEW! K56Flex support on ALL modems\nVoice: [+1 312 803-MCS1 x219]| EXCLUSIVE NEW FEATURE ON ALL PERSONAL ACCOUNTS\nFax: [+1 312 803-4929] | *SPAMBLOCK* Technology now included at no cost\n",
"msg_date": "Mon, 9 Mar 1998 22:04:58 -0600",
"msg_from": "Karl Denninger <[email protected]>",
"msg_from_op": true,
"msg_subject": "Heh heh, who broke pg_dump?"
}
] |
[
{
"msg_contents": "> \n> Bruce Momjian wrote:\n> > \n> <SNIP>\n> > \n> > Gee, I didn't think we were faster than Velocis. Hope you are planning\n> > on moving to PostgreSQL 6.3. It is even better and faster.\n> \n> That's what I'm using! (now) and yes, in my particular application\n> simple selects are about 15% faster than Velocis 1.41 on BSDi 3.1. I\n\nThat is surprising. I assume you have turned off fsync with the -F\noption.\n\n> didn't spend a *whole* lot of time trying to tweak the CGI programs (\n> didn't have too much time to spend actually ) for either platform tho. \n> Truth is, PostgreSQL *ROCKS!*, I just have a hard time selling customers\n> on the idea of 'free' software. Perhaps you guys should package 6.3\n> under a different name and sell it for $12,000 to $15,000 U.S. ;-)\n> \n> You guys do a fantastic job!\n\nGlad to hear you like it.\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",
"msg_date": "Mon, 9 Mar 1998 23:16:28 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: Why no Oracle, Sybase, Informix etc."
}
] |
[
{
"msg_contents": "\nwell, it looks like we're moving postgres to a linux machine, so I\nwon't be doing any more alpha debugging for a while (BTW, adjusting\nsome types on the attribute struct to make it 64-bit aligned had no\neffect) unless someone has some suggestions for me to try (aside from\nthe snapshot thing, I figure anyone with an alpha can do that. let me\nknow if I should). I haven't given up, however my free time is\nlimited and obviously I can't do it on my employer's penny anymore\n(beleive it or not, I did have permission), and I'd also like to work\non pl/perl & odbc stuff.\n\nalso, cast (var to type) is the same as type(var)? weird! why\ndoesn't it just call the output/input functions (I can't cast name to\ntext or vice-versa for comparisons)..\n",
"msg_date": "Mon, 9 Mar 1998 20:17:41 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "postgres/alpha problems"
},
{
"msg_contents": "> also, cast (var to type) is the same as type(var)? weird! why\n> doesn't it just call the output/input functions (I can't cast name to\n> text or vice-versa for comparisons)..\n\nIf there is an explicit conversion routine available, it might be/should\nbe faster and more correct than converting to and from strings. The last\nresort fallback could be the string strategy. I'd like to work on the\ntype conversion problem for v6.4...\n\n - Tom\n\n",
"msg_date": "Tue, 10 Mar 1998 05:18:21 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "On Mon, 9 Mar 1998, Brett McCormick wrote:\n\n>well, it looks like we're moving postgres to a linux machine, so I\n>won't be doing any more alpha debugging for a while (BTW, adjusting\n>some types on the attribute struct to make it 64-bit aligned had no\n>effect) unless someone has some suggestions for me to try (aside from\n>the snapshot thing, I figure anyone with an alpha can do that. let me\n>know if I should). I haven't given up, however my free time is\n>limited and obviously I can't do it on my employer's penny anymore\n>(beleive it or not, I did have permission), and I'd also like to work\n>on pl/perl & odbc stuff.\n\nOk, I could do the cvs stuff. No promises about time, though. Marc? How\ncan we do it?\n\n-------------------------------------------------------------------\nPedro José Lobo Perea Tel: +34 1 336 78 19\nCentro de Cálculo Fax: +34 1 331 92 29\nEUIT Telecomunicación - UPM e-mail: [email protected]\n\n",
"msg_date": "Tue, 10 Mar 1998 09:55:00 +0100 (MET)",
"msg_from": "\"Pedro J. Lobo\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "On Tue, 10 Mar 1998, Pedro J. Lobo wrote:\n\n> On Mon, 9 Mar 1998, Brett McCormick wrote:\n> \n> >well, it looks like we're moving postgres to a linux machine, so I\n> >won't be doing any more alpha debugging for a while (BTW, adjusting\n> >some types on the attribute struct to make it 64-bit aligned had no\n> >effect) unless someone has some suggestions for me to try (aside from\n> >the snapshot thing, I figure anyone with an alpha can do that. let me\n> >know if I should). I haven't given up, however my free time is\n> >limited and obviously I can't do it on my employer's penny anymore\n> >(beleive it or not, I did have permission), and I'd also like to work\n> >on pl/perl & odbc stuff.\n> \n> Ok, I could do the cvs stuff. No promises about time, though. Marc? How\n> can we do it?\n\n\t*pop head up* Do what? Only way of grabbing the current source\ntree is via CVSup...I could make a tarball of the CVS repository available\nas of the time of the v6.3 release, if that helps...?\n\n\n",
"msg_date": "Tue, 10 Mar 1998 07:47:57 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "The Hermit Hacker wrote:\n\n> On Tue, 10 Mar 1998, Pedro J. Lobo wrote:\n>\n> > On Mon, 9 Mar 1998, Brett McCormick wrote:\n> >\n> > >well, it looks like we're moving postgres to a linux machine, so I\n> > >won't be doing any more alpha debugging for a while (BTW, adjusting\n> > >some types on the attribute struct to make it 64-bit aligned had no\n> > >effect) unless someone has some suggestions for me to try (aside from\n> > >the snapshot thing, I figure anyone with an alpha can do that. let me\n> > >know if I should). I haven't given up, however my free time is\n> > >limited and obviously I can't do it on my employer's penny anymore\n> > >(beleive it or not, I did have permission), and I'd also like to work\n> > >on pl/perl & odbc stuff.\n> >\n> > Ok, I could do the cvs stuff. No promises about time, though. Marc? How\n> > can we do it?\n>\n> *pop head up* Do what? Only way of grabbing the current source\n> tree is via CVSup...I could make a tarball of the CVS repository available\n> as of the time of the v6.3 release, if that helps...?\n\nCVSup supports the \"snapshot\" idea directly. Use the \"*default date\"\nparameter:\n\n *default date=98.01.20.00.00.00\n\nto retrieve the source tree as of midnight, January 20. I have used this\nsuccessfully just recently in trying to track down a problem. The nice thing\nis that you can change this date and CVSup will roll back any more recent\nchanges without requiring you to download the entire tree, and afaik does all\nthe right things to make this work correctly.\n\nMy complete \"oldtree\" CVSup configuration file is enclosed below. Prepending\na \"#\" to the defaultdate parameter line will give you the usual behavior of\npicking up the latest version of the tree.\n\n - Tom\n\n\n# This file represents the standard CVSup distribution file\n# for the PostgreSQL ORDBMS project\n# Modified by [email protected] 1997-08-28\n# - Point to my local snapshot source tree\n#\n# Defaults that apply to all the collections\n*default host=postgresql.org\n*default compress\n*default release=cvs\n*default delete use-rel-suffix\n*default tag=.\n*default date=98.01.20.00.00.00\n\n# base directory points to where CVSup will store its 'bookmarks' file(s)\n*default base=/opt/postgres/oldtree # /usr/local/pgsql\n\n# prefix directory points to where CVSup will store the actual\ndistribution(s)\n*default prefix=/opt/postgres/oldtree # /usr/local/pgsql\n\n# complete distribution, including all below\npgsql\n\n# individual distributions vs 'the whole thing'\n# pgsql-doc\n# pgsql-perl5\n# pgsql-src\n\n",
"msg_date": "Tue, 10 Mar 1998 13:34:02 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "On Tue, 10 Mar 1998, The Hermit Hacker wrote:\n\n>On Tue, 10 Mar 1998, Pedro J. Lobo wrote:\n>\n>> On Mon, 9 Mar 1998, Brett McCormick wrote:\n>> \n>> >won't be doing any more alpha debugging for a while (BTW, adjusting\n>> >some types on the attribute struct to make it 64-bit aligned had no\n>> >effect) unless someone has some suggestions for me to try (aside from\n>> >the snapshot thing, I figure anyone with an alpha can do that. let me\n>> >know if I should). I haven't given up, however my free time is\n>> >limited and obviously I can't do it on my employer's penny anymore\n>> >(beleive it or not, I did have permission), and I'd also like to work\n>> >on pl/perl & odbc stuff.\n>> \n>> Ok, I could do the cvs stuff. No promises about time, though. Marc? How\n>> can we do it?\n>\n>\t*pop head up* Do what? Only way of grabbing the current source\n>tree is via CVSup...I could make a tarball of the CVS repository available\n>as of the time of the v6.3 release, if that helps...?\n\nOk, I thought I heard something about getting an account. If I can use\nCVSup to get the CVS tree, that's fine. I'll let you know when/if I find\nsomething interesting.\n\n-------------------------------------------------------------------\nPedro José Lobo Perea Tel: +34 1 336 78 19\nCentro de Cálculo Fax: +34 1 331 92 29\nEUIT Telecomunicación - UPM e-mail: [email protected]\n\n",
"msg_date": "Tue, 10 Mar 1998 15:25:09 +0100 (MET)",
"msg_from": "\"Pedro J. Lobo\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "On Tue, 10 Mar 1998, Thomas G. Lockhart wrote:\n\n>The Hermit Hacker wrote:\n>\n>> *pop head up* Do what? Only way of grabbing the current source\n>> tree is via CVSup...I could make a tarball of the CVS repository available\n>> as of the time of the v6.3 release, if that helps...?\n>\n>CVSup supports the \"snapshot\" idea directly. Use the \"*default date\"\n>parameter:\n>\n> *default date=98.01.20.00.00.00\n\n[...]\n\nThanks for the info. I'll start investigating as soon as I can.\n\n-------------------------------------------------------------------\nPedro José Lobo Perea Tel: +34 1 336 78 19\nCentro de Cálculo Fax: +34 1 331 92 29\nEUIT Telecomunicación - UPM e-mail: [email protected]\n\n",
"msg_date": "Tue, 10 Mar 1998 15:26:48 +0100 (MET)",
"msg_from": "\"Pedro J. Lobo\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "> \n> On Tue, 10 Mar 1998, Pedro J. Lobo wrote:\n> \n> > On Mon, 9 Mar 1998, Brett McCormick wrote:\n> > \n> > >well, it looks like we're moving postgres to a linux machine, so I\n> > >won't be doing any more alpha debugging for a while (BTW, adjusting\n> > >some types on the attribute struct to make it 64-bit aligned had no\n> > >effect) unless someone has some suggestions for me to try (aside from\n> > >the snapshot thing, I figure anyone with an alpha can do that. let me\n> > >know if I should). I haven't given up, however my free time is\n> > >limited and obviously I can't do it on my employer's penny anymore\n> > >(beleive it or not, I did have permission), and I'd also like to work\n> > >on pl/perl & odbc stuff.\n> > \n> > Ok, I could do the cvs stuff. No promises about time, though. Marc? How\n> > can we do it?\n> \n> \t*pop head up* Do what? Only way of grabbing the current source\n> tree is via CVSup...I could make a tarball of the CVS repository available\n> as of the time of the v6.3 release, if that helps...?\n\nHe needs a login to postgresql, and do cvs extracts based on date with\nthe -D option. He will test Jan 15, and go a month forward or backward,\nuntil he finds the date that initdb fails.\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",
"msg_date": "Tue, 10 Mar 1998 11:56:22 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "> CVSup supports the \"snapshot\" idea directly. Use the \"*default date\"\n> parameter:\n> \n> *default date=98.01.20.00.00.00\n> \n> to retrieve the source tree as of midnight, January 20. I have used this\n> successfully just recently in trying to track down a problem. The nice thing\n> is that you can change this date and CVSup will roll back any more recent\n> changes without requiring you to download the entire tree, and afaik does all\n> the right things to make this work correctly.\n\nCool. Much better than using cvs directly. Let's get those Alpha's\nfixed!\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",
"msg_date": "Tue, 10 Mar 1998 11:59:04 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": "On Tue, 10 Mar 1998, Bruce Momjian wrote:\n\n>> On Tue, 10 Mar 1998, Pedro J. Lobo wrote:\n>> \n>> > On Mon, 9 Mar 1998, Brett McCormick wrote:\n>> > \n>> > >well, it looks like we're moving postgres to a linux machine, so I\n>> > >won't be doing any more alpha debugging for a while (BTW, adjusting\n>> > >some types on the attribute struct to make it 64-bit aligned had no\n>> > >effect) unless someone has some suggestions for me to try (aside from\n>> > >the snapshot thing, I figure anyone with an alpha can do that. let me\n>> > >know if I should). I haven't given up, however my free time is\n>> > >limited and obviously I can't do it on my employer's penny anymore\n>> > >(beleive it or not, I did have permission), and I'd also like to work\n>> > >on pl/perl & odbc stuff.\n>> > \n>> > Ok, I could do the cvs stuff. No promises about time, though. Marc? How\n>> > can we do it?\n>> \n>> \t*pop head up* Do what? Only way of grabbing the current source\n>> tree is via CVSup...I could make a tarball of the CVS repository available\n>> as of the time of the v6.3 release, if that helps...?\n>\n>He needs a login to postgresql, and do cvs extracts based on date with\n>the -D option. He will test Jan 15, and go a month forward or backward,\n>until he finds the date that initdb fails.\n\nIt seems that I don't need such a thing. I am using cvsup right now, and\nit seems to work well... except that it's been running for more than an\nhour and I only have the 'COPYRIGHT' file :-( The overseas connection is\nterribly slow. At night (between 0:00 and 7:00 MET, which is 18:00 - 1:00\nEST, I think) it goes better, but that way I could only get (and try) one\nsnapshot each day. Slow, but... I am downloading now the Jan 1st tree.\nWe'll see what happens.\n\n-------------------------------------------------------------------\nPedro José Lobo Perea Tel: +34 1 336 78 19\nCentro de Cálculo Fax: +34 1 331 92 29\nEUIT Telecomunicación - UPM e-mail: [email protected]\n\n",
"msg_date": "Tue, 10 Mar 1998 18:55:59 +0100 (MET)",
"msg_from": "\"Pedro J. Lobo\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
},
{
"msg_contents": ">>>>> \"pjlobo\" == Pedro J Lobo <[email protected]> writes:\n\nStuff deleted.\n\n > It seems that I don't need such a thing. I am using cvsup right\n > now, and it seems to work well... except that it's been running\n > for more than an hour and I only have the 'COPYRIGHT' file :-(\n > The overseas connection is terribly slow. At night (between 0:00\n > and 7:00 MET, which is 18:00 - 1:00 EST, I think) it goes\n > better, but that way I could only get (and try) one snapshot\n > each day. Slow, but... I am downloading now the Jan 1st tree.\n > We'll see what happens.\n\nWhy get the snapshots instead of the entire RCS files using CVSUP?\nuse \n\n*default release=cvs.\n\n This allows you to create any possible snapshot, by checking out of\nCVS. I have found this a much better way to track bugs. This does\nrequire additional disk space since you have a local copy of the CVS\ntree, plus need space to check the actual source out from CVS.\n\n > -------------------------------------------------------------------\n > Pedro Jos� Lobo Perea Tel: +34 1 336 78 19 Centro de C�lculo\n > Fax: +34 1 331 92 29 EUIT Telecomunicaci�n - UPM e-mail:\n > [email protected]\n\n-- \nKent S. Gordon\nArchitect\niNetSpace Co.\nvoice: (972)851-3494 fax:(972)702-0384 e-mail:[email protected]\n",
"msg_date": "Wed, 11 Mar 1998 11:04:13 -0600 (CST)",
"msg_from": "\"Kent S. Gordon\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
}
] |
[
{
"msg_contents": "\nI'm not sure if I've reported this or not... selecting from a view\nthat uses count and group by (when data exists) crashes the back end.\n\nnice to see that I can order by the result of count(*) now :)\n\nbrett=> create table a (b text, c text);\nCREATE\nbrett=> create table d (e text, f text);\nCREATE\nbrett=> create view g as select b,count(d.oid) from b,d group by b; \nERROR: b: Table does not exist.\nbrett=> create view g as select b,count(d.oid) from a,d group by b;\nCREATE\nbrett=> select * From g;\nb|count\n-+-----\n(0 rows)\n\nbrett=> insert into a values ('foo', 'bar');\nINSERT 53151 1\nbrett=> select * From g;\nb|count\n-+-----\n(0 rows)\n\nbrett=> insert into d values ('baz', 'chicken');\nINSERT 53152 1\nbrett=> select * From g;\nPQexec() -- Request was sent to backend, but backend closed the channel before responding.\n This probably means the backend terminated abnormally before or while processing the request.\nbrett=> \n",
"msg_date": "Mon, 9 Mar 1998 20:50:11 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "views with group by/count segfault"
},
{
"msg_contents": "> \n> \n> I'm not sure if I've reported this or not... selecting from a view\n> that uses count and group by (when data exists) crashes the back end.\n> \n> nice to see that I can order by the result of count(*) now :)\n\nAlready in TODO:\n\t\n\t* Allow views of aggregate columns\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",
"msg_date": "Tue, 10 Mar 1998 09:10:06 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [BUGS] views with group by/count segfault"
}
] |
[
{
"msg_contents": "\nI decided to use the timestamp type, but this may have been a mistake!\nThere is no btree index for timestamp, but there is for datetime?!?!\nI suppose I could write one...\n",
"msg_date": "Mon, 9 Mar 1998 21:04:33 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "no btree index for timestamp?"
}
] |
[
{
"msg_contents": ">>There is simply very much that speaks against the use of oid's a la long.\n>Yes I agree that the way oids are implemented now has problems.\n>However I choose to see these problems as \"implementation details\".\n\nNo, here I disagree, a globally unique identifier like oid is an architecture and strategy thing.\nAs such it needs a lot of thought and care.\n\n>>Illustra defines a \"reference\" maybe we should dig into that ?\n>>create table person (\n>> name char (16),\n>> mother ref(person),\n>> father ref(person)\n>>)\n>>\n>Ok, if such a reference is unique within a table then we've got something\n>similar to tids if they are not they would more resemble oids.\n>I don't know which is the case for illustra but recalling it's\n>heritage...?!?\n\nIllustra uses oid's, and of course it suffers the same bottleneck on a multi CPU System.\n\n>I presume that illustra would allow me to extend the above like in the\n>following:\n\n>create table teacher (\n> course char (32)\n>) inherits (person);\n\nYup, syntax is: create table teacher (course char(32)) under person; \n\n>If the illustra system allows me to insert a teacher object as my\n>mother then the illustra reference is not likely to be implemented as\n>physical\n\nNo, in the above schema the teacher instance gets its own identity, same as postgresql.\n\n>reference (tid) but more likely with some logical reference (oid).\n\n>I really hope the last suggestion is the case as it would much\n>resemble that which I would like to see in postgresql.\n\n>>A unique pointer to a row for me is always:\n>>dbid + tableid + fileid + primary key (or even rowid)\n\n>In a relational system, yes. In a OO system not necesarily.\n>Because as in the example above, a lady who happens to be my mother\n>may also happen to be a teacher. The identity of \"my mother\" and the\n>identity of \"my mother the teacher\" should be the same.\n\nAgreed, but currently not the case. Let me explain further:\nIf you have parents, teachers under parents and petowners under parents :-).\nThen the teachers that are also petowners would get 2 oid's. One for teacher\nand one for petowner. I agree that this is not perfect, and can not be solved with the current\narchitecture :-(\n\n>It would be a pity is \"my mother\" would have two identities just because\n>of the way my database system stores it's data.\n>>\n>>I personally like the idea of a physical address as an alterntive to oid.\n>The problem\n>>with this is that physical position changes over time. As the past has\n>shown the\n>>same problem is also present for oid's. The problem could maybe be solved\n>>with a physical position tracking system, that gets reset at vacuum time.\n>Or maybe\n>>the existing logic for indexes could be reused in a somewhat modified\n>manner.\n\n>I'm not trying to say that the physical address approach to identity is\n>wrong.\n>I'll try to explain with an example.\n\n><EXAMPLE>\n>CREATE TABLE base (f1 int4);\n>CREATE TABLE derived (f2 int4) inherits(base);\n>\n>INSERT INTO base values(10);\n>INSERT INTO base values(20,20);\n\nHow would you know, that this is an entry for the derived subclass ?\n\n></EXAMPLE>\n> .....\n>\n>This of course as a result of the so-called \"is_a\" relation between\n>instances of the\n>derived class and instances of the base class.\n>So the query is polymorphic because it also operates on instances of classes\n>derived from the base class.\n>So to support polymorphism we need to have some form of identity which\n>is also valid between tables. As a result the current tids in postgresql\n>won't work because they are only valid within one table.\n\nsimply add the table id to the tid ?\n\n>You'll have noticed that my \"ideal\" system has different semantics\n>than postgresql. So as far as I concerned there is room for improvement\n>in postgresql.\n\n>According to me in the least triggers, indices and select/update/delete\n>statements should be polymorphic\n>(should work for instances of base classes and instances of derived\n>classes).\n\nThe successor of Illustra, the Informix Universial Server does this.\nWe have a comment in the code stating that the base* should probably be the default.\nLooks like others think that way too. :-) \n\nIn short:\n\t1. I think your work in the current direction is very valuable !\n\t2. I still suggest to implement it in a way that leaves the door open\n\t to not have an oid for every table/tuple per default.\n\t3. Tables without oid would simply not have all the OO functionality.\n\t tuples without the oid would not exist in the *OO world*\n\t4. I think it is valuable to have both OO and fast relational stuff. (ORDBMS)\n\nAndreas\n\n\n\n\n\n\n\n",
"msg_date": "Tue, 10 Mar 1998 10:13:08 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "WG: [HACKERS] newoid in invapi.c"
}
] |
[
{
"msg_contents": "I am using the following info to migrate a table from 6.2.1 to 6.3\n\n\\connect template1 \ndrop database magecal;\ncreate database magecal;\n\\connect magecal \n\\connect - julie\nCREATE TABLE dates (id int4, date char(10), news char(1024)) archive = none;\nCOPY dates FROM stdin;\n1\t1998Jan16\t<H3>Calendar Created</H3> !\n \n\\.\nCREATE INDEX dates_indx on dates using btree ( id int4_ops, date bpchar_ops );\n\n\nnow if I connect to the database and do\n\nselect attlen from pg_attribute where attname='news';\n\nI get\n-1\n\nThis doesn't seem quite right.\n\nJulie\n\n-- \n[ Julia Anne Case ] [ Ships are safe inside the harbor, ]\n[Programmer at large] [ but is that what ships are really for. ] \n[ Admining Linux ] [ To thine own self be true. ]\n[ Windows/WindowsNT ] [ Fair is where you take your cows to be judged. ]\n",
"msg_date": "Tue, 10 Mar 1998 10:00:56 +0000",
"msg_from": "\"Julia A.Case\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "attlen weirdness?"
},
{
"msg_contents": "Quoting The Hermit Hacker ([email protected]):\n> \n> Bruce? Guys? Any word on this one? I *thought* I remembered reading\n> soething about the 'attlen' becoming -1, but just checked the mailing list\n> archives and couldn't find anything :(\n>\n\n\tIt appears as if atttypmod now has the field length + 4 ???\n\n Julie\n\n-- \n[ Julia Anne Case ] [ Ships are safe inside the harbor, ]\n[Programmer at large] [ but is that what ships are really for. ] \n[ Admining Linux ] [ To thine own self be true. ]\n[ Windows/WindowsNT ] [ Fair is where you take your cows to be judged. ]\n",
"msg_date": "Wed, 11 Mar 1998 08:21:24 +0000",
"msg_from": "\"Julia A.Case\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "Quoting Bruce Momjian ([email protected]):\n> I assume you are asking of attlen for char() and varchar() have the max\n> length. The answer is now NO. They are varlena structures like text,\n> and so have a attlen of -1. Atttypmod for that pg_attribute row now has\n> the defined length.\n> \n\n\tEven for a char(4) field? ok, if attlen doesn't give the\nlength of a field, how do I find it?\n\nJulie\n\n-- \n[ Julia Anne Case ] [ Ships are safe inside the harbor, ]\n[Programmer at large] [ but is that what ships are really for. ] \n[ Admining Linux ] [ To thine own self be true. ]\n[ Windows/WindowsNT ] [ Fair is where you take your cows to be judged. ]\n",
"msg_date": "Wed, 11 Mar 1998 10:16:05 +0000",
"msg_from": "\"Julia A.Case\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "Quoting Bruce Momjian ([email protected]):\n> \n> pg_attribute.atttypmod.\n> \n\tthis field appears to be field size + 4\n in a char(1024) field it returns 1028\n in a char(10) it returns 14\n\n is it padded?\n\n Julie\n\n-- \n[ Julia Anne Case ] [ Ships are safe inside the harbor, ]\n[Programmer at large] [ but is that what ships are really for. ] \n[ Admining Linux ] [ To thine own self be true. ]\n[ Windows/WindowsNT ] [ Fair is where you take your cows to be judged. ]\n",
"msg_date": "Wed, 11 Mar 1998 10:21:29 +0000",
"msg_from": "\"Julia A.Case\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "\nBruce? Guys? Any word on this one? I *thought* I remembered reading\nsoething about the 'attlen' becoming -1, but just checked the mailing list\narchives and couldn't find anything :(\n\nThanks...\n\nOn Tue, 10 Mar 1998, Julia A.Case wrote:\n\n> I am using the following info to migrate a table from 6.2.1 to 6.3\n> \n> \\connect template1 \n> drop database magecal;\n> create database magecal;\n> \\connect magecal \n> \\connect - julie\n> CREATE TABLE dates (id int4, date char(10), news char(1024)) archive = none;\n> COPY dates FROM stdin;\n> 1\t1998Jan16\t<H3>Calendar Created</H3> !\n> !\n> !\n> \n> \\.\n> CREATE INDEX dates_indx on dates using btree ( id int4_ops, date bpchar_ops );\n> \n> \n> now if I connect to the database and do\n> \n> select attlen from pg_attribute where attname='news';\n> \n> I get\n> -1\n> \n> This doesn't seem quite right.\n> \n> Julie\n> \n> -- \n> [ Julia Anne Case ] [ Ships are safe inside the harbor, ]\n> [Programmer at large] [ but is that what ships are really for. ] \n> [ Admining Linux ] [ To thine own self be true. ]\n> [ Windows/WindowsNT ] [ Fair is where you take your cows to be judged. ]\n> \n\n",
"msg_date": "Wed, 11 Mar 1998 08:11:48 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "> \n> Quoting The Hermit Hacker ([email protected]):\n> > \n> > Bruce? Guys? Any word on this one? I *thought* I remembered reading\n> > soething about the 'attlen' becoming -1, but just checked the mailing list\n> > archives and couldn't find anything :(\n> >\n> \n> \tIt appears as if atttypmod now has the field length + 4 ???\n\nI am the atttypmod guy, but there is not enough context to answer this\nquestion.\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",
"msg_date": "Wed, 11 Mar 1998 09:41:07 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "> \n> Quoting The Hermit Hacker ([email protected]):\n> > \n> > Bruce? Guys? Any word on this one? I *thought* I remembered reading\n> > soething about the 'attlen' becoming -1, but just checked the mailing list\n> > archives and couldn't find anything :(\n> >\n> \n> \tIt appears as if atttypmod now has the field length + 4 ???\n\nI assume you are asking of attlen for char() and varchar() have the max\nlength. The answer is now NO. They are varlena structures like text,\nand so have a attlen of -1. Atttypmod for that pg_attribute row now has\nthe defined length.\n\nIn the old days, pre 6.3, pg_type.typlen != pg_attribute.attlen for\nchar() and varchar(), causing all sorts of type-conditional wierness in\nthe backend for these types. With the new code, pg_type.typlen =\npg_attribute.attlen all the time.\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",
"msg_date": "Wed, 11 Mar 1998 09:44:43 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "> \n> Quoting Bruce Momjian ([email protected]):\n> > I assume you are asking of attlen for char() and varchar() have the max\n> > length. The answer is now NO. They are varlena structures like text,\n> > and so have a attlen of -1. Atttypmod for that pg_attribute row now has\n> > the defined length.\n> > \n> \n> \tEven for a char(4) field? ok, if attlen doesn't give the\n> length of a field, how do I find it?\n\npg_attribute.atttypmod.\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",
"msg_date": "Wed, 11 Mar 1998 10:20:35 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "> \n> Quoting Bruce Momjian ([email protected]):\n> > \n> > pg_attribute.atttypmod.\n> > \n> \tthis field appears to be field size + 4\n> in a char(1024) field it returns 1028\n> in a char(10) it returns 14\n> \n> is it padded?\n> \n\nYes. All varlena sizes store the VARHDRSZ, variable header size of 4,\nin the length field.\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",
"msg_date": "Wed, 11 Mar 1998 10:53:10 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] attlen weirdness?"
}
] |
[
{
"msg_contents": "\nRedirected to 'the proper list' - [email protected]\n\n\nOn Tue, 10 Mar 1998, Gabriel Akos wrote:\n\n> Hi!\n> \n> If I download the CVS-Tree from the main site, then I'll have the most\n> recent version of PostgreSQL?\n\n\tUsing CVSup? Yes...\n\n\n",
"msg_date": "Tue, 10 Mar 1998 07:45:09 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] Development versions..."
}
] |
[
{
"msg_contents": "Hi, all.\n\nDoes anyone know if there is any PostgreSQL mirror that allows cvsup\naccess? I'm in Europe (in Spain), and the overseas connection with\npostgresql.org is sllllloooooowwwwwww...\n\n-------------------------------------------------------------------\nPedro José Lobo Perea Tel: +34 1 336 78 19\nCentro de Cálculo Fax: +34 1 331 92 29\nEUIT Telecomunicación - UPM e-mail: [email protected]\n\n",
"msg_date": "Tue, 10 Mar 1998 16:24:13 +0100 (MET)",
"msg_from": "\"Pedro J. Lobo\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Is there any cvsup mirror?"
}
] |
[
{
"msg_contents": "-----Original Message-----\nFrom: Zeugswetter Andreas <[email protected]>\nTo: 'pgsql-hackers' <[email protected]>\nDate: dinsdag 10 maart 1998 16:26\nSubject: WG: [HACKERS] newoid in invapi.c\n\n\n>>>There is simply very much that speaks against the use of oid's a la long.\n>>Yes I agree that the way oids are implemented now has problems.\n>>However I choose to see these problems as \"implementation details\".\n>\n>No, here I disagree, a globally unique identifier like oid is an\narchitecture and strategy thing.\n>As such it needs a lot of thought and care.\n\nYes, you are right, it does.\n\n>\n>>>Illustra defines a \"reference\" maybe we should dig into that ?\n>>>create table person (\n>>> name char (16),\n>>> mother ref(person),\n>>> father ref(person)\n>>>)\n>>>\n>>Ok, if such a reference is unique within a table then we've got something\n>>similar to tids if they are not they would more resemble oids.\n>>I don't know which is the case for illustra but recalling it's\n>>heritage...?!?\n>\n>Illustra uses oid's, and of course it suffers the same bottleneck on a\nmulti CPU System.\n>\n>>I presume that illustra would allow me to extend the above like in the\n>>following:\n>\n>>create table teacher (\n>> course char (32)\n>>) inherits (person);\n>\n>Yup, syntax is: create table teacher (course char(32)) under person;\n>\n>>If the illustra system allows me to insert a teacher object as my\n>>mother then the illustra reference is not likely to be implemented as\n>>physical\n>\n>No, in the above schema the teacher instance gets its own identity, same as\npostgresql.\n\nWhat a pity.\n\n\n>\n>>reference (tid) but more likely with some logical reference (oid).\n>\n>>I really hope the last suggestion is the case as it would much\n>>resemble that which I would like to see in postgresql.\n>\n>>>A unique pointer to a row for me is always:\n>>>dbid + tableid + fileid + primary key (or even rowid)\n>\n>>In a relational system, yes. In a OO system not necesarily.\n>>Because as in the example above, a lady who happens to be my mother\n>>may also happen to be a teacher. The identity of \"my mother\" and the\n>>identity of \"my mother the teacher\" should be the same.\n>\n>Agreed, but currently not the case. Let me explain further:\n>If you have parents, teachers under parents and petowners under parents\n:-).\n>Then the teachers that are also petowners would get 2 oid's. One for\nteacher\n>and one for petowner. I agree that this is not perfect, and can not be\nsolved with the current\n>architecture :-(\nI want to think about this for a while. Maybe there is some cleverness to be\nfound.\n>\n>>It would be a pity if \"my mother\" would have two identities just because\n>>of the way my database system stores it's data.\n>>>\n\n<snip>\n>>So to support polymorphism we need to have some form of identity which\n>>is also valid between tables. As a result the current tids in postgresql\n>>won't work because they are only valid within one table.\n>\n>simply add the table id to the tid ?\nMight work.\n>\n>>You'll have noticed that my \"ideal\" system has different semantics\n>>than postgresql. So as far as I concerned there is room for improvement\n>>in postgresql.\n>\n>>According to me in the least triggers, indices and select/update/delete\n>>statements should be polymorphic\n>>(should work for instances of base classes and instances of derived\n>>classes).\n>\n>The successor of Illustra, the Informix Universial Server does this.\n>We have a comment in the code stating that the base* should probably be the\ndefault.\n>Looks like others think that way too. :-)\nDoesn't the \"Postgresql Universal Server\" sound nice?\n>\n>In short:\n> 1. I think your work in the current direction is very valuable !\n> 2. I still suggest to implement it in a way that leaves the door open\n> to not have an oid for every table/tuple per default.\n> 3. Tables without oid would simply not have all the OO functionality.\n> tuples without the oid would not exist in the *OO world*\n> 4. I think it is valuable to have both OO and fast relational stuff.\n(ORDBMS)\n>\nYour points have been taken.\n\nThanks,\nMaurice.\n\n",
"msg_date": "Tue, 10 Mar 1998 19:53:46 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] newoid in invapi.c"
}
] |
[
{
"msg_contents": "Hi:\n\nLittle bug causes postgres to die:\nsc=> select count(*) from urlcont where (UPPER(url) LIKE '%APOLLO%');\nField| Value\n-- RECORD 0 --\ncount| 1\n(1 row)\n\nsc=> select count(*) from urlcont where (UPPER(content) LIKE '%APOLLO%');\nFATAL 1: palloc failure: memory exhausted\nsc=> select count(*) from urlcont where (UPPER(url) LIKE '%APOLLO%');\nBroken pipe\nbash$ psql -x sc\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: sc\n\nsc=> select count(*) from urlcont where (UPPER(url) LIKE '%APOLLO%');\nField| Value\n-- RECORD 0 --\ncount| 1\n(1 row)\nsc-> \\d urlcont\n\nTable = urlcont\n+----------------------------------+----------------------------------+-------+\n| Field | Type |\nLength|\n+----------------------------------+----------------------------------+-------+\n| url | varchar() |\n256 |\n| pos | int4 |\n4 |\n| content | varchar() |\n1024 |\n+----------------------------------+----------------------------------+-------+\n\n\nAlso, I suppose I should get on the postgres mailing list. You prolly\nalready have this done, so if you have a procmail filter set up for the\nlist, could you please clip it out and send it to me?\n\nthanks\n-Mike\n\n",
"msg_date": "Tue, 10 Mar 1998 18:04:52 -0400 (AST)",
"msg_from": "[email protected] (Michael Richards)",
"msg_from_op": true,
"msg_subject": "PostgreSQL bug in UPPER"
}
] |
[
{
"msg_contents": "> \n> For each test, I will list 2 numbers here, the first being with the default\n> number of buffers, the second with 256 buffers (ie. 2048K)\n> \n> btw. just to give all possible information: I start postgreSQL with the\n> following arguments:\n> \n> -i -b /usr/local/pgsql/bin/postgres -S 1024 -o \"-o mylogfile -F -d 1\"\n> \n> > OK, I meant to reply to this, but forgot to. Let's take the second query:\n> >\n> >\tselect count(*) from artist_fti where string ~ '^lling';\n> >\n> > Have you tried adding -B buffers. This, I think, would help. If you\n> > don't flush the cache, how long does a second identical query take? \n> \n> This is 3 times the above query without trashing in between and with\n> the default number of buffers:\n> \n> \t0.030u 0.020s 0:23.41 0.2% 0+0k 0+0io 201pf+0\n> \t0.040u 0.010s 0:01.06 4.7% 0+0k 0+0io 184pf+0w\n> \t0.030u 0.040s 0:00.70 10.0% 0+0k 0+0io 184pf+0w\n\nOK, I think this actually tells the whole story. The query goes from 22\nseconds to 0.75 seconds because all of the btree indexes are in the\nbuffer. Looks like the optimizer is clearly understanding both parts\nof the restriction, which is good. No problems there.\n\nAlso seems the index fits easily in the cache.\n\nNow my question is \"What is the performance problem?\" Doesn't the\nshared buffer cache keep these in memory, so the first one is slow, but\nthe rest are fast, or does the buffer get flushed a lot, and performance\nis terrible on the first query after that. Does going after other words\nflush the cache for previous words searched?\n\nI must say, I am surprised that the buffer cache causes it to speed up\nso much. Are these really slow disks?\n\nIn the Ingres case, the first query or two was not as quick, but they\nspeeded up, and kept pretty fast for the rest of the day. Also, the\nindexes where ISAM, which has a less sophisticated/less overhead way of\nindexing than btree.\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",
"msg_date": "Tue, 10 Mar 1998 17:20:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: your mail"
}
] |
[
{
"msg_contents": "[email protected] wrote:\n\n> Thomas G. Lockhart wrote:\n> >\n> [snip]\n> > a \"#\" to the defaultdate parameter line will give you the usual behavior of\n> > picking up the latest version of the tree.\n> >\n> > - Tom\n> >\n> >\n> > # This file represents the standard CVSup distribution file\n> > # for the PostgreSQL ORDBMS project\n> > # Modified by [email protected] 1997-08-28\n> > # - Point to my local snapshot source tree\n> > #\n> > # Defaults that apply to all the collections\n> > *default host=postgresql.org\n> > *default compress\n> > *default release=cvs\n> > *default delete use-rel-suffix\n> > *default tag=.\n> > *default date=98.01.20.00.00.00\n> >\n> > # base directory points to where CVSup will store its 'bookmarks' file(s)\n> > *default base=/opt/postgres/oldtree # /usr/local/pgsql\n> >\n> > # prefix directory points to where CVSup will store the actual\n> > distribution(s)\n> > *default prefix=/opt/postgres/oldtree # /usr/local/pgsql\n> >\n> > # complete distribution, including all below\n> > pgsql\n> >\n> > # individual distributions vs 'the whole thing'\n> > # pgsql-doc\n> > # pgsql-perl5\n> > # pgsql-src\n>\n> I am trying to get this working under Linux and behind a firewall. Do\n> you know of any cvsup documentation that might tell me how to set this\n> up? Or do you know how to set this up offhand?\n\nafaik, the only thing you need to do is to install the CVSup client (I had\nposted a static binary for linux; a dynamic binary would require the Modula-3\nlibraries and I wasn't sure how to link a partial dynamic binary with those\nincluded).\n\nThen, modify your configuration file (as was done in my example above) to point\nto the right places on your system. Only the base directory needs to exist for\nthis to work (e.g. \"/opt/postgres/oldtree\") and CVSup will create the rest. To\nexecute CVSup, type\n\n cvsup -L 2 postgres.cvsup\n\nwhere \"postgres.cvsup\" is the name of your configuration file. The \"-L 2\" is a\nverbosity setting. For my system, I include a path name in front of the\nconfiguration file name, and have the whole thing aliased to a command \"pgget\",\nso just typing that does the update from postgresql.org.\n\nBuilding CVSup from scratch is not trivial, since Modula-3 must be installed\n(~200MB to install, ~50MB installed) and then CVSup must be built, working\naround some BSD-specific makefiles. Don't know if anyone has built a glibc2\nversion for us, or whether it is even necessary.\n\nGood luck.\n\n - Tom\n\n",
"msg_date": "Wed, 11 Mar 1998 03:06:09 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] postgres/alpha problems"
}
] |
[
{
"msg_contents": "> > I'd like to know what are the table repair tools. I examined a table with\n> > od -c\n> > and boy, that does not look like what I put in. So other than good old\n> > fashioned backups, how do you rebuild a corrupted file ?\n> > Just wonderin'\n> We don't need no sinking repair tools. :-)\n> Actually, we very rarely hear about any corruption problems, so there\n> are no repair tools.\n\nWell, I've been wondering the same thing. I guess no one has been caught badly\nenough to write the tools. It seems like one project could be to write a \"recover\nmode\" into the backend, which would allow pg_dump to run to completion even though\na table is damaged. Would need a command line switch (or an SQL command) and then\nsome checks in the backend to return gracefully rather than throw an error and\nquit.\n\n - Tom\n\n",
"msg_date": "Wed, 11 Mar 1998 03:45:51 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] Basic table repair info."
}
] |
[
{
"msg_contents": "I think I've tracked down a problem I see using the BETWEEN operator:\n\npostgres=> select 1 where float8(1) between 0.0 and 2.0;\nERROR: transformExpr: does not know how to transform node 105\npostgres=> select 1 where float8(1) >= 0.0 and float8(1) <= 2.0;\n?column?\n--------\n 1\n(1 row)\n\nI'm pretty sure that the problem stems from the fact that in gram.y the\nBETWEEN clause is expanded into the AND clauses as in the second\nexample, _but_ the parse tree for the function call is reused! So,\ntransformExpr() is run twice on the same part of the parse tree, and\ndoes not know how to cope.\n\nProbably the strongest solution would be to copy the entire parse tree.\nIs there already a function to do that? It seems like it would be a lot\nof work to start this from scratch.\n\nAnother possible solution would be to have transformExpr() accept a\npreviously transformed parse tree without damaging it. A simple minded\nfix having transformExpr() ignore the T_Const (the \"105\" node in the\nexample) and T_Var nodes did not work.\n\nYet another solution would be to have transformExpr() replicate parse\ntrees instead of having gram.y do it.\n\nAny thoughts on this?\n\n - Tom\n\n",
"msg_date": "Wed, 11 Mar 1998 04:20:08 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Function call problems with BETWEEN"
},
{
"msg_contents": "> I think I've tracked down a problem I see using the BETWEEN operator:\n>\n> postgres=> select 1 where float8(1) between 0.0 and 2.0;\n> ERROR: transformExpr: does not know how to transform node 105\n>\n> I'm pretty sure that the problem stems from the fact that in gram.y the\n> BETWEEN clause is expanded into the AND clauses as in the second\n> example, _but_ the parse tree for the function call is reused! So,\n> transformExpr() is run twice on the same part of the parse tree, and\n> does not know how to cope.\n> ...\n\n> Another possible solution would be to have transformExpr() accept a\n> previously transformed parse tree without damaging it. A simple minded\n> fix having transformExpr() ignore the T_Const (the \"105\" node in the\n> example) and T_Var nodes did not work.\n\nHmm. I forgot to set the return value for this. I now get:\n\npostgres=> select 1 where float8(1) between 0.0 and 2.0;\n?column?\n--------\n 1\n(1 row)\n\nBut I'm not really happy putting in this change without knowing what the\nside effects might be. I'll go ahead and try the regression tests...\n\n - Tom\n\n",
"msg_date": "Wed, 11 Mar 1998 04:44:38 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Function call problems with BETWEEN"
},
{
"msg_contents": "> Probably the strongest solution would be to copy the entire parse tree.\n> Is there already a function to do that? It seems like it would be a lot\n> of work to start this from scratch.\n> \n> Another possible solution would be to have transformExpr() accept a\n> previously transformed parse tree without damaging it. A simple minded\n> fix having transformExpr() ignore the T_Const (the \"105\" node in the\n> example) and T_Var nodes did not work.\n> \n> Yet another solution would be to have transformExpr() replicate parse\n> trees instead of having gram.y do it.\n> \n> Any thoughts on this?\n\nI am sure copyfuncs.c has some function to do this, but these are\nparse-specific structures. Not sure if they do have such functions. I\nconsidered the reuse a possible problem when I wrote the code, but could\nnot find any problems, so I left it.\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",
"msg_date": "Tue, 10 Mar 1998 23:49:23 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: Function call problems with BETWEEN"
}
] |
[
{
"msg_contents": "Dear all,\n\nDoes anybody know where I can get the latest ODBC Drivers for PostgresQL\n6.1/6.2 ?\nthanks !\n\nregards,\n--\n/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\\n\\ Michael Yeung\n/ Alpha Network Shop\n\\ Tel:(02)94133886\n/ Mobile : 0411-233-597\n\\ Fax:(02)94133617\n/ mailto:[email protected]\n\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\n\n\n",
"msg_date": "Wed, 11 Mar 1998 14:47:46 +1000",
"msg_from": "Michael Yeung <[email protected]>",
"msg_from_op": true,
"msg_subject": "ODBC DRIVERS FOR 6.1/6.2"
},
{
"msg_contents": "\nRedirected to 'the proper list' - [email protected]\n\n\nOn Wed, 11 Mar 1998, Michael Yeung wrote:\n\n> Dear all,\n> \n> Does anybody know where I can get the latest ODBC Drivers for PostgresQL\n> 6.1/6.2 ?\n\n\tThe latest ODBC drivers are v6.3 now, and part of the\ndistribution...there were protocol changes that, I believe, cause the\ncurrent ODBC to not work under v6.[12]\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Wed, 11 Mar 1998 01:17:02 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] ODBC DRIVERS FOR 6.1/6.2"
},
{
"msg_contents": "Hi,\n\nIs that meant that no more ODBC Drivers supported v6.1/6.2\n\nThe Hermit Hacker wrote:\n\n> Redirected to 'the proper list' - [email protected]\n>\n> On Wed, 11 Mar 1998, Michael Yeung wrote:\n>\n> > Dear all,\n> >\n> > Does anybody know where I can get the latest ODBC Drivers for PostgresQL\n> > 6.1/6.2 ?\n>\n> The latest ODBC drivers are v6.3 now, and part of the\n> distribution...there were protocol changes that, I believe, cause the\n> current ODBC to not work under v6.[12]\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\\ Michael Yeung\n/ Alpha Network Shop\n\\ Tel:(02)94133886\n/ Mobile : 0411-233-597\n\\ Fax:(02)94133617\n/ mailto:[email protected]\n\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\n\n\n",
"msg_date": "Thu, 12 Mar 1998 07:51:45 +1000",
"msg_from": "Michael Yeung <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] ODBC DRIVERS FOR 6.1/6.2"
},
{
"msg_contents": "On Thu, 12 Mar 1998, Michael Yeung wrote:\n\n> Hi,\n> \n> Is that meant that no more ODBC Drivers supported v6.1/6.2\n\n\tSorry...the only non-commercial ODBC driver that I know of, that\nships with our source code, now only supports v6.3, due to drastic\ninternal server changes...Openlink sells drivers that are most likely\nstill v6.2 compliant...\n\n\n\n > > The Hermit Hacker wrote:\n> \n> > Redirected to 'the proper list' - [email protected]\n> >\n> > On Wed, 11 Mar 1998, Michael Yeung wrote:\n> >\n> > > Dear all,\n> > >\n> > > Does anybody know where I can get the latest ODBC Drivers for PostgresQL\n> > > 6.1/6.2 ?\n> >\n> > The latest ODBC drivers are v6.3 now, and part of the\n> > distribution...there were protocol changes that, I believe, cause the\n> > current ODBC to not work under v6.[12]\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> \\ Michael Yeung\n> / Alpha Network Shop\n> \\ Tel:(02)94133886\n> / Mobile : 0411-233-597\n> \\ Fax:(02)94133617\n> / mailto:[email protected]\n> \\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\\/\n> \n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Wed, 11 Mar 1998 22:08:54 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] ODBC DRIVERS FOR 6.1/6.2"
}
] |
[
{
"msg_contents": "I ran some timing tests to check the performance of varchar() vs char16.\nThe results of the test indicate that there is no difference in\nperformance (within the timing scatter of the tests):\n\nchar16\tvc(16)\n 0.99s\t 1.05s\t1 row (this measures startup time, not types)\n39.29s\t39.28s\t~65000 rows\n\nThe char2,4,8,16 types seem to have no value-added over the\nbetter-supported char(), varchar(), text types; I am considering\nremoving them from the backend, and instead have the parser\ntransparently translate the types into varchar() (or char() - I'm not\ncertain which is a better match for the types) for v6.4. Applications\nwould not have to be changed.\n\nComments?\n\n - Tom\n\nThe test is included below:\n\n-- create table c16 (c char16);\ncreate table c16 (c varchar(16));\n\ncopy c16 from 'c16.copy';\n\nselect count(*) from c16 where c = 'hi there';\n\nselect count(*) from c16 where c = 'test string';\n\nselect count(*) from c16 where c != 'hi there';\n\nselect count(*) from c16 where c != 'test string';\n\ndelete from c16;\n\ndrop table c16;\n",
"msg_date": "Wed, 11 Mar 1998 14:39:23 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "varchar() vs char16 performance"
},
{
"msg_contents": "> \n> I ran some timing tests to check the performance of varchar() vs char16.\n> The results of the test indicate that there is no difference in\n> performance (within the timing scatter of the tests):\n> \n> char16\tvc(16)\n> 0.99s\t 1.05s\t1 row (this measures startup time, not types)\n> 39.29s\t39.28s\t~65000 rows\n> \n> The char2,4,8,16 types seem to have no value-added over the\n> better-supported char(), varchar(), text types; I am considering\n> removing them from the backend, and instead have the parser\n> transparently translate the types into varchar() (or char() - I'm not\n> certain which is a better match for the types) for v6.4. Applications\n> would not have to be changed.\n\nFine, remove them. You may want to keep 'char' because of the reduced\noverhead compared to char(1), but the others certainly can be removed.\nSeems like you have not mentioned char, so you may no intension of\nremoving it.\n\nI would map to char(). They are fixed size, and the old types were\nfixed size too. I know char16 may be better for varchar(), but that is\nnot as clean a translation from the old type.\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",
"msg_date": "Wed, 11 Mar 1998 10:26:15 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": "> Date: Wed, 11 Mar 1998 14:39:23 +0000\n> From: \"Thomas G. Lockhart\" <[email protected]>\n...\n> The char2,4,8,16 types seem to have no value-added over the\n> better-supported char(), varchar(), text types; I am considering\n> removing them from the backend, and instead have the parser\n> transparently translate the types into varchar() (or char() - I'm not\n> certain which is a better match for the types) for v6.4. Applications\n> would not have to be changed.\n> \n> Comments?\n\nI'm not up on the details of PostgreSQL's differing character types,\nbut wonder - would the proposed change break any apps where trailing\n(or leading?) whitespace is significant? Not that I'm running any\n...\n\n",
"msg_date": "Wed, 11 Mar 1998 11:15:34 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": "\n\nn Wed, 11 March 1998, at 11:15:34, Hal Snyder wrote:\n\n> > The char2,4,8,16 types seem to have no value-added over the\n> > better-supported char(), varchar(), text types; I am considering\n> > removing them from the backend, and instead have the parser\n> > transparently translate the types into varchar() (or char() - I'm not\n> > certain which is a better match for the types) for v6.4. Applications\n> > would not have to be changed.\n> > \n> > Comments?\n> \n> I'm not up on the details of PostgreSQL's differing character types,\n> but wonder - would the proposed change break any apps where trailing\n> (or leading?) whitespace is significant? Not that I'm running any\n> ...\n> \n\nHeh.. migrating to 6.3. was a surprise for me.. I certainly wasn't\nexpecting whitespace pads, and there are some cases where it makes a\nbig difference!\n",
"msg_date": "Wed, 11 Mar 1998 12:29:14 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": ">I ran some timing tests to check the performance of varchar() vs char16.\n>The results of the test indicate that there is no difference in\n>performance (within the timing scatter of the tests):\n>\n>char16\tvc(16)\n> 0.99s\t 1.05s\t1 row (this measures startup time, not types)\n>39.29s\t39.28s\t~65000 rows\n>\n>The char2,4,8,16 types seem to have no value-added over the\n>better-supported char(), varchar(), text types; I am considering\n>removing them from the backend, and instead have the parser\n>transparently translate the types into varchar() (or char() - I'm not\n>certain which is a better match for the types) for v6.4. Applications\n>would not have to be changed.\n>\n>Comments?\n\nPlease do not remove char2! Some users uses it for making an array of\nchar.\n\ncreate table c(c char2[]);\n\nSeems strange? Yes. Actually what he wanted to do was:\n\ntest=> create table c(c char[]);\nERROR: parser: parse error at or near \"[\"\n--\nTatsuo Ishii\[email protected]\n",
"msg_date": "Mon, 16 Mar 1998 15:20:36 +0900",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance "
},
{
"msg_contents": "> \n> >I ran some timing tests to check the performance of varchar() vs char16.\n> >The results of the test indicate that there is no difference in\n> >performance (within the timing scatter of the tests):\n> >\n> >char16\tvc(16)\n> > 0.99s\t 1.05s\t1 row (this measures startup time, not types)\n> >39.29s\t39.28s\t~65000 rows\n> >\n> >The char2,4,8,16 types seem to have no value-added over the\n> >better-supported char(), varchar(), text types; I am considering\n> >removing them from the backend, and instead have the parser\n> >transparently translate the types into varchar() (or char() - I'm not\n> >certain which is a better match for the types) for v6.4. Applications\n> >would not have to be changed.\n> >\n> >Comments?\n> \n> Please do not remove char2! Some users uses it for making an array of\n> char.\n> \n> create table c(c char2[]);\n> \n> Seems strange? Yes. Actually what he wanted to do was:\n> \n> test=> create table c(c char[]);\n> ERROR: parser: parse error at or near \"[\"\n\nMaybe we just need to fix char[].\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",
"msg_date": "Mon, 16 Mar 1998 01:24:06 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": "\nhmm.. well until the grammar gets fixed, create table c(c _char)\nshould work, as _typename is the typename from an array of that type.\nAlthough I don't see what advantages a character array has over text?\n\nOn Mon, 16 March 1998, at 15:20:36, [email protected] wrote:\n\n> Please do not remove char2! Some users uses it for making an array of\n> char.\n> \n> create table c(c char2[]);\n> \n> Seems strange? Yes. Actually what he wanted to do was:\n> \n> test=> create table c(c char[]);\n> ERROR: parser: parse error at or near \"[\"\n> --\n> Tatsuo Ishii\n> [email protected]\n",
"msg_date": "Mon, 16 Mar 1998 00:15:08 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance "
},
{
"msg_contents": ">hmm.. well until the grammar gets fixed, create table c(c _char)\n>should work, as _typename is the typename from an array of that type.\n>Although I don't see what advantages a character array has over text?\n\nGood point. I asked him the same question. He needed a chararcter\nvector in that each char represents an individual value. Rather than\nusing substr(), an array might be more intuitive, he said.\n\n>On Mon, 16 March 1998, at 15:20:36, [email protected] wrote:\n>\n>> Please do not remove char2! Some users uses it for making an array of\n>> char.\n>> \n>> create table c(c char2[]);\n>> \n>> Seems strange? Yes. Actually what he wanted to do was:\n>> \n>> test=> create table c(c char[]);\n>> ERROR: parser: parse error at or near \"[\"\n--\nTatsuo Ishii\[email protected]\n",
"msg_date": "Mon, 16 Mar 1998 17:27:17 +0900",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance "
},
{
"msg_contents": "> > >The char2,4,8,16 types seem to have no value-added over the\n> > >better-supported char(), varchar(), text types; I am considering\n> > >removing them from the backend, and instead have the parser\n> > >transparently translate the types into varchar() or char()\n> Maybe we just need to fix char[].\n\nThere are notes in the source code from Jolly wondering whether arrays\nof char/varchar would work, and specifically disallowing it \"for now\".\n\nimho, even though there is one user forcing arrays of single characters\nby invoking char2, rather than using text and substrings, that is not\nsufficient to keep this obsolete capability in the backend forever.\n\nThere are two ways to smoothly transition to a system in which char2-16\nis not a native built-in type:\n\n1) modify the parser to automatically translate char2-16 into\nchar(2-16). This will not allow arrays of char2.\n\n2) put char2-16 into a user-loadable module, which could be configured\ninto the template1 database if the installer chooses. This would retain\nall current char2-16 capabilities.\n\nThese options are mutually exclusive, since implementing (1) would mean\nthe parser would not allow user-defined types for (2).\n\nI had thought that char2-16 add _no_ functionality over the char() and\nvarchar() types; Tatsuo points out at least one capability which they\nhave. Are there any others?\n\n - Tom\n",
"msg_date": "Mon, 16 Mar 1998 15:12:49 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": ">imho, even though there is one user forcing arrays of single characters\n>by invoking char2, rather than using text and substrings, that is not\n>sufficient to keep this obsolete capability in the backend forever.\n\nI agree with you. With the suggestion from Brett, the user now have\nbetter solution than using char2, I think.\n--\nTatsuo Ishii\[email protected]\n",
"msg_date": "Thu, 19 Mar 1998 10:12:48 +0900",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance "
}
] |
[
{
"msg_contents": "> \n> Hi,\n> \n> I have done a little more testing, and the performance bottleneck\n> seems definitely be memory related. Note that it does not really seems\n> to be dependend on buffer-settings, but really on disk caches.\n> \n> additional info:\n> \tthe index on this table is around 155 Megs big\n> \n> Now, if I do a count(*) on '^rol', after the second query, this takes\n> around 1 second, and returns 2528.\n> \n> On the other hand, if I do a count(*) on '^ric', his takes consequently\n> around 1:30 mins, no matter how often I run it. This returns 7866.\n> \n> A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> running it several times.\n\nWow, this makes no sense to me.\n> \n> Running different queries in between affect these times.\n> \n> My computer has 64 Megs of RAM, and I'm running X (linux 2.0.30)\n> \n> So, it seems to me that with this amount of memory, my system is usable\n> only for smaller tables (ie. no 550,000 rows in the main table, and no\n> 4,500,000 rows in the 'index' table). If I want better performance for this\n> setup, I need faster disks and more (how much more?) memory. Wish I could\n> test this somehow.... Maybe I can ask my sysop at the university if I may\n> test this on the dual PPro with 256 megs, but I don't think so....\n> \n> Maarten\n> \n> \n> _____________________________________________________________________________\n> | TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n> | Department of Electrical Engineering |\n> | Computer Architecture and Digital Technique section |\n> | [email protected] |\n> -----------------------------------------------------------------------------\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",
"msg_date": "Wed, 11 Mar 1998 09:48:47 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: your mail"
},
{
"msg_contents": "> \n> Hi,\n> \n> I have done a little more testing, and the performance bottleneck\n> seems definitely be memory related. Note that it does not really seems\n> to be dependend on buffer-settings, but really on disk caches.\n> \n> additional info:\n> \tthe index on this table is around 155 Megs big\n> \n> Now, if I do a count(*) on '^rol', after the second query, this takes\n> around 1 second, and returns 2528.\n> \n> On the other hand, if I do a count(*) on '^ric', his takes consequently\n> around 1:30 mins, no matter how often I run it. This returns 7866.\n> \n> A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> running it several times.\n\nWow, this makes no sense to me. How can this happen? 'rol' is fast, but\n'ric' which returns 3 times as many rows takes 1.5 minutes, no matter how\noften it is run? And 64MB is a lot of memory.\n\nDoes profiling show that 'rol' and 'ric' take the same amount of CPU\ntime in the backend? Is EXPLAIN identical for these two? If so, there\nmust be something different about the I/O demands of the two queries,\nbut I can't really understand what that difference could be. If I had\nto take a guess, it is that 'ric' is somehow causing a cache flush for\neach row it retrieves, while 'rol' fits in the cache.\n\nSeveral million rows is a good sized table, but certainly not too big\nfor PostgreSQL. I am sure people have larger tables.\n\nI am kind of tempted to test it on my machine here. I have BSD/OS with\na PP200 and 64MB RAM. Can you send me something I can recreate here? \nPerhaps put it on our ftp site?\n\n> \n> Running different queries in between affect these times.\n> \n> My computer has 64 Megs of RAM, and I'm running X (linux 2.0.30)\n> \n> So, it seems to me that with this amount of memory, my system is usable\n> only for smaller tables (ie. no 550,000 rows in the main table, and no\n> 4,500,000 rows in the 'index' table). If I want better performance for this\n> setup, I need faster disks and more (how much more?) memory. Wish I could\n> test this somehow.... Maybe I can ask my sysop at the university if I may\n> test this on the dual PPro with 256 megs, but I don't think so....\n> \n> Maarten\n> \n> \n> _____________________________________________________________________________\n> | TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n> | Department of Electrical Engineering |\n> | Computer Architecture and Digital Technique section |\n> | [email protected] |\n> -----------------------------------------------------------------------------\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",
"msg_date": "Wed, 11 Mar 1998 10:02:14 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "indexing words slow"
},
{
"msg_contents": "On Wed, 11 Mar 1998, Bruce Momjian wrote:\n\n> > \n> > Hi,\n> > \n> > I have done a little more testing, and the performance bottleneck\n> > seems definitely be memory related. Note that it does not really seems\n> > to be dependend on buffer-settings, but really on disk caches.\n> > \n> > additional info:\n> > \tthe index on this table is around 155 Megs big\n> > \n> > Now, if I do a count(*) on '^rol', after the second query, this takes\n> > around 1 second, and returns 2528.\n> > \n> > On the other hand, if I do a count(*) on '^ric', his takes consequently\n> > around 1:30 mins, no matter how often I run it. This returns 7866.\n> > \n> > A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> > running it several times.\n\nbtw. to make things clearer on what I mean with '^lling' and '^tones', I \nreally mean \"'^lling' *AND* '^tones'\", ie. a join :) actually pretty good \ndon't ya think? :)\n \n> Wow, this makes no sense to me. How can this happen? 'rol' is fast, but\n> 'ric' which returns 3 times as many rows takes 1.5 minutes, no matter how\n> often it is run? And 64MB is a lot of memory.\n> \n> Does profiling show that 'rol' and 'ric' take the same amount of CPU\n> time in the backend? Is EXPLAIN identical for these two? If so, there\n\nProfiling show different time (I guess you're talking about the \ncumulative time for the query?). 'rol' takes 0.68 secs cumulative, 'ric' \ntakes 1.69 secs cumulative. EXPLAIN's are identical.\n\n> I am kind of tempted to test it on my machine here. I have BSD/OS with\n> a PP200 and 64MB RAM. Can you send me something I can recreate here? \n> Perhaps put it on our ftp site?\n\nI *can* put something up through http on my universiy machine I guess. \nIt'll be a pg_dump of that table. Note however that this is generated \nfrom a commercial database, so please promise me to delete it after \ntesting :) don't wanna get into any legal trouble (ah what, who's gonna \nfind out :).\n\nI'll prepare a mail for you (ie. bruce) either tonight or tomorrow \nmorning (CET), with the source and a 'psql script' to set things up, as a \ntar-file. The data itself you'll have to get through the web, much to \nlarge to mail (over 10 megs gzip -9).\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n\n",
"msg_date": "Wed, 11 Mar 1998 22:27:00 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: indexing words slow"
},
{
"msg_contents": "Maarten Boekhold wrote:\n> \n> On Wed, 11 Mar 1998, Bruce Momjian wrote:\n> \n> > > \n> > > Hi,\n> > > \n> > > I have done a little more testing, and the performance bottleneck\n> > > seems definitely be memory related. Note that it does not really seems\n> > > to be dependend on buffer-settings, but really on disk caches.\n> > > \n> > > additional info:\n> > > \tthe index on this table is around 155 Megs big\n> > > \n> > > Now, if I do a count(*) on '^rol', after the second query, this takes\n> > > around 1 second, and returns 2528.\n> > > \n> > > On the other hand, if I do a count(*) on '^ric', his takes consequently\n> > > around 1:30 mins, no matter how often I run it. This returns 7866.\n> > > \n> > > A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> > > running it several times.\n> \n> btw. to make things clearer on what I mean with '^lling' and '^tones', I \n> really mean \"'^lling' *AND* '^tones'\", ie. a join :) actually pretty good \n> don't ya think? :)\n\nThis sounds like an unsatisfyable query, perhaps if the database\nfigured this out, it could return zero rows without even hitting the\nindex. If the first item matched, the first character is an 'l', if\nthe second matches, a 't'. It can't be both an 'l' and a 't'!\n\nWe could also do this for some other queries, like:\n\nselect ... from ... where a<b and b<c and c<d\n\nIf a and d are parameters to the query (constants), and d<a, then\nthere can never be a row returned.\n\nOcie\n",
"msg_date": "Wed, 11 Mar 1998 16:55:12 -0800 (PST)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: indexing words slow"
},
{
"msg_contents": "> > btw. to make things clearer on what I mean with '^lling' and '^tones', I \n> > really mean \"'^lling' *AND* '^tones'\", ie. a join :) actually pretty good \n> > don't ya think? :)\n> \n> This sounds like an unsatisfyable query, perhaps if the database\n> figured this out, it could return zero rows without even hitting the\n> index. If the first item matched, the first character is an 'l', if\n> the second matches, a 't'. It can't be both an 'l' and a 't'!\n\nHe has to instances of the table, and is looking for index entries with\nboth words. We have only been sharing part of the conversation.\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",
"msg_date": "Wed, 11 Mar 1998 22:28:32 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: indexing words slow"
},
{
"msg_contents": "On Wed, 11 Mar 1998 [email protected] wrote:\n\n> Maarten Boekhold wrote:\n> > \n> > On Wed, 11 Mar 1998, Bruce Momjian wrote:\n> > \n> > > > \n> > > > Hi,\n> > > > \n> > > > I have done a little more testing, and the performance bottleneck\n> > > > seems definitely be memory related. Note that it does not really seems\n> > > > to be dependend on buffer-settings, but really on disk caches.\n> > > > \n> > > > additional info:\n> > > > \tthe index on this table is around 155 Megs big\n> > > > \n> > > > Now, if I do a count(*) on '^rol', after the second query, this takes\n> > > > around 1 second, and returns 2528.\n> > > > \n> > > > On the other hand, if I do a count(*) on '^ric', his takes consequently\n> > > > around 1:30 mins, no matter how often I run it. This returns 7866.\n> > > > \n> > > > A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> > > > running it several times.\n> > \n> > btw. to make things clearer on what I mean with '^lling' and '^tones', I \n> > really mean \"'^lling' *AND* '^tones'\", ie. a join :) actually pretty good \n> > don't ya think? :)\n> \n> This sounds like an unsatisfyable query, perhaps if the database\n> figured this out, it could return zero rows without even hitting the\n> index. If the first item matched, the first character is an 'l', if\n> the second matches, a 't'. It can't be both an 'l' and a 't'!\n\nOK, I wasn't clear enough:\n\nselect count(*) from table t1, table t2 where t1.string ~ '^lling' and \nt2.string '^tones';\n\nOfcourse the can be done (note that this is a table with 4,500,000 rows).\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Thu, 12 Mar 1998 11:05:11 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: indexing words slow"
},
{
"msg_contents": "> \n> On Wed, 11 Mar 1998, Bruce Momjian wrote:\n> \n> > > \n> > > Hi,\n> > > \n> > > I have done a little more testing, and the performance bottleneck\n> > > seems definitely be memory related. Note that it does not really seems\n> > > to be dependend on buffer-settings, but really on disk caches.\n> > > \n> > > additional info:\n> > > \tthe index on this table is around 155 Megs big\n> > > \n> > > Now, if I do a count(*) on '^rol', after the second query, this takes\n> > > around 1 second, and returns 2528.\n> > > \n> > > On the other hand, if I do a count(*) on '^ric', his takes consequently\n> > > around 1:30 mins, no matter how often I run it. This returns 7866.\n> > > \n> > > A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> > > running it several times.\n\nOK, I have the artist_fti table with 4.5 million rows, with an index\nartist_fti_idx on artist_fti.string. I don't have a 'product' table\nbecause I don't have the disk space, but that is not really a problem\nfor testing. \n\nI used the product table to populate the artits_fti, then deleted the\ntable so I could create the artist_fti_idx index. Single table, no\njoin.\n\nI am running on BSD/OS 3.1 PP200 with SCSI Ultra Barracuda disks.\n\nI am seeing the same thing you are. 'lling' and 'tones' take 11-22\nseconds on the first few runs, then return almost immediately. If I do\nanother query and come back to the first one, I have the same slowness,\nwith the disk light being on almost constantly. Certain queries seem to\nlend themselves to speeding up, while 'rol' never seems to get really\nfast.\n\nI have to conclude that because of the way this table is created by\nslicing words, its layout is almost random. The table is 272MB, and\nwith 8k/page, that is 34,000 pages. If we are going for 2,500 rows,\nodds are that each row is in a separate page. So, to do the query, we\nare retrieving 2,500 8k pages, or 20MB of random data on the disk. How\nlong does it take to issue 2,500 disk requests that are scattered all\nover the disk. Probably 11-22 seconds.\n\nMy OS only lets me have 400 8k buffers, or 3.2MB of buffer. As we are\ntransferring 2,500 8k pages or 20MB of data, it is no surprise that the\nbuffer cache doesn't help much. Sometime, the data is grouped together\non the disk, and that is why some are fast, but often they are not, and\ncertainly in a situation where you are looking for two words to appear\non the same row, they certainly are not on adjacent pages.\n\nJust started running CLUSTER, and it appears to be taking forever. Does\nnot seem to be using the disk, but a lot of CPU. It appears to be\ncaught in an endless loop.\n\nI can attach to a running process, so when I attach to the backend, I\nsee:\n\n#$ gdb /u/pg/bin/postgres 29755 GDB is free software and you are welcome\nto distribute copies of it under certain conditions; type \"show copying\" to see the conditions.\nThere is absolutely no warranty for GDB; type \"show warranty\" for details.\nGDB 4.16 (i386-unknown-bsdi3.0), \nCopyright 1996 Free Software Foundation, Inc...\n\n/usr/local/src/pgsql/pgsql/src/backend/commands/29755: No such file or\ndirectory.\nAttaching to program `/u/pg/bin/postgres', process 29755\n0x94999 in WaitIO (buf=0x4c0ccc8, spinlock=0) at bufmgr.c:1107\n1107 S_LOCK(&(buf->io_in_progress_lock));\n\n\n---------------------------------------------------------------------------\n\nStepping to the next statement shows it is just hung here.\n\nThe backtrace shows:\n\n---------------------------------------------------------------------------\n\n#0 0x9499b in WaitIO (buf=0x4c0ccc8, spinlock=0) at bufmgr.c:1107\n#1 0x94109 in BufferAlloc (reln=0x19df90, blockNum=11324, \n foundPtr=0xefbfb903 \"\\001\\030������������=\\t\", bufferLockHeld=0) at bufmgr.c:400\n#2 0x93e91 in ReadBufferWithBufferLock (reln=0x19df90, blockNum=11324, \n bufferLockHeld=0) at bufmgr.c:255\n#3 0x93dee in ReadBuffer (reln=0x19df90, blockNum=11324) at bufmgr.c:174\n#4 0xb28f in heap_fetch (relation=0x19df90, seeself=0 '\\000', tid=0x1b2256, \n b=0xefbfb974) at heapam.c:1050\n#5 0x303d5 in rebuildheap (OIDNewHeap=11466400, OIDOldHeap=6424406, \n OIDOldIndex=11466368) at cluster.c:357\n#6 0x30111 in cluster (oldrelname=0x112790 \"artist_fti\", \n oldindexname=0x10cdd0 \"artist_fti_idx\") at cluster.c:160\n#7 0xa29c1 in ProcessUtility (parsetree=0x1127b0, dest=Remote)\n at utility.c:626\n\n\n---------------------------------------------------------------------------\n\nNothing else is happening on my machine, so it must be locked against\nitself. I will check into this.\n\nCLUSTER may be misbehaving on your machine too.\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",
"msg_date": "Fri, 13 Mar 1998 22:54:25 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: indexing words slow"
},
{
"msg_contents": "Bruce Momjian writes:\n> Just started running CLUSTER, and it appears to be taking forever. Does\n> not seem to be using the disk, but a lot of CPU. It appears to be\n> caught in an endless loop.\n... \n> Attaching to program `/u/pg/bin/postgres', process 29755\n> 0x94999 in WaitIO (buf=0x4c0ccc8, spinlock=0) at bufmgr.c:1107\n> 1107 S_LOCK(&(buf->io_in_progress_lock));\n> \n> ---------------------------------------------------------------------------\n> \n> Stepping to the next statement shows it is just hung here.\n> The backtrace shows:\n> \n> ---------------------------------------------------------------------------\n> \n> #0 0x9499b in WaitIO (buf=0x4c0ccc8, spinlock=0) at bufmgr.c:1107\n> #1 0x94109 in BufferAlloc (reln=0x19df90, blockNum=11324, \n> foundPtr=0xefbfb903 \"\\001\\030������������=\\t\", bufferLockHeld=0) at bufmgr.c:400\n> #2 0x93e91 in ReadBufferWithBufferLock (reln=0x19df90, blockNum=11324, \n> bufferLockHeld=0) at bufmgr.c:255\n> #3 0x93dee in ReadBuffer (reln=0x19df90, blockNum=11324) at bufmgr.c:174\n> #4 0xb28f in heap_fetch (relation=0x19df90, seeself=0 '\\000', tid=0x1b2256, \n> b=0xefbfb974) at heapam.c:1050\n> #5 0x303d5 in rebuildheap (OIDNewHeap=11466400, OIDOldHeap=6424406, \n> OIDOldIndex=11466368) at cluster.c:357\n> #6 0x30111 in cluster (oldrelname=0x112790 \"artist_fti\", \n> oldindexname=0x10cdd0 \"artist_fti_idx\") at cluster.c:160\n> #7 0xa29c1 in ProcessUtility (parsetree=0x1127b0, dest=Remote)\n> at utility.c:626\n\nThis is all just speculation from memory as I do not have the code in front\nof me, but here is what I think might be happening:\n\nheap_fetch -> ReadBuffer -> ReadBufferWithBufferLock\n-- trying to fetch the next row, needed to read a new page\n\n -> BufferAlloc\n -- to read a new page, we need a buffer, so grab one\n -- or possibly we found the buffer we were looking for\n\n -> WaitIO\n -- but the target buffer was already being read or written\n (by another process???)\n\n -> S_LOCK(&(buf->io_in_progress_lock))\n -- so spin on the buffer in_progress lock waiting for the I/O to\n complete and clear the lock\n\n Apparently, the I/O completion is never seen so the lock is\n never cleared so this will wait forever.\n\nSince the system is stuck at this point, somebody dropped the ball on\nclearing the io_in_progress_lock spinlock. I can't tell without looking\nat the code (and probably not even then) if this is I/O initiated by the\nstuck process and just forgotten about, or if some other process was involved.\n\nWere there any other active backends while this was running?\n\nIf so, did any of them exit abnormally (ie without cleaning up)?\n\nIs the stuck process holding any other spinlocks (like the bufmgr spinlock\nfor instance)? (There is an array of held spinlocks in the Proc structure\nthat gets set by SpinAcquire to record spinlocks owned by the current\nprocess).\n\nIf so, is it possible that another process is somehow deadlocked\nwith this one?\n\n eg: A has bufmgr spinlock, wants io_in_progress_lock spinlock\n B has io_in_progress_lock spinlock, wants bufmgr spinlock\n\nI hope this is useful...\n\n-dg\n\nps: what is CLUSTER anyhow? I have a guess, but...\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n",
"msg_date": "Fri, 13 Mar 1998 22:48:39 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: indexing words slow"
},
{
"msg_contents": " > I have to conclude that because of the way this table is created by\n> slicing words, its layout is almost random. The table is 272MB, and\n> with 8k/page, that is 34,000 pages. If we are going for 2,500 rows,\n> odds are that each row is in a separate page. So, to do the query, we\n> are retrieving 2,500 8k pages, or 20MB of random data on the disk. How\n> long does it take to issue 2,500 disk requests that are scattered all\n> over the disk. Probably 11-22 seconds.\n \n> My OS only lets me have 400 8k buffers, or 3.2MB of buffer. As we are\n> transferring 2,500 8k pages or 20MB of data, it is no surprise that the\n> buffer cache doesn't help much. Sometime, the data is grouped together\n> on the disk, and that is why some are fast, but often they are not, and\n> certainly in a situation where you are looking for two words to appear\n> on the same row, they certainly are not on adjacent pages.\n\nThanx bruce, this si a good report :) This confirms what I thought myself \ntoo btw.\n \n> Just started running CLUSTER, and it appears to be taking forever. Does\n> not seem to be using the disk, but a lot of CPU. It appears to be\n> caught in an endless loop.\n\nNote that cluster actually *did* something on my system. It created a \ntemp-table, which was populated, but very very slowly. I also had \ndisk-activity, but maybe you don't notice much of it because of your SCSI \ndisks....\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Sat, 14 Mar 1998 10:39:32 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: indexing words slow"
}
] |
[
{
"msg_contents": "> \n> The char2,4,8,16 types seem to have no value-added over the\n> better-supported char(), varchar(), text types; I am considering\n> removing them from the backend, and instead have the parser\n> transparently translate the types into varchar() (or char() - I'm not\n> certain which is a better match for the types) for v6.4. Applications\n> would not have to be changed.\n> \n> Comments?\n\nWouldn't bother me. I've got this mega-patch sitting here that would\nremove them and put them into a loadable module under contrib. Just\nhave to tidy up the section that creates the index ops.\n\nWill there be a warning about using a \"depreciated type\" in 6.4 or are\nwe going to have this gunking up the grammer forever? :)\n\ndarrenk\n",
"msg_date": "Wed, 11 Mar 1998 10:03:05 -0500",
"msg_from": "[email protected] (Darren King)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": "> Will there be a warning about using a \"depreciated type\" in 6.4 or are\n> we going to have this gunking up the grammer forever? :)\n\nGood idea. Then we can pull it out of the grammar sometime later. Now,\nif these types are in a loadable module, then we can't actually do\nanything in the parser anyway, since the loadable module would never\nwork. Are these character types worth keeping at all? Less support and\nno performance benefit leaves me thinking not...\n\n - Tom\n",
"msg_date": "Wed, 11 Mar 1998 16:15:39 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
}
] |
[
{
"msg_contents": "I ran some timing tests to check the performance of varchar() vs char16.\nThe results of the test indicate that there is no difference in\nperformance (within the timing scatter of the tests):\n\nchar16\tvc(16)\n 0.99s\t 1.05s\t1 row (this measures startup time, not types)\n39.29s\t39.28s\t~65000 rows\n\nThe char2,4,8,16 types seem to have no value-added over the\nbetter-supported char(), varchar(), text types; I am considering\nremoving them from the backend, and instead have the parser\ntransparently translate the types into varchar() (or char() - I'm not\ncertain which is a better match for the types) for v6.4. Applications\nwould not have to be changed.\n\nComments?\n\n>From the logic char(16) would be the correct replacement.\nI also support the idea to remove char-char16. The problems I see are:\n\t1. char2-16 does not have the 4 byte var header (this would be good for the char() type too)\n\t2. per definition 'a ' = 'a' for the char type but 'a ' <> 'a' for varchar\n\nNote the correct behavior of postgresql:\ntest=> select 'a '::char(4) = 'a'::char(4);\n?column?\n--------\nt\n(1 row)\n\ntest=> select 'a '::varchar(4) = 'a'::varchar(4);\n?column?\n--------\nf\n(1 row)\n\nbut note the \"incorrect\" (char-char16 beeing a fixed length type) behavior of the char-char16 code \ntest=> select 'a '::char4 = 'a'::char4;\n?column?\n--------\nf\n(1 row)\n\nso the current char2-16 code behaves more like varchar. I don't know if anybody does\nrely on this behavior. If not, I would vote to change the char-char16 with the char()\ntype and remove the varhdr from char(). I like getting a simple char * into my C function.\n(see point 1)\n\nAndreas\n\n\n\n\n",
"msg_date": "Wed, 11 Mar 1998 16:30:17 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": "> so the current char2-16 code behaves more like varchar. I don't know if anybody does\n> rely on this behavior. If not, I would vote to change the char-char16 with the char()\n> type and remove the varhdr from char(). I like getting a simple char * into my C function.\n> (see point 1)\n\nRemoving the header from char() types is possible now that we have\natttypmod, but I doubt atttypmod is available in all places that the\nlength of the type is needed. varlena is supported all over the place\nin the backend.\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",
"msg_date": "Wed, 11 Mar 1998 13:50:13 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: AW: [HACKERS] varchar() vs char16 performance"
}
] |
[
{
"msg_contents": "> On the other hand, if I do a count(*) on '^ric', his takes consequently\n> around 1:30 mins, no matter how often I run it. This returns 7866.\n> \n> A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> running it several times.\n\nSince postgres will still read the data row (even for this count(*)) I would guess,\nthat this is a data distribution problem. Maybe you could cluster your data ?\nMaybe the '^rol' rows stick pretty much together, whilst the '^ric' rows \nare evenly distributed on all datapages. Of course there would be room\nfor improvement if postgresql would not read the data pages, which are not needed \nfor any query that only selects columns that are part of the index.\n\nAndreas \n\n",
"msg_date": "Wed, 11 Mar 1998 16:43:06 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] indexing words slow"
},
{
"msg_contents": "> \n> > On the other hand, if I do a count(*) on '^ric', his takes consequently\n> > around 1:30 mins, no matter how often I run it. This returns 7866.\n> > \n> > A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> > running it several times.\n> \n> Since postgres will still read the data row (even for this count(*)) I would guess,\n> that this is a data distribution problem. Maybe you could cluster your data ?\n> Maybe the '^rol' rows stick pretty much together, whilst the '^ric' rows \n> are evenly distributed on all datapages. Of course there would be room\n> for improvement if postgresql would not read the data pages, which are not needed \n> for any query that only selects columns that are part of the index.\n> \n\nBrilliant. I had forgotten that the data pages are accessed as well as\nthe index pages. The CLUSTER command works in 6.3, so perhaps that will\nchange some times.\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",
"msg_date": "Wed, 11 Mar 1998 14:05:17 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] indexing words slow"
},
{
"msg_contents": "On Wed, 11 Mar 1998, Bruce Momjian wrote:\n\n> > \n> > > On the other hand, if I do a count(*) on '^ric', his takes consequently\n> > > around 1:30 mins, no matter how often I run it. This returns 7866.\n> > > \n> > > A search on count(*) of '^lling' and '^tones' takes around 2.5 secs after\n> > > running it several times.\n> > \n> > Since postgres will still read the data row (even for this count(*)) I would guess,\n> > that this is a data distribution problem. Maybe you could cluster your data ?\n> > Maybe the '^rol' rows stick pretty much together, whilst the '^ric' rows \n> > are evenly distributed on all datapages. Of course there would be room\n> > for improvement if postgresql would not read the data pages, which are not needed \n> > for any query that only selects columns that are part of the index.\n> > \n> \n> Brilliant. I had forgotten that the data pages are accessed as well as\n> the index pages. The CLUSTER command works in 6.3, so perhaps that will\n> change some times.\n\nRunning a 'cluster' right now....\n\nMaarten\n\n_____________________________________________________________________________\n| TU Delft, The Netherlands, Faculty of Information Technology and Systems |\n| Department of Electrical Engineering |\n| Computer Architecture and Digital Technique section |\n| [email protected] |\n-----------------------------------------------------------------------------\n\n",
"msg_date": "Wed, 11 Mar 1998 22:53:58 +0100 (MET)",
"msg_from": "Maarten Boekhold <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] indexing words slow"
}
] |
[
{
"msg_contents": "Quoting Bruce Momjian ([email protected]):\n> I assume you are asking of attlen for char() and varchar() have the max\n> length. The answer is now NO. They are varlena structures like text,\n> and so have a attlen of -1. Atttypmod for that pg_attribute row now has\n> the defined length.\n> \n\n\tEven for a char(4) field? ok, if attlen doesn't give the\nlength of a field, how do I find it?\n\nRight now: yes, but I would like to see the char() type changed to a fixed length type (which it really is)\nthen it would have the length in attlen. Right now the length for char and varchar is in atttypmod.\n\nAndreas\n\n",
"msg_date": "Wed, 11 Mar 1998 16:47:56 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "> \n> Quoting Bruce Momjian ([email protected]):\n> > I assume you are asking of attlen for char() and varchar() have the max\n> > length. The answer is now NO. They are varlena structures like text,\n> > and so have a attlen of -1. Atttypmod for that pg_attribute row now has\n> > the defined length.\n> > \n> \n> \tEven for a char(4) field? ok, if attlen doesn't give the\n> length of a field, how do I find it?\n> \n> Right now: yes, but I would like to see the char() type changed to a fixed length type (which it really is)\n> then it would have the length in attlen. Right now the length for char and varchar is in atttypmod.\n> \n\nAgain, we can do this, but we then have to have atttypmod available in\nall places where we check for length. I will add it to the TODO list.\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",
"msg_date": "Wed, 11 Mar 1998 14:08:04 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: AW: [HACKERS] attlen weirdness?"
}
] |
[
{
"msg_contents": "> \n> > Will there be a warning about using a \"depreciated type\" in 6.4 or are\n> > we going to have this gunking up the grammer forever? :)\n> \n> Good idea. Then we can pull it out of the grammar sometime later. Now,\n> if these types are in a loadable module, then we can't actually do\n> anything in the parser anyway, since the loadable module would never\n> work. Are these character types worth keeping at all? Less support and\n> no performance benefit leaves me thinking not...\n\nIMHO, not worth keeping if the performance benefit is gone and the only\nreal benefit though was the few bytes of header space they saved per field.\n\nThey should work as loadable if removed from the parser since they would\nbe created thru a series of CREATE TYPE, FUNCTION and OPERATOR statements.\nShould go thru the parser as just some user-created t[r]ype.\n\nBut any benefit from including them in the distribution as loadable would\nbe negligible. Could be, but no real point, so I vote to depreciate them.\n\ndarrenk\n",
"msg_date": "Wed, 11 Mar 1998 11:37:11 -0500",
"msg_from": "[email protected] (Darren King)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
}
] |
[
{
"msg_contents": "Hello,\n\nI was just reading new KUBL documentation (www.kubl.com) and I found an\ninteresting feature there.\n\nShortly speaking, it allows for checkpointing and reading without\nlocking at all. I was thinking about something similiar in Postgres.\nWhile there was timetravel it was relatively easy to implement, but now\nwe are going out of timetravel code.\n\nHowever, maybe it would be nice to leave some of the code. As far as I\nunderstand kubl (and probably other databases) allows for lock-free\nquering before-checkpoint data. For me it seems like a very powerful\nfeature. Especially if you have a database that is fed constantly and\nyou still want to retrieve large statistical data that would lock the\ntable for too long. By using checkpoint, one may query checkpointed data\nwithout interrupting updating processes. \n\nSo what if we leave some part of this code to give us more isolation of\nhistorical data. I think it's easier to implement than page/row level\nlocking, but it may increase performance of frequently updated\ndatabases.\n\nOf course I don't mean timetravel in previous sense. This was way too\nexpensive. But having two versions of each record, i.e. a checkpointed\nversion and working version seems to be cheap.\n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n",
"msg_date": "Thu, 12 Mar 1998 00:34:04 +0100",
"msg_from": "Michal Mosiewicz <[email protected]>",
"msg_from_op": true,
"msg_subject": "Timetravel vs checkpointing and no read-locking"
},
{
"msg_contents": "Timetravel is gone but postgres is still non-overwriting system:\nboth old and new versions of a changed tuple are in place.\n\nMichal Mosiewicz wrote:\n> \n> Hello,\n> \n> I was just reading new KUBL documentation (www.kubl.com) and I found an\n> interesting feature there.\n> \n> Shortly speaking, it allows for checkpointing and reading without\n> locking at all. I was thinking about something similiar in Postgres.\n> While there was timetravel it was relatively easy to implement, but now\n> we are going out of timetravel code.\n> \n> However, maybe it would be nice to leave some of the code. As far as I\n> understand kubl (and probably other databases) allows for lock-free\n> quering before-checkpoint data. For me it seems like a very powerful\n> feature. Especially if you have a database that is fed constantly and\n> you still want to retrieve large statistical data that would lock the\n> table for too long. By using checkpoint, one may query checkpointed data\n> without interrupting updating processes.\n> \n> So what if we leave some part of this code to give us more isolation of\n> historical data. I think it's easier to implement than page/row level\n> locking, but it may increase performance of frequently updated\n> databases.\n> \n> Of course I don't mean timetravel in previous sense. This was way too\n> expensive. But having two versions of each record, i.e. a checkpointed\n> version and working version seems to be cheap.\n\nVadim\n",
"msg_date": "Thu, 12 Mar 1998 14:06:04 +0700",
"msg_from": "\"Vadim B. Mikheev\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Timetravel vs checkpointing and no read-locking"
}
] |
[
{
"msg_contents": "Does anybody know why getrusage() in backend/port/getrusage.c is\nremoved (surrounded by #if 0)?\n\nSome ports including Solaris 2.4, NEWS OS 6 (I'm trying now) etc. need \ngetrusage().\n--\nTatsuo Ishii\[email protected]\n",
"msg_date": "Thu, 12 Mar 1998 11:13:54 +0900",
"msg_from": "[email protected]",
"msg_from_op": true,
"msg_subject": "port/getrusage.c?"
},
{
"msg_contents": "> From: [email protected]\n> Date: Thu, 12 Mar 1998 11:13:54 +0900\n\n> Does anybody know why getrusage() in backend/port/getrusage.c is\n> removed (surrounded by #if 0)?\n> \n> Some ports including Solaris 2.4, NEWS OS 6 (I'm trying now) etc. need \n> getrusage().\n\nI have the same question - have been wrestling with the SCO (ODT 5)\nport. Actually all we need here is client-side pgsql code to run on\nSCO so I butchered my way thru the build to make it happen.\n\nUsed gcc-2.8.1 to get past the monster macro - altho if there's C\ncode we can put in for easily confused compilers it would be nice to\nhave a port buildable with SCO native tools. Bruce?\n\nAdditions to wishlist - client-only builds, fix setup so Pg.pm builds\non first install.\n\nWill be happy to work with the experts on SCO port + these small\nitems - my hat is off to the 6.3 team.\n\n",
"msg_date": "Wed, 11 Mar 1998 22:04:01 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> I have the same question - have been wrestling with the SCO (ODT 5)\n> port. Actually all we need here is client-side pgsql code to run on\n> SCO so I butchered my way thru the build to make it happen.\n> \n> Used gcc-2.8.1 to get past the monster macro - altho if there's C\n> code we can put in for easily confused compilers it would be nice to\n> have a port buildable with SCO native tools. Bruce?\n\nOK, please tell me what macro is a monster, and what define I can check\nto see if I am using the SCO/Microsoft/piece-O-junk :-) compiler, and I\nwill take a crack at making a static function out of the macro and send\nyou a patch to test.\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",
"msg_date": "Wed, 11 Mar 1998 23:11:37 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> From: Bruce Momjian <[email protected]>\n> Date: Wed, 11 Mar 1998 23:11:37 -0500 (EST)\n\n> OK, please tell me what macro is a monster, and what define I can check\n> to see if I am using the SCO/Microsoft/piece-O-junk :-) compiler, and I\n> will take a crack at making a static function out of the macro and send\n> you a patch to test.\n\nsrc/include/access/heapam.h - heap_getattr is the one that blows the\nmind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\nGNU 2.8.1 was all I could get to handle it.\n\nBTW just so you won't think we're total idiots - we run the pgsql\nserver on FreeBSD where it belongs - but some of our hosts use\nhardware for which vendors refuse to allow FreeBSD support. Thus we\nhave SCO systems (plus hardware to reboot them when they wedge).\n\n",
"msg_date": "Wed, 11 Mar 1998 23:13:04 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> \n> > From: Bruce Momjian <[email protected]>\n> > Date: Wed, 11 Mar 1998 23:11:37 -0500 (EST)\n> \n> > OK, please tell me what macro is a monster, and what define I can check\n> > to see if I am using the SCO/Microsoft/piece-O-junk :-) compiler, and I\n> > will take a crack at making a static function out of the macro and send\n> > you a patch to test.\n> \n> src/include/access/heapam.h - heap_getattr is the one that blows the\n> mind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\n> GNU 2.8.1 was all I could get to handle it.\n\nWhat define can I use to check for SCO open-deathtrap cc?\n\n> \n> BTW just so you won't think we're total idiots - we run the pgsql\n> server on FreeBSD where it belongs - but some of our hosts use\n> hardware for which vendors refuse to allow FreeBSD support. Thus we\n> have SCO systems (plus hardware to reboot them when they wedge).\n\nThis is funny.\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",
"msg_date": "Thu, 12 Mar 1998 00:15:51 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "On Wed, 11 Mar 1998, Hal Snyder wrote:\n\n> > From: Bruce Momjian <[email protected]>\n> > Date: Wed, 11 Mar 1998 23:11:37 -0500 (EST)\n> \n> > OK, please tell me what macro is a monster, and what define I can check\n> > to see if I am using the SCO/Microsoft/piece-O-junk :-) compiler, and I\n> > will take a crack at making a static function out of the macro and send\n> > you a patch to test.\n> \n> src/include/access/heapam.h - heap_getattr is the one that blows the\n> mind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\n> GNU 2.8.1 was all I could get to handle it.\n\n\tI don't believe that this is a \"religious war\" issue...but why do\nppl even *use* the stock C compilers anymore? I know at work we us all\nGCC/G++ stuff, since its one helluva lot cheaper then picking up Sun's CC\ncompiler...I'm not advocating requiring GCC...just wondering one ppl\nactually do use the stock compiler..\n\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Thu, 12 Mar 1998 01:48:28 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> From: Bruce Momjian <[email protected]>\n> Date: Thu, 12 Mar 1998 00:15:51 -0500 (EST)\n\n> > src/include/access/heapam.h - heap_getattr is the one that blows the\n> > mind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\n> > GNU 2.8.1 was all I could get to handle it.\n> \n> What define can I use to check for SCO open-deathtrap cc?\n\nGood question. I don't see one, but the config stuff is not familiar\nground. There is reference to \"sco\" on line 85 of\n\n src/include/utils/memutils.h - \"#if ! defined(sco)\"\n\nbut I think nothing sets this any more. (Actually this is another\nthing on the list of SCO breakage in 6.3). I'd add something to the\ntemplate for sco.\n\n",
"msg_date": "Thu, 12 Mar 1998 00:18:13 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> Date: Thu, 12 Mar 1998 01:48:28 -0400 (AST)\n> From: The Hermit Hacker <[email protected]>\n> cc: [email protected], [email protected]\n\n> > src/include/access/heapam.h - heap_getattr is the one that blows the\n> > mind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\n> > GNU 2.8.1 was all I could get to handle it.\n> \n> \tI don't believe that this is a \"religious war\" issue...but why do\n> ppl even *use* the stock C compilers anymore? I know at work we us all\n> GCC/G++ stuff, since its one helluva lot cheaper then picking up Sun's CC\n> compiler...I'm not advocating requiring GCC...just wondering one ppl\n> actually do use the stock compiler..\n\n1. Corporate engineering management scared of stuff that only builds\n with GNU tools and not \"commercial quality\" native compiler. Feh.\n\n2. OS's so whacked-out that GNU tools don't build without major pain\n and suffering. Look at what it took to build pre-2.8 gcc on SCO.\n\nWhere I work we have two kinds of programmer using the same SCO\ndevelopment system - the GNU'ers who spend as much time as possible on\nFreeBSD and only use the SCO box when held at gunpoint, and the SCO\nusers who would just get all confused by enhanced warnings emitted by\ngcc -Wall -ansi -pedantic (prototype checking, printf arg checks, etc)\nand just sit catatonic in their cubes for weeks if their moldy old\n/bin/cc tools don't work on a module they pull out of CVS. Aargh.\n\n",
"msg_date": "Thu, 12 Mar 1998 01:04:57 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> \n> > From: Bruce Momjian <[email protected]>\n> > Date: Thu, 12 Mar 1998 00:15:51 -0500 (EST)\n> \n> > > src/include/access/heapam.h - heap_getattr is the one that blows the\n> > > mind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\n> > > GNU 2.8.1 was all I could get to handle it.\n> > \n> > What define can I use to check for SCO open-deathtrap cc?\n> \n> Good question. I don't see one, but the config stuff is not familiar\n> ground. There is reference to \"sco\" on line 85 of\n> \n> src/include/utils/memutils.h - \"#if ! defined(sco)\"\n> \n> but I think nothing sets this any more. (Actually this is another\n> thing on the list of SCO breakage in 6.3). I'd add something to the\n> template for sco.\n> \n> \n\nGo to src/tools and run ccsym with the SCO compiler. Send me the\noutput. Should be something unique in there.\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",
"msg_date": "Thu, 12 Mar 1998 08:40:43 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
}
] |
[
{
"msg_contents": "\nRedirected to 'the proper list' - [email protected]\n\nOn Wed, 11 Mar 1998, Chris Albertson wrote:\n\n> Also, is anyone working on storage mangers? I was thinking that\n> a raw partition manager would be good to have. Could be faster\n> then one that uses the file system. Give it two partitions and\n> it could do stripping and gain some real speed.\n\n\tstripping can be done from the operating system level to give you\nthat 'boost'...and Oracle, in fact, moved away from the raw partition\nlevel to just using the Unix file system...I believe it would\novercomplicate the backend, and give a negligible boost in performance, if\nwe had to build a 'low level drive interface'...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Wed, 11 Mar 1998 22:17:15 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] Does Storage Manager support >2GB tables?"
},
{
"msg_contents": "> \n> \n> I looked at the Magnetic Disk Storage manager code and it appears\n> that it at least tries to support tables larger then the maximum\n> file size. I heard some talk however about this feature being either\n> broken or untested for many years. Anyone know the real status?\n\nI is broken, but we are not sure how. Someone recently got over 2GB,\nand the table got messed up.\n\n> \n> Yes I can test it myself but I need to buy another drive first so\n> I can have 6GB of free space.\n> \n> Also, is anyone working on storage mangers? I was thinking that\n> a raw partition manager would be good to have. Could be faster\n> then one that uses the file system. Give it two partitions and\n> it could do stripping and gain some real speed.\n\nNo one is working on storage managers that I know of. The code works,\nbut it just uses the default storage manager.\n\n> \n> Next question is the storagemanager switch working?\n\nYes, I believe that code has not changed recently.\n\n> \n> I'll need this to work in about a year. I'll do the work\n> myself if that option is best but I would not want to put in the\n> effort unless others would use it. \n\nNo one has discussed working on that aspect, but some people have\nsuggested a raw storage manager.\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",
"msg_date": "Wed, 11 Mar 1998 22:08:24 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] Does Storage Manager support >2GB tables?"
},
{
"msg_contents": "The Hermit Hacker wrote:\n> \n> Redirected to 'the proper list' - [email protected]\n> \n> On Wed, 11 Mar 1998, Chris Albertson wrote:\n> \n> > Also, is anyone working on storage mangers? I was thinking that\n> > a raw partition manager would be good to have. Could be faster\n> > then one that uses the file system. Give it two partitions and\n> > it could do stripping and gain some real speed.\n> \n> stripping can be done from the operating system level to give you\n> that 'boost'...and Oracle, in fact, moved away from the raw partition\n> level to just using the Unix file system...I believe it would\n> overcomplicate the backend, and give a negligible boost in performance, if\n> we had to build a 'low level drive interface'...\n\nI know you must have looked at far more Postgresql code then I have but\nI was browsing the storage manager. Apparently it is fairly easy to\nassign a class to a manager as each class is tagged in the system catalog\nwith a storage method. What I really want is a >2GB table. I was trying\nto see if this was supported by reading the source. Looks like it may be.\nThe note in the To Do list includes testing. I would test it but for\nlack of disk space. (I'll have more in a while.) \n\nI need the >2GB feature bad enough that I'd implement it myself. My thought\nwas that I may to easier to write a new manager then understand and fix\na broken one. A manager is just given a class name and block number and\ntold to either fetch or get it. (well not quite so simple but close).\n\nI don't think it needs to look inside the 8K (adjustable now) blocks. \nAnyway, if I wrote such a beast my real motivation would be to have big\ntables. Faster big tables would be just a plus. What I really hope for is\nthat somebody else fixes the existing code :-)\n-- \n--Chris Albertson\n\n [email protected] Voice: 626-351-0089 X127\n Logicon RDA, Pasadena California Fax: 626-351-0699\n",
"msg_date": "Wed, 11 Mar 1998 19:31:28 -0800",
"msg_from": "Chris Albertson <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] Does Storage Manager support >2GB tables?"
},
{
"msg_contents": "> Redirected to 'the proper list' - [email protected]\n> \n> On Wed, 11 Mar 1998, Chris Albertson wrote:\n> \n> > Also, is anyone working on storage mangers? I was thinking that\n> > a raw partition manager would be good to have. Could be faster\n> > then one that uses the file system. Give it two partitions and\n> > it could do stripping and gain some real speed.\n> \n> \tstripping can be done from the operating system level to give you\n> that 'boost'...and Oracle, in fact, moved away from the raw partition\n> level to just using the Unix file system...I believe it would\n> overcomplicate the backend, and give a negligible boost in performance, if\n> we had to build a 'low level drive interface'...\n> \n> Marc G. Fournier \n> Systems Administrator @ hub.org \n> primary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\nI have had the pleasure to work on the guts of one of the major databases\nraw partition storage managers over the last ten years (hint, not my\ncurrent domain), and guess what? It implements a file system. And, not a\nparticularly good filesystem at that. Think about something like \"FAT\",\nbut not quite that nice. It was also a major source of pain in that it\nwas complex, heavily concurrent, and any errors show up as massive data\nloss or corruption. Be careful what you wish for.\n\nMost of the supposed benefit comes from integrating the buffer cache\nmanagement and the writeahead log so that you can defer or avoid I/O (as\nlong as the log records get to disk, there is no reason to ever write the\ndata page unless you need the buffer for something else). You can also\nconvert random I/O to semi sequential I/O if most writes are done by a page\ncleaner or by a checkpoint as this gives you lots of I/O to sort.\n\nI don't know the current state of Postgres so I cannot comment on it, but at\nleast with Illustra, the lack of a traditional writeahead log style\ntransaction system was a major performance hit as it forced an fsync at\nevery commit. A good WAL system gets many commits per log I/O, but \nIllusta was stuck with many writes per transaction. If Postgres still does\nthis (and the recent elimination of time travel suggests that it might not),\nit would be well worth fixing.\n\nA last point, the raw disk, implement our own filesystem architecture used\nby some systems is much more compelling if the filesystems are slow and\ninflexible, and the filesystem caching is ineffective. These things were\nmore true back in the early 80's when these systems were being designed.\nThings are not as bad now, in particular ext2 has quite good performance.\n\nSorry for the ramble...\n\n-dg\n\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Wed, 11 Mar 1998 22:00:45 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "> I have had the pleasure to work on the guts of one of the major databases\n> raw partition storage managers over the last ten years (hint, not my\n> current domain), and guess what? It implements a file system. And, not a\n> particularly good filesystem at that. Think about something like \"FAT\",\n> but not quite that nice. It was also a major source of pain in that it\n> was complex, heavily concurrent, and any errors show up as massive data\n> loss or corruption. Be careful what you wish for.\n\nInteresting.\n\n> \n> Most of the supposed benefit comes from integrating the buffer cache\n> management and the writeahead log so that you can defer or avoid I/O (as\n> long as the log records get to disk, there is no reason to ever write the\n> data page unless you need the buffer for something else). You can also\n> convert random I/O to semi sequential I/O if most writes are done by a page\n> cleaner or by a checkpoint as this gives you lots of I/O to sort.\n\nYes, I have heard that the standard file system read-ahead is often\nuseless for a database, so on a raw partition you know the next block\nthat is going to be requested, so you can prefetch there rather than\nhaving the file system prefetch the next sequential block.\n\nAlso nice so you can control what gets written to disk/fsync'ed and what doesn't\nget fsync'ed.\n\n> I don't know the current state of Postgres so I cannot comment on it, but at\n> least with Illustra, the lack of a traditional writeahead log style\n> transaction system was a major performance hit as it forced an fsync at\n> every commit. A good WAL system gets many commits per log I/O, but \n> Illusta was stuck with many writes per transaction. If Postgres still does\n> this (and the recent elimination of time travel suggests that it might not),\n> it would be well worth fixing.\n\nOur idea is to control when pg_log gets written to disk. We keep active\npg_log pages in shared memory, and every 30-60 seconds, we make a memory\ncopy of the current pg_log active pages, do a system sync() (which\nhappens anyway at that interval), update the pg_log file with the saved\nchanges, and fsync() the pg_log pages to disk. That way, after a crash,\nthe current database only shows transactions as committed where we are\nsure all the data has made it to disk.\n\nI have a more detailed posting if you are interested.\n\n> A last point, the raw disk, implement our own filesystem architecture used\n> by some systems is much more compelling if the filesystems are slow and\n> inflexible, and the filesystem caching is ineffective. These things were\n> more true back in the early 80's when these systems were being designed.\n> Things are not as bad now, in particular ext2 has quite good performance.\n> \n> Sorry for the ramble...\n\nNo ramble at all. It is not every day we get someone with real-world\nexperience in changing from a filesystem to a raw partition database\nstorage manager.\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",
"msg_date": "Thu, 12 Mar 1998 08:36:18 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "Here 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",
"msg_date": "Thu, 12 Mar 1998 09:10:57 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "Bruce Momjian wrote:\n> \n> > I have had the pleasure to work on the guts of one of the major databases\n> > raw partition storage managers over the last ten years (hint, not my\n> > current domain), and guess what? It implements a file system. And, not a\n> > particularly good filesystem at that. Think about something like \"FAT\",\n> > but not quite that nice. It was also a major source of pain in that it\n> > was complex, heavily concurrent, and any errors show up as massive data\n> > loss or corruption. Be careful what you wish for.\n> \n> Interesting.\n\nPerhaps we could:\n\na) Incorporate an existing filesystem into the code (ext2?). By\nIncorporate, I mean that we would just take the latest version of the\ncode and link it into the executable, or into a library and make calls\nto some of the lower level access and allocation routines.\n\nb) suggest that for higher performance, the user should format the\ndisk partition with ext2 (or whatever) and turn off caching and set the\nblock size to the maximum possible.\n\nI know for a fact that ext2 lets the user select the block size, and\nit looks like Linux at least supports a sync mount option which makes\nall I/O to this FS synchronous (which I assume would turn off write\ncaching at least). If caching could be disabled, then option b would\nseem to provide performance equivalent to a.\n\nOcie Mitchell\n",
"msg_date": "Thu, 12 Mar 1998 10:54:38 -0800 (PST)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "> \n> Bruce Momjian wrote:\n> > \n> > > I have had the pleasure to work on the guts of one of the major databases\n> > > raw partition storage managers over the last ten years (hint, not my\n> > > current domain), and guess what? It implements a file system. And, not a\n> > > particularly good filesystem at that. Think about something like \"FAT\",\n> > > but not quite that nice. It was also a major source of pain in that it\n> > > was complex, heavily concurrent, and any errors show up as massive data\n> > > loss or corruption. Be careful what you wish for.\n> > \n> > Interesting.\n> \n> Perhaps we could:\n> \n> a) Incorporate an existing filesystem into the code (ext2?). By\n> Incorporate, I mean that we would just take the latest version of the\n> code and link it into the executable, or into a library and make calls\n> to some of the lower level access and allocation routines.\n> \n> b) suggest that for higher performance, the user should format the\n> disk partition with ext2 (or whatever) and turn off caching and set the\n> block size to the maximum possible.\n> \n> I know for a fact that ext2 lets the user select the block size, and\n> it looks like Linux at least supports a sync mount option which makes\n> all I/O to this FS synchronous (which I assume would turn off write\n> caching at least). If caching could be disabled, then option b would\n> seem to provide performance equivalent to a.\n\nI checked, and under BSD/OS, the readahead call for ufs looks like:\n\n error = breadn(vp, lbn, size,\n (daddr_t *)&nextlbn, &nextsize, 1, NOCRED, &bp);\n\nThe '1' is requesting a read-ahead of one block past the requested\nblock. Clearly this is not tunable, tough a read-ahead of one is not a\nsignificant performance problem. In most cases, the block was already\nread as part of the disk scan, but this gives us the next block in cases\nwhere we are reading sequentially.\n\nThe sync option is not really desired because we do our own syncs on\ntransaction completion. Don't want a sync on every write. Don't think\nyou can disable caching.\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",
"msg_date": "Thu, 12 Mar 1998 15:26:56 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "Bruce Momjian writes: \n> Yes, I have heard that the standard file system read-ahead is often\n> useless for a database, so on a raw partition you know the next block\n> that is going to be requested, so you can prefetch there rather than\n> having the file system prefetch the next sequential block.\n\nAt least on the systems I am intimately familiar with, the prefetch that the\nOS does (assuming a modern OS like Linux) is pretty hard to beat. If you have\na table that was bulk loaded in key order, a sequential scan is going to\nresult in a sequential access pattern to the underlying file and the OS\nprefetch does the right thing. If you have an unindexed table with rows\ninserted at the end, the OS prefetch still works. If you are using a secondary\nindex on some sort of chopped up table with rows inserted willy-nilly, it\nthen, it may be worth doing async reads in a burst and let the disk request\nsort make the best of it.\n\nAs far as I am aware, Postgres does not do async I/O. Perhaps it should.\n\n> Also nice so you can control what gets written to disk/fsync'ed and what doesn't\n> get fsync'ed.\n\nThis is really the big win.\n \n> Our idea is to control when pg_log gets written to disk. We keep active\n> pg_log pages in shared memory, and every 30-60 seconds, we make a memory\n> copy of the current pg_log active pages, do a system sync() (which\n> happens anyway at that interval), update the pg_log file with the saved\n> changes, and fsync() the pg_log pages to disk. That way, after a crash,\n> the current database only shows transactions as committed where we are\n> sure all the data has made it to disk.\n\nOK as far as it goes, but probably bad for concurrancy if I have understood\nyou.\n\n> I have a more detailed posting if you are interested.\n\nThanks, I will read it. Probably should hold more comments until after that ;-)\n \n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Thu, 12 Mar 1998 12:35:34 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "> At least on the systems I am intimately familiar with, the prefetch that the\n> OS does (assuming a modern OS like Linux) is pretty hard to beat. If you have\n> a table that was bulk loaded in key order, a sequential scan is going to\n> result in a sequential access pattern to the underlying file and the OS\n> prefetch does the right thing. If you have an unindexed table with rows\n> inserted at the end, the OS prefetch still works. If you are using a secondary\n> index on some sort of chopped up table with rows inserted willy-nilly, it\n> then, it may be worth doing async reads in a burst and let the disk request\n> sort make the best of it.\n> \n> As far as I am aware, Postgres does not do async I/O. Perhaps it should.\n\nI am adding this to the TODO list:\n\n\t* Do async I/O to do better read-ahead of data\n\nBecause we are not threaded, we really can't do anything else while we\nare waiting for I/O, but we can pre-request data we know we will need.\n\n> \n> > Also nice so you can control what gets written to disk/fsync'ed and what doesn't\n> > get fsync'ed.\n> \n> This is really the big win.\n\nYep, and this is what we are trying to work around in our buffered\npg_log change. Because we have the transaction ids all compact in one\nplace, this seems like a workable solution to our lack of write-to-disk\ncontrol. We just control the pg_log writes.\n\n> \n> > Our idea is to control when pg_log gets written to disk. We keep active\n> > pg_log pages in shared memory, and every 30-60 seconds, we make a memory\n> > copy of the current pg_log active pages, do a system sync() (which\n> > happens anyway at that interval), update the pg_log file with the saved\n> > changes, and fsync() the pg_log pages to disk. That way, after a crash,\n> > the current database only shows transactions as committed where we are\n> > sure all the data has made it to disk.\n> \n> OK as far as it goes, but probably bad for concurrancy if I have understood\n> you.\n\nInteresed in hearing your comments.\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",
"msg_date": "Thu, 12 Mar 1998 16:13:53 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "Bruce Momjian writes:\n> I am adding this to the TODO list:\n> \n> \t* Do async I/O to do better read-ahead of data\n\nGood.\n \n> Because we are not threaded, we really can't do anything else while we\n> are waiting for I/O, but we can pre-request data we know we will need.\n\nThreading is a bit like raw devices. It sounds like a really good idea,\nparticularly with M$ banging the \"NT, now with threads\" drum, but in real\nlife there are some very good reasons not to thread. Particularly with an\nextensible product like Postgres where J-Random routine gets loaded at\nruntime. In a threaded system, J-Random routine needs to be pretty well\nperfect or the whole system comes down. In a process based system, unless\nit trashes something in the shared memory, only the one connection instance\nneeds to come down. My experience with Illustra says that this is fairly\nimportant.\n\nThe other big problem with threading is that now the stacks and all dynamically\nallocated data are in the shared memory and are not easily extendable. So,\nif some recursive procedure (like in the rewriter) uses a bit of extra stack\nsome other thread gets its stack trashed. This is never pretty.\n\nOr if some user function loads a giant large object (like an mpeg say), that\nmemory has to come out of the shared memory, now if they leak that memory,\nit is gone for good. In a per process system, it just ends up using a little\nmore swap space.\n\nThe other thing threading does is introduce new synchronization requirements\ninto things that never needed it before. For example, Postgres is pretty\nfree with memory allocation and deallocation (think Nodes!). With threading\neach palloc() and pfree() is going to have to take a round trip through a\nmutex. This makes an expensive operation even more costly. By and large,\nthe dbms's that are threaded have had pretty static (ie pre-allocate every\nthing in arrays at boot time) memory models. Postgres does not fit this\npicture very well.\n\nUltimately, threading may buy some performance, but not very much compared\nto how hard it is to get right and how brittle it tends to make the system.\n\nUnless I have misunderstood the state of Postgres, there is a vast amount\nof performance improvement to be had without even thinking about threading.\nIf it were me, I would pick up the easy stuff, then the middle-hard stuff\nwith the really big wins like a proper transaction log, and leave the very\nhard stuff like threading until last.\n\n-dg\n\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n\"Of course, someone who knows more about this will correct me if I'm wrong,\n and someone who knows less will correct me if I'm right.\"\n --David Palmer ([email protected])\n",
"msg_date": "Thu, 12 Mar 1998 14:00:18 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "David Gould wrote:\n> \n> Bruce Momjian writes:\n> > I am adding this to the TODO list:\n> > \n> > \t* Do async I/O to do better read-ahead of data\n> \n> Good.\n> \n> > Because we are not threaded, we really can't do anything else while we\n> > are waiting for I/O, but we can pre-request data we know we will need.\n> \n> Threading is a bit like raw devices. It sounds like a really good idea,\n> particularly with M$ banging the \"NT, now with threads\" drum, but in real\n> life there are some very good reasons not to thread. Particularly with an\n> extensible product like Postgres where J-Random routine gets loaded at\n> runtime. In a threaded system, J-Random routine needs to be pretty well\n> perfect or the whole system comes down. In a process based system, unless\n> it trashes something in the shared memory, only the one connection instance\n> needs to come down. My experience with Illustra says that this is fairly\n> important.\n\nI tend to agree. I think threads are more usefull in applications\nsoftware, where the programmer has at least some idea of the use of\nthe program.\n\nOn a more practical/implementation note, it might be nice to have one\nprocess that can perform prefetches for the others. I.E. If I know\nthat I will need page X, I queue a request to this process (in shared\nmemory?) and it will read in this page for us (or even write a page to\ndisk for us).\n\nAsynchronous I/O is interesting, but I don't know how widely it has\nbeen implemented (Linux has none, unless the development version is\nworking on it, and Solaris has aio, but not the POSIX flavor yet).\n\n\nOcie Mitchell\n",
"msg_date": "Thu, 12 Mar 1998 14:57:35 -0800 (PST)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "> Threading is a bit like raw devices. It sounds like a really good idea,\n> particularly with M$ banging the \"NT, now with threads\" drum, but in real\n> life there are some very good reasons not to thread. Particularly with an\n> extensible product like Postgres where J-Random routine gets loaded at\n> runtime. In a threaded system, J-Random routine needs to be pretty well\n> perfect or the whole system comes down. In a process based system, unless\n> it trashes something in the shared memory, only the one connection instance\n> needs to come down. My experience with Illustra says that this is fairly\n> important.\n\nYes, the threading topic has come up before, and I have never considered\nit a big win. We want to remove the exec() from the startup, so we just\ndo a fork. Will save 0.001 seconds of startup.\n\nThat is a very easy win for us. I hadn't considered the synchonization\nproblems with palloc/pfree, and that could be a real problem.\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",
"msg_date": "Thu, 12 Mar 1998 18:09:18 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "Bruce Momjian wrote:\n\n> Yes, the threading topic has come up before, and I have never considered\n> it a big win. We want to remove the exec() from the startup, so we just\n> do a fork. Will save 0.001 seconds of startup.\n\nAs I was always biased to threading I would note that in many cases it\nis a big win. First of all, today it's the IO which is usually the\nslowest part of the database. Most of good IO designs have different\noptimisations like tag queuing etc. But, to make use of all this stuff\nyou have to generate your IO requests as soon as possible. If you\nprocess your query in one thread, most of the time your processor waits\nfor the IO, but sometimes your IO waits for the processor. If you\npopulate your IO system with a bunch of IO requests paralelly,\ncontrollers may try to requeue them to optimise heads movements etc.\nHowever, if you do some IO, then some processing, then some IO.... you\nloose the capability of optimising your requests. Unless your system is\nheavily loaded. In this case, it actually doesn't matter how you\nschedule IO requests. But if your system is not loaded too heavily, it's\ngood to parallelize IO tasks. And the easiest way to accomplish it is to\nuse threads for parallel execution of tasks.\n\nBut I notice that many people still think of threads as a replacement\nfor fork. Of course, in such case it's pretty useless since fork is fast\nenough. But the key to the success is to parallelize single queries not\nonly to leverage processor usage, but also to push IO to it's maximum.\n\n> That is a very easy win for us. I hadn't considered the synchonization\n> problems with palloc/pfree, and that could be a real problem.\n\nFew months ago I was thinking about it. Actually I don't see much\nproblems with things like palloc/pfree. I don't see any problems with\nany heap data that is used locally. But it is a big problem when you\ntake a look at global variables and global data that is accessed and\nmodified in many places. This is potential source of troubles.\n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n",
"msg_date": "Fri, 13 Mar 1998 01:40:23 +0100",
"msg_from": "Michal Mosiewicz <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "Michal Mosiewicz writes: \n> As I was always biased to threading I would note that in many cases it\n> is a big win. First of all, today it's the IO which is usually the\n> slowest part of the database. ...\n...\n> However, if you do some IO, then some processing, then some IO.... you\n> loose the capability of optimising your requests. ...\n> ... But if your system is not loaded too heavily, it's\n> good to parallelize IO tasks. And the easiest way to accomplish it is to\n> use threads for parallel execution of tasks.\n\nAgreed, but what you are talking about here is decomposing a query into\nit's parallel components and executing them in parallel. This is a win\nof course, but the optimizer and executor have to support it. Also, you\nstart to want things like table fragmentation across devices to make this\nwork. A big job. As a shortcut, you can just do some lookahead on index scans\nand do prefetch. Doesn't buy as much, but could probably be done very\nquickly.\n \n> But I notice that many people still think of threads as a replacement\n> for fork. Of course, in such case it's pretty useless since fork is fast\n> enough. But the key to the success is to parallelize single queries not\n> only to leverage processor usage, but also to push IO to it's maximum.\n\nThis is indeed what I was thinking about. The process per connection scheme\nof Postgres is often criticised vs a thread per connection scheme as in\nSybase for example. I was responding to that criticism.\n\n> > That is a very easy win for us. I hadn't considered the synchonization\n> > problems with palloc/pfree, and that could be a real problem.\n> \n> Few months ago I was thinking about it. Actually I don't see much\n> problems with things like palloc/pfree. I don't see any problems with\n\nIf you have multiple threads each allocing memory at the same time, the\nallocator data structures have to be protected. \n\n> any heap data that is used locally. But it is a big problem when you\n> take a look at global variables and global data that is accessed and\n> modified in many places. This is potential source of troubles.\n\nToo right.\n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Thu, 12 Mar 1998 17:53:02 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
},
{
"msg_contents": "> Agreed, but what you are talking about here is decomposing a query into\n> it's parallel components and executing them in parallel. This is a win\n> of course, but the optimizer and executor have to support it. Also, you\n> start to want things like table fragmentation across devices to make this\n> work. A big job. As a shortcut, you can just do some lookahead on index scans\n> and do prefetch. Doesn't buy as much, but could probably be done very\n> quickly.\n\nYes, this is what I was thinking. Could be done pretty easily, and\nhence the TODO addition.\n\n> If you have multiple threads each allocing memory at the same time, the\n> allocator data structures have to be protected. \n> \n> > any heap data that is used locally. But it is a big problem when you\n> > take a look at global variables and global data that is accessed and\n> > modified in many places. This is potential source of troubles.\n> \n> Too right.\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",
"msg_date": "Thu, 12 Mar 1998 21:27:12 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?t"
},
{
"msg_contents": "Chris Albertson wrote:\n> \n> I know you must have looked at far more Postgresql code then I have but\n> I was browsing the storage manager. Apparently it is fairly easy to\n> assign a class to a manager as each class is tagged in the system catalog\n> with a storage method. What I really want is a >2GB table. I was trying\n> to see if this was supported by reading the source. Looks like it may be.\n> The note in the To Do list includes testing. I would test it but for\n> lack of disk space. (I'll have more in a while.)\n ^^^^^^^^^^^^^^^^^^\n\nsrc/backend/storage/smgr/md.c:\n\n#define RELSEG_SIZE ...\n\n- just decrease it!\n\nVadim\n",
"msg_date": "Mon, 16 Mar 1998 15:09:07 +0700",
"msg_from": "\"Vadim B. Mikheev\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
}
] |
[
{
"msg_contents": "unsubscribe [email protected]\n\n\n\n\n\n\n\nunsubscribe [email protected]",
"msg_date": "Thu, 12 Mar 1998 15:30:44 +0900",
"msg_from": "\"=?euc-kr?B?udqxzcXC?=\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "None"
}
] |
[
{
"msg_contents": "> From: [email protected] (David Gould)\n> Date: Wed, 11 Mar 1998 22:37:59 -0800 (PST)\n\n...\n> > Good question. I don't see one, but the config stuff is not familiar\n> > ground. There is reference to \"sco\" on line 85 of\n> > \n> > src/include/utils/memutils.h - \"#if ! defined(sco)\"\n> > \n> > but I think nothing sets this any more. (Actually this is another\n> > thing on the list of SCO breakage in 6.3). I'd add something to the\n> > template for sco.\n\n> What is the policy for defines on Postgres?\n> \n> It is often suggested that that 'ifdef defined(RANDOM_PLATFORM)' is not a\n> nice way to proceed. Far better to use for example 'defined(HAS_FEATURE_XYZ)'\n> than something like:\n> \n> 'defined(SOLARIS) or defined(SUNOS) or defined(PYRAMID) && !defined(WIN32)'.\n> \n> In the case at hand, perhaps something like 'if defined(BROKEN_CPP_GCC_271)',\n> and then define it in the SCO part of the build system. After all, SCO\n> might survive long enough to upgrade to a current compiler...\n\nAck. How about changing\n\n CFLAGS:\n\nto\n\n CFLAGS:-DLAME_CPP\n\nin src/templates/sco?\n\n",
"msg_date": "Thu, 12 Mar 1998 00:51:23 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
},
{
"msg_contents": "> > 'defined(SOLARIS) or defined(SUNOS) or defined(PYRAMID) && !defined(WIN32)'.\n> > \n> > In the case at hand, perhaps something like 'if defined(BROKEN_CPP_GCC_271)',\n> > and then define it in the SCO part of the build system. After all, SCO\n> > might survive long enough to upgrade to a current compiler...\n> \n> Ack. How about changing\n> \n> CFLAGS:\n> \n> to\n> \n> CFLAGS:-DLAME_CPP\n> \n> in src/templates/sco?\n\nBut we don't want that if they use a good compiler under SCO. I have\nasked them to run src/tools/ccsym and send the output. Should be\nsomething unique in there.\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",
"msg_date": "Thu, 12 Mar 1998 08:42:58 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] port/getrusage.c?]"
},
{
"msg_contents": "> But we don't want that if they use a good compiler under SCO. I have\n> asked them to run src/tools/ccsym and send the output. Should be\n> something unique in there.\n\nOn SCO, ccsym becomes /bin/cc -b elf -ii -E foo.c, which after filtering\ngives the symbols below. If I omit \"-b elf\", the last line is replaced\nby the two lines\n'-D_M_COFF'\n'-D_SCO_COFF'\n\nHow about _SCO_DS as the define when testing for SCO's native\ncompiler? (I confess ignorance as to SCO's intended use of this symbol.)\n\n'-D__i386'\n'-D_SCO_DS=1'\n'-D__unix'\n'-D_M_I386'\n'-D_M_XENIX'\n'-D_M_UNIX'\n'-D_SCO_C_DIALECT=1'\n'-D_STRICT_NAMES'\n'-D_M_I86'\n'-D_M_I86SM'\n'-D_M_SDATA'\n'-D_M_STEXT'\n'-D_M_BITFIELDS'\n'-D_M_INTERNAT'\n'-D_M_SYS5'\n'-D_M_SYSV'\n'-D_M_SYSIII'\n'-D_M_WORDSWAP'\n'-Di386'\n'-Dunix'\n'-DM_I386'\n'-DM_UNIX'\n'-DM_XENIX'\n'-D_SCO_XPG_VERS=4'\n'-D_SCO_ELF'\n\n",
"msg_date": "Thu, 12 Mar 1998 14:55:29 -0600 (CST)",
"msg_from": "Hal Snyder <[email protected]>",
"msg_from_op": true,
"msg_subject": "SCO vs. the monster macro"
},
{
"msg_contents": "> > But we don't want that if they use a good compiler under SCO. I have\n> > asked them to run src/tools/ccsym and send the output. Should be\n> > something unique in there.\n> \n> On SCO, ccsym becomes /bin/cc -b elf -ii -E foo.c, which after filtering\n> gives the symbols below. If I omit \"-b elf\", the last line is replaced\n> by the two lines\n> '-D_M_COFF'\n> '-D_SCO_COFF'\n> \n> How about _SCO_DS as the define when testing for SCO's native\n> compiler? (I confess ignorance as to SCO's intended use of this symbol.)\n> \n> '-D__i386'\n> '-D_SCO_DS=1'\n> '-D__unix'\n> '-D_M_I386'\n> '-D_M_XENIX'\n> '-D_M_UNIX'\n> '-D_SCO_C_DIALECT=1'\n> '-D_STRICT_NAMES'\n> '-D_M_I86'\n> '-D_M_I86SM'\n> '-D_M_SDATA'\n> '-D_M_STEXT'\n> '-D_M_BITFIELDS'\n> '-D_M_INTERNAT'\n> '-D_M_SYS5'\n> '-D_M_SYSV'\n> '-D_M_SYSIII'\n> '-D_M_WORDSWAP'\n> '-Di386'\n> '-Dunix'\n> '-DM_I386'\n> '-DM_UNIX'\n> '-DM_XENIX'\n> '-D_SCO_XPG_VERS=4'\n> '-D_SCO_ELF'\n> \n\nOr even: '-D_SCO_C_DIALECT=1' as in:\n\n#if _SCO_C_DIALECT == 1 ...\n\nCan't get much more specific than that.\n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Thu, 12 Mar 1998 13:12:41 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SCO vs. the monster macro"
},
{
"msg_contents": "> \n> > But we don't want that if they use a good compiler under SCO. I have\n> > asked them to run src/tools/ccsym and send the output. Should be\n> > something unique in there.\n> \n> On SCO, ccsym becomes /bin/cc -b elf -ii -E foo.c, which after filtering\n> gives the symbols below. If I omit \"-b elf\", the last line is replaced\n> by the two lines\n> '-D_M_COFF'\n> '-D_SCO_COFF'\n> \n> How about _SCO_DS as the define when testing for SCO's native\n> compiler? (I confess ignorance as to SCO's intended use of this symbol.)\n\nI am inclined to choose:\n\n\t-D_SCO_C_DIALECT=1\n\nNot to test for the value of one, but just to test if it is defined.\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",
"msg_date": "Thu, 12 Mar 1998 16:15:54 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: SCO vs. the monster macro"
}
] |
[
{
"msg_contents": "> Right now: yes, but I would like to see the char() type changed to a fixed length type (which it really is)\n> then it would have the length in attlen. Right now the length for char and varchar is in atttypmod.\n> \n\nAgain, we can do this, but we then have to have atttypmod available in\nall places where we check for length. I will add it to the TODO list.\n\nI am starting to see the problem, only atttypmod is passed around not attlen.\nTherefore even if we do this, the char() length of a column still has to be in atttypmod.\nSo it will stay the same for Julie, whether it is taken out or not. But since it is really \nhard for a generic application like ODBC, maybe we could change atttypmod to not \ninclude the VARHDRSZ, most of the time it is needed without VARHDRSZ anyway ?\nI am starting to think that this would be more important than to remove the 4 bytes.\n\nAndreas\n\n\n",
"msg_date": "Thu, 12 Mar 1998 09:56:36 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: AW: [HACKERS] attlen weirdness?"
},
{
"msg_contents": "> \n> > Right now: yes, but I would like to see the char() type changed to a fixed length type (which it really is)\n> > then it would have the length in attlen. Right now the length for char and varchar is in atttypmod.\n> > \n> \n> Again, we can do this, but we then have to have atttypmod available in\n> all places where we check for length. I will add it to the TODO list.\n> \n> I am starting to see the problem, only atttypmod is passed around not attlen.\n> Therefore even if we do this, the char() length of a column still has to be in atttypmod.\n> So it will stay the same for Julie, whether it is taken out or not. But since it is really \n> hard for a generic application like ODBC, maybe we could change atttypmod to not \n> include the VARHDRSZ, most of the time it is needed without VARHDRSZ anyway ?\n> I am starting to think that this would be more important than to remove the 4 bytes.\n\nThose four bytes have always been there. Check psql or pg_dump. See\nthem subtracting VARHDRSZ all the time.\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",
"msg_date": "Thu, 12 Mar 1998 08:48:44 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: AW: AW: [HACKERS] attlen weirdness?"
}
] |
[
{
"msg_contents": "Redirected to 'the proper list' - [email protected]\n\nOn Wed, 11 Mar 1998, Chris Albertson wrote:\n\n> Also, is anyone working on storage mangers? I was thinking that\n> a raw partition manager would be good to have. Could be faster\n> then one that uses the file system. Give it two partitions and\n> it could do stripping and gain some real speed.\n\n\tstripping can be done from the operating system level to give you\nthat 'boost'...and Oracle, in fact, moved away from the raw partition\nlevel to just using the Unix file system...I believe it would\novercomplicate the backend, and give a negligible boost in performance, if\nwe had to build a 'low level drive interface'...\n\nActually Oracle *is* suggesting the use of RAW devices for heavy IO databases.\nRAW devices have the advantage that the operating system does not cache\nthe pages. This can at once be a drastic disadvantage if the database does not\nhave it's own read/write optimization (only read/write whole pages or better multiples thereof)\nan intelligent buffer cache and intelligent read ahead (here only the db engine can be really intelligent about it).\nIf you use file system files and have the above features in the database engine, \nthen all pages will be in memory twice, once in the file system buffer cache and \nonce in the database cache. This results in only half of the pages beeing cached and also \nadds the additional overhead of 2 memcpy's.\n\nI think the first step in the direction of RAW devices needs to be the implementation of the \ntablespace idea, i.e. more than one table or index in one file. On the driver side,\nif there is a logical volume manager, then accessing a raw device is the same as \naccessing a file system file (read/write/seek).\n\nAndreas\n\n\n",
"msg_date": "Thu, 12 Mar 1998 10:33:19 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
}
] |
[
{
"msg_contents": "> > \n> > > Will there be a warning about using a \"depreciated type\" in 6.4 or are\n> > > we going to have this gunking up the grammer forever? :)\n> > \n> > Good idea. Then we can pull it out of the grammar sometime later. Now,\n> > if these types are in a loadable module, then we can't actually do\n> > anything in the parser anyway, since the loadable module would never\n> > work. Are these character types worth keeping at all? Less support and\n> > no performance benefit leaves me thinking not...\n> \n> IMHO, not worth keeping if the performance benefit is gone and the only\n> real benefit though was the few bytes of header space they saved per field.\n> \n\nIf char2 et al are going completely from 6.4, I think it would be sensible\nfor pg_dump to filter these types and change them to char(2) et al when\nit writes the CREATE statments.\n\nBest wishes,\n\nAndrew\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) [email protected] (Home) [email protected]\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n",
"msg_date": "Thu, 12 Mar 1998 11:08:59 GMT",
"msg_from": "Andrew Martin <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
},
{
"msg_contents": "> If char2 et al are going completely from 6.4, I think it would be sensible\n> for pg_dump to filter these types and change them to char(2) et al when\n> it writes the CREATE statments.\n> \n\nThat is an interesting idea, but what about applications that use them? \nI think we can have the parser change them for one release, then drop\nthem completely.\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",
"msg_date": "Thu, 12 Mar 1998 09:07:35 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
}
] |
[
{
"msg_contents": "I have a xpg_user file all my database directories.\n\nSeems initdb is creating it. Any way to prevent it from being created?\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",
"msg_date": "Thu, 12 Mar 1998 09:31:43 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "initdb and xpg_user"
},
{
"msg_contents": "> \n> I have a xpg_user file all my database directories.\n\n Hands off - this is my bug :-)\n\n> \n> Seems initdb is creating it. Any way to prevent it from being created?\n> \n\n Add a \n\n mv $PGDATA/base/template1/xpg_user $PGDATA/base/template1/pg_user\n\n after the UPDATE pg_class renaming xpg_user to pg_user in initdb.\n\n But this showed up another FEATURE! If you remove a tables file,\n the next time it is accessed it is recreated empty! I think this\n is not right, because after a system crash, a database file (I\n know, very unlikely) could reside in lost+found. It's IMHO not\n good to assume the table is empty and recreate the file silent.\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": "Thu, 12 Mar 1998 17:26:48 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] initdb and xpg_user"
},
{
"msg_contents": "> \n> > \n> > I have a xpg_user file all my database directories.\n> \n> Hands off - this is my bug :-)\n> \n> > \n> > Seems initdb is creating it. Any way to prevent it from being created?\n> > \n> \n> Add a \n> \n> mv $PGDATA/base/template1/xpg_user $PGDATA/base/template1/pg_user\n> \n> after the UPDATE pg_class renaming xpg_user to pg_user in initdb.\n> \n> But this showed up another FEATURE! If you remove a tables file,\n> the next time it is accessed it is recreated empty! I think this\n> is not right, because after a system crash, a database file (I\n> know, very unlikely) could reside in lost+found. It's IMHO not\n> good to assume the table is empty and recreate the file silent.\n> \n\nPatch applied.\n\n---------------------------------------------------------------------------\n\n*** ./bin/initdb/initdb.sh.orig\tMon Mar 16 00:55:12 1998\n--- ./bin/initdb/initdb.sh\tMon Mar 16 00:57:25 1998\n***************\n*** 367,375 ****\n--- 367,382 ----\n \t valuntil\tabstime);\" |\\\n \tpostgres -F -Q -D$PGDATA template1 2>&1 > /dev/null |\\\n \tgrep -v \"'DEBUG:\"\n+ \n+ #move it into pg_user\n echo \"UPDATE pg_class SET relname = 'pg_user' WHERE relname = 'xpg_user';\" |\\\n \tpostgres -F -Q -D$PGDATA template1 2>&1 > /dev/null |\\\n \tgrep -v \"'DEBUG:\"\n+ echo \"UPDATE pg_type SET typname = 'pg_user' WHERE typname = 'xpg_user';\" |\\\n+ \tpostgres -F -Q -D$PGDATA template1 2>&1 > /dev/null |\\\n+ \tgrep -v \"'DEBUG:\"\n+ mv $PGDATA/base/template1/xpg_user $PGDATA/base/template1/pg_user\n+ \n echo \"CREATE RULE _RETpg_user AS ON SELECT TO pg_user DO INSTEAD\t\\\n \t SELECT usename, usesysid, usecreatedb, usetrace,\t\t\\\n \t usesuper, usecatupd, '********'::text as passwd,\t\\\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",
"msg_date": "Mon, 16 Mar 1998 00:58:48 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] initdb and xpg_user"
}
] |
[
{
"msg_contents": "> > > OK, please tell me what macro is a monster, and what define I can check\n> > > to see if I am using the SCO/Microsoft/piece-O-junk :-) compiler, and I\n> > > will take a crack at making a static function out of the macro and send\n> > > you a patch to test.\n> > \n> > src/include/access/heapam.h - heap_getattr is the one that blows the\n> > mind of SCO ODT5 /bin/cc as well as their Skunkware gcc-2.7.2.1. Pure\n> > GNU 2.8.1 was all I could get to handle it.\n> \n> \tI don't believe that this is a \"religious war\" issue...but why do\n> ppl even *use* the stock C compilers anymore? I know at work we us all\n> GCC/G++ stuff, since its one helluva lot cheaper then picking up Sun's CC\n> compiler...I'm not advocating requiring GCC...just wondering one ppl\n> actually do use the stock compiler..\n\nBecause it was there...\n\ndarrenk\n",
"msg_date": "Thu, 12 Mar 1998 09:49:35 -0500",
"msg_from": "[email protected] (Darren King)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] port/getrusage.c?"
}
] |
[
{
"msg_contents": "Hi,\n\n as I proposed, I'm now starting on the PL/pgSQL loadable\n procedural language. As far as I'm now I have a pl_handler\n with an independent flex/bison parser that can parse a\n rudimentary implementation of the language. The next step is\n to start on the PL/pgSQL executor and look if the generated\n instruction tree can be used (up to now the pl_handler only\n dumps the instruction tree and returns a 0 Datum.\n\n If that works I'll expand the scanner/parser to the full\n PL/plSQL language including trigger procedures.\n\n But I would like to have some discussion on language itself.\n So I wrote down what I had in mind. The document is appended\n below.\n\n Please comment/suggest !\n\n Someone gave a hint about global variables existing during a\n session. What is a session than? One transaction? The\n backends lifetime? And should global variables be visible by\n more than one function? I vote for NO! In that case we need\n something like packages of functions that share globals.\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 PL/pgSQL\n A procedural language for the PostgreSQL RDBMS\n 1st draft\n\n Jan Wieck <[email protected]>\n\n\n\n Notice\n\n This document is for the postgresql-hackers community for\n completing the syntax specification of PL/pgSQL. The\n extension module described here doesn't exist up to now!\n\n\n Preface\n\n PL/pgSQL is a procedural language based on SQL designed for\n the PostgreSQL database system.\n\n The extensibility features of PostgreSQL are mostly based on\n the ability to define functions for various operations.\n Functions could have been written in PostgreSQL's SQL dialect\n or in the C programming language. Functions written in C are\n compiled into a shared object and loaded by the database\n backend process on demand. Also the trigger features of\n PostgreSQL are based on functions but required the use of the\n C language.\n\n Since version 6.3 PostgreSQL supports the definition of\n procedural languages. In the case of a function or trigger\n procedure defined in a procedural language, the database has\n no builtin knowlege how to interpret the functions source\n text. Instead, the function and trigger calls are passed into\n a handler that knows the details of the language. The\n handler itself is function compiled into a shared object and\n loaded on demand.\n\n\n Overview\n\n The PL/pgSQL language is case insensitive. All keywords and\n identifiers can be used in upper-/lowercase mixed.\n\n PL/pgSQL is a block oriented language. A block is defined as\n\n [<<label>>]\n [DECLARE\n -- declarations]\n BEGIN\n -- statements\n END;\n\n There can be any number of subblocks in the statements\n section of a block. Subblocks can be used to hide variables\n from outside a block of statements (see Scope and visability\n below).\n\n\n Comments\n\n There are two types of comments in PL/pgSQL. A double dash\n '--' starts a comment that extends to the end of the line. A\n '/*' starts a block comment that extends to the next '*/'.\n Block comments cannot be nested, but double dash comments can\n be enclosed into a block comment.\n\n\n Declarations\n\n All variables, rows and records used in a block or it's\n subblocks must be declared in the declarations section of the\n block. The parameters given to the function are\n automatically declared with the usual identifiers $n. The\n declarations have the following syntax:\n\n <name> [CONSTANT] <type> [NOT NULL]\n [DEFAULT | := <value>];\n\n Declares a variable of the specified type. If the\n variable is declared as CONSTANT, the value cannot be\n changed. If NOT NULL is specified, an assignment of a\n NULL value results in a runtime error. Since the\n default value of a variable is the SQL NULL value,\n all variables declared as NOT NULL must also have a\n default value.\n\n The default value is evaluated at the actual function\n call. So assigning 'now' to an abstime varable causes\n the variable to have the time of the actual function\n call, not when the function was compiled.\n\n <name> <class>%ROWTYPE;\n\n Declares a row with the structure of the given class.\n Class must be an existing table- or viewname of the\n database. The fields of the row are accessed in the\n dot notation. Parameters to a procedure could be\n tuple types. In that case the corresponding\n identifier $n will be a rowtype. Only the user\n attributes and the oid of a tuple are accessible in\n the row. There must be no whitespaces between the\n classname, the percent and the ROWTYPE keyword.\n\n <name> RECORD;\n\n Records are similar to rowtypes, but they have no\n predefined structure and it's impossible to assign a\n value into them. They are used in selections and FOR\n loops to hold one actual database tuple from a select\n operation. One and the same record can be used in\n different selections (but not in nested ones).\n\n <name> ALIAS FOR $n;\n\n For better readability of the code it's possible to\n define an alias for a positional parameter to the\n function.\n\n Datatypes\n\n The type of a variable can be any of the existing data types\n of the database. <type> above is defined as:\n\n postgesql-basetype\n or variable%TYPE\n or rowtype.field%TYPE\n or class.field%TYPE\n\n As for the rowtype declaration, there must be no whitespaces\n between the classname, the percent and the TYPE keyword.\n\n Expressions\n\n All expressions used in PL/pgSQL statements are processed\n using the backends executor. Since even a constant looking\n expression can have a totally different meaning for a\n particular data type (as 'now' for abstime), it is impossible\n for the PL/pgSQL parser to identify real constant values\n other than the NULL keyword. The expressions are evaluated by\n internally executing a query\n\n SELECT <expr>\n\n over the SPI manager. In the expression, occurences of\n variable identifiers are substituted by parameters and the\n actual values from the variables are passed to the executor\n as query parameters. All the expressions used in a PL/pgSQL\n function are only prepared and saved once.\n\n Statements\n\n Anything not understood by the parser as specified below will\n be put into a query and sent down to the database engine to\n execute. The resulting query should not return any data.\n\n Assignment\n\n An assignment of a value to a variable or rowtype field\n is written as:\n\n <identifier> := <expr>;\n\n If the expressions result data type doesn't match the\n variables data type, or the variables atttypmod value is\n known (as for char(20)), the result value will be\n implicitly casted by the PL/pgSQL executor using the\n result types output- and the variables type input-\n functions. Note that this could potentially result in\n runtime errors generated by the types input functions.\n\n An assignment of a complete selection into a record or\n rowtype can be done as:\n\n SELECT targetlist INTO <recname|rowname> FROM fromlist;\n\n If a rowtype is used as target, the selected targetlist\n must exactly match the structure of the rowtype or a\n runtime error occurs. The fromlist can be followed by\n any valid qualification, grouping, sorting etc. There is\n a special condition [NOT] FOUND that can be used\n immediately after a SELECT INTO to check if the data has\n been found.\n\n SELECT * INTO myrec FROM EMP WHERE empname = myname;\n IF NOT FOUND THEN\n ELOG ERROR 'employee %s not found' myname;\n END IF;\n\n In addition, the select statement must not return more\n that one row. If multiple rows are returned, a runtime\n error will be generated.\n\n\n Returning from the function\n\n RETURN <expr>;\n\n The function terminates and the value of <expr> will be\n returned to the upper executor. The return value of a\n function cannot be undefined. If control reaches the end\n of the toplevel block of the function without hitting a\n RETURN statement, a runtime error will occur.\n\n\n Aborting and messages\n\n As indicated above there is an ELOG statement that can\n throw messages into the PostgreSQL elog mechanism.\n\n ELOG level 'format' [identifiers];\n\n Inside the format, only %s might be used as a placeholder\n for the following identifiers. The identifiers must\n specify an existing variable or row/record field.\n\n\n Conditionals\n\n IF <expr> THEN\n -- statements\n [ELSE\n -- statements]\n END IF;\n\n The expression <expr> must return a value that at least\n can be casted into a boolean.\n\n\n Loops\n\n There are multiple types of loops.\n\n [<<label>>]\n LOOP\n -- statements\n END LOOP;\n\n An unconditional loop that must be terminated explicitly\n by an EXIT statement. The optional label can be used by\n EXIT statements of nested loops to specify which level of\n nesting should be terminated.\n\n [<<label>>]\n WHILE <expr> LOOP\n -- statements\n END LOOP;\n\n A conditional loop that is executed as long as the\n evaluation of <expr> returns true.\n\n [<<label>>]\n FOR <name> IN [REVERSE] <expr>..<expr> LOOP\n -- statements\n END LOOP.\n\n A loop that iterates over a range of integer values. The\n variable <name> is automatically created as type integer\n and exists only inside the loop. The two expressions\n giving the lower and upper bound of the range are\n evaluated only when entering the loop. The iteration step\n is 1.\n\n FOR <recname|rowname> IN <select_clause> LOOP\n -- statements\n END LOOP;\n\n The record or row is assigned all the rows resulting from\n the select clause and the statements executed for each.\n If the loop is terminated with an EXIT statement, the\n last accessed row is still accessible in the record or\n rowtype.\n\n EXIT [label] [WHEN <expr>];\n\n If no label given, the innermost loop is terminated and\n the statement following END LOOP is executed next. If\n label is given, it must be the label of the current or an\n upper level of nested loops or blocks. Then the named\n loop or block is terminated and control continues with\n the statement after the loops/blocks corresponding END.\n\n",
"msg_date": "Thu, 12 Mar 1998 16:38:32 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": true,
"msg_subject": "PL/pgSQL - for discussion"
},
{
"msg_contents": "> as I proposed, I'm now starting on the PL/pgSQL loadable\n> procedural language.\n> Please comment/suggest !\n> Someone gave a hint about global variables existing during a\n> session. What is a session than? One transaction? The\n> backends lifetime? And should global variables be visible by\n> more than one function? I vote for NO! In that case we need\n> something like packages of functions that share globals.\n\nThis looks nice. SQL92 (and presumably SQL3) has the concept of global\nand local, temporary and permanent, tables. I believe that it also has\nthe concept of variables with the same possible combinations of\nbehaviors, but I'm not finding that in my books at the moment.\n\nClearly if we have these features in the backend someday, then there\nwould need to be hooks for your PL to use also. No need for you to\nprovide these if the backend doesn't help you imho.\n\nI haven't had a chance to read the main portion of your document yet...\n\n - Tom\n",
"msg_date": "Thu, 12 Mar 1998 16:21:08 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "> \n> Hi,\n> \n> as I proposed, I'm now starting on the PL/pgSQL loadable\n> procedural language. As far as I'm now I have a pl_handler\n> with an independent flex/bison parser that can parse a\n> rudimentary implementation of the language. The next step is\n> to start on the PL/pgSQL executor and look if the generated\n> instruction tree can be used (up to now the pl_handler only\n> dumps the instruction tree and returns a 0 Datum.\n> \n> If that works I'll expand the scanner/parser to the full\n> PL/plSQL language including trigger procedures.\n> \n> But I would like to have some discussion on language itself.\n> So I wrote down what I had in mind. The document is appended\n> below.\n> \n> Please comment/suggest !\n\nGee, it looks really nice. I have never used such an advanced language\nINSIDE a database engine.\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",
"msg_date": "Thu, 12 Mar 1998 12:40:44 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "Jan Wieck wrote:\n> \n> Hi,\n> \n> as I proposed, I'm now starting on the PL/pgSQL loadable\n> procedural language. As far as I'm now I have a pl_handler\n> with an independent flex/bison parser that can parse a\n> rudimentary implementation of the language. The next step is\n> to start on the PL/pgSQL executor and look if the generated\n> instruction tree can be used (up to now the pl_handler only\n> dumps the instruction tree and returns a 0 Datum.\n> \n> If that works I'll expand the scanner/parser to the full\n> PL/plSQL language including trigger procedures.\n> \n> But I would like to have some discussion on language itself.\n> So I wrote down what I had in mind. The document is appended\n> below.\n> \n> Please comment/suggest !\n> \n> Someone gave a hint about global variables existing during a\n> session. What is a session than? One transaction? The\n> backends lifetime? And should global variables be visible by\n> more than one function? I vote for NO! In that case we need\n> something like packages of functions that share globals.\n> \n> \n> Jan\n\nThis looks good. \n\nI especially like the \"for x in select ...\" part, it looks a lot more\nelegant than cursors, but we might want to provide a cursor with a\n\"get next row\" and \"get previous row\" function, as the for loop only\ngoes one way.\n\nAnother suggestion related to parameters:\n\n> <name> ALIAS FOR $n;\n> \n> For better readability of the code it's possible to\n> define an alias for a positional parameter to the\n> function.\n> \n\nWhat is the defined behavior if the user leaves out this parameter?\nDo we generate a runtime error? If I might suggest the following:\n\n<name> ALIAS FOR $n;\nsets up name as an alias for $n, name is null if that parameter was\nnot given.\n\n<name> REQUIRED ALIAS FOR $n;\nsets up name as an alias for $n, generate a runtime error if that\nparameter was not given.\n\nActually, an assignment might be a better way to do this. I.E. Define\nfoo as int not null, assign $2 to foo and if there is an error, the user is notified.\n\nOcie\n",
"msg_date": "Thu, 12 Mar 1998 12:16:13 -0800 (PST)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "\nI haven't read the PL/SQL proposal yet so please do not take this as\ncriticism of the proposal. It is just that I have sometimes wondered (having\nused and maintained a couple of them) if there is a real need to invent\nanother procedural language inside a dbms. Who really needs yet another\nlanguage that only works in certain special circumstances? Why not just\nadapt an existing language implementation and graft SQL integration into\nit? For example in Perl:\n\nsub find_prospects_to_contact ()\n # note that sql select looks like a file handle/split() combo to perl\n while (<select ($name, $phone) from prospects p, \n where prospect.interests ~= /computer/ or /electronics/;\n >) {\n # some stuff to be done per row\n ...\n if ($should_contact) {\n insert into contacts_todo values ($name, $phone);\n }\n }\n\n\nOf course there are probably a zillion great reasons why this would be hard\nor the wrong thing to do, but still...\n\n Wouldn't it be cool?\n\n-dg\n\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Thu, 12 Mar 1998 13:31:33 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "David Gould wrote:\n> \n> I haven't read the PL/SQL proposal yet so please do not take this as\n> criticism of the proposal. It is just that I have sometimes wondered (having\n> used and maintained a couple of them) if there is a real need to invent\n> another procedural language inside a dbms. Who really needs yet another\n> language that only works in certain special circumstances?\n\nBut Jan has already adopted an existing language interpreter (i.e. TCL).\nNow he speaks about server side programing using native SQL.\n\nActually existance of SQL server programming in SQL database seems to be\nquite expected feature. You may consider that most SQL developers\ndoesn't really need other languages but SQL, so it's not inventing\nanother language. It's just a wider, more flexible implementation of\ninternal SQL.\n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n",
"msg_date": "Thu, 12 Mar 1998 22:54:20 +0100",
"msg_from": "Michal Mosiewicz <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "Jan Wieck wrote:\n> \n> Hi,\n> \n> as I proposed, I'm now starting on the PL/pgSQL loadable\n> procedural language. As far as I'm now I have a pl_handler\n> with an independent flex/bison parser that can parse a\n> rudimentary implementation of the language. The next step is\n> to start on the PL/pgSQL executor and look if the generated\n> instruction tree can be used (up to now the pl_handler only\n> dumps the instruction tree and returns a 0 Datum.\n> \n> If that works I'll expand the scanner/parser to the full\n> PL/plSQL language including trigger procedures.\n\nWhy PL/pgSQL should be loadable PL? Why not built-in ?\nWould it be possible to add dirrect support for PL/pgSQL syntax\nto current parser ?\nTyping procedure body inside ' is not nice thing, imho.\n\n> Someone gave a hint about global variables existing during a\n> session. What is a session than? One transaction? The\n> backends lifetime? And should global variables be visible by\n ^^^^^^^^^^^^^^^^^\nThis.\n\n> more than one function? I vote for NO! In that case we need\n> something like packages of functions that share globals.\n\nLet's leave packages for future, but why session-level variables\nshouldn't be visible inside procedures right now? \n\n> \n> PL/pgSQL is a block oriented language. A block is defined as\n> \n> [<<label>>]\n> [DECLARE\n> -- declarations]\n> BEGIN\n> -- statements\n> END;\n\nSomeday we'll have nested transactions...\nHow about disallow using BEGIN/END as transaction control statements\nright now ?\nSTART/COMMIT/ROLLBACK/ABORT and nothing more...\n\n> <name> <class>%ROWTYPE;\n> \n> Declares a row with the structure of the given class.\n> Class must be an existing table- or viewname of the\n> database. The fields of the row are accessed in the\n> dot notation. Parameters to a procedure could be\n> tuple types. In that case the corresponding\n> identifier $n will be a rowtype. Only the user\n> attributes and the oid of a tuple are accessible in\n> the row. There must be no whitespaces between the\n> classname, the percent and the ROWTYPE keyword.\n> \n> <name> RECORD;\n> \n> Records are similar to rowtypes, but they have no\n> predefined structure and it's impossible to assign a\n> value into them. They are used in selections and FOR\n> loops to hold one actual database tuple from a select\n> operation. One and the same record can be used in\n> different selections (but not in nested ones).\n\nDo we really need in both ROWTYPE & RECORD ?\nI would get rid of RECORD and let ROWTYPE variables be \n'with yet undefined type of row' (make <class> optional). More of that,\nwhy not treat ROWTYPE like structures in C and let the following:\n\nname %ROWTYPE {a\tint4, b text};\n\n?\n\n> SELECT * INTO myrec FROM EMP WHERE empname = myname;\n ^^^^^ ^^^^^^\nHow about $-prefix ?\n\n> As indicated above there is an ELOG statement that can\n> throw messages into the PostgreSQL elog mechanism.\n> \n> ELOG level 'format' [identifiers];\n ^^^^^^^^^^\nNO, pls - too postgres-ish! Just let ABORT to have 'format' etc and add \nPRINT (or something like this) to put some messages to application (via NOTICE).\nWhat are used in Oracle, Sybase etc here ?\n\nVadim\n",
"msg_date": "Fri, 13 Mar 1998 11:40:07 +0700",
"msg_from": "\"Vadim B. Mikheev\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "\nOcie wrote:\n>\n> This looks good.\n>\n> I especially like the \"for x in select ...\" part, it looks a lot more\n> elegant than cursors, but we might want to provide a cursor with a\n> \"get next row\" and \"get previous row\" function, as the for loop only\n> goes one way.\n\n We don't have real cursors up to now. The SPI_exec() and\n SPI_execp() functions return the complete set of tuples\n selected. And I'm not sure if that what PostgreSQL calls a\n cursor can be used in the backend over SPI. It requires named\n portals and them in turn require a transaction block (BEGIN\n ... COMMIT). But I think it would be easy to build something\n that looks like cursors on top of the complete set of tuples.\n\n>\n> Another suggestion related to parameters:\n>\n> > <name> ALIAS FOR $n;\n> >\n> > For better readability of the code it's possible to\n> > define an alias for a positional parameter to the\n> > function.\n> >\n>\n> What is the defined behavior if the user leaves out this parameter?\n> Do we generate a runtime error? If I might suggest the following:\n>\n> <name> ALIAS FOR $n;\n> sets up name as an alias for $n, name is null if that parameter was\n> not given.\n\n The backends main parser chooses functions not only by name.\n The number and datatypes of the given parameters must also\n match (function overloading - possible with our PL\n implementation). If a query execution reaches the function\n call, be sure that all parameters are given.\n\n I thought about it just as a way to make the code more\n readable. The parameters might also be accessed by the usual\n $n notation. So if you have\n\n empname ALIAS FOR $n;\n\n in the declarations, empname and $n are identical. Thats how\n I understand the word ALIAS.\n\n>\n> <name> REQUIRED ALIAS FOR $n;\n> sets up name as an alias for $n, generate a runtime error if that\n> parameter was not given.\n>\n> Actually, an assignment might be a better way to do this. I.E. Define\n> foo as int not null, assign $2 to foo and if there is an error, the user is notified.\n\n Does make sense. But for sake a function cannot identify a\n null value in one of the parameters. The passed in isNull\n flag only says that one of all parameters is null. Not which\n one. We have to change this someday, and I have already some\n ideas on that. But that's another topic.\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, 13 Mar 1998 09:59:36 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "\nMike wrote:\n>\n> David Gould wrote:\n> >\n> > I haven't read the PL/SQL proposal yet so please do not take this as\n> > criticism of the proposal. It is just that I have sometimes wondered (having\n> > used and maintained a couple of them) if there is a real need to invent\n> > another procedural language inside a dbms. Who really needs yet another\n> > language that only works in certain special circumstances?\n>\n> But Jan has already adopted an existing language interpreter (i.e. TCL).\n> Now he speaks about server side programing using native SQL.\n\n For version 6.3 look into .../pgsql/src/pl/tcl.\n\n>\n> Actually existance of SQL server programming in SQL database seems to be\n> quite expected feature. You may consider that most SQL developers\n> doesn't really need other languages but SQL, so it's not inventing\n> another language. It's just a wider, more flexible implementation of\n> internal SQL.\n\n Even if \"most SQL developers don't need (or know) another\n language\" is reason enough for an SQL based PL, my main\n reason is another one.\n\n PL/Tcl at least requires that the Tcl library got built on a\n system. And I have tested only that it works with Tcl7.5 and\n Tcl8.0. I expect that the PL/perl implementation (I hope\n Brett McCormick is still working on that) will need a perl\n library too.\n\n So there is no PL implementation up to now, that is\n independent from another software package. PL/pgSQL will be!\n PL/pgSQL will be the first language that can get installed\n and enabled by default and then be available in all\n PostgreSQL installations.\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, 13 Mar 1998 10:31:01 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "\nVadim wrote:\n>\n> Jan Wieck wrote:\n> >\n> > Hi,\n> >\n> > as I proposed, I'm now starting on the PL/pgSQL loadable\n> > procedural language. As far as I'm now I have a pl_handler\n> > with an independent flex/bison parser that can parse a\n> > rudimentary implementation of the language. The next step is\n> > to start on the PL/pgSQL executor and look if the generated\n> > instruction tree can be used (up to now the pl_handler only\n> > dumps the instruction tree and returns a 0 Datum.\n> >\n> > If that works I'll expand the scanner/parser to the full\n> > PL/plSQL language including trigger procedures.\n>\n> Why PL/pgSQL should be loadable PL? Why not built-in ?\n> Would it be possible to add dirrect support for PL/pgSQL syntax\n> to current parser ?\n> Typing procedure body inside ' is not nice thing, imho.\n\n Well, PL/pgSQL could be compiled in and the pl_handler\n function and language tuples set at bootstrap. But\n incorporating the parser into the backends main parser isn't\n nesseccary then either. Not that I think it's impossible, but\n the current main parser is complex enough for me.\n\n The typing of the procedure body inside of ' is damned. I\n know :-) I think it might be possible to allow {} or the like\n to be used instead and then only quote \\} inside the body.\n This stuff might be easy done in the scanner (haven't looked\n at the code yet).\n\n>\n> > Someone gave a hint about global variables existing during a\n> > session. What is a session than? One transaction? The\n> > backends lifetime? And should global variables be visible by\n> ^^^^^^^^^^^^^^^^^\n> This.\n\n OK.\n\n>\n> > more than one function? I vote for NO! In that case we need\n> > something like packages of functions that share globals.\n>\n> Let's leave packages for future, but why session-level variables\n> shouldn't be visible inside procedures right now?\n\n For security. At least I would like the visibility of global\n variables depend on the functions owner. So users A and B can\n use the same global name in their functions but the variables\n are different.\n\n>\n> >\n> > PL/pgSQL is a block oriented language. A block is defined as\n> >\n> > [<<label>>]\n> > [DECLARE\n> > -- declarations]\n> > BEGIN\n> > -- statements\n> > END;\n>\n> Someday we'll have nested transactions...\n> How about disallow using BEGIN/END as transaction control statements\n> right now ?\n> START/COMMIT/ROLLBACK/ABORT and nothing more...\n\n Right now!\n\n>\n> Do we really need in both ROWTYPE & RECORD ?\n> I would get rid of RECORD and let ROWTYPE variables be\n> 'with yet undefined type of row' (make <class> optional). More of that,\n> why not treat ROWTYPE like structures in C and let the following:\n>\n> name %ROWTYPE {a int4, b text};\n\n Hmmm. Or doing it the Oracle way\n\n DECLARE\n TYPE myrectype IS RECORD (\n field1 integer NOT NULL,\n field2 text);\n\n myrec myrectype;\n BEGIN\n ...\n END\n\n But I would like to let the RECORD of unspecified structure\n in. It doesn't need much declarations typing.\n\n>\n> ?\n>\n> > SELECT * INTO myrec FROM EMP WHERE empname = myname;\n> ^^^^^ ^^^^^^\n> How about $-prefix ?\n\n I don't like the $'s. But I have seen the problem that\n without blowing up my parser I cannot do it the oracle way\n where a field name of a selected table precedes a local\n varname and the local varname if identical to a tables\n fieldname must be prefixed with the label of the block. This\n is what Oracle does:\n\n <<outer>>\n DECLARE\n emp emp%ROWTYPE;\n empname emp.empname%TYPE\n salary emp.salary%TYPE\n BEGIN\n ...\n SELECT * INTO outer.emp FROM emp WHERE empname = outer.empname;\n -- ^^^^^^^^^ ^^^ ^^^^^^^ ^^^^^^^^^^^^^\n -- PLs rowtype table table- PLs variable\n -- field\n\n salary := emp.salary;\n -- ^^^^^^^^^^^^^^^^^\n -- Outside of SELECT stmt - all identifiers in PL\n\n ...\n END\n\n\n>\n> > As indicated above there is an ELOG statement that can\n> > throw messages into the PostgreSQL elog mechanism.\n> >\n> > ELOG level 'format' [identifiers];\n> ^^^^^^^^^^\n> NO, pls - too postgres-ish! Just let ABORT to have 'format' etc and add\n> PRINT (or something like this) to put some messages to application (via NOTICE).\n> What are used in Oracle, Sybase etc here ?\n\n Oracle uses RAISE EXCEPTION ... with some numbers specifying\n the message in the message catalog and other information.\n\n What about\n\n RAISE EXCEPTION 'format' [identifiers]; -- elog(ERROR, ...)\n RAISE NOTICE 'format' [identifiers]; -- elog(NOTICE, ...)\n RAISE DEBUG 'format' [identifiers]; -- elog(DEBUG, ...)\n\n The first is somewhat compatible and the two otheres can be\n easyly commented out. Since the language is somewhat\n PostgreSQL specific anyway (arguments are unnamed and\n identified by position with $n), PL procedures must be ported\n when moving to another DB. But who ever wants to use another\n DB, once he used PostgreSQL?\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, 13 Mar 1998 13:10:30 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
},
{
"msg_contents": "Jan and Vadim (I think) discuss:\n> >\n> > Why PL/pgSQL should be loadable PL? Why not built-in ?\n> > Would it be possible to add dirrect support for PL/pgSQL syntax\n> > to current parser ?\n...\n> Well, PL/pgSQL could be compiled in and the pl_handler\n> function and language tuples set at bootstrap. But\n> incorporating the parser into the backends main parser isn't\n> nesseccary then either. Not that I think it's impossible, but\n> the current main parser is complex enough for me.\n\nI agree, for a language this simple, it would be nice to have it\nin the main parser. This might not be too hard since the existing statements\nare not really changing, only some new ones are added. eg:\n\nSelect_Stmt := SELECT _name_list FROM _from_clause WHERE _where_clause_\n...\n\n+ If_Stmt := IF _expression_ THEN _statement_list END\n+ Loop_Stmt := ...\n\nOr whatever the grammar looks like.\n\nA bit harder is that the executor now has to sequence through the statements\nand handle branches and storing variables etc.\n\n> > > Someone gave a hint about global variables existing during a\n> > > session. What is a session than? One transaction? The\n> > > backends lifetime? And should global variables be visible by\n> > ^^^^^^^^^^^^^^^^^\n> > This.\n\nAgree. \n\n> OK.\n> \n> > > more than one function? I vote for NO! In that case we need\n> > > something like packages of functions that share globals.\n> >\n> > Let's leave packages for future, but why session-level variables\n> > shouldn't be visible inside procedures right now?\n> \n> For security. At least I would like the visibility of global\n> variables depend on the functions owner. So users A and B can\n> use the same global name in their functions but the variables\n> are different.\n\nNot a problem. If the global variables are global only to the session they\ncan be stored in allocated memory, not in the shared memory. Automatically\nthen they are private to the user of that session and are destroyed when\nthe session ends.\n\n> > > [<<label>>]\n> > > [DECLARE\n> > > -- declarations]\n> > > BEGIN\n> > > -- statements\n> > > END;\n> >\n> > Someday we'll have nested transactions...\n> > How about disallow using BEGIN/END as transaction control statements\n> > right now ?\n> > START/COMMIT/ROLLBACK/ABORT and nothing more...\n> \n> Right now!\n\nHmmm, I like BEGIN TRAN/END TRAN/ABORT TRAN. I suppose there is a standard\nwe should be following... Also, we probably should not start breaking\nexisting applications, users get very fussy about that.\n \n> > > throw messages into the PostgreSQL elog mechanism.\n> > >\n> > > ELOG level 'format' [identifiers];\n> > ^^^^^^^^^^\n> > NO, pls - too postgres-ish! Just let ABORT to have 'format' etc and add\n> > PRINT (or something like this) to put some messages to application (via NOTICE).\n\nWhat is wrong with ELOG? It seems concise and does what is wanted, yes?\nWhy add syntax?\n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Fri, 13 Mar 1998 23:41:44 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
}
] |
[
{
"msg_contents": "subscribe\n\n",
"msg_date": "Fri, 13 Mar 1998 08:43:19 +0900",
"msg_from": "Kyungsoo Jeong <[email protected]>",
"msg_from_op": true,
"msg_subject": "None"
}
] |
[
{
"msg_contents": "\nI've noticed recently that after making certain changes to the\ndatabase (adding indices, attributes) they aren't immediately known\nabout. for example, I add an attribute, then try to update that\nattribute it tells me it doesn't exist. if I exist psql and start a\nnew connection all is well. but it doesn't happen consistently..\n",
"msg_date": "Thu, 12 Mar 1998 15:55:28 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "suspected problem with cache updates"
},
{
"msg_contents": "> \n> \n> I've noticed recently that after making certain changes to the\n> database (adding indices, attributes) they aren't immediately known\n> about. for example, I add an attribute, then try to update that\n> attribute it tells me it doesn't exist. if I exist psql and start a\n> new connection all is well. but it doesn't happen consistently..\n> \n> \n\n Looks like problems in the syscache/catcache invalidation\n routines.\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",
"msg_date": "Fri, 13 Mar 1998 18:17:17 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] suspected problem with cache updates"
}
] |
[
{
"msg_contents": "On Thu, 12 Mar 1998, Steve Tarkalson wrote:\n\n> Hi,\n> \n> thanks for your response to my questions. We are\n> reviewing our backup/vacuum procedures in light of\n> your information.\n> \n> I/O subsys is SCSI, comprised of (8) 4Gb Seagates on\n> (2) DPT controllers in a RAID 5 configuration.\n\n\tOuch...I'm just curious here, but what does an 'ls -l' look like\nfor your database? \n\n> Our database is made up of approximately 50 tables with\n> 6 of these incurring relentless \"hits\". The problem table\n> I referred to is the largest(+300K) and is read/written\n> to most frequently. We could possibly do further normalization\n> here.\n\n\tThis might help, as, depending on how it is done, you can reduce\nthe overall size of the database requiring the vacuum's...on other thing\nto try is vacuum'ng each table seperately.\n\n\tThis might sound odd, but, if, for instance, you stagger your\nvacuum's, you might find that the impact is a little less on your system.\n\n\tBasically, create a file with all your tables listed in it, and do\nsomething like:\n\n========================\n#!/bin/sh\nfor i in `cat <table file>`\ndo\n\tpsql -c \"vacuum analyze $i\" <database>\n\tsleep 30\ndone\n========================\n\n\tThe problem (one which we hope to have addressed in v6.4) is that\nwhen vacuum runs, it locks the pg_class table, which essentially locks the\ncomplete database. The above will run a vacuum on a particular table,\nlocking the pg_class file, then, when complete, will release that lock so\nthat other readers/writers can grab the lock...30 seconds later, the next\ntable gets done. This basically allows other aspects of the system to\ngrab access to the database, even for a little while, so that the whole\nsystem deadlocks.\n\n\tThe locking manager in v6.3 has been considerably enhanced to\nimprove this...\n\n> I was curious about your thoughts on the problem \n> with the corrupted index \"FATAL 1:btree BTP_CHAIN was expected\".\n> Was this a known bug in pre-6.3 versions? We have experienced\n> spurious postgres process core dumps. Can this corrupt the\n> indexes/db?\n\n\tI believe so, but I've CC'd [email protected] on this,\nso that those better informed/more knowledgeable may respond...\n\n> You suggested that we upgrade to 6.3 and I agree\n> that we will do this in the very near future. One\n> question regarding this. Is regular vacuuming still\n> a requirement in 6.3?\n\t\n\tregular vacuuming will always be a requirement, as it updates\nvarious statistics...but, one of the things that I believe is being looked\ninto is having the database overwrite delete records, so that a vacuum\nisn't required to remove those records and shrink the database...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Thu, 12 Mar 1998 23:03:04 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: vacuum..."
},
{
"msg_contents": "The Hermit Hacker wrote:\n> \n> psql -c \"vacuum analyze $i\" <database>\n ^^^^^^^\nIt takes additional time and shouldn't be used in each run.\n\n> > I was curious about your thoughts on the problem\n> > with the corrupted index \"FATAL 1:btree BTP_CHAIN was expected\".\n> > Was this a known bug in pre-6.3 versions? We have experienced\n> > spurious postgres process core dumps. Can this corrupt the\n> > indexes/db?\n\nYes. Core dumps and power off-s can break indices in all versions.\n\n> > You suggested that we upgrade to 6.3 and I agree\n> > that we will do this in the very near future. One\n> > question regarding this. Is regular vacuuming still\n> > a requirement in 6.3?\n> \n> regular vacuuming will always be a requirement, as it updates\n> various statistics...but, one of the things that I believe is being looked\n> into is having the database overwrite delete records, so that a vacuum\n> isn't required to remove those records and shrink the database...\n\nRequired, currently.\n\nVadim\n",
"msg_date": "Fri, 13 Mar 1998 10:35:15 +0700",
"msg_from": "\"Vadim B. Mikheev\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: vacuum..."
}
] |
[
{
"msg_contents": "\nHas anyone peeked that this to see if it can help make postgresql more\nstable in the face of a system crash or power loss?\n\nhttp://www.coda.cs.cmu.edu/doc/html/rvm_manual.html\n\n\"RVM provides an unstructured recoverable virtual memory.\nThe recoverable storage is represented by Unix files\nor disk partitions that applications can map at page granularity\ninto the address space of a process. Simple, non-nested atomic\ntransactions guarantee permanence of changes to recoverable storage\nacross system crashes. Applications can schedule transaction logging\nactions to enhance performance. The design stresses simplicity, ease of use,\nand high performance. Unix compatibility is standard, while optional\nMach-specific extensions are supported for additional flexibility\nand performance. RVM has been extensively used in the clients and\nservers of the Coda File System, and in the Venari system.\"\n\n--Michael\n",
"msg_date": "Thu, 12 Mar 1998 20:30:50 -0800 (PST)",
"msg_from": "Michael Graff <[email protected]>",
"msg_from_op": true,
"msg_subject": "RVM -- recoverable virtual memory"
},
{
"msg_contents": "> \n> Has anyone peeked that this to see if it can help make postgresql more\n> stable in the face of a system crash or power loss?\n> \n> http://www.coda.cs.cmu.edu/doc/html/rvm_manual.html\n> \n> \"RVM provides an unstructured recoverable virtual memory.\n> The recoverable storage is represented by Unix files\n> or disk partitions that applications can map at page granularity\n> into the address space of a process. Simple, non-nested atomic\n> transactions guarantee permanence of changes to recoverable storage\n> across system crashes. Applications can schedule transaction logging\n> actions to enhance performance. The design stresses simplicity, ease of use,\n> and high performance. Unix compatibility is standard, while optional\n> Mach-specific extensions are supported for additional flexibility\n> and performance. RVM has been extensively used in the clients and\n> servers of the Coda File System, and in the Venari system.\"\n> \n> --Michael\n> \n\nVery interesting. \n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Fri, 13 Mar 1998 23:22:43 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] RVM -- recoverable virtual memory"
}
] |
[
{
"msg_contents": "\nWhat do you all think about the fact that cast(anytype as varchar)\nresults in a call to a procedure that is not creatable with 'create\nfunction'? Perhaps we should re-think our casting implementation,\nmaybe one that isn't based on just rewriting itself into a function\ncall :) If I wanted to call a function, I would :)\n\nI can, however, do a create function with a different name, then\nupdate that to varchar. the reason I can't, of course, is because the\ngrammar expects varchar(number), not varchar(argument types)..\n\nthere was probably a much better way of saying all that :)\n",
"msg_date": "Thu, 12 Mar 1998 23:13:45 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "casting & type comments"
},
{
"msg_contents": "On Thu, 12 Mar 1998, Brett McCormick wrote:\n\n> \n> What do you all think about the fact that cast(anytype as varchar)\n> results in a call to a procedure that is not creatable with 'create\n> function'? Perhaps we should re-think our casting implementation,\n> maybe one that isn't based on just rewriting itself into a function\n> call :) If I wanted to call a function, I would :)\n> \n> I can, however, do a create function with a different name, then\n> update that to varchar. the reason I can't, of course, is because the\n> grammar expects varchar(number), not varchar(argument types)..\n> \n> there was probably a much better way of saying all that :)\n> \nI agree with you.\nMany databases as SOLID and MySQL don't have CAST statement because they\ndon't need it.\nUnfortunately we need a CAST statement to convert data from one to another\ntype but it is incomplete, seems that CAST can translate only a few types\nlike:\n int to float\n float to int\n int to text\n float to text\n\nWe can't CAST a char to a varchar, for example.\nIt would be interesting al least to have a list of possible conversions.\n\n Ciao, Jose'\n\n",
"msg_date": "Fri, 13 Mar 1998 17:57:33 +0000 (GMT)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "> What do you all think about the fact that cast(anytype as varchar)\n> results in a call to a procedure that is not creatable with 'create\n> function'? \n\nNot too nice...\n\n> Perhaps we should re-think our casting implementation,\n> maybe one that isn't based on just rewriting itself into a function\n> call :) If I wanted to call a function, I would :)\n\nBut, this is the real strength of Postgres, everything is treated uniformly\nand everything can be extended by defining functions. To hardcode certain\ntypes would be to lose the one of the most creative and desireable aspects\nof the system.\n \n> I can, however, do a create function with a different name, then\n> update that to varchar. the reason I can't, of course, is because the\n> grammar expects varchar(number), not varchar(argument types)..\n\nPerhaps the grammar could be fixed to allow this?\n \n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Fri, 13 Mar 1998 15:19:24 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "On Fri, 13 March 1998, at 15:19:24, David Gould wrote:\n\n> > What do you all think about the fact that cast(anytype as varchar)\n> > results in a call to a procedure that is not creatable with 'create\n> > function'? \n> \n> Not too nice...\n> \n> > Perhaps we should re-think our casting implementation,\n> > maybe one that isn't based on just rewriting itself into a function\n> > call :) If I wanted to call a function, I would :)\n> \n> But, this is the real strength of Postgres, everything is treated uniformly\n> and everything can be extended by defining functions. To hardcode certain\n> types would be to lose the one of the most creative and desireable aspects\n> of the system.\n\nI'm certainly not saying that this aspect of the postgres system\nshould be changed, but rather a different way of mapping casts to\nfunctions, so we don't run into problems like this, and perhaps a\nfall-back to a straight string cast (i.e. call the destination types\ninput function on the return value of the source types output\nfunction)..\n\nOne downside to all this is if I already have a function called\nwhatever, and suddenly someone wants to add a type called whatever,\nthat function would be used for casting when it really shouldn't, and\ncould have unexpected results..\n\n--brett\n\n> \n> > I can, however, do a create function with a different name, then\n> > update that to varchar. the reason I can't, of course, is because the\n> > grammar expects varchar(number), not varchar(argument types)..\n> \n> Perhaps the grammar could be fixed to allow this?\n> \n> -dg\n> \n> David Gould [email protected] 510.628.3783 or 510.305.9468 \n> Informix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n> - I realize now that irony has no place in business communications.\n",
"msg_date": "Fri, 13 Mar 1998 15:41:29 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "> > > What do you all think about the fact that cast(anytype as varchar)\n> > > results in a call to a procedure that is not creatable with \n> > > 'create function'?\n> > > Perhaps we should re-think our casting implementation,\n> > > maybe one that isn't based on just rewriting itself into a \n> > > function call :) If I wanted to call a function, I would :)\n\nNo you wouldn't. There are too many functions with wildly varying names\nto keep all of them in your head.\n\n> > But, this is the real strength of Postgres, everything is treated uniformly\n> > and everything can be extended by defining functions. To hardcode certain\n> > types would be to lose the one of the most creative and desireable aspects\n> > of the system.\n> \n> I'm certainly not saying that this aspect of the postgres system\n> should be changed, but rather a different way of mapping casts to\n> functions, so we don't run into problems like this, and perhaps a\n> fall-back to a straight string cast (i.e. call the destination types\n> input function on the return value of the source types output\n> function)..\n\nThat already happens if you do not specify an explicit cast.\n\n> One downside to all this is if I already have a function called\n> whatever, and suddenly someone wants to add a type called whatever,\n> that function would be used for casting when it really shouldn't, and\n> could have unexpected results..\n\nNot likely. Postgres does match up the types correctly, since it allows\nfunction overloading. So the only problem comes if you have a function\nname with the same input arguments, and the function name happens to be\nthe same name as the new type.\n\n> > > I can, however, do a create function with a different name, then\n> > > update that to varchar. the reason I can't, of course, is because \n> > > the grammar expects varchar(number), not varchar(argument types)..\n> >\n> > Perhaps the grammar could be fixed to allow this?\n\nI'm planning on working on the type conversion stuff. Not certain yet\nhow to handle varchar; SQL92 has this ad-hoc type syntax for some types\nwhich makes it difficult to generalize. We might have to have a few\nspecial cases to handle the grungy SQL92 stuff, and try to leave the\nrest of it generic.\n\nOther types, like numeric and decimal, have the same problem with extra\nparameters associated with the type.\n\n - Tom\n",
"msg_date": "Sat, 14 Mar 1998 02:59:36 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "On Sat, 14 March 1998, at 02:59:36, Thomas G. Lockhart wrote:\n\n> No you wouldn't. There are too many functions with wildly varying names\n> to keep all of them in your head.\n\nI like to use IP numbers, too :)\n\n> > I'm certainly not saying that this aspect of the postgres system\n> > should be changed, but rather a different way of mapping casts to\n> > functions, so we don't run into problems like this, and perhaps a\n> > fall-back to a straight string cast (i.e. call the destination types\n> > input function on the return value of the source types output\n> > function)..\n> \n> That already happens if you do not specify an explicit cast.\n\nMy mistake, is this new in 6.3?\n\n> \n> > One downside to all this is if I already have a function called\n> > whatever, and suddenly someone wants to add a type called whatever,\n> > that function would be used for casting when it really shouldn't, and\n> > could have unexpected results..\n> \n> Not likely. Postgres does match up the types correctly, since it allows\n> function overloading. So the only problem comes if you have a function\n> name with the same input arguments, and the function name happens to be\n> the same name as the new type.\n\nThat's the situation I was citing, sorry for not being clear.\n\n> I'm planning on working on the type conversion stuff. Not certain yet\n> how to handle varchar; SQL92 has this ad-hoc type syntax for some types\n> which makes it difficult to generalize. We might have to have a few\n> special cases to handle the grungy SQL92 stuff, and try to leave the\n> rest of it generic.\n> \n> Other types, like numeric and decimal, have the same problem with extra\n> parameters associated with the type.\n\nSpeaking of which, how is the atttypmod passed?\n",
"msg_date": "Fri, 13 Mar 1998 19:11:48 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "> > I'm planning on working on the type conversion stuff. Not certain yet\n> > how to handle varchar; SQL92 has this ad-hoc type syntax for some types\n> > which makes it difficult to generalize. We might have to have a few\n> > special cases to handle the grungy SQL92 stuff, and try to leave the\n> > rest of it generic.\n> > \n> > Other types, like numeric and decimal, have the same problem with extra\n> > parameters associated with the type.\n> \n> Speaking of which, how is the atttypmod passed?\n\nIt is used internally, but not passed to functions. Not sure how to\naddress that.\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",
"msg_date": "Fri, 13 Mar 1998 22:55:36 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "\nSorry, I meant at table creation time.. What are the implications of\natttypmod in any case? it is just a general purpose number to attach\nto attributes (for precision etc), correct?\n\nOn Fri, 13 March 1998, at 22:55:36, Bruce Momjian wrote:\n\n> It is used internally, but not passed to functions. Not sure how to\n> address that.\n> \n> -- \n> Bruce Momjian | 830 Blythe Avenue\n> [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",
"msg_date": "Fri, 13 Mar 1998 21:45:35 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "> \n> \n> Sorry, I meant at table creation time.. What are the implications of\n> atttypmod in any case? it is just a general purpose number to attach\n> to attributes (for precision etc), correct?\n> \n\nIt is passed to all input and output functions for each type as the\nthird argument. You can see its use in char() and varchar().\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",
"msg_date": "Sat, 14 Mar 1998 01:08:38 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] casting & type comments"
},
{
"msg_contents": "> I agree with you.\n> Many databases as SOLID and MySQL don't have CAST statement because \n> they don't need it.\n\nThey would need it if they were SQL92 conformant :)\n\n> Unfortunately we need a CAST statement to convert data from one to \n> another type but it is incomplete, seems that CAST can translate only \n> a few types like:\n> int to float\n> float to int\n> int to text\n> float to text\n> \n> We can't CAST a char to a varchar, for example.\n> It would be interesting al least to have a list of possible \n> conversions.\n\nThe only thing lacking is for someone to go through and implement more\nof the conversion functions. In conjunction with the improvements I\nwould like to do for automatic type conversion in the parser, this would\ngive us a more complete \"transparent\" conversion capability. CAST uses\nthe same underlying functions, but allows you to force the resultant\ntype, overriding any automatic conversion.\n\nI tried selecting a list of all possible conversions, and narrowed it\ndown to 218 entries. But most are not actual straight conversions...\n\n select proname, prorettype, proargtypes \n from pg_proc\n where pronargs = 1 \n and proname not like '%in' \n and proname not like '%out'\n\n - Tom\n",
"msg_date": "Mon, 16 Mar 1998 15:31:21 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] casting & type comments"
}
] |
[
{
"msg_contents": "> Someone gave a hint about global variables existing during a\n> session. What is a session than? One transaction? The\n> backends lifetime? \n\nThat was me. A session is one client connection to the database from\nthe connect to the disconnect (the lifetime of one postgres process).\nThe same user connecting a second time will get a new global session \nvariable pool. \n\n> And should global variables be visible by more than one function? \nYes, if the variable is only seen by one function it is not global but\nstatic ?\n\n> In that case we need something like packages of functions that share globals. \nOf course this would be elegant, but since it is hard to implement and the commercial dbms's\ndont't do that I guess we could also do without it (only one default package).\nBut I think all of this is specified in SQL92 under \"session variables\" (or was it another standard, I forgot).\n\nI currently use this for a Web perl CGI application that uses the same user\non a given IP address for all connections within 30 minutes without a password.\nBasically I have 2 procedures loginip and myuser:\n\ncreate dba procedure \"dns\".loginip(name char(16), pwd char(16), ip char(15))\ndefine global authuser char(16) default USER;\nif name <> authuser then\n let authuser = (select uname from passwd where uname=name and fromip=ip\n and lastlogin > current year to minute - 30 units minute);\n if authuser is null then\n let authuser = (select uname from passwd where uname=name and passwd=pwd);\n end if;\nend if;\nif authuser is null then\n let authuser = USER;\n raise exception -952,-61,'You entered an invalid login name or password.';\nelse\n update passwd set lastlogin=current year to minute, fromip=ip where uname=name;\nend if;\nend procedure;\ngrant execute on \"dns\".loginip to \"nobody\" as \"dns\";\n\ncreate procedure \"dns\".myuser()\nreturning char(16);\ndefine global authuser char(16) default USER;\nreturn authuser;\nend procedure;\ngrant execute on \"dns\".myuser to public as \"dns\"; \n\nI then use myuser() to do authentication with instead triggers checking if the\ncurrent myuser() is allowed to change certain values (domains).\n\nOne example:\n\ncreate dba procedure \"dns\".checkadmin()\ndefine auth integer;\nlet auth = (select count(*) from passwd where uname = myuser() and admin > 0);\nif auth < 1 then\n raise exception -273,0,'You are not user administrator.';\nend if;\nend procedure;\ngrant execute on \"dns\".checkadmin to public as \"dns\";\n\ncreate trigger \"dns\".passwd_utr\nupdate on passwd\nreferencing old as o new as n\nfor each row\nwhen (o.uname <> n.uname or o.uname <> myuser())\n ( execute procedure checkadmin() ); \n\nAndreas\n\n",
"msg_date": "Fri, 13 Mar 1998 09:58:17 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion (session global variables)"
}
] |
[
{
"msg_contents": "> Please comment/suggest !\n\nI think it would be good, if Oracle PL/SQL could be easily converted to this new language.\nSince this PL/SQL is what binds people to Oracle most. But only thinking about this makes\nme shudder ...\n\nAndreas\n\n",
"msg_date": "Fri, 13 Mar 1998 10:16:25 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] PL/pgSQL - for discussion"
}
] |
[
{
"msg_contents": ">> As indicated above there is an ELOG statement that can\n>> throw messages into the PostgreSQL elog mechanism.\n>> \n>> ELOG level 'format' [identifiers];\n> ^^^^^^^^^^\n>NO, pls - too postgres-ish! Just let ABORT to have 'format' etc and add \n>PRINT (or something like this) to put some messages to application (via NOTICE).\n>What are used in Oracle, Sybase etc here ?\n\nYes I was going to comment on this too:\nInformix:\traise exception -273,0,'You are not user administrator.'; --or\n\t\traise exception -100;\nOracle:\t\traise program_error; -- where program_error is one of many Orcl predefined error numbers\n\nI think we will need error numbers in the future that map directly to some text.\nThis would also help Michael with the ESQL/C preprocessor.\n\nAndreas\n\n",
"msg_date": "Fri, 13 Mar 1998 11:03:12 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion (ELOG)"
}
] |
[
{
"msg_contents": "> Returning from the function\n>\n> RETURN <expr>;\n>\n> The function terminates and the value of <expr> will be\n> returned to the upper executor. The return value of a\n> function cannot be undefined. If control reaches the end\n> of the toplevel block of the function without hitting a\n> RETURN statement, a runtime error will occur.\n\n1. I think we should allow functions/procedures, that don't have a return value.\n\t(think about a proc that only does an update, why return anything except an error)\n2. I think the RETURN will need a WITH RESUME, so a procedure can return more than one row.\n\ncreate dba procedure \"dns\".getusers()\nreturning varchar(16), smallint;\ndefine gotuser varchar(16);\ndefine gotadmin smallint;\nexecute procedure checkadmin();\nforeach select uname, admin into gotuser, gotadmin from passwd\n return gotuser, gotadmin with resume;\nend foreach;\nend procedure;\n\nAndreas\n\n\n",
"msg_date": "Fri, 13 Mar 1998 11:12:21 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion (RETURN)"
},
{
"msg_contents": "\nAndreas wrote:\n>\n> > Returning from the function\n> >\n> > RETURN <expr>;\n> >\n> > The function terminates and the value of <expr> will be\n> > returned to the upper executor. The return value of a\n> > function cannot be undefined. If control reaches the end\n> > of the toplevel block of the function without hitting a\n> > RETURN statement, a runtime error will occur.\n>\n> 1. I think we should allow functions/procedures, that don't have a return value.\n> (think about a proc that only does an update, why return anything except an error)\n> 2. I think the RETURN will need a WITH RESUME, so a procedure can return more than one row.\n>\n> create dba procedure \"dns\".getusers()\n> returning varchar(16), smallint;\n> define gotuser varchar(16);\n> define gotadmin smallint;\n> execute procedure checkadmin();\n> foreach select uname, admin into gotuser, gotadmin from passwd\n> return gotuser, gotadmin with resume;\n> end foreach;\n> end procedure;\n\n Yup. Looks nice. But the executor neither supports sets of\n tuples nor single tuple to be returned from functions (except\n for 'sql' language functions). I hacked around on that with\n the PL/Tcl and even if I return a heap tuple or a tuple table\n slot to the upper executor, it doesn't work :-(\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, 13 Mar 1998 13:36:39 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion (RETURN)"
}
] |
[
{
"msg_contents": "> > If char2 et al are going completely from 6.4, I think it would be sensible\n> > for pg_dump to filter these types and change them to char(2) et al when\n> > it writes the CREATE statments.\n> > \n> \n> That is an interesting idea, but what about applications that use them? \n> I think we can have the parser change them for one release, then drop\n> them completely.\n> \nNot just an interesting idea, but essential!\n\nThere are two separate issues here:\n1. Migrating data between PostgreSQL versions\n2. Migrating applications between PostgreSQL versions\n\nMy suggestion for modifying pg_dump addresses the first of these. Without\nthis fix, it will become impossible to migrate data without editing the\ndump file. The problem of column names mentioned before is not so\nserious as one can change the column name in the old version of the\ndatabase before dumping (providing you know this is necessary! A list\nof depracated column names in the upgrade procedure would be useful,\nor better pg_dump would tell you...). With changes of type, there is\nno option but to edit the dump file. If that happens to be bigger than\nyour total VM you are stuffed! (I guess you could mess around with\nhead and tail to split the file up or write a Perl script to modify\nthe schemae, but we shouldn't be expecting users to do this...)\n\nAn alternative workaround would be for pg_dumpall to have an option\nwhere it makes use of the pg_dump's ability to dump only the data\nor only the schema. That way one could edit just the schemae which\nwill be small and not have to touch the actual data dumps. However,\nthis really isn't necessary if pg_dump does the filtering.\n\nIssue 2 really isn't a problem. Providing there are adequate warnings\nin the upgrade information, developers can quite simply alter the\napplications they write which use CREATE statements. These warnings\nshould *stay in the upgrade information* for evermore... We must\nnot assume that people who are happy with a release they have of\nPostgreSQL will upgrade everytime a new version comes out.\n\n\n\nAndrew\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) [email protected] (Home) [email protected]\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n",
"msg_date": "Fri, 13 Mar 1998 10:18:04 GMT",
"msg_from": "Andrew Martin <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] varchar() vs char16 performance"
}
] |
[
{
"msg_contents": "I'm currently trying to get ecpg more in line with the backends parser.\nDoing so I found ecpg accepts a syntax the backend doesn't. It's 'commit\nwork release' which is accepted for compatibility with Oracle. How about\nallowing this in the backend, too?\n\nMichael\n-- \nDr. Michael Meskes, Project-Manager | topsystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Go Rhein Fire! | Tel: (+49) 2405/4670-44\nUse Debian GNU/Linux! | Fax: (+49) 2405/4670-10\n",
"msg_date": "Fri, 13 Mar 1998 11:22:08 +0100 (CET)",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Keyword"
},
{
"msg_contents": "> I'm currently trying to get ecpg more in line with the backends\n> parser. Doing so I found ecpg accepts a syntax the backend doesn't.\n> It's 'commit work release' which is accepted for compatibility with\n> Oracle. How about allowing this in the backend, too?\n\nI don't want to sound too fussy here, but does this statement provide\nany functionality missing from the SQL92 syntax 'commit [work]'? If not,\nthen perhaps Oracle users could start writing portable code if they\ncared to?\n\n - Tom\n",
"msg_date": "Fri, 13 Mar 1998 15:04:21 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Keyword"
}
] |
[
{
"msg_contents": "I forgot one more question. What exactly happens inside the backend when\nBEGIN is entered? Right now ecpg doesn't know about this at all. If there is\nsome action I think I should add this to ecpg, i.e. a BEGIN causes a PQexec\ncall with the 2nd parameter = 'begin'.\n\nMichael\n-- \nDr. Michael Meskes, Project-Manager | topsystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Go Rhein Fire! | Tel: (+49) 2405/4670-44\nUse Debian GNU/Linux! | Fax: (+49) 2405/4670-10\n",
"msg_date": "Fri, 13 Mar 1998 11:24:35 +0100 (CET)",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Too quick hitting the button"
}
] |
[
{
"msg_contents": "Forget about my last question. I found the begin call in ecpglib.c. It doesn\nwhat you expect from a embedded SQL preprocessor, it starts a new\ntransaction as soon as one ends. Nevertheless I thought about accepting\nexplicit begin calls in the new version. But they will always generate a\nwarning message as the code's always inside a transaction. So I could as\nwell accept the begin call but not give it to the backend.\n\nComments?\n\nMichael\n\n-- \nDr. Michael Meskes, Project-Manager | topsystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Go Rhein Fire! | Tel: (+49) 2405/4670-44\nUse Debian GNU/Linux! | Fax: (+49) 2405/4670-10\n",
"msg_date": "Fri, 13 Mar 1998 11:58:56 +0100 (CET)",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": true,
"msg_subject": "Begin statement again"
}
] |
[
{
"msg_contents": "I think we should depreciate the BEGIN/END keywords in SQL to allow them\nto be used for the new PL/SQL. So definitely leave them out of ecpg now.\nOnly accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit work; begin work) \nBTW.: why is a transaction always open ? A lot of programs would never need a \ntransaction. Is it because of cursors ?\n\nAndreas\n\n\nMichael Meskes wrote:\nForget about my last question. I found the begin call in ecpglib.c. It doesn\nwhat you expect from a embedded SQL preprocessor, it starts a new\ntransaction as soon as one ends. Nevertheless I thought about accepting\nexplicit begin calls in the new version. But they will always generate a\nwarning message as the code's always inside a transaction. So I could as\nwell accept the begin call but not give it to the backend.\n\n\n\n\n",
"msg_date": "Fri, 13 Mar 1998 12:50:27 +-100",
"msg_from": "Zeugswetter Andreas <[email protected]>",
"msg_from_op": true,
"msg_subject": "AW: [HACKERS] Begin statement again"
},
{
"msg_contents": "\nAndreas wrote:\n>\n> I think we should depreciate the BEGIN/END keywords in SQL to allow them\n> to be used for the new PL/SQL. So definitely leave them out of ecpg now.\n> Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit work; begin work)\n> BTW.: why is a transaction always open ? A lot of programs would never need a\n> transaction. Is it because of cursors ?\n\n BEGIN/END in PL/SQL and PL/pgSQL doesn't mean transactions!\n It's just to group statements to a block. You cannot commit\n something inside a PostgreSQL function. All changes made by a\n function are covered by the statements transaction or the\n upper transaction block.\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, 13 Mar 1998 13:43:29 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: AW: [HACKERS] Begin statement again"
},
{
"msg_contents": "Jan Wieck wrote:\n> \n> Andreas wrote:\n> >\n> > I think we should depreciate the BEGIN/END keywords in SQL to allow them\n> > to be used for the new PL/SQL. So definitely leave them out of ecpg now.\n> > Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit work; begin work)\n> > BTW.: why is a transaction always open ? A lot of programs would never need a\n> > transaction. Is it because of cursors ?\n> \n> BEGIN/END in PL/SQL and PL/pgSQL doesn't mean transactions!\n> It's just to group statements to a block. You cannot commit\n> something inside a PostgreSQL function. All changes made by a\n> function are covered by the statements transaction or the\n> upper transaction block.\n\nThis will be changed - there is a way to implement nested transaction!\nAnd so, some day we will need in something to start/end transaction\nblock inside functions.\n\nVadim\n",
"msg_date": "Wed, 25 Mar 1998 10:43:44 +0700",
"msg_from": "\"Vadim B. Mikheev\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: AW: [HACKERS] Begin statement again"
},
{
"msg_contents": "Andreas wrote:\n>\n> I think we should depreciate the BEGIN/END keywords in SQL to allow them\n> to be used for the new PL/SQL. So definitely leave them out of ecpg now.\n> Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit work; begin work)\n> BTW.: why is a transaction always open ? A lot of programs would never need a\n> transaction. Is it because of cursors ?\n\nBecause without transactions it is darn near impossible to build a database \nthat can guarantee data consistancy. Transactions are _the_ tool used to\nbuild robust systems that remain usable even after failures.\n\nFor example take the simple single statment:\n\ninsert into customers values(\"my name\", customer_number(\"my name\"));\n\nAssuming that there is an index on the name and id # columns, what happens\nif the system dies after the name index is updated, but the id # index \nis not? Your indexes are corrupt. With transactions, the whole thing just\nrolls back and remains consistant.\n\nSince PostgreSQL is more powerful than many databases, it is just about\nimpossible for a client application to tell what is really happening and\nwhether a transaction is needed even if the client only is using very\nsimple SQL that looks like it doesn't need a transaction.\n\nTake the SQL statement above and add a trigger or rule on the customers\ntable like so:\n\ncreate rule new_cust on insert to customers do after\n insert into daily_log values (\"new customer\", new.name);\n update statistics set total_customers = total_customers + 1 ...\n\nNow you really need a transaction.\n\nOh, but lets look at the customer_number() function:\n\nbegin\n return (select unique max(cust_no) + 1 from customers);\nend \n\nThis needs to lock the whole table and cannot release those locks until\nthe insert to customer is done. This too must be part of the transaction.\n\nFortunately, unlike say 'mySQL', posgreSQL does the right thing and always\nhas a transaction wrapped around any statement.\n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - Linux. Not because it is free. Because it is better.\n\n",
"msg_date": "Tue, 24 Mar 1998 22:16:16 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: AW: [HACKERS] Begin statement again"
},
{
"msg_contents": "\nVadim wrote:\n>\n> Jan Wieck wrote:\n> >\n> > Andreas wrote:\n> > >\n> > > I think we should depreciate the BEGIN/END keywords in SQL to allow them\n> > > to be used for the new PL/SQL. So definitely leave them out of ecpg now.\n> > > Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit work; begin work)\n> > > BTW.: why is a transaction always open ? A lot of programs would never need a\n> > > transaction. Is it because of cursors ?\n> >\n> > BEGIN/END in PL/SQL and PL/pgSQL doesn't mean transactions!\n> > It's just to group statements to a block. You cannot commit\n> > something inside a PostgreSQL function. All changes made by a\n> > function are covered by the statements transaction or the\n> > upper transaction block.\n>\n> This will be changed - there is a way to implement nested transaction!\n> And so, some day we will need in something to start/end transaction\n> block inside functions.\n\n What exactly is planned for this?\n\n Will it be possible to begin/commit a subtransaction so that\n updates done before and after it could still get rolled back\n while things inside remain persistent?\n\n Or would it be possible to commit up to now and resume (maybe\n a new transaction)?\n\n What would the syntax be for these statements, or must the\n function/trigger call special functions in the backend on the\n C level?\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, 25 Mar 1998 10:36:36 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: AW: [HACKERS] Begin statement again"
}
] |
[
{
"msg_contents": "I agree with BEGIN/END being used for PL/pgSQL. But that means I have to\nadd it to ecpg to start a block of PL/pgSQL, e.g. to call a stored\nprocedure in ORACLE I need PL/SQL in C.\n\nAFAIK there is no transaction start command in SQL standard anymore.\nJust issue commit and your transaction is flushed. So either I have to\nremove that feature and go back to explicit transaction start commands\nor BEGIN WORK resp. BEGIN TRANSACTION are useless.\n\nMichael\n\n--\nDr. Michael Meskes, Project-Manager | topsystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Go Rhein Fire! | Tel: (+49) 2405/4670-44\nUse Debian GNU/Linux! | Fax: (+49) 2405/4670-10\n\n> -----Original Message-----\n> From:\tZeugswetter Andreas [SMTP:[email protected]]\n> Sent:\tFriday, March 13, 1998 1:50 PM\n> To:\t'Michael Meskes'\n> Cc:\t'[email protected]'\n> Subject:\tAW: [HACKERS] Begin statement again\n> \n> I think we should depreciate the BEGIN/END keywords in SQL to allow\n> them\n> to be used for the new PL/SQL. So definitely leave them out of ecpg\n> now.\n> Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit\n> work; begin work) \n> BTW.: why is a transaction always open ? A lot of programs would never\n> need a \n> transaction. Is it because of cursors ?\n> \n> Andreas\n> \n> \n> Michael Meskes wrote:\n> Forget about my last question. I found the begin call in ecpglib.c. It\n> doesn\n> what you expect from a embedded SQL preprocessor, it starts a new\n> transaction as soon as one ends. Nevertheless I thought about\n> accepting\n> explicit begin calls in the new version. But they will always generate\n> a\n> warning message as the code's always inside a transaction. So I could\n> as\n> well accept the begin call but not give it to the backend.\n> \n> \n> \n",
"msg_date": "Fri, 13 Mar 1998 12:57:42 +0100",
"msg_from": "\"Meskes, Michael\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Begin statement again"
},
{
"msg_contents": "> \n> I agree with BEGIN/END being used for PL/pgSQL. But that means I have to\n> add it to ecpg to start a block of PL/pgSQL, e.g. to call a stored\n> procedure in ORACLE I need PL/SQL in C.\n> \n> AFAIK there is no transaction start command in SQL standard anymore.\n> Just issue commit and your transaction is flushed. So either I have to\n> remove that feature and go back to explicit transaction start commands\n> or BEGIN WORK resp. BEGIN TRANSACTION are useless.\n\n But there is AUTOCOMMIT ON/OFF\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": "Fri, 13 Mar 1998 13:38:17 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Begin statement again"
},
{
"msg_contents": "Jan Wieck writes:\n> But there is AUTOCOMMIT ON/OFF\n\nHmm, but not in ecpg. It seems this has to be added.\n\nAnyway, here's what ecpg allows for transaction handling:\n\nABORT TRANSACTION | ROLLBACK WORK | ABORT | ROLLBACK;\n\nEND TRANSCACTION | COMMIT | COMMIT RELEASE | COMMIT WORK RELEASE;\n\nBEGIN | BEGIN TRANSACTION | BEGIN WORK;\n\nMichael\n\n-- \nDr. Michael Meskes, Project-Manager | topsystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Go Rhein Fire! | Tel: (+49) 2405/4670-44\nUse Debian GNU/Linux! | Fax: (+49) 2405/4670-10\n",
"msg_date": "Fri, 13 Mar 1998 13:44:13 +0100 (CET)",
"msg_from": "Michael Meskes <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Begin statement again"
}
] |
[
{
"msg_contents": "How do you set autocommit off in ORACLE embedded SQL? I only know this\nfrom ODBC.\n\nMichael\n\n\n--\nDr. Michael Meskes, Project-Manager | topsystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Go Rhein Fire! | Tel: (+49) 2405/4670-44\nUse Debian GNU/Linux! | Fax: (+49) 2405/4670-10\n\n> -----Original Message-----\n> From:\[email protected] [SMTP:[email protected]]\n> Sent:\tFriday, March 13, 1998 1:38 PM\n> To:\[email protected]\n> Cc:\[email protected]; [email protected];\n> [email protected]\n> Subject:\tRe: [HACKERS] Begin statement again\n> \n> > \n> > I agree with BEGIN/END being used for PL/pgSQL. But that means I\n> have to\n> > add it to ecpg to start a block of PL/pgSQL, e.g. to call a stored\n> > procedure in ORACLE I need PL/SQL in C.\n> > \n> > AFAIK there is no transaction start command in SQL standard anymore.\n> > Just issue commit and your transaction is flushed. So either I have\n> to\n> > remove that feature and go back to explicit transaction start\n> commands\n> > or BEGIN WORK resp. BEGIN TRANSACTION are useless.\n> \n> But there is AUTOCOMMIT ON/OFF\n> \n> \n> Jan\n> \n> -- \n> \n> #=====================================================================\n> =#\n> # It's easier to get forgiveness for being wrong than for being right.\n> #\n> # Let's break this rule - forgive me.\n> #\n> #======================================== [email protected] (Jan Wieck)\n> #\n",
"msg_date": "Fri, 13 Mar 1998 13:42:38 +0100",
"msg_from": "\"Meskes, Michael\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Begin statement again"
}
] |
[
{
"msg_contents": "[email protected] (Jan Wieck) wrote:\n\n> But I would like to have some discussion on language itself.\n> So I wrote down what I had in mind. The document is appended\n> below.\n>\n> Please comment/suggest !\n>\n> Someone gave a hint about global variables existing during a\n> session. What is a session than? One transaction? The\n> backends lifetime?\n\nI think it is a time from connect to disconnect, which currently equals to backends\nlifetime, but may in future be shorter, if we will implement a backend pool for\nquick-starting servers.\n\n> And should global variables be visible by\n> more than one function?\n\nHow are they global then ?\n\nI think that global variables are something like DateStyle is now.\n\n> I vote for NO! In that case we need\n> something like packages of functions that share globals.\n\nOr we need local temporary tables.\n\n> PL/pgSQL\n> A procedural language for the PostgreSQL RDBMS\n> 1st draft\n>\n> Jan Wieck <[email protected]>\n>\n> Notice\n>\n> This document is for the postgresql-hackers community for\n> completing the syntax specification of PL/pgSQL. The\n> extension module described here doesn't exist up to now!\n>\n> Preface\n>\n> PL/pgSQL is a procedural language based on SQL designed for\n> the PostgreSQL database system.\n>\n> The extensibility features of PostgreSQL are mostly based on\n> the ability to define functions for various operations.\n> Functions could have been written in PostgreSQL's SQL dialect\n> or in the C programming language. Functions written in C are\n> compiled into a shared object and loaded by the database\n> backend process on demand. Also the trigger features of\n> PostgreSQL are based on functions but required the use of the\n> C language.\n>\n> Since version 6.3 PostgreSQL supports the definition of\n> procedural languages. In the case of a function or trigger\n> procedure defined in a procedural language, the database has\n> no builtin knowlege how to interpret the functions source\n> text. Instead, the function and trigger calls are passed into\n> a handler that knows the details of the language. The\n> handler itself is function compiled into a shared object and\n> loaded on demand.\n>\n> Overview\n>\n> The PL/pgSQL language is case insensitive. All keywords and\n> identifiers can be used in upper-/lowercase mixed.\n>\n> PL/pgSQL is a block oriented language. A block is defined as\n>\n> [<<label>>]\n> [DECLARE\n> -- declarations]\n> BEGIN\n> -- statements\n> END;\n\n\n\n>\n>\n> There can be any number of subblocks in the statements\n> section of a block. Subblocks can be used to hide variables\n> from outside a block of statements (see Scope and visability\n> below).\n\nI think that subblock should also be used as the extent of IF, FOR ... LOOP, WHILE and\nother such statements.\n\nThen we would not need the END IF, END LOOP etc.\n\n>\n>\n> Comments\n>\n> There are two types of comments in PL/pgSQL. A double dash\n> '--' starts a comment that extends to the end of the line. A\n> '/*' starts a block comment that extends to the next '*/'.\n> Block comments cannot be nested, but double dash comments can\n> be enclosed into a block comment.\n\nAnd vice versa : block comment delimiters can be commented out by --\n\n>\n>\n> Declarations\n>\n> All variables, rows and records used in a block or it's\n> subblocks must be declared in the declarations section of the\n> block. The parameters given to the function are\n> automatically declared with the usual identifiers $n.\n\nHow hard would it bet to have named parameters, or why must we use alias?\n\n> The declarations have the following syntax:\n>\n> <name> [CONSTANT] <type> [NOT NULL]\n> [DEFAULT | := <value>];\n>\n> Declares a variable of the specified type. If the\n> variable is declared as CONSTANT, the value cannot be\n> changed. If NOT NULL is specified, an assignment of a\n> NULL value results in a runtime error. Since the\n> default value of a variable is the SQL NULL value,\n> all variables declared as NOT NULL must also have a\n> default value.\n>\n> The default value is evaluated at the actual function\n> call. So assigning 'now' to an abstime varable causes\n> the variable to have the time of the actual function\n> call, not when the function was compiled.\n>\n> <name> <class>%ROWTYPE;\n>\n> Declares a row with the structure of the given class.\n> Class must be an existing table- or viewname of the\n> database. The fields of the row are accessed in the\n> dot notation. Parameters to a procedure could be\n> tuple types. In that case the corresponding\n> identifier $n will be a rowtype. Only the user\n> attributes and the oid of a tuple are accessible in\n> the row. There must be no whitespaces between the\n> classname, the percent and the ROWTYPE keyword.\n>\n> <name> RECORD;\n>\n> Records are similar to rowtypes, but they have no\n> predefined structure and it's impossible to assign a\n> value into them. They are used in selections and FOR\n> loops to hold one actual database tuple from a select\n> operation. One and the same record can be used in\n> different selections (but not in nested ones).\n>\n> <name> ALIAS FOR $n;\n>\n> For better readability of the code it's possible to\n> define an alias for a positional parameter to the\n> function.\n>\n> Datatypes\n>\n> The type of a variable can be any of the existing data types\n> of the database. <type> above is defined as:\n>\n> postgesql-basetype\n> or variable%TYPE\n> or rowtype.field%TYPE\n> or class.field%TYPE\n>\n> As for the rowtype declaration, there must be no whitespaces\n> between the classname, the percent and the TYPE keyword.\n>\n> Expressions\n>\n> All expressions used in PL/pgSQL statements are processed\n> using the backends executor. Since even a constant looking\n> expression can have a totally different meaning for a\n> particular data type (as 'now' for abstime), it is impossible\n> for the PL/pgSQL parser to identify real constant values\n> other than the NULL keyword. The expressions are evaluated by\n> internally executing a query\n>\n> SELECT <expr>\n>\n> over the SPI manager. In the expression, occurences of\n> variable identifiers are substituted by parameters and the\n> actual values from the variables are passed to the executor\n> as query parameters. All the expressions used in a PL/pgSQL\n> function are only prepared and saved once.\n>\n> Statements\n>\n> Anything not understood by the parser as specified below will\n> be put into a query and sent down to the database engine to\n> execute. The resulting query should not return any data.\n>\n> Assignment\n>\n> An assignment of a value to a variable or rowtype field\n> is written as:\n>\n> <identifier> := <expr>;\n>\n> If the expressions result data type doesn't match the\n> variables data type, or the variables atttypmod value is\n> known (as for char(20)), the result value will be\n> implicitly casted by the PL/pgSQL executor using the\n> result types output- and the variables type input-\n> functions. Note that this could potentially result in\n> runtime errors generated by the types input functions.\n>\n> An assignment of a complete selection into a record or\n> rowtype can be done as:\n>\n> SELECT targetlist INTO <recname|rowname> FROM fromlist;\n>\n> If a rowtype is used as target, the selected targetlist\n> must exactly match the structure of the rowtype or a\n> runtime error occurs. The fromlist can be followed by\n> any valid qualification, grouping, sorting etc. There is\n> a special condition [NOT] FOUND that can be used\n> immediately after a SELECT INTO to check if the data has\n> been found.\n>\n> SELECT * INTO myrec FROM EMP WHERE empname = myname;\n> IF NOT FOUND THEN\n> ELOG ERROR 'employee %s not found' myname;\n> END IF;\n>\n> In addition, the select statement must not return more\n> that one row. If multiple rows are returned, a runtime\n> error will be generated.\n>\n> Returning from the function\n>\n> RETURN <expr>;\n>\n> The function terminates and the value of <expr> will be\n> returned to the upper executor. The return value of a\n> function cannot be undefined. If control reaches the end\n> of the toplevel block of the function without hitting a\n> RETURN statement, a runtime error will occur.\n\nWhat can <expr> be?\n\nPossibilities: null, single value, record, recordset\n\nAFAIK, recordsets are usually returned by more strange constructs, like haveing some\nkinds of breakpoints inside the loop that either returns a record or some EOF token.\n\n> A conditional loop that is executed as long as the\n> evaluation of <expr> returns true.\n>\n> [<<label>>]\n> FOR <name> IN [REVERSE] <expr>..<expr> LOOP\n> -- statements\n\n> END LOOP.\n\nPerhaps PL/PgSQL should have some notation for specifying immediate lists of other kinds\nof values as well,perhaps like ['aa','bb','cc'], so that one cold loop over not only\nintegers. Perhaps even with type specifiers, like [::time '22.01', '13.47.05', '15.20']\n\n>\n>\n> A loop that iterates over a range of integer values. The\n> variable <name> is automatically created as type integer\n> and exists only inside the loop. The two expressions\n> giving the lower and upper bound of the range are\n> evaluated only when entering the loop. The iteration step\n> is 1.\n>\n> FOR <recname|rowname> IN <select_clause> LOOP\n> -- statements\n> END LOOP;\n>\n> The record or row is assigned all the rows resulting from\n> the select clause and the statements executed for each.\n> If the loop is terminated with an EXIT statement, the\n> last accessed row is still accessible in the record or\n> rowtype.\n>\n> EXIT [label] [WHEN <expr>];\n>\n> If no label given, the innermost loop is terminated and\n> the statement following END LOOP is executed next. If\n> label is given, it must be the label of the current or an\n> upper level of nested loops or blocks. Then the named\n> loop or block is terminated and control continues with\n> the statement after the loops/blocks corresponding END.\n>\n\nKeep up the nice work!\n\nHannu\n\n",
"msg_date": "Fri, 13 Mar 1998 16:46:08 +0200",
"msg_from": "Hannu Krosing <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: PL/PgSQL discussion"
},
{
"msg_contents": "\nHannu Krosing wrote:\n>\n> [email protected] (Jan Wieck) wrote:\n>\n> > Someone gave a hint about global variables existing during a\n> > session. What is a session than? One transaction? The\n> > backends lifetime?\n>\n> I think it is a time from connect to disconnect, which currently equals to backends\n> lifetime, but may in future be shorter, if we will implement a backend pool for\n> quick-starting servers.\n\n Hmmm - how does a language handler then notice that a new\n session began?\n\n> >\n> > PL/pgSQL is a block oriented language. A block is defined as\n> >\n> > [<<label>>]\n> > [DECLARE\n> > -- declarations]\n> > BEGIN\n> > -- statements\n> > END;\n>\n>\n>\n> >\n> >\n> > There can be any number of subblocks in the statements\n> > section of a block. Subblocks can be used to hide variables\n> > from outside a block of statements (see Scope and visability\n> > below).\n>\n> I think that subblock should also be used as the extent of IF, FOR ... LOOP, WHILE and\n> other such statements.\n>\n> Then we would not need the END IF, END LOOP etc.\n\n The LOOP ... END LOOP etc. syntax is just what I saw in\n Oracles PL/SQL documentation. I could also live with\n BEGIN...END, but what is it good for to be different?\n\n>\n> >\n> >\n> > Comments\n> >\n> > There are two types of comments in PL/pgSQL. A double dash\n> > '--' starts a comment that extends to the end of the line. A\n> > '/*' starts a block comment that extends to the next '*/'.\n> > Block comments cannot be nested, but double dash comments can\n> > be enclosed into a block comment.\n>\n> And vice versa : block comment delimiters can be commented out by --\n\n Right - works already :-)\n\n>\n> >\n> >\n> > Declarations\n> >\n> > All variables, rows and records used in a block or it's\n> > subblocks must be declared in the declarations section of the\n> > block. The parameters given to the function are\n> > automatically declared with the usual identifiers $n.\n>\n> How hard would it bet to have named parameters, or why must we use alias?\n\n That isn't subject to the PL handler. All the PL handler\n knows about the function is in pg_proc and pg_type. There are\n no parameter names, and that's what the ALIAS idea came from.\n\n If we sometimes implement a new function call interface, this\n might be possible. Some details about what I have in mind:\n\n Add a field to pg_proc that tells the backend the call\n interface the function uses.\n\n Create a new catalog pg_parameter, that holds the\n parameter names and other information (like notnull,\n atttypmod etc.). So a function can be defined to expect a\n VARCHAR(20) NOT NULL.\n\n The new call interface then hands out more information to\n the function than now. It's the functions Oid, the\n parameter Datums, a character array telling which of the\n Datums are NULL and the usual bool pointer where the\n function can tell that it's return value is NULL.\n\n> > Returning from the function\n> >\n> > RETURN <expr>;\n> >\n> > The function terminates and the value of <expr> will be\n> > returned to the upper executor. The return value of a\n> > function cannot be undefined. If control reaches the end\n> > of the toplevel block of the function without hitting a\n> > RETURN statement, a runtime error will occur.\n>\n> What can <expr> be?\n>\n> Possibilities: null, single value, record, recordset\n>\n> AFAIK, recordsets are usually returned by more strange constructs, like haveing some\n> kinds of breakpoints inside the loop that either returns a record or some EOF token.\n>\n\n Currently only 'null' and 'single value'. The executor\n doesn't accept anything else for non-sql language functions.\n PL functions are treated by the executor like 'C' functions.\n\n> > A conditional loop that is executed as long as the\n> > evaluation of <expr> returns true.\n> >\n> > [<<label>>]\n> > FOR <name> IN [REVERSE] <expr>..<expr> LOOP\n> > -- statements\n>\n> > END LOOP.\n>\n> Perhaps PL/PgSQL should have some notation for specifying immediate lists of other kinds\n> of values as well,perhaps like ['aa','bb','cc'], so that one cold loop over not only\n> integers. Perhaps even with type specifiers, like [::time '22.01', '13.47.05', '15.20']\n\n Good issue. Could be done.\n\n>\n> Keep up the nice work!\n>\n> Hannu\n>\n>\n\n Already reached the point of no return. The first tiny\n function ran without problems:\n\n CREATE FUNCTION f1(int4, int4) RETURNS int4 AS '\n BEGIN\n RETURN $1 + $2;\n END;\n ' LANGUAGE 'plpgsql';\n\n I set up a little test table with 2 int4 fields containing\n some thousand records. Then I wrote equivalent functions in\n 'sql', 'plpgsql' and 'pltcl'. The execution times for a\n query\n\n SELECT sum(func(a, b)) FROM t1;\n\n are:\n\n Builtin SQL language 100%\n PL/Tcl 180%\n PL/pgSQL 230%\n\n PL/Tcl is slower than builtin SQL because the internals of it\n require the two parameters to be converted to their external\n representation, than calling the Tcl interpreter who parses\n them back to numbers, calculates the result, returns it as\n string and then it's parsed back to int4 internal value.\n\n In the PL/pgSQL case I haven't expected that big performance\n loss. The calculation is internally done with a saved query\n plan (made on the first call) that does a\n\n SELECT $1 + $2\n\n with two int4 parameters. This is exactly what the SQL\n version of the above does! And >95% of the execution time\n for the function call are spent in SPI_execp(). Since\n SPI_execp() calls ExecutorRun() directly, I think the\n querydesc creation and/or plan copying on each invocation is\n the time consuming part. I assume that there are some\n optimizable corners in SPI where we can gain more speed. So I\n continue with PL/pgSQL as it is now and speed it up later by\n tuning SPI.\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, 13 Mar 1998 17:16:27 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: PL/PgSQL discussion"
},
{
"msg_contents": "Jan Wieck wrote:\n\n> Hannu Krosing wrote:\n> >\n> > [email protected] (Jan Wieck) wrote:\n> >\n> > > Someone gave a hint about global variables existing during a\n> > > session. What is a session than? One transaction? The\n> > > backends lifetime?\n> >\n> > I think it is a time from connect to disconnect, which currently equals to backends\n> > lifetime, but may in future be shorter, if we will implement a backend pool for\n> > quick-starting servers.\n>\n> Hmmm - how does a language handler then notice that a new\n> session began?\n\nProbably by defining the reset_session or init_session functions for the language handler and\ncalling it at the connect time.\n\n> > >\n> > > PL/pgSQL is a block oriented language. A block is defined as\n> > >\n> > > [<<label>>]\n> > > [DECLARE\n> > > -- declarations]\n> > > BEGIN\n> > > -- statements\n> > > END;\n> >\n> >\n> >\n> > >\n> > >\n> > > There can be any number of subblocks in the statements\n> > > section of a block. Subblocks can be used to hide variables\n> > > from outside a block of statements (see Scope and visability\n> > > below).\n> >\n> > I think that subblock should also be used as the extent of IF, FOR ... LOOP, WHILE and\n> > other such statements.\n> >\n> > Then we would not need the END IF, END LOOP etc.\n>\n> The LOOP ... END LOOP etc. syntax is just what I saw in\n> Oracles PL/SQL documentation. I could also live with\n> BEGIN...END, but what is it good for to be different?\n\nthen again we don't need the block delimiters ;)\n\n> > >\n> > > Declarations\n> > >\n> > > All variables, rows and records used in a block or it's\n> > > subblocks must be declared in the declarations section of the\n> > > block. The parameters given to the function are\n> > > automatically declared with the usual identifiers $n.\n> >\n> > How hard would it bet to have named parameters, or why must we use alias?\n>\n> That isn't subject to the PL handler. All the PL handler\n> knows about the function is in pg_proc and pg_type. There are\n> no parameter names, and that's what the ALIAS idea came from.\n\nI just meant them as an automatic way to declare and use aliases for $N, like for example C\ndoes currently.\n\nThe calling function does not know the local names of called function in C either.\n\n> If we sometimes implement a new function call interface, this\n> might be possible. Some details about what I have in mind:\n>\n> Add a field to pg_proc that tells the backend the call\n> interface the function uses.\n>\n> Create a new catalog pg_parameter, that holds the\n> parameter names and other information (like notnull,\n> atttypmod etc.). So a function can be defined to expect a\n> VARCHAR(20) NOT NULL.\n\nThis again can be done by automatically rewriting these to additional statements to check for\nthis as first things in the function body. It makes only a theoretical difference if the error\nis reportid before the call or just after it.\n\n> Currently only 'null' and 'single value'. The executor\n> doesn't accept anything else for non-sql language functions.\n> PL functions are treated by the executor like 'C' functions.\n\nActually what I understood from the docs was thatit is 'terribly complicated' and 'beyond the\nscope of this tutorial', but not impossible ;)\n\n---------\n\nHannu\n\n",
"msg_date": "Fri, 13 Mar 1998 18:53:08 +0200",
"msg_from": "Hannu Krosing <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: PL/PgSQL discussion"
},
{
"msg_contents": "\nHannu Krosing wrote:\n>\n> Jan Wieck wrote:\n>\n> > Hannu Krosing wrote:\n> > >\n> > > [email protected] (Jan Wieck) wrote:\n> > >\n> >\n> > Hmmm - how does a language handler then notice that a new\n> > session began?\n>\n> Probably by defining the reset_session or init_session functions for the language handler and\n> calling it at the connect time.\n\n Not in the current implementation of the backend/pl_handler\n interface. But a global variable SessionId might help.\n\n> > > Then we would not need the END IF, END LOOP etc.\n> >\n> > The LOOP ... END LOOP etc. syntax is just what I saw in\n> > Oracles PL/SQL documentation. I could also live with\n> > BEGIN...END, but what is it good for to be different?\n>\n> then again we don't need the block delimiters ;)\n\n Forgotten in the 1st draft. Variables are initialized to\n their default values every time a block/subblock is entered.\n Not only when the function is entered.\n\n ...\n FOR i IN 1..10 LOOP\n DECLARE\n flag bool DEFAULT false;\n n integer DEFAULT 0;\n BEGIN\n ...\n END;\n END LOOP;\n ...\n\n> > Create a new catalog pg_parameter, that holds the\n> > parameter names and other information (like notnull,\n> > atttypmod etc.). So a function can be defined to expect a\n> > VARCHAR(20) NOT NULL.\n>\n> This again can be done by automatically rewriting these to additional statements to check for\n> this as first things in the function body. It makes only a theoretical difference if the error\n> is reportid before the call or just after it.\n\n But this rewriting must be done when the function is created.\n At this time, the pl_handler and it's parser aren't called.\n It is done by the backends main parser. For loadable\n procedural languages, the main parser doesn't know anything\n about the languages syntax or if the string given after AS is\n a program text at all. It only creates the pg_proc tuple.\n\n>\n> > Currently only 'null' and 'single value'. The executor\n> > doesn't accept anything else for non-sql language functions.\n> > PL functions are treated by the executor like 'C' functions.\n>\n> Actually what I understood from the docs was thatit is 'terribly complicated' and 'beyond the\n> scope of this tutorial', but not impossible ;)\n\n I tried that really hard with no luck. And I know\n ExecMakeFunctionResult() pretty good. But I'll give it\n another shot when PL/pgSQL reached a state where it makes\n sense.\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, 13 Mar 1998 18:50:23 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: PL/PgSQL discussion"
},
{
"msg_contents": "Jan Wieck wrote:\n\n> > > Hmmm - how does a language handler then notice that a new\n> > > session began?\n> >\n> > Probably by defining the reset_session or init_session functions for the language handler and\n> > calling it at the connect time.\n>\n> Not in the current implementation of the backend/pl_handler\n> interface. But a global variable SessionId might help.\n\nNeither do we have backend pooling now.\n\n> > > > Then we would not need the END IF, END LOOP etc.\n> > >\n> > > The LOOP ... END LOOP etc. syntax is just what I saw in\n> > > Oracles PL/SQL documentation. I could also live with\n> > > BEGIN...END, but what is it good for to be different?\n> >\n> > then again we don't need the block delimiters ;)\n>\n> Forgotten in the 1st draft. Variables are initialized to\n> their default values every time a block/subblock is entered.\n> Not only when the function is entered.\n>\n> ...\n> FOR i IN 1..10 LOOP\n> DECLARE\n> flag bool DEFAULT false;\n> n integer DEFAULT 0;\n> BEGIN\n> ...\n> END;\n> END LOOP;\n> ...\n\nOk, this makes sense.\n\n> > This again can be done by automatically rewriting these to additional statements to check for\n> > this as first things in the function body. It makes only a theoretical difference if the error\n> > is reportid before the call or just after it.\n>\n> But this rewriting must be done when the function is created.\n> At this time, the pl_handler and it's parser aren't called.\n\nOk, I understand it now.\n\n<name> ALIAS FOR $n;\n\nis to overcome the split between the two parser.\n\n> It is done by the backends main parser. For loadable\n> procedural languages, the main parser doesn't know anything\n> about the languages syntax or if the string given after AS is\n> a program text at all. It only creates the pg_proc tuple.\n\nAnd another thing:\n\nDoes'nt the Oracle PL/SQL have both input and input/output parameters ?\n\nIt's more than a year since i last used it but I think that this is the case.\n\nHannu\n\n",
"msg_date": "Fri, 13 Mar 1998 20:43:11 +0200",
"msg_from": "Hannu Krosing <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: PL/PgSQL discussion"
},
{
"msg_contents": "> \n> [email protected] (Jan Wieck) wrote:\n> \n> > But I would like to have some discussion on language itself.\n> > So I wrote down what I had in mind. The document is appended\n> > below.\n> >\n> > Please comment/suggest !\n\nA question. Will it be possible to call functions from the PL/pgSQL? This\ncovers a fair bit, for example to evaluate expressions and casts etc as\nwell as calling user functions.\n\nOr was this covered and I just missed it somehow?\n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n",
"msg_date": "Fri, 13 Mar 1998 23:47:24 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: PL/PgSQL discussion"
},
{
"msg_contents": "[email protected] (Jan Wieck) wrote:\n> Hannu Krosing wrote:\n> > [email protected] (Jan Wieck) wrote:\n...\n> > I think it is a time from connect to disconnect, which currently equals to backends\n> > lifetime, but may in future be shorter, if we will implement a backend pool for\n> > quick-starting servers.\n> \n> Hmmm - how does a language handler then notice that a new\n> session began?\n\nEither when the backend exits, or is re-initted to go back to the idle pool.\n \n\n> > > PL/pgSQL is a block oriented language. A block is defined as\n> > > [<<label>>]\n> > > [DECLARE\n> > > -- declarations]\n> > > BEGIN\n> > > -- statements\n> > > END;\n...\n> > I think that subblock should also be used as the extent of IF, FOR ... LOOP, WHILE and\n> > other such statements.\n> > Then we would not need the END IF, END LOOP etc.\n> \n> The LOOP ... END LOOP etc. syntax is just what I saw in\n> Oracles PL/SQL documentation. I could also live with\n> BEGIN...END, but what is it good for to be different?\n\nI am not convinced that imitating Oracle is necessarily the best possible way\nto proceed. If there is a standard, and I think there is at least a proposal\n(called PSM, if memory serves), we might (but only might) want to do that.\nOtherwise, either lets steal something worth stealing or invent something\nbetter than whats out there.\n\n\n> > How hard would it bet to have named parameters, or why must we use alias?\n> \n> That isn't subject to the PL handler. All the PL handler\n> knows about the function is in pg_proc and pg_type. There are\n> no parameter names, and that's what the ALIAS idea came from.\n... \n> > > Returning from the function\n> > >\n> > > RETURN <expr>;\n> > >\n> > What can <expr> be?\n> >\n> > Possibilities: null, single value, record, recordset\n> >\n> > AFAIK, recordsets are usually returned by more strange constructs, like haveing some\n> > kinds of breakpoints inside the loop that either returns a record or some EOF token.\n> >\n> Currently only 'null' and 'single value'. The executor\n> doesn't accept anything else for non-sql language functions.\n> PL functions are treated by the executor like 'C' functions.\n\nThis limitation suggests that PL/pgSQL functions want to _be_ SQL functions,\nnot 'C' functions. Handy to be able to write:\n\nbegin\n if $1 = \"totals\" then\n select sum(qty), sum(qty) * price from sales_orders where ...;\n else if $1 = \"details\"\n select qty, price from sales_orders where ...\n\nOk, lousy example, but I have seen this kind of thing in apps ...\n\n\n> Already reached the point of no return. The first tiny\n> function ran without problems:\n\nI am impressed.\n \n> CREATE FUNCTION f1(int4, int4) RETURNS int4 AS '\n> BEGIN\n> RETURN $1 + $2;\n> END;\n> ' LANGUAGE 'plpgsql';\n> \n> I set up a little test table with 2 int4 fields containing\n> some thousand records. Then I wrote equivalent functions in\n> 'sql', 'plpgsql' and 'pltcl'. The execution times for a\n> query\n> \n> SELECT sum(func(a, b)) FROM t1;\n> \n> are:\n> \n> Builtin SQL language 100%\n> PL/Tcl 180%\n> PL/pgSQL 230%\n> \n> PL/Tcl is slower than builtin SQL because the internals of it\n> require the two parameters to be converted to their external\n> representation, than calling the Tcl interpreter who parses\n> them back to numbers, calculates the result, returns it as\n> string and then it's parsed back to int4 internal value.\n>\n> In the PL/pgSQL case I haven't expected that big performance\n> loss. The calculation is internally done with a saved query\n> plan (made on the first call) that does a\n> \n> SELECT $1 + $2\n> \n> with two int4 parameters. This is exactly what the SQL\n> version of the above does! And >95% of the execution time\n> for the function call are spent in SPI_execp(). Since\n> SPI_execp() calls ExecutorRun() directly, I think the\n> querydesc creation and/or plan copying on each invocation is\n> the time consuming part. I assume that there are some\n> optimizable corners in SPI where we can gain more speed. So I\n> continue with PL/pgSQL as it is now and speed it up later by\n> tuning SPI.\n\nI think you might want to do some profiling to find where the time is\nreally going. Not to throw water on your parade, but being slower than\ntcl is something of an achievement ;-).\n\nI wish I could be of more use than this, but that will have to wait\nuntil I get some time to look over the postgres code a bit. It is similar\nto Illustra in many ways, but the two lines branched apart quite a while\nago and have gone in different directions especially in the language areas.\n \n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n",
"msg_date": "Sat, 14 Mar 1998 00:21:24 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: PL/PgSQL discussion"
},
{
"msg_contents": "\nHannu Krosing wrote:\n>\n> And another thing:\n>\n> Does'nt the Oracle PL/SQL have both input and input/output parameters ?\n>\n> It's more than a year since i last used it but I think that this is the case.\n\n Right, they have. But PostgreSQL doesn't and so PL/pgSQL\n can't. PL/pgSQL cannot do any thing that a C function in\n PostgreSQL cannot do. The PL handler is written in C and that\n is what the executor calls instead of the compiled function.\n From the backends point of view, a PL handler is just one C\n function with a bunch of personalities.\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, 16 Mar 1998 08:28:38 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: PL/PgSQL discussion"
},
{
"msg_contents": "David Gould wrote:\n> \n> > \n> > [email protected] (Jan Wieck) wrote:\n> > \n> > > But I would like to have some discussion on language itself.\n> > > So I wrote down what I had in mind. The document is appended\n> > > below.\n> > >\n> > > Please comment/suggest !\n> \n> A question. Will it be possible to call functions from the PL/pgSQL? This\n> covers a fair bit, for example to evaluate expressions and casts etc as\n> well as calling user functions.\n> \n> Or was this covered and I just missed it somehow?\n\n It is covered automagically by the way I'm implementing the\n assignement. The assign code internally does a SELECT. And\n thus:\n\n n := count(*) from t1;\n\n works (even if it's far from standard I think).\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": "Mon, 16 Mar 1998 08:39:22 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: PL/PgSQL discussion"
}
] |
[
{
"msg_contents": "> How can I use composite types?\n> If I create two tables like this:\n> create table A ( a int, b int);\n> create table D ( a A, b int);\n> I would expect to be able to\n> insert into D (a.a, a.b, b) values (1,2,3);\n> and\n> select a.a, a.b, b from D;\n> That doesn't work, at least not with postgresql 6.3. Is that the way\n> it's supposed to work? What is the proper way to do it?\n\nI'm not sure if this is supported. The old tutorial does not have an\nexample exactly like this (at least that I could find), but it seems\nlike it _should_ work, and the backend accepts the syntax.\n\nWhat you want to do can probably be accomplished with inheritance, but\nthat only seems to work if you have unique names for more of your\nfields:\n\n create table A ( a int, b int);\n create table D ( c int) inherits (A);\n\nBut, if you try\n\n create table D ( b int) inherits (A);\n\nthen the new column in D gets lost without warning!\n\n - Tom\n",
"msg_date": "Fri, 13 Mar 1998 15:37:22 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] using composite types"
},
{
"msg_contents": "> What you want to do can probably be accomplished with inheritance, but\n> that only seems to work if you have unique names for more of your\n> fields:\n> \n> create table A ( a int, b int);\n> create table D ( c int) inherits (A);\n> \n> But, if you try\n> \n> create table D ( b int) inherits (A);\n> \n> then the new column in D gets lost without warning!\n\n\nPatient: Ow, doctor, it hurts when I do this.\nDoctor: Well, then don't do that.\n\n---------------------------------------------------------------------------\n\nThis is a pretty bizarre example, and I am not sure how the system will\nhandle this. My guess is that it finds only the first b. Seems like we\nneed a check somewhere when making inheritance. I will add it to the\nTODO.\n\n\t* disallow inherited columns with the same name as new columns\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",
"msg_date": "Fri, 13 Mar 1998 11:14:14 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] using composite types"
},
{
"msg_contents": "On Fri, 13 Mar 1998, Thomas G. Lockhart wrote:\n\n> > How can I use composite types?\n> > If I create two tables like this:\n> > create table A ( a int, b int);\n> > create table D ( a A, b int);\n> > I would expect to be able to\n> > insert into D (a.a, a.b, b) values (1,2,3);\n> > and\n> > select a.a, a.b, b from D;\n> > That doesn't work, at least not with postgresql 6.3. Is that the way\n> > it's supposed to work? What is the proper way to do it?\n\nI've just tried it here, and it seems that the value to insert into D is\nthe OID of a row in table A.\n\n> I'm not sure if this is supported. The old tutorial does not have an\n> example exactly like this (at least that I could find), but it seems\n> like it _should_ work, and the backend accepts the syntax.\n> \n> What you want to do can probably be accomplished with inheritance, but\n> that only seems to work if you have unique names for more of your\n> fields:\n> \n> create table A ( a int, b int);\n> create table D ( c int) inherits (A);\n> \n> But, if you try\n> \n> create table D ( b int) inherits (A);\n> \n> then the new column in D gets lost without warning!\n\nThis would give a table based on the parent table, but I think he was\ntrying to use table D to refer to values in table A _as_ a type.\n\n[I may be wrong here, but that's what it looks like]\n\n-- \nPeter T Mount [email protected] or [email protected]\nMain Homepage: http://www.demon.co.uk/finder\nWork Homepage: http://www.maidstone.gov.uk Work EMail: [email protected]\n\n",
"msg_date": "Sun, 15 Mar 1998 13:16:52 +0000 (GMT)",
"msg_from": "Peter T Mount <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] using composite types"
},
{
"msg_contents": "\n> On Fri, 13 Mar 1998, Thomas G. Lockhart wrote:\n> \n> > > How can I use composite types?\n> > > If I create two tables like this:\n> > > create table A ( a int, b int);\n> > > create table D ( a A, b int);\n> > > I would expect to be able to\n> > > insert into D (a.a, a.b, b) values (1,2,3);\n> > > and\n> > > select a.a, a.b, b from D;\n> > > That doesn't work, at least not with postgresql 6.3. Is that the way\n> > > it's supposed to work? What is the proper way to do it?\n> \n> I've just tried it here, and it seems that the value to insert into D is\n> the OID of a row in table A.\n> \n> > I'm not sure if this is supported. The old tutorial does not have an\n> > example exactly like this (at least that I could find), but it seems\n> > like it _should_ work, and the backend accepts the syntax.\n> > \n> > What you want to do can probably be accomplished with inheritance, but\n> > that only seems to work if you have unique names for more of your\n> > fields:\n> > \n> > create table A ( a int, b int);\n> > create table D ( c int) inherits (A);\n> > \n> But, if you try\n> > \n> > create table D ( b int) inherits (A);\n> > \n> > then the new column in D gets lost without warning!\n\n> This would give a table based on the parent table, but I think he was\n> trying to use table D to refer to values in table A _as_ a type.\n\nYes, I am trying to use A as a type. I was hoping there was some syntax\nsugar that I was missing; the C-like struct traversal syntax would have\nbeen nice, but oh well. I see how an oid could work, but the idea was to\nmake my life as a programmer easier. I'll just prepend the A_ to a and b\nin A and inherit, so I can \"select A_a, A_b, b from D\". I got lucky and\nall my prefixed names fit in under the 32 character name limit.\n\nThanks,\nBrian\n",
"msg_date": "Sun, 15 Mar 1998 16:50:15 -0700",
"msg_from": "Brian Grossman <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] using composite types "
}
] |
[
{
"msg_contents": "How do you add entries to existing groups???\nIt seems that inserts replaces the grolist completly.\n\nI'm using perl to extract the whole grolist, add to it as a perl list\nthen replace it in the db using an insert command. Is there a better\nway, if so what's the systax.\n\nMy intention is control student access using group permissions - is this\nresonable??\n\nAre there limitations on the size of the grolist??\n\n\n\tThanks\n\n\t\tArt Van Hook\n\t\[email protected]\n\t\[email protected]\n",
"msg_date": "Fri, 13 Mar 1998 09:38:21 -0600",
"msg_from": "\"A. R. Van Hook\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "group control"
}
] |
[
{
"msg_contents": "> >I'll have to guess:\n> >You are running glibc2 on Linux, and it is earlier than full 2.0.7.\n> >The solution, if you are running Debian or RH, is to upgrade your\n> >library. The last time I looked a RH rpm was available at\n> >//ftp.redhat.com/home/gafton/glibc/ (or something like that).\n> >There is also a chance that your binary needs to be built without one\n> >of the library optimizations which inlines math functions, but I'm\n> >not certain about that.\n> >If you are running something else, can't help you :)\n> I'm running RH 5.0 postgres 6.3, but the same with 6.2.\n> \n> I've installed glibc 2.0.7 as you suggested, i reconfigure and\n> recompile, but the results where the same i insert 4 hour and the\n> record display @ 3 hours 59 minutes 60 seconds.\n> \n> You suggested recompilation without optimization of the inline math\n> functions, but where i have to make changes to do that?\n\nWell, I have to admit that my first tries on my plain-but-updated RH5.0\nbox did not succeed either, but when I tried again with a new binary\nfrom the RH site things magically started working. You say you\n\"reconfigured and recompiled\"; so you are not using the RH rpm? I can't\nremember which compiler flags might have been set or cleared, and I\ncan't find the message which I am vaguely recalling contained that info.\n\nCan someone else remember??\n\n - Tom\n",
"msg_date": "Fri, 13 Mar 1998 15:51:41 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] timespan"
}
] |
[
{
"msg_contents": "> Andreas wrote:\n> >\n> > > Returning from the function\n> > >\n> > > RETURN <expr>;\n> > >\n> > > The function terminates and the value of <expr> will\n> be\n> > > returned to the upper executor. The return value of\n> a\n> > > function cannot be undefined. If control reaches the\n> end\n> > > of the toplevel block of the function without hitting\n> a\n> > > RETURN statement, a runtime error will occur.\n> >\n> > 1. I think we should allow functions/procedures, that don't have a\n> return value.\n> > (think about a proc that only does an update, why return anything\n> except an error)\n> > 2. I think the RETURN will need a WITH RESUME, so a procedure can\n> return more than one row.\n> >\n> > create dba procedure \"dns\".getusers()\n> > returning varchar(16), smallint;\n> > define gotuser varchar(16);\n> > define gotadmin smallint;\n> > execute procedure checkadmin();\n> > foreach select uname, admin into gotuser, gotadmin from passwd\n> > return gotuser, gotadmin with resume;\n> > end foreach;\n> > end procedure;\n> \n> Yup. Looks nice. But the executor neither supports sets of\n> tuples nor single tuple to be returned from functions (except\n> for 'sql' language functions). I hacked around on that with\n> the PL/Tcl and even if I return a heap tuple or a tuple table\n> slot to the upper executor, it doesn't work :-(\n> \n> \n> Jan\n> \nBut what about Triggers?!.\nDEJ\n> --\n> \n> #=====================================================================\n> =#\n> # It's easier to get forgiveness for being wrong than for being right.\n> #\n> # Let's break this rule - forgive me.\n> #\n> #======================================== [email protected] (Jan Wieck)\n> #\n> \n> \n",
"msg_date": "Fri, 13 Mar 1998 10:29:22 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] PL/pgSQL - for discussion (RETURN)"
},
{
"msg_contents": "> But what about Triggers?!.\n> DEJ\n\n That's a completely different call interface. They return\n a HeapTuple that can be a modified one or just NULL to skip\n the whole action. But first I run for functions, then for\n triggers.\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": "Fri, 13 Mar 1998 17:55:26 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion (RETURN)"
}
] |
[
{
"msg_contents": "REGARDING postgre install/perl interface\n\nHello,\n\nI'm having trouble installing the perl interface to postgresql. Postgres\nseems to have installed correctly. The regression tests worked as documented\nafter the install.\n\nOne quirk with psql though is that if I don't specify an existing database to\nuse (e.g., template1) on the command line, I get the following error:\n\nConnection to database 'postgres' failed.\nFATAL 1: Database postgres does not exist in pg_database\n\nIs the 'postgres' database suppose to be created? How was it suppose to be\ncreated (i.e., did initdb not work correctly)?\n\nBack to the perl problem. The make file seems to be created appropriately\nand the make works as documented. The 'make test' fails on the second test\nwhich attempts to connect to postgres using the 'template1' database. The\nfollowing error is reported:\n\nnot ok 2: 0, 1\nconnectDB() failed: Is the postmaster running and accepting TCP/IP(with -i)\nconnections at 'localhost' on port '5432'?\n \nThe options in the pg_hba.conf file are:\nlocal all trust\nhost all 199.4.254.38 255.255.255.255 trust\n\nI'm running Perl5.00403, Solaris 2.5.1 on a Sun Ultra 10 and using PostGreSQL\nv6.3.\n\nAny help would be sincerely appreciated,\nAndy\n\n",
"msg_date": "13 Mar 1998 16:34:52 U",
"msg_from": "\"Andy Farrell\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "postgre install/perl interf"
},
{
"msg_contents": "Andy Farrell wrote:\n> \n> REGARDING postgre install/perl interface\n> \n> Hello,\n> \n> I'm having trouble installing the perl interface to postgresql. Postgres\n> seems to have installed correctly. The regression tests worked as documented\n> after the install.\n> \n> One quirk with psql though is that if I don't specify an existing database to\n> use (e.g., template1) on the command line, I get the following error:\n> \n> Connection to database 'postgres' failed.\n> FATAL 1: Database postgres does not exist in pg_database\n> \n> Is the 'postgres' database suppose to be created? How was it suppose to be\n> created (i.e., did initdb not work correctly)?\n> \n> Back to the perl problem. The make file seems to be created appropriately\n> and the make works as documented. The 'make test' fails on the second test\n> which attempts to connect to postgres using the 'template1' database. The\n> following error is reported:\n> \n> not ok 2: 0, 1\n> connectDB() failed: Is the postmaster running and accepting TCP/IP(with -i)\n> connections at 'localhost' on port '5432'?\n> \n> The options in the pg_hba.conf file are:\n> local all trust\n> host all 199.4.254.38 255.255.255.255 trust\n> \n> I'm running Perl5.00403, Solaris 2.5.1 on a Sun Ultra 10 and using PostGreSQL\n> v6.3.\n> \n> Any help would be sincerely appreciated,\n> Andy\n\n\nthe test-script relies on the fact, that you are\nusing TCP/IP sockets. Starting the postmater with\n-i should solve your problem.\n\nEdmund\n-- \nEdmund Mergl mailto:[email protected]\nIm Haldenhau 9 http://www.bawue.de/~mergl\n70565 Stuttgart fon: +49 711 747503\nGermany gsm: +49 171 2645325\n",
"msg_date": "Sat, 14 Mar 1998 07:47:07 +0100",
"msg_from": "Edmund Mergl <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] postgre install/perl interf"
}
] |
[
{
"msg_contents": "> PL/pgSQL\n> A procedural language for the PostgreSQL RDBMS\n> 1st draft\n>\n> Jan Wieck <[email protected]>\n>\n\nI see the EXCEPTION handling feature is missing from the draft. It is very\nuseful feature for writing robust application. How about PROCEDURE and FUNTION?\n\nDavid Phan <[email protected]>\nSystems/Internet Programmer\nComSource, Inc. http://www.comsource.net/\n____________________________________________\n\n",
"msg_date": "Fri, 13 Mar 1998 10:54:36 -0600",
"msg_from": "David Phan <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] PL/pgSQL - for discussion"
}
] |
[
{
"msg_contents": "****************************************************************\nAny idea when postgresql v6.3 will work on Alpha Digital Unix?\n****************************************************************\n-David\n\n",
"msg_date": "Fri, 13 Mar 1998 11:07:40 -0600",
"msg_from": "David Phan <[email protected]>",
"msg_from_op": true,
"msg_subject": "Postgresql v6.3 for Alpha Digital Unix"
},
{
"msg_contents": "> \n> ****************************************************************\n> Any idea when postgresql v6.3 will work on Alpha Digital Unix?\n> ****************************************************************\n> -David\n> \n\nSomeone is testing it now to find out on what day it broke, so we can\nfix it.\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",
"msg_date": "Fri, 13 Mar 1998 14:05:12 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgresql v6.3 for Alpha Digital Unix"
},
{
"msg_contents": "\nShould I be the one doing this? I know someone else volunteered, but\nI don't know if that has started happening -- I'm on the net all day,\nand I have a fast alpha, so it might be easier for me.. Just let me\nknow! I know not a thing about CVSup though..\n\nOn Fri, 13 March 1998, at 14:05:12, Bruce Momjian wrote:\n\n> > \n> > ****************************************************************\n> > Any idea when postgresql v6.3 will work on Alpha Digital Unix?\n> > ****************************************************************\n> > -David\n> > \n> \n> Someone is testing it now to find out on what day it broke, so we can\n> fix it.\n> \n> -- \n> Bruce Momjian | 830 Blythe Avenue\n> [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",
"msg_date": "Fri, 13 Mar 1998 15:31:12 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgresql v6.3 for Alpha Digital Unix"
},
{
"msg_contents": "> \n> \n> Should I be the one doing this? I know someone else volunteered, but\n> I don't know if that has started happening -- I'm on the net all day,\n> and I have a fast alpha, so it might be easier for me.. Just let me\n> know! I know not a thing about CVSup though..\n> \n\nHere are some postings. cvsup is on the ftp site for various platforms.\n\n---------------------------------------------------------------------------\n",
"msg_date": "Fri, 13 Mar 1998 18:57:25 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Postgresql v6.3 for Alpha Digital Unix"
}
] |
[
{
"msg_contents": "On Fri, 13 Mar 1998, Zeugswetter Andreas wrote:\n\n> I think we should depreciate the BEGIN/END keywords in SQL to allow them\n> to be used for the new PL/SQL. So definitely leave them out of ecpg now.\n> Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of commit work; begin work) \n\nApologies for intrusion.\n\nI think we don't need BEGIN/END at all, these statements aren't SQL standard.\nEND is an alias for COMMIT.\n(why do we need two statements to do the same thing?).\n\nfrom man commit:\n \"...\n This commands commits the current transaction. All\n changes made by the transaction become visible to others\n and are guaranteed to be durable if a crash occurs.\n COMMIT is functionally equivalent to the END command\"\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n\nfrom man begin:\n \"...\n commands commits the current transaction. All\n changes made by the transaction become visible to others\n and are guaranteed to be durable if a crash occurs.\"\n\nand BEGIN should be changed to SQL standard SET TRANSACTION statement.\n-------\nPS:\n I think PL/pgSQL is an eccellent idea. Go for it.\n Ciao, Jose'\n\n\n",
"msg_date": "Fri, 13 Mar 1998 17:28:31 +0000 (GMT)",
"msg_from": "[email protected]",
"msg_from_op": true,
"msg_subject": "Re: AW: [HACKERS] Begin statement again"
}
] |
[
{
"msg_contents": "> > \n> > I've noticed recently that after making certain changes to the\n> > database (adding indices, attributes) they aren't immediately known\n> > about. for example, I add an attribute, then try to update that\n> > attribute it tells me it doesn't exist. if I exist psql and start a\n> > new connection all is well. but it doesn't happen consistently..\n> > \n> > \n> \n> Looks like problems in the syscache/catcache invalidation\n> routines.\n> \n\nThis existed in an older version, but I could almost swear that someone\nfixed it.\n\nBruce, do you have old versions of the TODO lists around to check on\nthis one? I specifically remember the part about adding an attribute\nand it not being there until the next session.\n\ndarrenk\n",
"msg_date": "Fri, 13 Mar 1998 12:39:24 -0500",
"msg_from": "[email protected] (Darren King)",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] suspected problem with cache updates"
},
{
"msg_contents": "\nDarren wrote:\n>\n> > >\n> > > I've noticed recently that after making certain changes to the\n> > > database (adding indices, attributes) they aren't immediately known\n> > > about. for example, I add an attribute, then try to update that\n> > > attribute it tells me it doesn't exist. if I exist psql and start a\n> > > new connection all is well. but it doesn't happen consistently..\n> > >\n> > >\n> >\n> > Looks like problems in the syscache/catcache invalidation\n> > routines.\n> >\n>\n> This existed in an older version, but I could almost swear that someone\n> fixed it.\n\n I fixed something there lately when it came to pg_shadow,\n where the required update on pg_class during revoke/grant\n just outdated the syscache tuple for pg_shadow, but the\n permission checking code needed that.\n\n But this problem looks different. This time the pg_class\n tuple for a user table is updated where that table had to be\n flushed from the relcache and reopened. Cannot reproduce the\n error here. If I alter table add field, they show up\n immediately in the same session.\n\n>\n> Bruce, do you have old versions of the TODO lists around to check on\n> this one? I specifically remember the part about adding an attribute\n> and it not being there until the next session.\n>\n> darrenk\n>\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": "Fri, 13 Mar 1998 19:43:48 +0100 (MET)",
"msg_from": "[email protected] (Jan Wieck)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] suspected problem with cache updates"
},
{
"msg_contents": "> \n> This existed in an older version, but I could almost swear that someone\n> fixed it.\n> \n> Bruce, do you have old versions of the TODO lists around to check on\n> this one? I specifically remember the part about adding an attribute\n> and it not being there until the next session.\n> \n\nI think that was a very old bug. It involved ALTER TABLE ADD and the\nthing not appearing until the next session, I think. It has to do with\nthe attribute cache. The only old TODO versions I have are in the old\nreleases.\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",
"msg_date": "Fri, 13 Mar 1998 17:20:52 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] suspected problem with cache updates"
},
{
"msg_contents": "\nIt must be reappearing again. It happened under 6.3\n\nOn Fri, 13 March 1998, at 17:20:52, Bruce Momjian wrote:\n\n> I think that was a very old bug. It involved ALTER TABLE ADD and the\n> thing not appearing until the next session, I think. It has to do with\n> the attribute cache. The only old TODO versions I have are in the old\n> releases.\n> \n> -- \n> Bruce Momjian | 830 Blythe Avenue\n> [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",
"msg_date": "Fri, 13 Mar 1998 14:56:49 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] suspected problem with cache updates"
},
{
"msg_contents": "\nIt certainly didn't happen very consistently, in fact, only once.\n\nOn Fri, 13 March 1998, at 19:43:48, Jan Wieck wrote:\n\n> But this problem looks different. This time the pg_class\n> tuple for a user table is updated where that table had to be\n> flushed from the relcache and reopened. Cannot reproduce the\n> error here. If I alter table add field, they show up\n> immediately in the same session.\n> \n> >\n> > Bruce, do you have old versions of the TODO lists around to check on\n> > this one? I specifically remember the part about adding an attribute\n> > and it not being there until the next session.\n> >\n> > darrenk\n> >\n> >\n> \n> \n> Jan\n",
"msg_date": "Fri, 13 Mar 1998 15:33:30 -0800 (PST)",
"msg_from": "Brett McCormick <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] suspected problem with cache updates"
}
] |
[
{
"msg_contents": "Continuing the thread about checkpointing. While I was browsing through\nSQL standard documents I noticed that there are clauses like:\n\nSET TRANSACTION {READ UNCOMMITED|READ COMMITED|READ\nREPEATABLE|SERIALIZABLE}\n\nI was wondering... If we have non-overwriting feature of postgres, we\nwould accomplish no-lock reads for at least first two isolation levels.\n\nAlso, by adding 'checkpointed' flag to each record we would allow for:\nSET TRANSACTION READ CHECKPOINTED \n(it's out of SQL Standard, but I couldn't find the right command)\n\nDuring CHECKPOINT we would mark all the current records with\n'checkpointed' flag. Also checkpointing would do VACUUM, so it would be\nguaranteed that each checkpointed record would be the first in it's\nmodification chain.\n\nThen for READ CHECKPOINTED transaction mode we would accomplish no-lock\nreading which is especially usefull when you have to do a very long\nstatistical query on your data being constantly updated. Also, it would\nbe guaranteed that checkpointed data are consistent.\n\nIs this hard to accomplish?\n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n",
"msg_date": "Fri, 13 Mar 1998 20:20:44 +0100",
"msg_from": "Michal Mosiewicz <[email protected]>",
"msg_from_op": true,
"msg_subject": "SET TRANSACTION * proposal"
},
{
"msg_contents": "Michal Mosiewicz wrote:\n> \n> Continuing the thread about checkpointing. While I was browsing through\n> SQL standard documents I noticed that there are clauses like:\n> \n> SET TRANSACTION {READ UNCOMMITED|READ COMMITED|READ\n> REPEATABLE|SERIALIZABLE}\n> \n> I was wondering... If we have non-overwriting feature of postgres, we\n> would accomplish no-lock reads for at least first two isolation levels.\n> \n> Also, by adding 'checkpointed' flag to each record we would allow for:\n> SET TRANSACTION READ CHECKPOINTED \n> (it's out of SQL Standard, but I couldn't find the right command)\n> \n> During CHECKPOINT we would mark all the current records with\n> 'checkpointed' flag. Also checkpointing would do VACUUM, so it would be\n> guaranteed that each checkpointed record would be the first in it's\n> modification chain.\n> \n> Then for READ CHECKPOINTED transaction mode we would accomplish no-lock\n> reading which is especially usefull when you have to do a very long\n> statistical query on your data being constantly updated. Also, it would\n> be guaranteed that checkpointed data are consistent.\n\nFor that matter, if we could work this into the vacuum code, that\nwould be sweet. We could update the statistics without locking out\nwriters.\n\nOcie\n",
"msg_date": "Fri, 13 Mar 1998 11:47:28 -0800 (PST)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SET TRANSACTION * proposal"
},
{
"msg_contents": "> SET TRANSACTION {READ UNCOMMITED|READ COMMITED|READ\n> REPEATABLE|SERIALIZABLE}\n<snip> \n> Is this hard to accomplish?\n\nThis is one of Vadim's favorite projects. We diverted him for v6.3 to\nimplement subselects, but I think he plans to pick it up again soon (for\nv6.4 or v6.5; he has other things on his list too :).\n\n - Tom\n",
"msg_date": "Sat, 14 Mar 1998 02:32:26 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] SET TRANSACTION * proposal"
}
] |
[
{
"msg_contents": "test=> create sequence mytest MINVALUE -234234 CYCLE;\nERROR: parser: parse error at or near \"-\"\n\nIs this the current desired behavior? If so the doc's need to be\nchanged.\nRunning\n\tRH5.0\n\tpg6.3 - rpm from 1998-03-01\n\tgcc 2.7.2.3\n\n\t\t-DEJ\n",
"msg_date": "Fri, 13 Mar 1998 16:26:23 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Possible Bug."
},
{
"msg_contents": "> test=> create sequence mytest MINVALUE -234234 CYCLE;\n> ERROR: parser: parse error at or near \"-\"\n> \n> Is this the current desired behavior?\n\nWell, it is the current, but not desired, behavior. I looked at it but\ndon't yet understand why it is happening. Seems like the parser\nshouldn't be breaking the number into two pieces.\n\n - Tom\n",
"msg_date": "Sat, 14 Mar 1998 03:53:24 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Possible Bug."
}
] |
[
{
"msg_contents": "Hi !\n\nI've been playing PostgreSQL today, and there are still some places\nwhere you can find Postgres95 mentionned. One place I remember is in\nthe pg_hba.conf copied from src/backend/lib/pg_hba.conf.sample (I\nthink - I say this from memory).\n\nI know it's quite unimportant (hence the subject title), but some\nnewcomers might get startled, since it's one of the first files one\nhas to deal with when installing the software.\n\nAnyway, it's a very very very (ad libitum...) nice release :) Thank\nyou all...\n\nOff-topic for ``hackers'' : I saw that the tutorial and some other\nplaces did not present the current laying of psql for the examples. If\nnobody does this in the next few days, I'll volonteer on updating\nthat, and also try to provide corrections in the docs where there are\ninaccurate leftovers from the pg95 days... I wondered also if it would\nbe interesting to collect (most) questions & answers from the\nquestions mailing lists to do a (more) exhaustive faq ?? I'll try to\ndo that also (gathering from the archives and updating each day after\nthat), if it interests someone (Tom ?). The problem is I won't be able\nto do all this before April, since I won't have any computer from\nMarch 16th to March 31st. :(( [I'll still try to follow the mailing\nlists during lunch time at work ;) ]\n\nPatrice\n\n--\nPatrice HÉDÉ --------------------------------- [email protected] -----\n... Ásólfr hljóp upp á skip Hrúts ok varð fjögurra manna bani, áðr\nHrútr varð varr við. Sneri hann þá í móti honum. En er þeir fundust,\nlagði Ásólfr í skjöld Hrúts ok í gegnum, en Hrútr hjó til Ásólfs, ok\nvarð þat banahögg. --- Njáls Saga\n\n",
"msg_date": "Fri, 13 Mar 1998 21:17:47 -0500 (EST)",
"msg_from": "Patrice HEDE <[email protected]>",
"msg_from_op": true,
"msg_subject": "Minor detail..."
},
{
"msg_contents": "> \n> Hi !\n> \n> I've been playing PostgreSQL today, and there are still some places\n> where you can find Postgres95 mentionned. One place I remember is in\n> the pg_hba.conf copied from src/backend/lib/pg_hba.conf.sample (I\n> think - I say this from memory).\n\nFixed.\n\n> \n> I know it's quite unimportant (hence the subject title), but some\n> newcomers might get startled, since it's one of the first files one\n> has to deal with when installing the software.\n\nYep.\n\n> that, and also try to provide corrections in the docs where there are\n> inaccurate leftovers from the pg95 days... I wondered also if it would\n> be interesting to collect (most) questions & answers from the\n> questions mailing lists to do a (more) exhaustive faq ?? I'll try to\n> do that also (gathering from the archives and updating each day after\n> that), if it interests someone (Tom ?). The problem is I won't be able\n> to do all this before April, since I won't have any computer from\n> March 16th to March 31st. :(( [I'll still try to follow the mailing\n> lists during lunch time at work ;) ]\n\nSure. You can have the FAQ if you wish to maintain it.\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",
"msg_date": "Fri, 13 Mar 1998 23:16:08 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Minor detail..."
},
{
"msg_contents": "On Fri, 13 Mar 1998, Bruce Momjian wrote:\n\n> > that, and also try to provide corrections in the docs where there are\n> > inaccurate leftovers from the pg95 days... I wondered also if it would\n> > be interesting to collect (most) questions & answers from the\n> > questions mailing lists to do a (more) exhaustive faq ?? I'll try to\n> > do that also (gathering from the archives and updating each day after\n> > that), if it interests someone (Tom ?). The problem is I won't be able\n> > to do all this before April, since I won't have any computer from\n> > March 16th to March 31st. :(( [I'll still try to follow the mailing\n> > lists during lunch time at work ;) ]\n> \n> Sure. You can have the FAQ if you wish to maintain it.\n\nOk, I can take it if noone else objects, but I'll confirm it to you\nwhen I am ready (in two or three weeks...).\n\nI tried this morning to get the questions-digests from the ftp site as\na single tarball using :\n\n get <<directory-name>>.tar.gz\n\nbut the ftp-server complained that it couldn't find ``tar'' nor\n``gzip''. Can it be fixed ? Or are the digests included on the CD ??\n\nAnyway, I guess I still can do an mget * :)\n\nPatrice\n\n--\nPatrice HÉDÉ --------------------------------- [email protected] -----\nNous sommes au monde. [...] La croyance en un esprit absolu ou en un\nmonde en soi détaché de nous n'est qu'une rationalisation de cette\nfoi primordiale. --- Merleau-Ponty, Phénoménologie de la Perception\n\n",
"msg_date": "Sat, 14 Mar 1998 15:14:22 -0500 (EST)",
"msg_from": "Patrice HEDE <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Minor detail..."
},
{
"msg_contents": "On Sat, 14 Mar 1998, Patrice HEDE wrote:\n\n> Ok, I can take it if noone else objects, but I'll confirm it to you\n> when I am ready (in two or three weeks...).\n> \n> I tried this morning to get the questions-digests from the ftp site as\n> a single tarball using :\n> \n> get <<directory-name>>.tar.gz\n> \n> but the ftp-server complained that it couldn't find ``tar'' nor\n> ``gzip''. Can it be fixed ? Or are the digests included on the CD ??\n\n\tThe digests *and* the complete mhonarc archives are included on\nthe CD...\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Sat, 14 Mar 1998 23:19:02 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Minor detail..."
}
] |
[
{
"msg_contents": "Added to TODO:\n\n\t* recover or force failure when disk space is exhausted\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",
"msg_date": "Sat, 14 Mar 1998 00:53:35 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "disk space"
}
] |
[
{
"msg_contents": "> I have not had a chance yet to browse the postgres code, some silliness about\n> my day job keeps interfering, so it may be a few weeks before I get to this.\n> \n> My thought is that comitting transactions can't release locks until the log\n> is written. So, if the log is not flushed except every so often, locks are\n> held longer harming concurrancy.\n> \n> Of course, I am probably thinking about \"normal\" WAL type transactions not\n> the \"unique\" postgres log thing which truth to tell I may never have\n> really understood.\n\nYou have good point here. We are telling an application that the\ntransaction is committed, even though if there is a crash in the next 10\nseconds, on reboot, it will show it as not committed.\n\nIt is my understanding that all buffered logging database systems have\nthis problem, so I think we will just have to live with it. I believe\nwe will still continue to offer the original fsync() after commit\nbehavior.\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",
"msg_date": "Sat, 14 Mar 1998 12:15:46 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] Does Storage Manager support >2GB\n\ttables?"
}
] |
[
{
"msg_contents": "I wholeheartly agree. AFAIK it does nothing else, but ORACLE likes it.\nAnd most users have experience in ORACLE.\n\nMichael\n--\nDr. Michael Meskes, Projekt-Manager | topystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Use Debian GNU/Linux! | Tel: (+49) 2405/4670-44\n\n> ----------\n> From: \tThomas G. Lockhart[SMTP:[email protected]]\n> Sent: \tFreitag, 13. März 1998 16:04\n> To: \tMichael Meskes\n> Cc: \tPostgreSQL Hacker\n> Subject: \tRe: [HACKERS] Keyword\n> \n> > I'm currently trying to get ecpg more in line with the backends\n> > parser. Doing so I found ecpg accepts a syntax the backend doesn't.\n> > It's 'commit work release' which is accepted for compatibility with\n> > Oracle. How about allowing this in the backend, too?\n> \n> I don't want to sound too fussy here, but does this statement provide\n> any functionality missing from the SQL92 syntax 'commit [work]'? If\n> not,\n> then perhaps Oracle users could start writing portable code if they\n> cared to?\n> \n> - Tom\n> \n",
"msg_date": "Sat, 14 Mar 1998 19:20:39 +0100",
"msg_from": "\"Meskes, Michael\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Keyword"
}
] |
[
{
"msg_contents": "We, that is ecpg, will need BEGIN and END to include blocks of Pl/pgSQL.\nI think this can be modelled after ORACLE's PL/SQL. in your embedded SQL\ncode you say:\n\n\t...\n\texec sql begin;\n\t\t<some PL/SQL code, for instance the call of a stored\nprocedure>\n\texec sql end;\n\t...\n\nAnd I'd really like to call a stored procedure from my C program.\n\nMichael\n--\nDr. Michael Meskes, Projekt-Manager | topystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Use Debian GNU/Linux! | Tel: (+49) 2405/4670-44\n\n> ----------\n> From: \[email protected][SMTP:[email protected]]\n> Sent: \tFreitag, 13. März 1998 18:28\n> To: \tZeugswetter Andreas\n> Cc: \t'Michael Meskes'; '[email protected]'\n> Subject: \tRe: AW: [HACKERS] Begin statement again\n> \n> On Fri, 13 Mar 1998, Zeugswetter Andreas wrote:\n> \n> > I think we should depreciate the BEGIN/END keywords in SQL to allow\n> them\n> > to be used for the new PL/SQL. So definitely leave them out of ecpg\n> now.\n> > Only accept BEGIN WORK and BEGIN TRANSACTION. (do a sequence of\n> commit work; begin work) \n> \n> Apologies for intrusion.\n> \n> I think we don't need BEGIN/END at all, these statements aren't SQL\n> standard.\n> END is an alias for COMMIT.\n> (why do we need two statements to do the same thing?).\n> \n> from man commit:\n> \"...\n> This commands commits the current transaction. All\n> changes made by the transaction become visible to others\n> and are guaranteed to be durable if a crash occurs.\n> COMMIT is functionally equivalent to the END command\"\n> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n> \n> from man begin:\n> \"...\n> commands commits the current transaction. All\n> changes made by the transaction become visible to others\n> and are guaranteed to be durable if a crash occurs.\"\n> \n> and BEGIN should be changed to SQL standard SET TRANSACTION statement.\n> -------\n> PS:\n> I think PL/pgSQL is an eccellent idea. Go for it.\n> Ciao,\n> Jose'\n> \n> \n",
"msg_date": "Sat, 14 Mar 1998 19:20:41 +0100",
"msg_from": "\"Meskes, Michael\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: AW: [HACKERS] Begin statement again"
}
] |
[
{
"msg_contents": "It has input, input/output and output variables.\n\nMichael\n--\nDr. Michael Meskes, Projekt-Manager | topystem Systemhaus GmbH\[email protected] | Europark A2, Adenauerstr. 20\[email protected] | 52146 Wuerselen\nGo SF49ers! Use Debian GNU/Linux! | Tel: (+49) 2405/4670-44\n\n> ----------\n> From: \tHannu Krosing[SMTP:[email protected]]\n> Sent: \tFreitag, 13. März 1998 19:43\n> To: \tJan Wieck\n> Cc: \[email protected]\n> Subject: \t[HACKERS] Re: PL/PgSQL discussion\n> \n> Does'nt the Oracle PL/SQL have both input and input/output parameters\n> ?\n> \n> It's more than a year since i last used it but I think that this is\n> the case.\n> \n> Hannu\n> \n> \n",
"msg_date": "Sat, 14 Mar 1998 19:20:43 +0100",
"msg_from": "\"Meskes, Michael\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] Re: PL/PgSQL discussion"
}
] |
[
{
"msg_contents": "Hi,\n\nWhile finding out more about powers of postgresql I encountered the\nfollowing:\n\n<PSQL SESSION>\ncreate table t(f1 int, f2 t);\n</PSQL SESSION>\n\nThis gives an error message:\n\n\"type name lookup of t failed\".\n\nHowever according to the comments for the function \"BuildDescForRelation\",\nthis should succeed. It seems like some functions in parse_type.c are\ndon't consider this case as a valid one.\n\nThe following case doesn't give any errors.\n\n<PSQL SESSION>\ncreate table t1(f1 int);\ncreate table t2(f1 int, f2 t);\n</PSQL SESSION>\n\nDoes anyone have experience using these complex types in tables?\n\nThanks for any hints,\nMaurice\n\n",
"msg_date": "Sat, 14 Mar 1998 22:02:13 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Bug in create table?"
}
] |
[
{
"msg_contents": "> 2. I still suggest to implement it in a way that leaves the door open\n> to not have an oid for every table/tuple per default.\n> 3. Tables without oid would simply not have all the OO functionality.\n> tuples without the oid would not exist in the *OO world*\n> 4. I think it is valuable to have both OO and fast relational stuff.\n(ORDBMS)\n>\n\nHow about supporting CREATE TABLE and CREATE CLASS?\nCREATE table would give the features for the relational model (like no oid?)\nwhile create class would provide identity, polymorphism, inheritance as\nwould be preferred in the OO world.\n\nWhat do you think?\n\nI'm also considering trying to store tuples for derived class in the same\nfile as the tuples for the base class. This might make it easier\nto provide inheritance of triggers/indices etc.\nThis seems relatively easy to do if the class id of a tuple is stored in\nthe tuple, otherwise supporting inheritance of triggers etc seems like\nit's going to be an expensive operation which will also be fundamentally\nbroken from an OO perspective. (specifically, more than one identity\n for a single object).\n\nWhat do you think? Is it \"easy\" to do?\n\nThanks,\nMaurice\n\n\n",
"msg_date": "Sat, 14 Mar 1998 22:25:57 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "OO and relation models (was: newoid in invapi,c)"
}
] |
[
{
"msg_contents": "-----Original Message-----\nFrom: Peter T Mount <[email protected]>\nTo: Jan Wieck <[email protected]>\nCc: Bruce Momjian <[email protected]>;\[email protected] <[email protected]>;\[email protected] <[email protected]>\nDate: zondag 15 maart 1998 21:10\nSubject: Re: [QUESTIONS] Re: [HACKERS] text should be a blob field\n\n\n>On Wed, 4 Mar 1998, Jan Wieck wrote:\n>\n>> Bruce wrote:\n>>\n>> > > 1. Is there a call made by the backend to each datatype when a row is\n>> > > deleted? I can't see one.\n>> >\n>> > Well, you could have a RULE that deletes the large object at row\n>> > deletion time. However, if two rows point to the same large object,\nthe\n>> > first one deleting it would delete the large object for the other. The\n>> > only solution to this is to have a separate large object table, and use\n>> > reference counts so only the last user of the object deletes it.\n>>\n>> I think triggers are more appropriate.\n>>\n>> On INSERT check that the large object referenced exists.\n>>\n>> On UPDATE if large object reference changes, check that new\n>> large object exists and check if old large object isn't\n>> referenced any more in which case drop the old large object.\n>>\n>> On DELETE check if large object isn't referenced any more ...\n>>\n>> Yes - I like triggers :-)\n>\n>I'm begining to agree with you here.\n>\n>So far, I've got the trigger to work, so if a row of a table is deleted,\n>or an oid referencing a BLOB is updated, then the old BLOB is deleted.\n>This removes the orphaned BLOB problem.\n>\n>The only problem I have now, is:\n>\n> How to get a trigger to be automatically created on a table when the\n> table is created. This would be required, so the end user doesn't have\n> to do this (normally from within an application).\n>\n>This would be required, esp. for expanding the text type (or memo, or\n>whatever).\n>\n\n\nSince triggers are not inherited this doesn't seem appropiate too me -:(.\nWon't user have to do some magic on inherited tables?\n\nI think many things will be fixed when triggers, indices etc. are inherited\nproperly by derived classes.\n\nWith regards from,\nMaurice.\n\n\n",
"msg_date": "Sun, 15 Mar 1998 15:29:06 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] Re: [HACKERS] text should be a blob field"
},
{
"msg_contents": "> Since triggers are not inherited this doesn't seem appropiate to me\n> -:(. Won't user have to do some magic on inherited tables?\n> \n> I think many things will be fixed when triggers, indices etc. are\n> inherited properly by derived classes.\n\nThis is an interesting point. We have perhaps been neglecting to discuss\nhow inheritance should behave with new features. Certainly at least some\nthings (like default values and constraints) do inherit as you would\nexpect. It's not clear to me that inheriting indices is appropriate,\nsince indices are basically just an artifact for performance\nimprovement; the results of a query do not change with the presence of\nan index. Also, if a table is inherited, then the access pattern of the\nresulting table might be different, requiring a different index anyway.\n\nI haven't worked with triggers; what is their current behavior with\ninheritance? Maurice implies that they are not inherited at all. If not,\nshould they be?\n\n - Tom\n",
"msg_date": "Sun, 15 Mar 1998 15:19:48 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] Re: [HACKERS] text should be a blob field"
}
] |
[
{
"msg_contents": "-----Original Message-----\nFrom: Thomas G. Lockhart <[email protected]>\nTo: PostgreSQL-development <[email protected]>\nCc: Maurice Gittens <[email protected]>\nDate: zondag 15 maart 1998 22:24\nSubject: Re: [QUESTIONS] Re: [HACKERS] text should be a blob field\n\n\n>> Since triggers are not inherited this doesn't seem appropiate to me\n>> -:(. Won't user have to do some magic on inherited tables?\n>>\n>> I think many things will be fixed when triggers, indices etc. are\n>> inherited properly by derived classes.\n>\n>This is an interesting point. We have perhaps been neglecting to discuss\n>how inheritance should behave with new features. Certainly at least some\n>things (like default values and constraints) do inherit as you would\n>expect. It's not clear to me that inheriting indices is appropriate,\n>since indices are basically just an artifact for performance\n>improvement;\nYes I agree.\n> the results of a query do not change with the presence of\n>an index. Also, if a table is inherited, then the access pattern of the\n>resulting table might be different, requiring a different index anyway.\n\nIn general OO terms code reuse (in our case trigger reuse? -:) ) is achieved\nby inheriting as much as sanely possible.\nSince (from an OO perspective) every instance of a derived class is also an\ninstance of it's base class(es), inheriting indices seems like a natural\nthing\nto me.\n\nHowever it may be possible that an OO database makes different choices\nthan an OO programming language.\n\n>\n>I haven't worked with triggers; what is their current behavior with\n>inheritance? Maurice implies that they are not inherited at all. If not,\n>should they be?\n\nWell if we want to support polymorphism the answer to this question is a\ndefinite yes.\n\nIn general any operation defined on an instance of a base class should be\ndefined on an instance of a derived class. This is the main reason\nwe inherit in the first place. The user has the option of redefining\n(specializing)\nan operation for instances of derived classes if she/he so pleases.\n\nI think proper support for this would be really cool.\n\n\nWith regards from Maurice.\n\n\n",
"msg_date": "Sun, 15 Mar 1998 18:07:12 +0100",
"msg_from": "\"Maurice Gittens\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] Re: [HACKERS] text should be a blob field"
}
] |
[
{
"msg_contents": "\nIf I create a table with a datetime field with a default of 'now', every insert\nthe value is the time of table creation instead of the time of insert, which is how it behaved\nin previous releases (I think this was even documented).\n",
"msg_date": "Sun, 15 Mar 1998 16:46:54 -0800",
"msg_from": "Brett McCormickS <[email protected]>",
"msg_from_op": true,
"msg_subject": "datetime default 'now' broken?"
},
{
"msg_contents": "> If I create a table with a datetime field with a default of 'now', \n> every insert the value is the time of table creation instead of the \n> time of insert, which is how it behaved in previous releases (I think \n> this was even documented).\n\nI can't recall it ever working that way, though before we discovered\nthat it didn't we all assumed that it _did_ work that way :)\n\nThe workaround is to define it as\n\n ... default datetime('now'::text)\n\nwhich forces the string to be evaluated at runtime. The SQL symbol\nCURRENT_TIMESTAMP also misbehaves in \"default\" clauses, and I'm\nconsidering changing it a bit to get around the problem.\n\n - Tom\n",
"msg_date": "Mon, 16 Mar 1998 05:59:53 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] datetime default 'now' broken?"
},
{
"msg_contents": "At 9:59 PM -0800 3/15/98, Thomas G. Lockhart wrote:\n>> If I create a table with a datetime field with a default of 'now',\n>> every insert the value is the time of table creation instead of the\n>> time of insert, which is how it behaved in previous releases (I think\n>> this was even documented).\n>\n>I can't recall it ever working that way, though before we discovered\n>that it didn't we all assumed that it _did_ work that way :)\n>\n\nI'm running 6.1.1 and I *depend* on 'now' giving me the real now in an\ninsert. I guess I'm glad I never upgraded.\n\nSignature failed Preliminary Design Review.\nFeasibility of a new signature is currently being evaluated.\[email protected], or [email protected]\n\n\n",
"msg_date": "Mon, 16 Mar 1998 12:48:07 -0800",
"msg_from": "\"Henry B. Hotz\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] datetime default 'now' broken?"
},
{
"msg_contents": "> >> If I create a table with a datetime field with a default of 'now',\n> >> every insert the value is the time of table creation instead of the\n> >> time of insert, which is how it behaved in previous releases (I \n> >> think this was even documented).\n> >\n> >I can't recall it ever working that way, though before we discovered\n> >that it didn't we all assumed that it _did_ work that way :)\n> \n> I'm running 6.1.1 and I *depend* on 'now' giving me the real now in an\n> insert. I guess I'm glad I never upgraded.\n\nHi Henry. You're missing a lot by not upgrading. DEFAULT clauses on\ntables which Brett was asking about aren't even available pre-v6.2. The\n'now' behavior in default clauses was an unexpected side-effect of the\noptimizer, which evaluates things which look like constants before\nexecution.\n\nGive me a call during the day at x47797 if you have any other\nreservations about upgrading...\n\n - Tom\n",
"msg_date": "Tue, 17 Mar 1998 02:35:32 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] datetime default 'now' broken?"
}
] |
[
{
"msg_contents": "Nice to see the MySQL crashme page has updated to its PostgreSQL title\nto PostgreSQL 6.3, but it seems nonee of the items haven't been updated\nfor this release, http://www.tcx.se/crash-me.html.\n\nI see no mention of the new subselects, and it says we don't support\ncomments as --. And I see an item called \"atomic updates\" which I think\nwe decided was a very bizarre test, but still no mention of\ntransactions, which is a very basic database capability. We also have\nlarge objects. Is that the same as blobs? We have \"Function\nconcatenation with ||\". We have \"Function CURRENT_DATE\". We have\n\"Function LEFT\". We have \"Function IN on strings in SELECT\". We have\n\"Function = ALL\". We have \"Update with sub select\". We have \"case\ninsensitive compare\". We have \"lock table\".\n\nMany of these we have had for over a year.\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",
"msg_date": "Sun, 15 Mar 1998 22:31:29 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On 15 Mar 98 at 22:31, Bruce Momjian wrote:\n\n> Nice to see the MySQL crashme page has updated to its PostgreSQL title\n> to PostgreSQL 6.3, but it seems nonee of the items haven't been updated\n> for this release, http://www.tcx.se/crash-me.html.\n> \n> I see no mention of the new subselects, and it says we don't support\n> comments as --. And I see an item called \"atomic updates\" which I think\n> we decided was a very bizarre test, but still no mention of\n> transactions, which is a very basic database capability. We also have\n> large objects. Is that the same as blobs? We have \"Function\n> concatenation with ||\". We have \"Function CURRENT_DATE\". We have\n> \"Function LEFT\". We have \"Function IN on strings in SELECT\". We have\n> \"Function = ALL\". We have \"Update with sub select\". We have \"case\n> insensitive compare\". We have \"lock table\".\n> \n\nI have looked into crash-me and screened the items you mentioned. \nHere is the result:\n\nabout the large objects. I looked in the manual of postgres and \ncouldn't find that fast anything about large objects.\nblobs is a type where you can put binary or other big things in.\n\nfirst the concatenation with ||\nthe test is select 'abc' || 'def'\nand that's not working in postgres. In other db's it's working.\n\nfunction left:\nthe test is select left('abcd',2)\nlooked in the manual of postgres and couldn't find the function left.\nthis isn't working and I tried some other ways but couldn't find any\nworking query.\n\nfunction current_date:\nthe test is select current_date()\nthat's postgres not supporting only select current_date (without ()).\nthat's also the case with current_time, current_timestamp. I will make\npatch so it's tested both ways.\n\nFunction IN on strings in SELECT:\nthe test is: select 'monty' in ('david','monty','allan')\nand that's not working. it gives: using an explicit cast\n\nFunction = ALL:\nquery: select a from crash_me where b =all (select b from crash_me) \nyes \nso it's supported\n\nUpdate with sub select:\nquery: create table crash_q (a integer,b char(10)) \nquery: insert into crash_q values(1,'c') \nquery: update crash_q set crash_q.b= (select b from crash_me\nwhere crash_q.a = crash_me.a) \nGot error from query: 'update crash_q\nset crash_q.b= (select b from crash_me whe re crash_q.a = crash_me.a)'\nERROR: parser: parse error at or near \"=\" query: drop table crash_q \nno \nI think this test shows that it isn't possible todo an update with a \nsub select\n\ncase insensitive compare:\nquery: select b from crash_me where b = 'A' \nquery: select b from crash_me where b = 'A' didn't return any\nresult \nno\n(field b in table crash_me has a value of 'a'.)\n\nthe transaction are still not inserted in the crash-me test because \nmonty and I didn't have time to insert it. But what I mentioned \nbefore if you would like to see some other tests in the crash-me test \nyou can add them by your self and send us the patch. We will add it \nto the release of crash-me.\n\nI am not sure the results of postgres in the crash-me page is from \nthe final release of postgres. Because I have so much problems with \npostgres it's at this moment hard to test. Further it's taking a very \nlot of time to test the postgres db because it's eating my memory. \nWhen I test the limit of the number of joins I can do it's eating my \nmemory till there is no memory left anymore. And that's a lot (96 mb \nreal + 470mb swap). \n\nGreetz...\n\nLuuk\n|-----------------------------------------------------------------------|\n| Luuk de Boer -> <[email protected]> |\n| When you want to go fast ..... |\n| Drive a cobra on the road and take a T3 line for internet ...... :-) |\n| http://www.luuk.com/ -> The AC Cobra site |\n|-----------------------------------------------------------------------| \n",
"msg_date": "Mon, 16 Mar 1998 12:16:21 +1.00",
"msg_from": "\"Luuk de Boer\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "> I have looked into crash-me and screened the items you mentioned. \n> Here is the result:\n> \n> about the large objects. I looked in the manual of postgres and \n> couldn't find that fast anything about large objects.\n> blobs is a type where you can put binary or other big things in.\n\nWe call them large objects.\n\n> \n> first the concatenation with ||\n> the test is select 'abc' || 'def'\n> and that's not working in postgres. In other db's it's working.\n\nLooks like we have a type problem here. That type system again.\n\n> \n> function left:\n> the test is select left('abcd',2)\n> looked in the manual of postgres and couldn't find the function left.\n> this isn't working and I tried some other ways but couldn't find any\n> working query.\n\nYep, we don't have it. We have lots of them in the oracle_compat\nmodule, but not that one.\n\n> function current_date:\n> the test is select current_date()\n> that's postgres not supporting only select current_date (without ()).\n> that's also the case with current_time, current_timestamp. I will make\n> patch so it's tested both ways.\n\nDoes the standard say we need those parens?\n\n> Function IN on strings in SELECT:\n> the test is: select 'monty' in ('david','monty','allan')\n> and that's not working. it gives: using an explicit cast\n\nDarn type system again.\n\n> Function = ALL:\n> query: select a from crash_me where b =all (select b from crash_me) \n> yes \n> so it's supported\n\nCool.\n\n> \n> Update with sub select:\n> query: create table crash_q (a integer,b char(10)) \n> query: insert into crash_q values(1,'c') \n> query: update crash_q set crash_q.b= (select b from crash_me\n> where crash_q.a = crash_me.a) \n> Got error from query: 'update crash_q\n> set crash_q.b= (select b from crash_me whe re crash_q.a = crash_me.a)'\n> ERROR: parser: parse error at or near \"=\" query: drop table crash_q \n> no \n> I think this test shows that it isn't possible todo an update with a \n> sub select\n\nYes, we do not support subqueries in the target list. We do support\nsubqueries in the WHERE clause of all queries, including UPDATE, and I\nmistakenly thought that was the item. Adding this feature is on our\nTODO list, because subqueries legally should be allowed in any target\nlist, including UPDATE and SELECT.\n\n> \n> case insensitive compare:\n> query: select b from crash_me where b = 'A' \n> query: select b from crash_me where b = 'A' didn't return any\n> result \n> no\n> (field b in table crash_me has a value of 'a'.)\n\nWe use ~* for this. How do you control case-sensitivity? Are all\nstrings compared for lower and upper case. We also have many of those\nmath functions, but they are operations, not names like exp(). Try the\nnew psql \\df and \\do commands.\n\nAlso, it mentioned that MySQL ignores the CREATE INDEX operation. What\ndoes that mean? Does MySQL only support one index on a table?\n\n> the transaction are still not inserted in the crash-me test because \n> monty and I didn't have time to insert it. But what I mentioned \n> before if you would like to see some other tests in the crash-me test \n> you can add them by your self and send us the patch. We will add it \n> to the release of crash-me.\n> \n> I am not sure the results of postgres in the crash-me page is from \n> the final release of postgres. Because I have so much problems with \n> postgres it's at this moment hard to test. Further it's taking a very \n> lot of time to test the postgres db because it's eating my memory. \n> When I test the limit of the number of joins I can do it's eating my \n> memory till there is no memory left anymore. And that's a lot (96 mb \n> real + 470mb swap). \n> \n> Greetz...\n> \n> Luuk\n> |-----------------------------------------------------------------------|\n> | Luuk de Boer -> <[email protected]> |\n> | When you want to go fast ..... |\n> | Drive a cobra on the road and take a T3 line for internet ...... :-) |\n> | http://www.luuk.com/ -> The AC Cobra site |\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",
"msg_date": "Mon, 16 Mar 1998 09:28:06 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "> > first the concatenation with ||\n> > the test is select 'abc' || 'def'\n> > and that's not working in postgres. In other db's it's working.\n> \n> Looks like we have a type problem here. That type system again.\n\n\"that type system\" is A Good Thing(tm).\n\ntgl=> select 'abc' || 'def';\nERROR: There is more than one operator '||'\n for types 'unknown' and 'unknown'\n You will have to retype this query using an explicit cast\n\nWe are planning on working on the automatic type conversion\ncapabilities; I'll add this to the list of things to address. However,\nnote that this capability is in Postgres; it is misleading to state that\nit is not:\n\n tgl=> select cast('abc' as text) || 'def';\n --------\n abcdef\n (1 row)\n\nOh, that's right, we don't support CAST() :)\n\n> > function left:\n> > the test is select left('abcd',2)\n> > looked in the manual of postgres and couldn't find the function \n> > left.\n> > this isn't working and I tried some other ways but couldn't find any\n> > working query.\n> Yep, we don't have it. We have lots of them in the oracle_compat\n> module, but not that one.\n\nWouldn't this old Oracle/Ingres function be superceded by the SQL92\nfunction SUBSTRING()? We do have that, in all its glory. Doesn't appear\nin you comparison table though.\n\n> > function current_date:\n> > the test is select current_date()\n> > that's postgres not supporting only select current_date\n> > that's also the case with current_time, current_timestamp. I will \n> > make patch so it's tested both ways.\n> Does the standard say we need those parens?\n\nSQL92 defines the syntax to be CURRENT_DATE or CURRENT_DATE(integer),\nwhere the integer defines a precision for the seconds field. The parens\nwithout the integer is non-standard afaik. Postgres has all of the forms\nin the SQL92 standard.\n\n> > Function IN on strings in SELECT:\n> > the test is: select 'monty' in ('david','monty','allan')\n> > and that's not working. it gives: using an explicit cast\n> Darn type system again.\n\nNow, Bruce, learn to say \"that great type system again\", with feeling :)\n\n> > case insensitive compare:\n> > query: select b from crash_me where b = 'A'\n> > query: select b from crash_me where b = 'A' didn't return any\n> > result no (field b in table crash_me has a value of 'a'.)\n> \n> We use ~* for this. How do you control case-sensitivity? Are all\n> strings compared for lower and upper case.\n> > the transaction are still not inserted in the crash-me test because\n> > monty and I didn't have time to insert it. But what I mentioned\n> > before if you would like to see some other tests in the crash-me \n> > test you can add them by your self and send us the patch. We will\n> > add it to the release of crash-me.\n\nHow about if we send the full Postgres regression test? That should\nuncover a few differences :)\n\n> > I am not sure the results of postgres in the crash-me page is from\n> > the final release of postgres. Because I have so much problems with\n> > postgres it's at this moment hard to test.\n\n??? Better stick with MySQL then...\n\n> > Further it's taking a very\n> > lot of time to test the postgres db because it's eating my memory.\n> > When I test the limit of the number of joins I can do it's eating my\n> > memory till there is no memory left anymore. And that's a lot (96 mb\n> > real + 470mb swap).\n\nYup. Those joins do like to chew up memory at the moment.\n\nIt seems that the result of the test is to verify MySQL compatibility,\nrather than SQL92 compatibility or some other standard. It's OK if\nPostgres doesn't pass every test on non-standard features, but it's\nannoying that the test isn't labeled more appropriately. In fact, it\nactually claims to be trying to, in an unbiased way, help users select a\ndatabase product. Sorry, doesn't quite conform to labeling.\n\nOK, some feedback on specific information in \"crash-me\":\n\nSQL92 specifies that the allowed range in precision and scale for\nnumeric() and decimal() types is implementation-defined. By definition\nPostgres does support both of these ANSI types.\n\nFor \"Other Types\", Postgres has:\n\nabstime\nbool\nbox\nchar16\nchar2\nchar4\nchar8\ncircle\ndatetime\nfilename\nfloat4\nfloat8\nint2\nint28\nint4\nline\nlseg\nmoney\nname\npath\npoint\npolygon\nreltime\ntext\ntimespan\ntinterval\nunknown\n\nPlus the SQL92 ones already listed and some others which are\nPostgres-specific and not oriented toward application support; we\nwouldn't expect you to need to list those :)\n\nWe support the CAST() function; I had the syntax wrong prior to v6.3.\nWe support the \"||\" operator, even though your specific test causes\ntrouble because of the unspecified types in the test.\nWe support CHAR_LENGTH() and CHARACTER_LENGTH(), but which have the same\nproblem in your test as \"||\".\n\nWe support CURRENT_DATE, CURRENT_TIME, CURRENT_TIMESTAMP.\nWe support CURRENT_USER, which does not show up in your comparison chart\nat all. My reference books do not mention SYSDATE; what standard is that\nfrom?\n\nIn \"Other Functions\", ~800 of the 835 Postgres functions are not listed.\nAs you might know, most capabilities in Postgres are implemented as\nfunctions, so perhaps only ~300 of these unlisted ones are oriented\ntoward application support. Shall I send you a list? :)\n\nMy reference books do not list COUNT in a context other than COUNT(*)\nand COUNT(column). What is the Oracle COUNT and why is it an ANSI\nfunction?\n\nIn the \"Function Use\" table, why would PostgreSQL be listed as not\nsupporting \"update of column = -column\"?\n\n tgl=> update t set i = -i;\n UPDATE 2\n\nseems to work for me...\n\nIn \"Join Methods\", Postgres supports subqueries.\n\nIn \"String Handling\", Postgres does have extensive comparision support,\nincluding \"case insensitive compare\". We do not allow double-quote to\ndelimit strings, since that character is defined for other uses in\nSQL92. Do you show \"Has Delimited Identifiers\" as a feature? Postgres\nhas them, as called for in SQL92.\n\nHow about binary and hexadecimal integer input? Postgres has all of\nthese SQL92 features using SQL92 syntax. Is this the same comparison as\nthe non-standard \"binary items (0x41)\" in the Type Limits table?\n\nIn the \"Index Limits\" table, what do you mean by \"null in index\"?\nPostgres supports indices containing null values. I guess you mean\nsomething more specific than is indicated in the label.\n\nIn the \"Type Limits\" table, Postgres does support both items \"YY-MM-DD\ndates\" and \"0000-00-00 dates\", as well as other permutations not\nappearing in your table. Postgres supports \"European dates\" (e.g.\n\"16-03-1998\") as well as \"German dates\" and \"free-form text\" dates (e.g.\n\"Mon Mar 16 17:55:54 1998 MET\"). Postgres has full support for time\nzones, not just the limited and brain-damaged support called for in\nSQL92.\n\nIn \"Comments\", Postgres supports \"--\" as a comment delimiter.\n\nIn \"Other Features\", Postgres supports \"foreign key syntax\", although it\ndoes not yet support foreign keys. Postgres supports \"LOCK table\", but\ndoes not have \"LOCK TABLE table\". Neither are SQL92 standard anyway.\n\nIn \"Other Limits\", Postgres' maximum number of connections can be\nspecified at compile-time. There is no theoretical limit to these other\nthan operating system limits. Those maximums have a different meaning\nfor Open Source software than for commercial packages, where you cannot\nchange these kinds of parameters.\n\n\"crash-me\" is a specific test which does not warrant extending its\nresults to the tabular summary you are posting. It may be a helpful tool\nto elicit many of the entries in that summary, but any other use is\nmisleading at best.\n\nThanks for your time...\n\n - Tom\n",
"msg_date": "Mon, 16 Mar 1998 17:22:20 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On Mon, 16 Mar 1998, Thomas G. Lockhart wrote:\n\n> > > first the concatenation with ||\n> > > the test is select 'abc' || 'def'\n> > > and that's not working in postgres. In other db's it's working.\n> > \n> > Looks like we have a type problem here. That type system again.\n> \n> \"that type system\" is A Good Thing(tm).\n> \n> tgl=> select 'abc' || 'def';\n> ERROR: There is more than one operator '||'\n> for types 'unknown' and 'unknown'\n> You will have to retype this query using an explicit cast\n\nThis sentence doesn't work on MySQL too, take a look:\n\nmysql> select 'abc' || 'def';\n1 row in set (0.00 sec)\n+----------------+\n| 'abc' || 'def' |\n+----------------+\n| 0 |\n+----------------+\nThe result should be 'abcdef' not 0 !!\n\n",
"msg_date": "Tue, 17 Mar 1998 09:56:58 +0000 (GMT)",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On 17 Mar 98 at 9:56, [email protected] wrote:\n> This sentence doesn't work on MySQL too, take a look:\n> \n> mysql> select 'abc' || 'def';\n> 1 row in set (0.00 sec)\n> +----------------+\n> | 'abc' || 'def' |\n> +----------------+\n> | 0 |\n> +----------------+\n> The result should be 'abcdef' not 0 !!\n\nAnd that's why it's printed on the crash-me page that it gives an \nerror (wrong output).\n\nGreetz..\nLuuk.\n|-----------------------------------------------------------------------|\n| Luuk de Boer -> <[email protected]> |\n| When you want to go fast ..... |\n| Drive a cobra on the road and take a T3 line for internet ...... :-) |\n| http://www.luuk.com/ -> The AC Cobra site |\n|-----------------------------------------------------------------------| \n",
"msg_date": "Wed, 18 Mar 1998 00:56:31 +1.00",
"msg_from": "\"Luuk de Boer\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On 16 Mar 98 at 17:22, Thomas G. Lockhart wrote:\n\n> > > first the concatenation with ||\n> > > the test is select 'abc' || 'def'\n> > > and that's not working in postgres. In other db's it's working.\n> > \n> > Looks like we have a type problem here. That type system again.\n> \n> \"that type system\" is A Good Thing(tm).\n> \n> tgl=> select 'abc' || 'def';\n> ERROR: There is more than one operator '||'\n> for types 'unknown' and 'unknown'\n> You will have to retype this query using an explicit cast\n> \n> We are planning on working on the automatic type conversion\n> capabilities; I'll add this to the list of things to address. However,\n> note that this capability is in Postgres; it is misleading to state that\n> it is not:\n> \n> tgl=> select cast('abc' as text) || 'def';\n> --------\n> abcdef\n> (1 row)\n> \n> Oh, that's right, we don't support CAST() :)\n\ncrash-me is intended to test how compatible (ANSI SQL, ODBC or some\nother standard) a SQL server is. If the server doesn't follow the ANSI\nSQL syntax for a specific function it can't pass that test. MySQL also\nfails a couple of test because of this (like the above test). \n\nThe only exception to the above is the 'extra functions' section where\nwe try to add all possible useful functions from all possible\ndatabases.\n\n> \n> > > function left:\n> > > the test is select left('abcd',2)\n> > > looked in the manual of postgres and couldn't find the function \n> > > left.\n> > > this isn't working and I tried some other ways but couldn't find any\n> > > working query.\n> > Yep, we don't have it. We have lots of them in the oracle_compat\n> > module, but not that one.\n> \n> Wouldn't this old Oracle/Ingres function be superceded by the SQL92\n> function SUBSTRING()? We do have that, in all its glory. Doesn't appear\n> in you comparison table though.\n\nleft() is an ODBC function and substring() is a SQL92 function. The \nleft() function is tested in the odbc section and substring is tested \nin the ansi sql section.\n\n> \n> > > function current_date:\n> > > the test is select current_date()\n> > > that's postgres not supporting only select current_date\n> > > that's also the case with current_time, current_timestamp. I will \n> > > make patch so it's tested both ways.\n> > Does the standard say we need those parens?\n> \n> SQL92 defines the syntax to be CURRENT_DATE or CURRENT_DATE(integer),\n> where the integer defines a precision for the seconds field. The parens\n> without the integer is non-standard afaik. Postgres has all of the forms\n> in the SQL92 standard.\n\nis fixed in the next release of crash-me.\n\n> > > case insensitive compare:\n> > > query: select b from crash_me where b = 'A'\n> > > query: select b from crash_me where b = 'A' didn't return any\n> > > result no (field b in table crash_me has a value of 'a'.)\n> > \n> > We use ~* for this. How do you control case-sensitivity? Are all\n> > strings compared for lower and upper case.\n\nMySQL has two classes of strings. Depending on context all\nstrings are compared binary are case sensitively. ALL compare\nfunctions works with binary and case sensitive strings. The ~* is\nmore like a shortcut for LOWER(string) = LOWER(string).\n\n> > > the transaction are still not inserted in the crash-me test because\n> > > monty and I didn't have time to insert it. But what I mentioned\n> > > before if you would like to see some other tests in the crash-me \n> > > test you can add them by your self and send us the patch. We will\n> > > add it to the release of crash-me.\n> \n> How about if we send the full Postgres regression test? That should\n> uncover a few differences :)\n\nAs the PostgreSQL regression test doesn't test any limits of\nPostgreSQL they aren't really that important. It would be nice if\nsome of the PostgreSQL developers really put some time into looking at\ncrash-me and add some interesting tests to get a better and fairer\ncomparison, not only between MySQL and PostgreSQL but also against\nother databases.\n\n> \n> > > I am not sure the results of postgres in the crash-me page is from\n> > > the final release of postgres. Because I have so much problems with\n> > > postgres it's at this moment hard to test.\n> \n> ??? Better stick with MySQL then...\n\nthe problem is is that postgres isn't very stable and the releases \nare not well tested. Approx 6 months ago some postgres developers \nmentioned something about including crash-me with postgres and \ntesting postgres with crash-me before releasing it, but I think that \ndidn't happen ...\n\n> \n> It seems that the result of the test is to verify MySQL compatibility,\n> rather than SQL92 compatibility or some other standard. It's OK if\n> Postgres doesn't pass every test on non-standard features, but it's\n> annoying that the test isn't labeled more appropriately. In fact, it\n> actually claims to be trying to, in an unbiased way, help users select a\n> database product. Sorry, doesn't quite conform to labeling.\n\nI had the same discussion a while ago in the questions mailinglist. \nAs the crash-me page states, there is still a lot todo with crash-me. \nAnd I think a lot of developers will agree with me that documention \ncomes always at last while developing something. We looked around on \nthe net and didn't find any other test which gives us a good base for \na selection. It shows what doesn't work with different database and \nalso how compatible a database is to ' standard sql'.\n\n> \n> OK, some feedback on specific information in \"crash-me\":\n> \n> SQL92 specifies that the allowed range in precision and scale for\n> numeric() and decimal() types is implementation-defined. By definition\n> Postgres does support both of these ANSI types.\n\nAccording to ODBC, the precision of numeric() and decimal() must\nbe at least 15, which is making postgres to fail the above test.\n\n> \n> Plus the SQL92 ones already listed and some others which are\n> Postgres-specific and not oriented toward application support; we\n> wouldn't expect you to need to list those :)\n\ncrash-me already list quite many of the above.\n\n> \n> We support the CAST() function; I had the syntax wrong prior to v6.3.\n> We support the \"||\" operator, even though your specific test causes\n> trouble because of the unspecified types in the test.\n> We support CHAR_LENGTH() and CHARACTER_LENGTH(), but which have the same\n> problem in your test as \"||\".\n\nwhen we test that sort of functions we test it the ANSI SQL way and \nnot all the ways the different databases are doing it else we could \nwrite thousands of queries for all different db's.\n\n> \n> We support CURRENT_DATE, CURRENT_TIME, CURRENT_TIMESTAMP.\nthat's fixed as I mailed before.\n> We support CURRENT_USER, which does not show up in your comparison chart\n> at all. My reference books do not mention SYSDATE; what standard is that\n> from?\n\nwe looked in our references and couldn't find it that fast we shall \ntake a look at it where we found it.\n\n> \n> In \"Other Functions\", ~800 of the 835 Postgres functions are not listed.\n> As you might know, most capabilities in Postgres are implemented as\n> functions, so perhaps only ~300 of these unlisted ones are oriented\n> toward application support. Shall I send you a list? :)\n> \n> My reference books do not list COUNT in a context other than COUNT(*)\n> and COUNT(column). What is the Oracle COUNT and why is it an ANSI\n> function?\n\nIt's an old label which was changed while developing. It will be \nchanged to the good label.\n\n> \n> In the \"Function Use\" table, why would PostgreSQL be listed as not\n> supporting \"update of column = -column\"?\n> \n> tgl=> update t set i = -i;\n> UPDATE 2\n> \n> seems to work for me...\n\nit didn't in the latest test with me ... I will take a look at it \nwhat is happening ...\n\n> \n> In \"Join Methods\", Postgres supports subqueries.\n\nthe crash-me test output which is now in html on crash-me.html is \nfrom a snapshot of postgres (beta version). I have almost run all \ntests on all db's again and soon you will see the new crash-me.html \npage with the output of the official release of postgres 6.3. There \nis subqueries ok ...\n\n> \n> In \"String Handling\", Postgres does have extensive comparision support,\n> including \"case insensitive compare\". We do not allow double-quote to\n> delimit strings, since that character is defined for other uses in\n> SQL92. Do you show \"Has Delimited Identifiers\" as a feature? Postgres\n> has them, as called for in SQL92.\n\nNo, this isn't tested in crash-me yet, Maybe you could send us some \ntests?\n\n> \n> How about binary and hexadecimal integer input? Postgres has all of\n> these SQL92 features using SQL92 syntax. Is this the same comparison as\n> the non-standard \"binary items (0x41)\" in the Type Limits table?\n\nthe 0x41 is legal according to the ODBC standard.\n\n> \n> In the \"Index Limits\" table, what do you mean by \"null in index\"?\n> Postgres supports indices containing null values. I guess you mean\n> something more specific than is indicated in the label.\n\nThis is indead a test if one can have a NULL value in an index \ncolumn.\nWe test this by creating a index, adding a null value and selecting \nthis. Postgres isn't returning the NULL value.\n\n> \n> In the \"Type Limits\" table, Postgres does support both items \"YY-MM-DD\n> dates\" and \"0000-00-00 dates\", as well as other permutations not\n> appearing in your table. Postgres supports \"European dates\" (e.g.\n> \"16-03-1998\") as well as \"German dates\" and \"free-form text\" dates (e.g.\n> \"Mon Mar 16 17:55:54 1998 MET\"). Postgres has full support for time\n> zones, not just the limited and brain-damaged support called for in\n> SQL92.\n\nthe tests weren't run on postgres yet. But in the run of mine it is \nand the test for YY-MM-DD isn't giving the good value back and \nthe 000-00-00 test is failing.\n\n> \n> In \"Comments\", Postgres supports \"--\" as a comment delimiter.\n\nYep I know. I tested it on the prompt but I had todo an return \nbefore I could do \\g. I don't know why but the same query is failing \nin the test while it is just parsed as all other queries to the \npostgres db.\n\n> \n> In \"Other Features\", Postgres supports \"foreign key syntax\", although it\n> does not yet support foreign keys. Postgres supports \"LOCK table\", but\n> does not have \"LOCK TABLE table\". Neither are SQL92 standard anyway.\n\nmaybe not the full foreign key syntax is supported.\nEmpress,oracle,mysql is supporting lock table and that's why we added \nthis test to crash-me.\n\n> \n> In \"Other Limits\", Postgres' maximum number of connections can be\n> specified at compile-time. There is no theoretical limit to these other\n> than operating system limits. Those maximums have a different meaning\n> for Open Source software than for commercial packages, where you cannot\n> change these kinds of parameters.\n\nYou can change most important limits on all commercial SQL servers.\ncrash-me only tests a default installation if not noted elsewhere. The\ncrash-me page also states very clearly that some limits may be\nchangeable with startup options.\n\nForget the 'Open source software' argument. This isn't true as long as\nthe default user can't change the code. A lot of things are also\nbound by algorithms that can't be changed very easily.\n\n> \n> \"crash-me\" is a specific test which does not warrant extending its\n> results to the tabular summary you are posting. It may be a helpful tool\n> to elicit many of the entries in that summary, but any other use is\n> misleading at best.\n> \n\ncrash-me is a open test suite and we encourage anyone to \nadd new tests to it. crash-me is also the only SQL server test that\ntruly shows the most common limits of some of the big database\nservers. (We hope to add Sybase, Informix and Adabas shortly).\nIf you are trying to decide how to code an compatible application,\ncrash-me is the only source of information you can depend on!\n\ncrash-me has also found numerous bugs in the postgreSQL code (and some\nother SQL servers) and this also variants its existence :)\n\nGreetz....\nLuuk\n\n|-----------------------------------------------------------------------|\n| Luuk de Boer -> <[email protected]> |\n| When you want to go fast ..... |\n| Drive a cobra on the road and take a T3 line for internet ...... :-) |\n| http://www.luuk.com/ -> The AC Cobra site |\n|-----------------------------------------------------------------------| \n",
"msg_date": "Wed, 18 Mar 1998 00:56:31 +1.00",
"msg_from": "\"Luuk de Boer\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "What does it mean when MySQL says \"Ignored\" in the CREATE INDEX entry?\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",
"msg_date": "Tue, 17 Mar 1998 20:13:58 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On 17 Mar 98 at 20:13, Bruce Momjian wrote:\n\n> What does it mean when MySQL says \"Ignored\" in the CREATE INDEX entry?\n> \n\nmysql isn't supporting create index like the ansi sql way. You must \ncreate an index in the create of the table with mysql. It's accepting \ncreate index statements but it isn't doing anything with it so that's \nwhy it's ignored.\n\nGreetz...\n\nLuuk\n|-----------------------------------------------------------------------|\n| Luuk de Boer -> <[email protected]> |\n| When you want to go fast ..... |\n| Drive a cobra on the road and take a T3 line for internet ...... :-) |\n| http://www.luuk.com/ -> The AC Cobra site |\n|-----------------------------------------------------------------------| \n",
"msg_date": "Wed, 18 Mar 1998 09:09:44 +1.00",
"msg_from": "\"Luuk de Boer\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "> \n> On 17 Mar 98 at 20:13, Bruce Momjian wrote:\n> \n> > What does it mean when MySQL says \"Ignored\" in the CREATE INDEX entry?\n> > \n> \n> mysql isn't supporting create index like the ansi sql way. You must \n> create an index in the create of the table with mysql. It's accepting \n> create index statements but it isn't doing anything with it so that's \n> why it's ignored.\n\nSo you can only have one index on a table? Yikes.\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",
"msg_date": "Wed, 18 Mar 1998 10:09:30 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "> > What does it mean when MySQL says \"Ignored\" in the CREATE INDEX \n> > entry?\n> \n> mysql isn't supporting create index like the ansi sql way. You must\n> create an index in the create of the table with mysql. It's accepting\n> create index statements but it isn't doing anything with it so that's\n> why it's ignored.\n\nYou are holding MySQL and Postgres to different standards when\ndeveloping your comparison table. Frankly, it seems that we should meet\nsomewhere in the middle on this if you are hoping for further\ncollaboration on developing a test suite :(\n\n - Tom\n",
"msg_date": "Wed, 18 Mar 1998 15:18:31 +0000",
"msg_from": "\"Thomas G. Lockhart\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On Wed, 18 Mar 1998, Luuk de Boer wrote:\n\n> > > > I am not sure the results of postgres in the crash-me page is from\n> > > > the final release of postgres. Because I have so much problems with\n> > > > postgres it's at this moment hard to test.\n> > \n> > ??? Better stick with MySQL then...\n> \n> the problem is is that postgres isn't very stable and the releases \n> are not well tested. \n\n\tWhat *are* you talking about? This is based on extensive\nexperience with which version?? Which platform \"isn't very stable\", as\nI'm running on Solaris/sparc and FreeBSD here, and haven't seen any\ninstability problems since v6.0 ... Vadim is running it on a FreeBSD box\nwith a several GB(?) database ... hell, someone I know from my past sent\nme email the other day saying he was switching from msql to postgresql\nbecause he was seeing 10x faster performance doing the same thing on Karl\nDenninger's server...\n\n> Approx 6 months ago some postgres developers \n> mentioned something about including crash-me with postgres and \n> testing postgres with crash-me before releasing it, but I think that \n> didn't happen ...\n\n\tActually, it couldn't have been more then 4 months ago, since the\n\"some postgres developers\" was me, and i know it was *since* v6.2.1 was\nreleased, as I was hoping to get it into v6.3...\n\n\n",
"msg_date": "Wed, 18 Mar 1998 11:44:05 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "Luuk de Boer wrote:\n> \n> On 17 Mar 98 at 20:13, Bruce Momjian wrote:\n> \n> > What does it mean when MySQL says \"Ignored\" in the CREATE INDEX entry?\n> >\n> \n> mysql isn't supporting create index like the ansi sql way. You must\n> create an index in the create of the table with mysql. \n\nNot necesarrily. You may use ALTER TABLE to ADD, REMOVE, CHANGE\nKEY(s)/INDEX(es).\n\nMike\n\n-- \nWWW: http://www.lodz.pdi.net/~mimo tel: Int. Acc. Code + 48 42 148340\nadd: Michal Mosiewicz * Bugaj 66 m.54 * 95-200 Pabianice * POLAND\n",
"msg_date": "Thu, 19 Mar 1998 01:05:53 +0100",
"msg_from": "Michal Mosiewicz <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On 18 Mar 98 at 15:18, Thomas G. Lockhart wrote:\n\n> > > What does it mean when MySQL says \"Ignored\" in the CREATE INDEX \n> > > entry?\n> > \n> > mysql isn't supporting create index like the ansi sql way. You must\n> > create an index in the create of the table with mysql. It's accepting\n> > create index statements but it isn't doing anything with it so that's\n> > why it's ignored.\n> \n> You are holding MySQL and Postgres to different standards when\n> developing your comparison table. Frankly, it seems that we should meet\n> somewhere in the middle on this if you are hoping for further\n> collaboration on developing a test suite :(\n> \n\nYep that's true but because it's open for everyone you can't expect \nfrom us that we will write all those changes. Hopefully someone of \nthe postgres developers will say .... let's take a look at it and \nlet's add some things for postgres in crash-me. And maybe it's \ninteresting for the postgres people to add an sub select benchmark in \nthe benchmark. We aren't going to write it till mysql is supporting \nit.\n\nGreetz...\n\nLuuk\n|-----------------------------------------------------------------------|\n| Luuk de Boer -> <[email protected]> |\n| When you want to go fast ..... |\n| Drive a cobra on the road and take a T3 line for internet ...... :-) |\n| http://www.luuk.com/ -> The AC Cobra site |\n|-----------------------------------------------------------------------| \n",
"msg_date": "Sat, 21 Mar 1998 11:38:35 +1.00",
"msg_from": "\"Luuk de Boer\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [QUESTIONS] MySQL benchmark page"
}
] |
[
{
"msg_contents": "How are you finding the speed of the 6.3 release. People are reporting\nmuch better performance with it, and I can see you ran your speed test\non the 6.3 final release.\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",
"msg_date": "Sun, 15 Mar 1998 22:45:37 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [QUESTIONS] MySQL benchmark page"
},
{
"msg_contents": "On 15 Mar 98 at 22:45, Bruce Momjian wrote:\n\n> How are you finding the speed of the 6.3 release. People are reporting\n> much better performance with it, and I can see you ran your speed test\n> on the 6.3 final release.\n> \n\nThe speed is much better in 6.3 final as in 6.2.1 with all patches. \nBut I also tested the snapshot of 2 feb 1998 (6.3) and the \nperformance difference between that one and the final release isn't \nthat big but noticable. The snapshot was faster. But in the final \nrelease you can see some things are added such as the group by \nthings. The performance of that test is much much better in the final \nrelease. Further it's now possible to compile postgres 6.3 with pgcc \n1.0.1 with -O6 -mpentium which is giving a little performance gain I \nthink. (compiled postgres 6.3 final release with it and all test are \ndone on it).\n\nFor the crash-me test I have some problems with the 6.3 final \nrelease. One bug is fixed by a patch which was mailed to me by Vadim \nbut one bug is still crapping the crash-me test. It's the bug when \nthe postgres server is crashed while you are doing a create table. \nThen there is a file placed in the db dir and you can't create that \ntable with the same name again after a connect. So several tests \naren't run then anymore or are set to 0 because nothing can be done. \nI have read in postgres-questions that there will be this week a big \npatch for 6.3 and I hope it will fix the problems I have with 6.3.\n\nI will look into your other e-mail to see why it isn't supported by \ncrash-me and it's in the functionlist of postgres.\n\nGreetz...\n\n\nLuuk \n|-----------------------------------------------------------------------|\n| Luuk de Boer -> <[email protected]> |\n| When you want to go fast ..... |\n| Drive a cobra on the road and take a T3 line for internet ...... :-) |\n| http://www.luuk.com/ -> The AC Cobra site |\n|-----------------------------------------------------------------------| \n",
"msg_date": "Mon, 16 Mar 1998 09:29:00 +1.00",
"msg_from": "\"Luuk de Boer\" <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [QUESTIONS] MySQL benchmark page"
}
] |
[
{
"msg_contents": "> \n> On Sun, 15 Feb 1998, Bruce Momjian wrote:\n> \n> > > \tI'm not quite sure I follow this...in a multi-cpu environment,\n> > > would process_yield() introduce a problem? *raised eyebrow*\n> > \n> > Probably. I would leave the code as-is for multi-cpu systems.\n> \n> \tSo...how would we determine which is which? *raised eyebrow*\n> \n> > Yep, but we need to check for multiple cpu's first before enabling it. \n> > That would be a good trick from configure.\n> \n> \tI'm curious, still, as to whether this function would help\n> performance on a multi-cpu environment as well...what if 2 processes are\n> running on one of two CPUs, and another 2 on the other? *raised eyebrow*\n\nGood point. You would almost need to know if the one holding the lock\nwas currently running. But it wouldn't be un-runnable while it was\nholding the spinlock, so it should be run-able, even if it is not\ncurrently running.\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",
"msg_date": "Mon, 16 Mar 1998 00:08:03 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] spin locks"
},
{
"msg_contents": "> > On Sun, 15 Feb 1998, Bruce Momjian wrote:\n> > \n> > > > \tI'm not quite sure I follow this...in a multi-cpu environment,\n> > > > would process_yield() introduce a problem? *raised eyebrow*\n> > > \n> > > Probably. I would leave the code as-is for multi-cpu systems.\n\nMy experience with exactly this suggests that it doesn't.\n\n> > \tSo...how would we determine which is which? *raised eyebrow*\n> > \n> > > Yep, but we need to check for multiple cpu's first before enabling it. \n> > > That would be a good trick from configure.\n> > \n> > \tI'm curious, still, as to whether this function would help\n> > performance on a multi-cpu environment as well...what if 2 processes are\n> > running on one of two CPUs, and another 2 on the other? *raised eyebrow*\n\nUnless you have more CPU's than proccesses, having the spinlocks back off\nis a win. Since there are not too many lightly loaded 64 cpu machines out\nthere, for all pratical cases it is a win.\n \n> Good point. You would almost need to know if the one holding the lock\n> was currently running. But it wouldn't be un-runnable while it was\n> holding the spinlock, so it should be run-able, even if it is not\n> currently running.\n\nBut, it can't run until everyone else has run out their timeslice pounding\non the spinlock. Even worse, the effect is to convoy all the backends together\nso it becomes even more likely that a spinlock will be contended in the future.\n\nThe cpu_yield() or (more portably) select() style of backoff works pretty well.\n\n> Bruce Momjian | 830 Blythe Avenue\n> [email protected] | Drexel Hill, Pennsylvania 19026\n\n-dg\n\nDavid Gould [email protected] 510.628.3783 or 510.305.9468 \nInformix Software (No, really) 300 Lakeside Drive Oakland, CA 94612\n - I realize now that irony has no place in business communications.\n\n",
"msg_date": "Mon, 16 Mar 1998 15:38:10 -0800 (PST)",
"msg_from": "[email protected] (David Gould)",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] spin locks"
}
] |
[
{
"msg_contents": "Does anyone want to comment on this?\n\n> \n> \n> ============================================================================\n> POSTGRESQL BUG REPORT TEMPLATE\n> ============================================================================\n> \n> \n> Your name\t\t: Adam Levine\n> Your email address\t: [email protected]\n> \n> Category\t\t: runtime: back-end\n> Severity\t\t: non-critical\n> \n> Summary: Grant select to other user removes owner select permission\n> \n> System Configuration\n> --------------------\n> Operating System : Linux 2.0.33 Elf Slackware\n> \n> PostgreSQL version : 6.3\n> \n> Compiler used : gcc 2.7.2\n> \n> Hardware:\n> ---------\n> 486 PCI, 64M ram\n> \n> Versions of other tools:\n> ------------------------\n> gmake 3.74\n> \n> --------------------------------------------------------------------------\n> \n> Problem Description:\n> --------------------\n> If a user creates a table, and then grants select privs to another user, the owner of the table can no longer select withouth first granting himself select privs.\n> No other permissions testing was done, fyi.\n> \n> --------------------------------------------------------------------------\n> \n> Test Case:\n> ----------\n> user1 (in their database) creates table1; user1 selects from table1 (success); user2 selects from table1 (failure); user1 grants select on table1 to user2; user2 selects from table1 (success); user1 selects from table1 (failure);\n> user1 grants select on table1 to user1; user1 selects from table1 (success);\n> \n> --------------------------------------------------------------------------\n> \n> Solution:\n> ---------\n> \n> \n> --------------------------------------------------------------------------\n> \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",
"msg_date": "Mon, 16 Mar 1998 00:45:31 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [PORTS] Port Bug Report: Grant select to other user removes owner\n\tselect permission"
},
{
"msg_contents": "On Mon, 16 Mar 1998, Bruce Momjian wrote:\n\n> Does anyone want to comment on this?\n\n\tIts a bug <- short, Bruce-like comment *grin*\n\n\n> \n> > \n> > \n> > ============================================================================\n> > POSTGRESQL BUG REPORT TEMPLATE\n> > ============================================================================\n> > \n> > \n> > Your name\t\t: Adam Levine\n> > Your email address\t: [email protected]\n> > \n> > Category\t\t: runtime: back-end\n> > Severity\t\t: non-critical\n> > \n> > Summary: Grant select to other user removes owner select permission\n> > \n> > System Configuration\n> > --------------------\n> > Operating System : Linux 2.0.33 Elf Slackware\n> > \n> > PostgreSQL version : 6.3\n> > \n> > Compiler used : gcc 2.7.2\n> > \n> > Hardware:\n> > ---------\n> > 486 PCI, 64M ram\n> > \n> > Versions of other tools:\n> > ------------------------\n> > gmake 3.74\n> > \n> > --------------------------------------------------------------------------\n> > \n> > Problem Description:\n> > --------------------\n> > If a user creates a table, and then grants select privs to another user, the owner of the table can no longer select withouth first granting himself select privs.\n> > No other permissions testing was done, fyi.\n> > \n> > --------------------------------------------------------------------------\n> > \n> > Test Case:\n> > ----------\n> > user1 (in their database) creates table1; user1 selects from table1 (success); user2 selects from table1 (failure); user1 grants select on table1 to user2; user2 selects from table1 (success); user1 selects from table1 (failure);\n> > user1 grants select on table1 to user1; user1 selects from table1 (success);\n> > \n> > --------------------------------------------------------------------------\n> > \n> > Solution:\n> > ---------\n> > \n> > \n> > --------------------------------------------------------------------------\n> > \n> > \n> > \n> \n> \n> -- \n> Bruce Momjian | 830 Blythe Avenue\n> [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",
"msg_date": "Mon, 16 Mar 1998 08:15:58 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] Port Bug Report: Grant select to other user\n\tremoves owner select permission"
}
] |
[
{
"msg_contents": "-----BEGIN PGP SIGNED MESSAGE-----\n\n(I lost a block of mail a while back, so if this is 'old hat,',\nplease let me know)\n\nWell, my eyes are about crossed, but I believe that I've found\nPART of the problem with initdb in 6.3 on the DEC Alpha. (I'm\nrunning Digital Unix 3.2, rather than 4.0, but I don't think that\nmakes a difference.) This does NOT fix the problem, but I\nbelieve that it moves us closer. I'm hoping that this will\ntrigger somebody else to know where else to look.\n\nIn backend/utils/adt/oid.c, the routine oid8in() makes the\nassumption that an Oid is the same size as a pointer. Actually,\nI'm not quite sure why this code was written this way at all. It\ndeclares an array of pointers to Oids, and then fills in the\npointers with the actual Oids. Sorry I can't provide a true\ndiff output at this time - the code is too hacked up. Simply\nreplacing '*result' with 'result' throughout the routine, except\non the declaration, which should be\n\n\tOid\t*result;\nrather than\n\tOid\t(*result)[];\n\nAnd then changing the cast of the palloc accordingly takes care\nof this invalid assumption.\n\nI found this because I was attempting to do a lookup of\n'mkoidname' explicitly, much earlier in the run than when\nspecified in the input file. It failed. I could lookup up\nint4in or int4out, but not mkoidname, even when I set all keys\nexplicitly. With this fix, the explicit lookup works correctly.\nHowever, it still fails during index creation, where it's\nactually used. This is because the Oid of attrelid and attname\nare both being returned as 0, rather than 26 and 19,\nrespectively. With a hack to 'hard code' those values in the\nlookup routine, the mkoidname lookup passes. (The index is\nstill not created, however. It does later.)\n\nI'm still trying to find other instances where this assumption has\nbeen made. As an interim solution, I'm experimenting with using\nthe -xtaso_short flag to DEC's cc compiler, which supposedly\ntells it to generate only 32-bit pointers. This of course won't\nfix the problems, but it may get 6.3 functional on Alphas.\n\n- --\nDwayne Bailey + WHAT is your name? Sir Galahad\nMIKA Systems, Bingham Farms, MI + WHAT is your quest? I Seek the Holy Grail\[email protected] + What is your favorite color?\nhttp://www.mika.com/~dwayne + Blue ... no, Yelloooooooooooooooooow\n finger [email protected] for PGP Public Key\n\n-----BEGIN PGP SIGNATURE-----\nVersion: 2.6.2\n\niQB1AwUBNQ0SNqA2uleK7maRAQFnwwL9HaRnTmb0dWYUi6IarASXjNuGyijiNtVR\nj6Vtzi58HI/BtwKaFzwtbCj5koXIxVmm1Vu8IYUzS1S3T2gditcRR04i6QMYuLT6\nIYABZt38IWcmdbBJ+yVdYaZfE7tUECUv\n=jmf2\n-----END PGP SIGNATURE-----\n\n",
"msg_date": "Mon, 16 Mar 1998 06:51:11 -0500 (EST)",
"msg_from": "Dwayne Bailey <[email protected]>",
"msg_from_op": true,
"msg_subject": "DEC Alpha initdb partial fix"
},
{
"msg_contents": "> \n> -----BEGIN PGP SIGNED MESSAGE-----\n> \n> (I lost a block of mail a while back, so if this is 'old hat,',\n> please let me know)\n> \n> Well, my eyes are about crossed, but I believe that I've found\n> PART of the problem with initdb in 6.3 on the DEC Alpha. (I'm\n> running Digital Unix 3.2, rather than 4.0, but I don't think that\n> makes a difference.) This does NOT fix the problem, but I\n> believe that it moves us closer. I'm hoping that this will\n> trigger somebody else to know where else to look.\n> \n> In backend/utils/adt/oid.c, the routine oid8in() makes the\n> assumption that an Oid is the same size as a pointer. Actually,\n> I'm not quite sure why this code was written this way at all. It\n> declares an array of pointers to Oids, and then fills in the\n> pointers with the actual Oids. Sorry I can't provide a true\n> diff output at this time - the code is too hacked up. Simply\n> replacing '*result' with 'result' throughout the routine, except\n> on the declaration, which should be\n\nTry the oid.c from 6.2.1. I see no changes I can see in that code from\n6.2.1 -> 6.3, and we are told 6.2.1 worked for Alpha. I believe this is\nour last big item in 6.3 open issues.\n\nIs the lack of index use in certain cases still true for some people? I\nremember two people complaining about it.\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",
"msg_date": "Mon, 16 Mar 1998 10:17:42 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] DEC Alpha initdb partial fix"
}
] |
[
{
"msg_contents": "> Does anyone want to comment on this?\n> \n> > Summary: Grant select to other user removes owner select permission\n> > \n> > Operating System : Linux 2.0.33 Elf Slackware\n> > \n> > PostgreSQL version : 6.3\n> > \n> > If a user creates a table, and then grants select privs to another user, the owner of the table can no longer select withouth first granting himself select privs.\n> > No other permissions testing was done, fyi.\n> > \n\nSame here under Irix....\n\nAndrew\n\n----------------------------------------------------------------------------\nDr. Andrew C.R. Martin University College London\nEMAIL: (Work) [email protected] (Home) [email protected]\nURL: http://www.biochem.ucl.ac.uk/~martin\nTel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775\n",
"msg_date": "Mon, 16 Mar 1998 12:48:16 GMT",
"msg_from": "Andrew Martin <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] Re: [PORTS] Port Bug Report: Grant select to other user\n\tremoves owner select permission"
},
{
"msg_contents": "> \n> > Does anyone want to comment on this?\n> > \n> > > Summary: Grant select to other user removes owner select permission\n> > > \n> > > Operating System : Linux 2.0.33 Elf Slackware\n> > > \n> > > PostgreSQL version : 6.3\n> > > \n> > > If a user creates a table, and then grants select privs to another user, the owner of the table can no longer select withouth first granting himself select privs.\n> > > No other permissions testing was done, fyi.\n> > > \n> \n> Same here under Irix....\n\nAdded to TODO.\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",
"msg_date": "Mon, 16 Mar 1998 09:33:21 -0500 (EST)",
"msg_from": "Bruce Momjian <[email protected]>",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] Re: [PORTS] Port Bug Report: Grant select to other user"
}
] |
[
{
"msg_contents": "I'm having trouble testing the pre-6.3 sources. I've tried the Jan, 15 and\nJan, 1 sources, and both give the following error:\n\ngmake[3]: Entering directory\n`/usr/local/pgsql.beta/src/pgsql/src/backend/access\n/common'\ncc -I../../../include -I/usr/local/include -DNOFIXADE -Dalpha -I../..\n-I../../port/alpha -I../../../include -c heaptuple.c -o heaptuple.o\n/usr/lib/cmplrs/cc/cfe: Error: heaptuple.c, line 878: Unacceptable operand\nof conditional operator\n &isNull) ? (*(\n ----------^\n\nI've taken a look at the code, and the error's cause seems to be an\nincorrect macro. Now I'm going to try the Dec, 15 source tree, but I am\nsuspecting that there is a problem with the sources. I am using cvsup (the\nprecompiled binary that is stored at the main ftp site) with the following\nconfiguration file:\n\n---------------------------------------------------------------------\n# This file represents the standard CVSup distribution file\n# for the PostgreSQL ORDBMS project\n# Modified by [email protected] 1997-08-28\n# - Point to my local snapshot source tree\n#\n# Defaults that apply to all the collections\n*default host=postgresql.org\n*default compress\n*default release=cvs\n*default delete use-rel-suffix\n*default tag=.\n*default date=97.12.01.00.00.00\n\n# base directory points to where CVSup will store its 'bookmarks' file(s)\n*default base=/usr/local/pgsql.beta/src\n\n# prefix directory points to where CVSup will store the actual distribution(s)\n*default prefix=/usr/local/pgsql.beta/src\n\n# complete distribution, including all below\npgsql\n\n# individual distributions vs 'the whole thing'\n# pgsql-doc\n# pgsql-perl5\n# pgsql-src\n----------------------------------------------------------------\n\nThere shouldn't be any problem with this configuration file, right?\n\n-------------------------------------------------------------------\nPedro José Lobo Perea Tel: +34 1 336 78 19\nCentro de Cálculo Fax: +34 1 331 92 29\nEUIT Telecomunicación - UPM e-mail: [email protected]\n\n",
"msg_date": "Mon, 16 Mar 1998 15:15:31 +0100 (MET)",
"msg_from": "\"Pedro J. Lobo\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "Problems testing previous snapshots on alpha"
}
] |
[
{
"msg_contents": "\nThis is going to be slightly painful to do, since it has to be done\nmanually, but I've put up a patch against v6.3 based on reports that have\ncome in since it was released 2 weeks ago.\n\nThe patch is purely a *test*...do not use it production yet, I just want\nto know whether it works fine or not, and whether there are any problems\nwith it.\n\nThe patch is a tar file, which contains a 'cvs diff -rrelease-6-3', as\nwell as those files that are new since that tag was written...\n\nThe file can be found in\n\n\tftp.postgresql.org:/pub/patches/.test/v6.3p1.tar.gz\n\nand is hidden from public view... \n\n\nBruce/Vadim/Thomas...over the next couple of days, please be careful with\nwhat you commit...I'm going to do a proper tar.gz up on Thursday morning,\nand if all goes well, will make the patch public on Sunday evening...\n\n*NOTE: This is not to be discussed in any other mailing list, period, or\n you will lose your secret decoder ring, and we'll have to cut off\n your hands so that you can't do the secret handshake anymore*\n\n\n",
"msg_date": "Mon, 16 Mar 1998 10:35:43 -0500 (EST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": true,
"msg_subject": "First mega-patch..."
},
{
"msg_contents": ">This is going to be slightly painful to do, since it has to be done\n>manually, but I've put up a patch against v6.3 based on reports that have\n>come in since it was released 2 weeks ago.\n>\n>The patch is purely a *test*...do not use it production yet, I just want\n>to know whether it works fine or not, and whether there are any problems\n>with it.\n>\n>The patch is a tar file, which contains a 'cvs diff -rrelease-6-3', as\n>well as those files that are new since that tag was written...\n>\n>The file can be found in\n>\n>\tftp.postgresql.org:/pub/patches/.test/v6.3p1.tar.gz\n>\n>and is hidden from public view... \n\nIt seems that some files in my \"multi-byte patch\" are missing. Without\nthem the patch is meaningless and I cannot test v6.3p1.\n--\nTatsuo Ishii\[email protected]\n",
"msg_date": "Wed, 18 Mar 1998 11:40:23 +0900",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] First mega-patch... "
},
{
"msg_contents": "On Wed, 18 Mar 1998 [email protected] wrote:\n\n> >This is going to be slightly painful to do, since it has to be done\n> >manually, but I've put up a patch against v6.3 based on reports that have\n> >come in since it was released 2 weeks ago.\n> >\n> >The patch is purely a *test*...do not use it production yet, I just want\n> >to know whether it works fine or not, and whether there are any problems\n> >with it.\n> >\n> >The patch is a tar file, which contains a 'cvs diff -rrelease-6-3', as\n> >well as those files that are new since that tag was written...\n> >\n> >The file can be found in\n> >\n> >\tftp.postgresql.org:/pub/patches/.test/v6.3p1.tar.gz\n> >\n> >and is hidden from public view... \n> \n> It seems that some files in my \"multi-byte patch\" are missing. Without\n> them the patch is meaningless and I cannot test v6.3p1.\n\n\tPlease let me know which files, and possibly submit a new patch\nfor it to be fixed tomorrow?\n\nMarc G. Fournier \nSystems Administrator @ hub.org \nprimary: [email protected] secondary: scrappy@{freebsd|postgresql}.org \n\n",
"msg_date": "Thu, 19 Mar 1998 00:10:01 -0400 (AST)",
"msg_from": "The Hermit Hacker <[email protected]>",
"msg_from_op": true,
"msg_subject": "Re: [HACKERS] First mega-patch... "
},
{
"msg_contents": ">> It seems that some files in my \"multi-byte patch\" are missing. Without\n>> them the patch is meaningless and I cannot test v6.3p1.\n>\n>\tPlease let me know which files, and possibly submit a new patch\n>for it to be fixed tomorrow?\n\nHere is a list of files included in the multi-byte patch. For your\nconvenience, I attached a patch for v6.3p1 (missing files are in it).\n\nIndex: postgresql-6.3/doc/README.mb\nIndex: postgresql-6.3/doc/README.mb.jp\nIndex: postgresql-6.3/src/backend/regex/Makefile\nIndex: postgresql-6.3/src/backend/regex/engine.c\nIndex: postgresql-6.3/src/backend/regex/regcomp.c\nIndex: postgresql-6.3/src/backend/regex/regerror.c\nIndex: postgresql-6.3/src/backend/regex/regexec.c\nIndex: postgresql-6.3/src/backend/regex/regfree.c\nIndex: postgresql-6.3/src/backend/regex/utftest.c\nIndex: postgresql-6.3/src/backend/regex/utils.c\nIndex: postgresql-6.3/src/backend/regex/wstrcmp.c\nIndex: postgresql-6.3/src/backend/regex/wstrncmp.c\nIndex: postgresql-6.3/src/backend/utils/adt/Makefile\nIndex: postgresql-6.3/src/backend/utils/adt/like.c\nIndex: postgresql-6.3/src/include/regex/pg_wchar.h\nIndex: postgresql-6.3/src/include/regex/regex.h\nIndex: postgresql-6.3/src/include/regex/regex2.h\nIndex: postgresql-6.3/src/include/regex/utils.h\nIndex: postgresql-6.3/src/test/regress/GNUmakefile\nIndex: postgresql-6.3/src/test/regress/regress.sh\nIndex: postgresql-6.3/src/test/regress/expected/euc_cn.out\nIndex: postgresql-6.3/src/test/regress/expected/euc_jp.out\nIndex: postgresql-6.3/src/test/regress/expected/mule_internal.out\nIndex: postgresql-6.3/src/test/regress/sql/euc_cn.sql\nIndex: postgresql-6.3/src/test/regress/sql/euc_jp.sql\nIndex: postgresql-6.3/src/test/regress/sql/mule_internal.sql\n-----------------------------cu here----------------------------\nbegin 644 p1.patch.gz\nM'XL(\"+6K$#4``W`Q+G!A=&-H`.P\\:5?;2+:?G5]165['!MLL26^XR7O&*,'=\nMQF:\\A,[I[L<(N6PTD26U2H8PZ?SWN??6HI(L&\\BDY[PW9S@)6*6JN]7=JV#J\nMSV:LX;%&GS42%L_%[T$S2OSYCDB\\G4O7>\\_#Z4[\"Y_S#SJG[GL_\\@,M9&R8\\\nMVMK:NA^HROAJR7YT0\\;VV.[W![N[\\(_M??_M[J-&HW$G'EI]ZB:P@NV].'CQ\nM]<'^'GS^_CLDP/XB@G89?43`>_47WS/XT'BTS9XVOM07`L/_3-,GAQA\\&='-\nMHH01$V9R]]CIC[NONYWVN#OHZP7/3K@[Y<D!VUF*9\">(/#?8\\:Y%$D7ISEUB\nMJ5^SO>9+%,2W.WO[._N[#.2Z_\\W!U]\\QX25N'-\\RYT/,GFD2OJ0`X-]HV#GN\nM#MDA:S9WFDT8\\$,O6$ZY?#94-N=!=.D&<DVWW^E-CIV+P=D8%C:ZS2:3+SJO\nM>^TWH^W#9U5K2LUZT3@^&XRZ/U^<=D?C]D^.6C8X^G$$@$`N7K2(FQ')/$FB\nM1'W\\P#WY:99PWHR0R-F4S]CID5Z\\?<B6J1\\(F'8CTL0C*/@II(^&`IS8.#X]\nM`@I/CY`PV`]_)JEP@^\"`C29'HW<C6H'\"44\\'[%D5\\>\"*RK-J[[B&YM>(S`3K\nM/?R;\\AC@XH\\#N:#3@06GIRPO%[;5][email protected]]`+NA@?XJ9(L6&.V`E\\)S)_Q\nMWUE5KJP_J][XP=1SDZE\"7*M9VVC`&TZG]_<A/)S[(0<JURFQGG\"W#]$S/\\^'\nMY%9_G@_9V]]]J;W(SA9^9UNL$\\6W0.]5RJI>#0'MU_'[\"_K^DIWP,+EE(Y\"@\nMQY/F?9;0G`J0R=D0*`]3P:(92^%Q$OK7/!%^>HLC'3?PP;N$OMMDK!T$C$`*\nM4'+!DVL^E<@DQO&5+Y@7P6;\"3W`S`&?*9DFT8\"*:I3=NPN%MF\";^Y3*%-VG$\nMCGCRG@?\\EEW>2A`EC,@70S[UA5SJ1R%S06V7`A\"%`'N9>)Q&+OW0A>5`[T+4\nMV8V?7C%PC/@S6J82SB(\"M?(]%Z'4&9(4\\V3AITA0G$37_A0INW)3DL4L\"H+H\nMQ@_G2/C4QT5\"PL&5\"YX>R*>]9H%`DJ:BC\"2R6(H4A):Z0#%\"=B^C:WRE]DAN\nM!V-AE/H>K\\,4$&$``!%.AINXS!,&6+W`]1=FW_=7:0&<EF@T+<#N=`GT_4GD\nM,,FI!C6-O.4\"],S5^[<#6Q/!A([email protected]=PC99D0Q^*))BNA.04U3\nM7R#6#`*B@*DX..-NN@0]11U`72']!CZT,FK*2!R`)P[<VP(W8-MA=!/PZ9PC\nMX`-C-3[1\"N)+M0/+X(+J7_,@BH&-R]M5DU(P,L.J9V:`\\O13D9E)E`C%]<LF\nMZW.?Y(4@0W?!2RPVC++7I(-%:(9IX/62HUS($,&#P4N.H@*^%E'*-7\\%2RZ5\nMH+(P)F+NH77!6A\\-+T&[\"J6%\"6%V4'F+D^Z(C0:OQ^?MH</@\\]EP\\!;2EF-V\nM]`Y>.FSHO($<9L3:_6/6&?3'P^[19#P8CMA?_]H>P8+GS_&5!-;NOV/.SV=#\nM9S1B@R'KGI[UN@`)0`_;D`<YH[I*!KK]-R#OR9CU!V/6ZYYVQS!M/*@C1@EJ\nM=2T;O&:GSK!S`H_MHVZO.WY'1+WNCON(\\#5@;+.S]G#<[4QZ[2$[FPPA?5#P\nMD+OC[JC3:W=/G>-FI=L'W,QY\"[RQT4F[U\\LQ\"Z!RO!XY0&7[J*>`$2K@%5(A\nMIS-&IK)/'<KYVKTZ&YTYG2Y^<'YV@)OV\\%U=`1XY?YE@8MCN27C'[=/V&^\"P\nM>H=T8&\\ZDZ%SBD2#-##<C[OCR=AA;P:#XY$$!BA&SO!MM^.,6JPW&)'@)B.G\nM#FC&;2(!X(#4X#5\\/IJ,NB2_;G_L#(>3,\\Q4:RH.#,Y!0D!R&]8?D[0'?>(<\nMA#48OD/(*!?:C#H[/W%@'/:]+Z771HF,0(J=L81GS07,(-FQQ3+K.V]Z79!_\nMQ\\&W`P1UWATY-=B[[@@G=\"7N\\_8[Q>F$Q(`;!_3)CY8VUVE[6?<U:Q^_[2(#\nM:C)HQ:BK-&CP6L(:33HG:A\\L\\ZA4_J?ZM&;2B>^:7[.J=A,U]@+2[QT5Q'=D\nMJK5C@C!'=^A=H?^2Z\\FMS!(?C%Q`%\"?O164#_'RJW!?Y*IW#>BK&S<\"K,K'T\nM4_<29C^%5!:@D5MQV;4+`&6.L'\"])!+2EX!_XCY$5A%1%)6`,).#\\1`\\!/A_\nM3K$'\\@<5#,!!09JEG9'V)=-E'%\"<)O<N!%>!EP+0E#=MWI_*-'O4;Y\\ZJ(N*\nM4BD'9$%]L%[-7)%6!'ZW!@6X_8K`[]8@1`;!P<\\+]<%ZA;E?`HB%^F\"#2GD,\nMH.\"[-0AN,02L],,:=M-*1;AID?`*DHV#.B-6;/;6L1FL83,H8S-8PV:PGLU@\nM#9M!*9M!&9M!\"9M!@4U29M!9F1Q`^(<HM,1$#;4`TKZ8]'D:W808N=SKR)^R\nMO_M!H-.NV$T@^H'J\"JD@$H)$!H\\?L6I18PF_F\"\\AQ&W-6SB*9%<J?!:X<T$#\nM8!\"T[\")E6S%]:L$$H.Z74\"POM_=^8]5=!B:)R0%;AJC`-8FT$L\\O;KPKJ`!`\nM3:/9+%8+X:/@@/$F2M[+B(H64[+FDH/IZE5@-@GE7I@,@4%`H7#M)^G2#5A_\nMT@.!<X]C\"K(*!H2J@6\"IMQ;$&BJ\\*##KE96&D12E)`I!77)(9/@Z$%L!:%\\<\nMB9827!PL!4I.SH0OJ+#'SMOV<-22.P,>0H!!I(B6\\\"Z3S'=H'O4TJ+3%528E\nMG(#-$)?1\"R6X_))TH3E*^2*.$LR+\\S-@.+W54L//#/9,RBXCX5/+Z.JA_F*T\nM:R`)[+(P**MXXJ;2MS9W%N^OK)D$0UGTQ86'4L'_6`!_`-T-*Y4GG2=*70NF\nM@<M-Q6M@$?60>TFC0]DI=U`M*CN;UYG9H2VI#^\"Q_;]ST/.05M45\"/JRK$`:\nMP2^_U1$-DY92:UG8,[@*@G(G5=L*L0[+DP\"[E!^(XAP)(HI![VB>2#&ZR,<H\nM%NEF],IE_7/HV0;T^BG@UYLI02?\\YY&Q&3?Z^L_`?5^$%64Y&AN$@U*EL\\!9\nM?%E@Z]K\"P*5(%0-5U&.0CA!2'3V.!KU*I0JIV/9>S1IV:!A>YH=A`-_0BWW[\nM!:2Z)YCCT9L7^27G\"M++/`(]_+4]W!D<.Z?MGYEZ([W'%533',I+Z@%0?E/(\nM6:@U^)0Q'S^\",;/32<^YH(RX3TDZ]6T-J7TH0H95KU:I8E/G%=O[YMMOO]W?\nM^Z:&R/0#@MG_W_V7^/,%^![CLG*P%+!*51'>,+#D3!X8FIQ)Y^+',_;'']9S\nMIY]__FF8?QZ?WTG\\9'0R'%\\`ZMJ=M!6G6L1-^EV<]:=BTPD8#P270ZMH-!Y\\\nMMM86D618<A,M#Z\\T8^@<.T>3-Y998Z:CS(9RJXT6[;FQ['$IZT'[U0;UNMMS\nMV-946E,AN]R(UMV,,_73@#_$I<$FW.5B[J9OQ=O%%_A8E=SF@:R+N9_4.STO\nM%]0Q<;IW2\"^C5+N(T5D56:PS5!>YA2`_-80[-87*Q_8I[3$NB/?@_SZ\\AY]B\nMOX:[()<57N2=FH-;!7@^HF^I+AJO9+3^:NB\\N<`*&<I;(F%6?7+X7^+7\\$F=\nMT8I:BZ1A=+V,=O)MF*!C^B==RT:*U\\V69!+`U3FYI`?+O8:I=!JR:>E1^EJH\nM=G$F;$:6#;%U>5!I&O2K7`]1+)</L?LD0;K>T)@KFK%=J*(]CPL0'XJ_/SAM\nMCZ'JG[E^L-1)LZ82XZ2B1>..U4^%*JMI\\O6+9@;-2O&#PY*12D5\"PY&,EXKF\nM!8>+Y<]'.Y6OF&HBJY3\\EE5+28^PN%X=VUJP0_:5>I.#IZ!Y4+FERO3G,Y@\\\nM;[R:^8E(9;M@F^VUT\"*W]RBW'SC]8Y7RYA8&<B&6&[1N%1UYF;+A*%;)/)41\nM_B*&\\\");M\\('':.6]0W5-U#(\"!^[(9(`-\"W`Z9',V%=J=T<3.K/3(@>Z=EMZ\nM-K>G0NDS'`,[-)O$79$5RZ$NT[:UMNW^UDP6%R)JJ5E1O'82EY/0BBO*BDN!\nMMPIIB`46IV4*=0'S`QY6:5PY5`6WN*1DIC)C20^*@);\\()=(.24<*OR055$F\nMW3[4@F]JUH9@L\\A+.,>N?DMVH*<1QQHZG&*5/Z.F-XR\"9Z:6`62[,N/*;1+U\nM^1\\?8KG;LP2.JZM3PT.+325QH!/P<7M;4DA`MG`:*1G\"`FFSK[ZBF>\"0X-4K\nM^0KXA_$5V4(]QA?>(@9<=0T\"72Y99TU2B$NWR&\"C655+OU:K(=;=6D[N=X!3\nMT.R5>A^P*DJX^UYJ$O(EV4)&5-IZ@_V@:?@\\9>\"FIQJ\\*=GM[5*^K&943FV:\nM[A&0'X#R>1FKY1\")YFBI+?6@#.*067T7&(ZUZ>!VZ4'52BB,8GLEI]0X*/LG\nM9E<ULE#MM+1XV7L8.>/)&8;4EY(+]=QXI:L<,T!-A<)8NH@+(]0RD&.=G@/9\nM86Y,R4<>[email protected]+41C<PC?JC?*4CM<8_^#Q.\"7U5F6L[GZ0SK9:EA8KQJC(\nM7-15C:5JJCG44O.@ENVWG&Q90N6C\"54NP[,:L]4DH#$P<3PX[U<7\"LAZ#9!6\nM3EBT\\P,%1#MY#%JDV5`V)=70!'\\V6R9DP'XXBUC(^=2JEZ1>H@_^;TT<]@.3\nM%*5+;2ICVZW,KI6,E)`JE&\\\\09U&MTE\">J)9TJJ!I3)(4$,M%6(F1=N?Y`UK\nMA;P?F%)`#42_V=Y>+[J]+R6Z\"&+S;1WR6';CXFD_!;?E)?^`WA7/YD2SV=22\nM51EC;,BP>+1MLVHW2&M@\\T$0>55<\"A1CBQ1C=\\TDYB;K5QXN6UVSM',-XH_W\nM5DAG=(;IK2U44@8?*(9,PF<_'+*,1!@P3MZ@_L57X9;1U-P@Q\\'&7D:PO3VT\nM75FVK>+\\4;OST[!6HHJJ.P;::/20M%!WS7**B+I34$3)G0D*'S/M1.ZPN(&*\nME,S/]IYYG<WYU6J6$5L;:H!MW,YL:<VVDWM3HJC'GB[/M$\"#6K?SZ[=>24>)\nM6I]B*-'FY:W;A&OEC:$5T\\P:1-.BM:ZX`>,%M/$MKQK1%=K7#5ZR6`99:'7S\nM1M@(\\.Z`SIMS;L[V`2V*L!B2+^7Q#(0-.F&#&@/B\"3HW/UU=+[>!'/(??]C[\nM<+?GS/.)R\\G]24LB>>5=H'1,D/YP-\\L;LHV`D'VWXZ47>\"5JG?,]+'.^I9@Q\nMX\":W*!2\\?,+$593@V:;,YOT`KT[HAGR(.5O?-!GN[3NT]F9N?\\65H-LP:KMF\nM'B_,^U3(W7(R_,+*K$C*)(O[+)^LN%50>2G=6;24VJPDBZF,%OZ*.4RYB/T4\nM;V]!H@,5O!O'W$U<U%Z\\:X1U%B1'YJ#).EC2$=P%C\\?<P$T6FG5=VIAPBU4C\nM(DNX=^L%/#NN\"O\"R4,$XY.H?5!)<S&5G?A\"HZTQ`>PI5N\\\"F2L)_7W*1ZDAK\nM1^Y7,NE62E$HX6PJ&TPEL*W5F11HE!'8LSZM(-NSD!7UJLRX,VU6/09;D1%R\nM0<^UA1E-+81%;<FF./E87&`XUX&S\\\"X75\"6'>@/R.4'.TU*DL.,.*X0-M;;H\nMY]8O5C-K6750C#@ZW$C3443J!I4^16^`SLRQM8-7&6[P.J%L[TSIH8ZZ77#;\nMCW)M*TE1^=D9F'6NNVHZ5@\\YN<FUJC(9J(1258/@)F3;J@:V=N/>JK+`2DYR\nM=4:&3>)YE#][!\\I+NE2J.\"MMR!#5RK1%FAM!!*9'E>M&J1E\")\\?2M&&8KH[H\nML^3EY=#1YZ9R!3<KL-^[87Z>U'5']'J_+V^S4%Q@,FZIE24+/#<,HY3N/)0=\nMK!,$R!G25AXY#EE$E\"Q\"[Y7#NPPMS'I]43@\"@UV)/&&X5)1FMB7+PMR\\,$3<\nMNDN,&]=SM3Y_R>&>J\\$3&P(@$[OV(TB38*1L[M2T\"MMCF;\\_N<,,I%KF6Q'D\nMA8708[\"/@$^UG=0#N'YAN74@S9_BY=;9K6'2X@=DCL\"$\"H8W/AI<=7\"F:C&4\nM1_P+%[_EBQ#/A3`Z..M-1A<']LA?)LYHK(<0]/8A3.L?KT#3OMRN>R6(SHE9\nM?W.%E\\W*B$&'/+CHG.CX\\D!4JOH1LH#68BIZ7C]SODI6:\\0CRL7C](\\U)RJV\nMAE',-[+>'NH5PE3W93./!KV<W)W\"\\]'@O/!>/Y<`:_??'12>!Z_O20?4IWD-\nMH)+UOFS+VT$\"6[+DJP(>SO'N_4PY?DCFKO#7/M)(%CTZ,2Q3-G\"+=H.N4BA'\nMLM2\"CM>C&Q9$=#T?BBK94\\/,\\S)K$E&!*%*KP!!DH_@-LDTN#$>:39INDB9I\nM>7A!?)HORQ3;\"C&7KE/.T(\\V$>AW+2+PM788F/7:OJ*BDC\"Y1.6D^H6N<@#W\nM+1>/\\]1`C&\\TJ-+),G%)DNF1VXFYEC:1#H5641+X]A7@CV7!J:1`5]<,B$]Z\nMW[#;@QZ($F_C.6@0%,.\"C@4`@/'#$'1\"2/HL2=$)@;552E(`\"24%N6BQYV84\nMHM@Y*5U=Y%&6-SC-O-(LJ716RC63F1;.ZD*AA;G>TFQ?^Q]=_[?7=2'COME?\nMF6G0VK5[KO@C3XGEB.(*?YE#F8S%$[=;)V*SQI-E\\5QQKQYA89G0\\7P:HF:4\nMF``:+J$2SM#GV)%M9KJ4&_(/*6V-1:F<Q^.BE16H*E-]DH(J16;+8$4WE10R\nMD2IL*J\\K(-0;SBT#-OTT_N'*71*VRUP.;:VSA0U`\\KZIZ!)6?%+9JD)+2!W'\nM\\?@ACL7*M_[C5O[]W`H-KN3%F+':6!72E4Q[>?D;J>9@,,P:D)O\\#MV^R&[6\nM7&)SSMZVNX.T-H02H5L)N0ONA'Y++;\\)]`N8*&8\\B^33.NT'.96\\[!_$+XDR\nMRX+OLWB_9GD4Y=[EH58.Z)J:!0'EM'$S*B,I'&\\T,@S9D?\\]B+XP)[&98CTH\nM,5J?%]W#\"UU@?I,O)\"B]SPW!IA2>]_-+.B</*[AZ9^VAT]=KL+E9:BD%#G>A\nMWO;Q0*K8[-3S2D\\#Q4K/MHRBX;^&(GX?BJ9\\YBZ#]*\"BCJ+P*+@8U-8).->(\nMS26^6:-<-T1%7-81U<=N=W5$Z2T=7(7W:8V6W^O_9UNCQ<O\\:[NDI3U2\\*>4\nM5U75O3[UJT#6<<RF/E&=L#[Z%_1,U0\"@TQU/M%KPKT*>C]`1))'^!;JKG]\\M\nM_7_2>E1W&/4MRX\"'K4QJV/;0@D,/J@QF#K8*D_&7P=QK>5'1$SPE>)ZP&XRH\nM/)_18,018&G.4^8*R.,3_(&'SVXH?[>>NP+82I>SF>*)&C3ZHF))>_(Q3<#[\nM;OD^I3XVLGIJ@FGGM;&O9K?)*)\\P?@4KDRWL6&$.F=V#DSY4U$RDU)['.M[:\nMW!@KX-D,:'/++-==\\P3=<)4\\\\U3\\HBC5!V,E_#WNG'3[50_R5N+TP4QEK4,\"\nMKJ!G%]]*[Z'T!V-85ZL!`1H=+C0WDW#55I4\"RAZE$\\]_#9_+RXPTF:D(EK_O\nMZISWNGVG5BMFDK=<K\":Y=C9S3U:==:QJDDL9=>YD5,+Y/\\%BUNB5+'[V=K*'\nM[>=F;HMP7^4(RH`\\[H[.!\\/CJD&%5U:S\"2LDZ>F0,?R)2K-.H@_2FA)YEJO-\nMPR29@_@X)X^\"X/(R9R62_I,$F$O<2]YC%B\\QRD`I9'/H$M([O'\\HL`\\DWOMQ\nM1HP0F4,MB1&F91)ICC3%6=$C2BLS4W]EG8A/+#MU*HM'YN3)ZDJ'^&=/S-_^\nM06HQ@.(?-L!S6#F*:S*.UJ?95Q#740(+]SUG+F2ED>]9E:N*MGLEXC>W.RCB\nMRA/`BOS2E?$-!',_-8T+%5P>YX):R2X7$G6)20@H-0UX=_HW['[-U)^1@9_/\nM\\6]%A\"[>?_$2^8ONYEXIT</EV8Z=2V#B@H-/=*<H4JVB7,*R1@/L1*)F'8#:\nM)U5:%IQ=PR;CG^_A<7J5Z;TIN(Q\"W+/\"RMTU*=[(4GM59C_K+WT]MB]SX?W^\nMT@NEC95!_9L>62,I4G6><@:O#O&<S2:<O(6:AZ]*Z56*#MEQM)Q?P;Q9*O74\nM[LO)]JF^6K^]GCB2E_P]!-II_$:_=O!8_=K!.H']H[UW[V_B2!:&_Y8_1>,E\nML60DVQ*0@(W)\"EN`3GQ;2PZ;`UYG+(UM/4@:K4;\">$G.9W_KUK>9EBP3DK/[\nM'O@EH)GIKKY55U=5UT5V9P8!L%H+%IZO'\\NJ9ZB;0SLT.%>NPDY\"NQ;74$?C\nM2U:Z96>K>?>/@E,)DS'8`#A+>D*R=G4N6N>;<B8G;R*NN]\\=V<7(.E;,;DFP\nM^@[#<V^<+)J&+3]U`0\"(+3A[!\"U'';MUJ?Y6\"IZZEQRW#\\+O.AD5S^R^JU&R\nM7+3?A=.2N24@@YGP;<6=^U5Q^B4&I#HJ\"36C(6<ZX\\[%7+RIS%Y-'X.VPQCT\nM60-Q.`$7$V?,/MXJR-%`>F%2RU-0P608E]$(,AK>.&0CK\\#.Z:]35V-]%P6N\nMKZ_^5'#[=8N^>MXR:`6H-_$S]FY@RNZ@KIZO`G;-8#(D6PX[;1;KPO1UVG=2\nM:<_0:\"^HT%Y<G^VKLUU.]\"[*[-]F8*CH?H431>9E.NPF\"D#W+H?$KO3$+.Z\"\nM['9_-XL@2IO`.6Y.Q445QW_0J3*G$Z[&*7PPYR;X^-]G@N-9$SQ+#_ZG3G!\\\nMYPEV1(:\\:CZ@F;=5'>/TY4Z$WB17:$0_7);JN<I0$*3;XT9]YW5#>T]+[S8\"\nM6GOT(X1_,)(+4)XQL8HD%+%J%$4D8#/3$0AZ2P'M?\"#6S1]KM:Q]`]6$H\\I]\nMD.!BJ`1&@JQ5\\U0_Z\"3YOV:\\;`->,<;+]9H7X(J_<*PK]^-D('PZQK3\"#UY/\nM?)-35DAWT,=+?\"4<_5))_4!!#3=1741?D6=PZB&#TQ'%NC&1[6C'!RJ\"B@^.\nM+^!=&7A=TO'$#+LF009!'(`Y-AX?(.#J,P+^J%4D.?'X.KJQLB=[UCH>NM4B\nMKQY-MXB8[.L>C^A\\\"=RXX/\\2_P<?N%:]U6J^.BC2;-NWK:/B,CN,<K55H0/L\nMQN$X).=]<Y')1V8`9R'J.,XG-*E0MR/TEOSO'\"V5LRI\"T9`V-?Y&(V5?9\"%/\nMNA?&.)GNV5/A&J(A1D7207=?P,_S>'(=@S#*/<!X>AUC:<'KJ%#%M;$BS9J;\nM`5:F2<>-%DS-T7T9U9>MA6.:J]#T;@M,?^\"3)M%&=S`\\3_J.WH2[=_>NW=JM\nMQNQN%?4OZB&N&<>:@F5K.#U^8+H<Y[O<<P1F#9_P\"`[)YS#UJE>I&%%X$:SF\nM/EGT+1`\"PV1!XX+!'<?W2Z,,6@-%YWA%&Z&12E>=HU]7-+[YP54V^2/&VPQ9\nMVWMV$D512>]=K6;1+26%.K-7_$UN<;,M^0W9=FP?&]+'4/\\ZI5EM-_)MN\\-^\nM@R#=-MYX8'[7(EW'R?6<1<(8TQBA`00!;UF:+2\"#1><,PRYF+[L\\QW%$;/4\"\nMN((?:5X\\A_%A4L96.#IV'/5U3&Z,\\9XE9$(RX11R!B,O+:WBU\\*:Z,4P[GNW\nMS!?![CC@<:K@\\,C,DT`G&KDXW:>_!(*^]<M8/01\"#!@G/J#@=*[1T1=:\"#_*\nM\"?DF,@@C#>4XQ0Q#1C%-9C-D@X3BCO=[YQ2>JG\\3XLH\"40#_:%\\RRYCU)LR+\nM26%M._:?P()Q\"%`^E2DRZ'\\T\"T9S.=IR6;\")\\%^1RS1;:H#J?%RPA>P0%N'+\nMB)<*,5.?Q[#QB/XD'NPK<_65N0J?V^E7[NH_@+M*OQQ[I8/&:/*3E]$XZM._\nM#1OF!*;ZHFQ8^F?P89I#XNEV>22734+^\".-$CC(ANV&Z8;HP<)]$*R<>!4T4\nMXHGSD;01/N]DXJ<&@AIG8AI[D8QM(,G;0QIC<Q(3LU#(Q#7VOF6\"&V?K^1&.\nMO:_Y,,>9REZLXTRC7L!C[UL@ZO%Z/NIQOC-SP^;Z9>='SO683Q.&FE9+8YSF\nM5V@%</9QVI=F!-%T><:,D2L%H\\3MAUFG\\*I\"VK8\\I5-A$NO\\&:0IG557N$CH\nMFE<W@+!47CC(JRW-7%L:@\"I.GBDG]J&&#UW)P(_ZT$+W1F&\"HV%V!SC6OQF[\nM5,^0548^8MXI00-B@%U`MDCX5#2I*>L$(=@44R[*7N39[?:3Y'V6;S6<W*BC\nMV6C,<]-L%[&%LAH!L8&_M.$+_G[P@#+A4`'7>`2-3<P%TZ@CMD9Y4Y/;/+]'\nM'6/#:;CS0M@=RQJWDEM<_\\:N%6<\\R;A\"20OWS-ZXHH.#0YX$C&!UVWC<\\\"DY\nMQU+VY9O=(JRL[(%Y'<_8E7I\\B'E`ANWNH!LAT(TO`CIC/VG8B,_J90!4X[-`\nM90R/W:7]/&A9B^-YYL8>(@%/9NR-/Z]U;7B%HL_E$`A2UXO+%`H;8(#37_.`\nMLXV*!)(:7T?CKN2<@__((`VV\"TOA9KLL#IPM2`CX>3*YTBUPQD#T-+D5)DI$\nMQ/WAZ+A$=A?:+\\'/M\"`]9JB#<$*NF<D5\\'Z2K`\\C'?<P`CF%##U/NJYO+5\"E\nMRK9K3**;+62HI;S^;>9<&5LHM#BX3L20`KTKO:F[=<;\\]]G9F&EL2NW^[@4W\nM)@%B7847U93T$9BO`,[ZOEJ+-R.V,60`.W[/]I)DAN).W.W]MT80S@%EC64\"\nMAAMWG5QKJ8LR#<B>W+FR->GQK5I%V,'@5T`RJ!%[GKHX2H<SGMRN*6NAH,3F\nMUA\\/EK-VMTYI`F`M>*T?Y'RS7W\\6$,@\"^(VNA@71H\"6CZ!*Y,ES(E906T?/W\nMGDD-V/)EUFIE0MJ8R;K[.M^RT+-'B0Z4!;N7/F,C^<;,\"?X9N0%B[^8XJ$4W\nM+>[H;`O9/`E:BJ.\\!.;?R!?EM'R0K>R(;)02+)0BPZB[0]DQC,`62I-!3>HP\nMC)[,(>DQ3#H'!S1+&]U%%-E2:\\OAV5D+;3A^^(T=8I\\S1W6=D6%$>9?3`8N\"\nMF`B4[%9ZC0=33B?'^2'<6PO#CTO.\"#B@E[_!T%?2<5%4,\\-TC]6;$F;/J:&H\nMRHC2<R`;8I6UK`2EB(1D#D1SX$0E]!4PO8Q6R>O4-VAW7J21HKYP^=UD66TJ\nM>+>,%057]41L..CJ@GDW7`Y=PT06+8VSH\\X0L\"ABYI*H?$X.%8NM]L^M]S2W\nMX;`D%N$>_-ZK&0*3>_N%`O[=$6MUHI-O4L#`RK(9(V,B=TJPT2FZ[!1(1MGO\nM_6X%_N>\\*1A`HV0GR3R;J;<4C\\++'B%COGMX0%F+)!>*>2?<WS#^`.P>'%&]\nM(:^5H8Z$!A7K6J(U`-0U%.(7P$1&I5$V44X&198X(VXJS#C%G^M$?=2;C>-+\nMJEQ2R0=.]MR-SZ>7:YT?\"F_B?K^,?DAKA1>HZV8H.J&G)`JDO(T8%^P#&91Q\nM[4OX70&*0IH(G8TT@=6B+,4,)[)FOMPB,,F<AYKE!>\"4/\\3#'G1UK=\"<4(2)\nM;B_E$+.8\"INA4)+D9%@A$%!G,.KU=6KJF^OH!K</1>M(XA3-ZZ!-B@0R[5QY\nMJ4?M'L)9!*3N<7I/GM22(*R;\"4(IQLUB!UTX*\")V!PTA3[16BTPC!4J'L4ZG\nM?3!U<]]ER7[SNI2_BZT\\SR1?Y8$8BIPY<.2(\\8&-SJ<7;ZL;I^X1X@S<ZBWP\nMGDC)*9#*OL'*2*4[RV6-<_82/%/HW;MO$J>8N3Z'K]ZFLNFYG\"Q/:#%VH5,@\nMVA<Z%2H_Z1RH_.1D/N473KY3*2^)3OG))#CEQVB2:7<)\\]ZJ\"@AG!ZH\"\\W:9\nM_K._EHQ[E^OIN+..P*'#ZX3F^#<BX5J'B\\TKL;2ZNKH@L$+[:JK^\"V0I554;\nM3S<W-N`_57WZ_<92I5*YO26JO@]+7GVJJ@\\W'S[>K%7A]],GV`7W#W5I0]%/\nMA%PM5Y\\\\?:C@9X4(5X4WW8[.+Z]0YPN`:F7\\^R']_4B]CH?C&]4:8=`'G<U^\nM?I4EG8\\]5L?0=<QKD,^)#F]LLO4U1<GC\"20J68&)_1!WO;SD=+6\"&M1>)O>Y\nM2?!N$JIS\\G23OOW\\AD$$!L(?CF,WGSV1+<Q,CSKA9#KN<*;F\\]X04Y)\"?P<@\nM=]-U#R93Y[S(#&>0=$F@9483NT0YUB?8(1!L/O3P_MYX,MI\\]M#Q+N46D63*\nM6',02UY[55W+=)!F4WI&,R+*B$DDP;\"C<R#^\\$G6:$GRP8/0W>O$HL+I]SBD\nMK&U;R+7;,6BUTX]Z`[/NM7Q?,#\"(G1K=%TY3'_]1W9&PWQI4-^E,!X9VRIT_\nMNP0-,*QX+^JG=@7DJHZ2*YN!R``?KA$B1EU`TTDOE8`3`@&;@*+X$L/83\\>8\nM(69,N$+X#>/0R*A[QA>$0(3[T4UF-+\"YA\\EU/^Y>DO_FIMDU/>IK%]DX20-N\nMX0+J?XC[(-U1`(S<EA(8=F.5[3;`^>Q-4KM-DG$JHWZTI@[B'LT7G<!X#Y#?\nML</$?DXY/I0/S0P:QJH3A\\-&!!(&'V.<*AC7`-UY97R9G1R<09UY/!W%'=Q=\nMR$_AQAOCOAKR#J,D#1ZU</.^*_A]='SX4W.WL:M>4*YZ9$P;!^T6I;\"GW/3-\nM%R?MP^.6^N67>@LJK*S@)P:&.>8;?S\\Z;K1:F(B^N7^TUP1(`/JX?M!N-EIT\nMN<%)ZV&^3]IX>:KVFOO--A1K'Y:Q1>'9<G4Q!_U^XQBXC(-V77+08Z=>-ML'\nMV.!+:+&NCNK'[>;.R5[]6!V='!\\=M@0>CFZWV=K9JS?W&[O`61U`VY3FOJU:\nMK^M[>]Y@`90WUA<-Z&7]Q9X`HZ9@K+O-X\\9.&P=E?^W`Y$$']\\JJ==38:>*/\nMQM\\;,)KZ\\<]E`=QJ_.T$\"C4I02K`VZWOUU_!\"(NWS`ZLS<[)<6,?.PVST3IY\nMT6HWVR?MAGIU>+C;8F#01*MQ_%-SI]':4GN'+9JXDU:C#,VTZ]0%@`.S!I_A\nM]XN35I/FCW*V'I\\<M9N'!R4Y!P[?P`Q!E^M0?Y=F^_\"`1@Z3=7C\\,T+&>:'%\nM**LWKQOP'M;]@&>OCC/2@EG<:3,\\IRRT##/;=H:L#AJO]IHP_SL-_'J(H-XT\nM6XT2K%VSA06:W/:;^L\\RTA.:!EPXZ!__=+\"Y3,NKFB]5??>G)@Y`\"@-6M)J\"\nM08<O&58+V%=9!V=[%`I_+?ZE9/F))VN/55'3B9)ZN%[;6)=3?-U(2)(RM5O<\nM:[[8.6OM[+3H_N*>?MT'KJMDF5SB2=-.)^UUWZ+OY?(B32YON;RCR,_K.,/2\nM)!$Q5!YC:T[OF$RJ9^E-NCZY&8&,<_7<_S#I]I+\\2[32S+[M((#LRWX/SO$0\nMU'[O//N6=7#\\UOM@^+CX8[8.?X*3J)]KQ*E5\"W_KP*F8SJC706IM^K(N]'$4\nM(4%F7<&4KB&PT\\#VC&B&NWCY#)2;I$%UV4_.0;K\\$,$I\"!(@G<P,AH]8:&\\`\nM)V4,$%)'A4%Z\"&IH*9.T$G4+:/:G;Z-]\"T`XVG,AG23+I1-\"R@:$*H+T='\"R\nMA\\<3)D:Z*7FVE_%XG(REWE64T@T&OH(#\"DZ/%/[ZP5YY%TR.3AT1BU,MQ5W7\nM.L\"&P.K]*PZ7PR^J2.\\P+6TI4Q&C407K28CN(AZ=_CB&G92*LI?.<#HX1^,\"\nM8)CPIE&9EG1#(1L*K=)0!W2U`M+'ANROZYB\"@[`\"81`3[\"+P]96G)LW=DE;_\nMD^./D_1.6QN0N>Q;!GU*B4%A78JJB/D8IT-W0+H&+*,I+U)P*5#^MRV-NTY2\nM7VIM\\?2_P>S\"@';Q>%@H+.\\L:X6`E]P6@2@K`1I@!4^??1:/XZ*+[;B]V*3(\nM!N/+U3B+/XYRM4)EH<Q\"Y<YG]@+&5#6_:EY=1\"UHHC<8G<T91`3;I4LL?J!R\nM)YE:??XM/41Q=L&R9[`T@T\"/$-W1Z\"5<B0GAG:O%=ZIF-2Y0-;T9G\"?]^6,R\nMA3O`^Y_!@1<:F*Q/)U^3B\"S>(+D:06\\,>`E,@?YGP`W5T4MZARK#9#B,K^'D\nMS:-*OC`FS(SRBYVQS,.FD/7G7Y,DAV\"P`$\"N9\\V7/UFT6$0)TX70##,-A4K.\nM6'?L#E;YUYTKX:5&Y^HN58:+%Z>A##I1MSNGAKE@\"TS#H-,;?N!8-G=ID##R\nM#D-\"81K/JIRUIKH4E,LO/@J:<ZL(8>O4\\CW$X_`R&<-!'$+\"'$`/`.,HJL9O\nM0V!^@&.UEP9V3#>)@6T,PU!T@</9S`%(-U0;9RRX,('ZNB\\F398/ZN(ZA!^F\nM\"H/\\$/6G<;YV/.Q'X\\O@//)<0\"_RM8BC28?3R[O//V_/WK\"+ZI//73X\\Y<.G\nMTZT`S*5!MW?9TY<&P0)1?W05S2LP'8WB\\;P\"_>3:*S\"+6\\FJU3UV\"#GBA9FA\nMI?Q-Q'#:3^GR0*PP^E&'XJ2-$F(!,.T5<^;Q![Q<373&:F$'F:/6P+60(?:E\nM%Z(B([5;1O#@@B\\:*-AN%N\"@@X]BCRI:TDP%)6:J7?7+:$7=NW?/\"!SFPJ[1\nM^+%8*A171^AV\\I%D4O=;#1C;XFI1OCZHEMP\"^X?'#:PL7]4S-2)_HVP9`F)`\nM!(NU&@TTI\"XR2!25N6=D$MHI98JVWQP6H[(Z]RKHEORZD7E1DS?G'K1&O4T-\nM%[D'>-E>/&C\\O5TLH5F)VI3$X[:TTW;1[8JI60M79:!FLM@FP/WHSA*:$M6R\nM!8;%8:E0<(O`\"[?0JT9;6K$+ZK?3:J!^Y_\"X&)<*FEN`#5Z,/3C'C;^=-&%:\nM.TD9N`:YJ.XD,$2GOKO&)ZTV31\\5U[5G+\"66R=2E-9A1UXQI=O6#0Z=U`^*>\nMP/CU5]W^O1\"`QGZS7=0.HB!$R2$TXO2B)?A4LAGIBUS(K=\\\\:#6.&0*>)?88\nM\\B&\\;F!G*D4L\\Z#*91TH]=>-^BY]1!K!YP]#]*MZ=6#BC@^P#097*-!8Z%GJ\nM9%KAMP7&H52BZ^F/[=Q7L@[/EM#%G$(>HNX>'QX!6BK[W:#I4B`UK<O%D:T`\nM6K,4M!&MOEAB'5&ZI=(K/(!(G34=68V67+[J/A\"D@A;7TV3@JIE2#FMY/KW$\nMZ)Z2$\\D799?(1$%D6?@%S<7C\"R3RV\">BL6-VB;17[EAG6X1D.JZD>I$42YA:\nMO(S:_W0BO\"6Q8F*N8/U$.,]VJ[E3*&S`G\\!7V`L-F+U=_%X-?&_NU%L-JET+\nM?#TX;)V\\H*^/0E_9ZQ&_5T-M0^VC!G6M%OI\\!!W#CX]\"'W=/]H_@8XT'Y?B!\nM:*WEADZ.4V89[KJ7QF01@^LWN3)6C:/+IX_/].2.X`>JX]\"N80@33)8TB&MZ\nMULE_&1YPHMWIERJ.<1I5S9JD\\7DZBK9FJHG\\DN0K_:V4]V+.^Y'5/5L*J[=;\nMO>Z,MC+HF#.`,8HIU/>_W*N_:A4O2@4%?P6V0K80_/WM_^CU*&6:DDF`(=A:\nM,J7&,J[H^<EJ;\"12[7]B9,E$#2`,/?BI?OQ*<V]L;N=61#0R7B_N1*&ZD?[`\nM+*&G(BQKY3DLI(U5158<]!IU1I3;&PMGHLEF.J$CFOH5*\\JLA;;L\"$$7//J\\\nM%APDM-FQR8H/V\\P9W13TR-U\\OSJ[/:`A-%+4_7F`*>ZS.>W%^DMAL+,B@9)(\nM$(SSRICFY//0*^*HX1]OUJ$O,DL8.QKED!*T-3BO47/%XG2(9NO`?DIG]4:]\nMUI*U4C(STL,S&88NH\"??+2M%BCKAL1YR?AKQ7S(8)]Z89RI5<33NWZ!EE&2X\nM@/[email protected]*20QSX>[XD8(HRM?[WDGK@LOK`#^YPU'*S%\"!GGVDK\nM/5R%2S\\NYJSIQ\\.4M.04+U>M&XY$U3#2@OQ^J![8#U76?6.,#>.+.Q*C;1H9\nM2+!V.*8!T]/46@O*2:X-36D?&$A.]]F2E5-2Z]4147'FR)QU0ET($?K4'.[8\nM-J[#999H%C07O)W;I_:+WE^*$G2)15TGWF+/:+H9(R,4FZ_#X@^+*`A$L]M*\nM9\\7`3R3\"Z0G!(G+AX&>?<`R\"675OC8%YKJ\"F7`50.G(QY,6WJ.YWWM'NPR@%\nMJ.%A1#C8V9*7=*GA1([@5^>]W-MA1XINZ#>&XMOS#U_W]&M3$,,H>,^Q_TRF\nM&WYK`Q=I)#ZB]ZR53I3VMRILG[R\\`88`]N,RLG(W?/SC`NO8@-AUM_:W_\"(=\nM`9?VME)D,\\3F@7B3#>(!:B$M4MK\"9;51AJD,[U`H)^9S9BXL33$Y7HCM1C]+\nM@&7JD1J3/6:W-6]M=D_X#.6UIQL28.&-/SN=!;E:<KQ*';SW&&4,$?F6@R$Y\nMX(*]Q=`(@<[*8\">][@UNR7Z20%<`,]G<Z*+7[^L-J^U[\"XXB$9HVBB*MW++O\nMC,+*OM(QFG'7&764]WD07?9PC^W77S5W:F+++(>JH)<@FO_-T!#[R@-5%?HR\nM]%DM[7QE['K)HI9C58YC.$42,2%C/Y(23H-8_!;-A212>;.COE4OZK+21I;&\nMY:RW4(STSS#IP#4`3=!G+HTQBL+*F,,Z..`-0:(0(;R-L+AV=-$L,U%FG(\",\nM+:I4#_D*$#K\"OX`26O!!OV%*KT0\"#:[WT%B2J:B/%-=U);CM^B][^Q?T2]&;\nM0@>?R(\"P;+SKP&XOR<E4\"H>07E&>1J3LT<1U$$<C_@*[@.L+UW'\\@3PK-<5R\nMWH-XGGV-4^%P_=9'Q0;:X6@[XE]G)NJ##M!A3HULS!\\,5`ORZI#SI]HY![XJ\nM_HAVT\"E9`,IBXW=HV*$X.DZ]5:I@-!O1BI!:9.77%3;A]M,OV+M7'2]6*U<8\nM.A;'YLI,Q_:/VC]+H.=Q_,]I;TS6Z8[CEN&B[Z'&!QK5?G*+!1*A^\"KD$4./\nM/#\\%4<\"@`QPC9(G.>XQA\"_P$S/GU.'$\\XNSRF17SUMIY&W*O*8C6!?T1RTK7\nMT]$^+!B7XA=$N\\,ME[1ZH_?11L\"2WNK]:OKH0F'\"?7A<(\\(M3FUFE'A,>D/5\nMC!7-MYTV-PT'\\+^5F`CYQ^G((%\"FM]ZHSW9>9T>=\"3'GJ-U0*6>=3D*$!7&+\nMY\"`R>`2:!H2[@49;$5\"323)`O5;:.^_?B\"$HJ^DCNBZ=D.&K2D:SZ$S`:.`6\nM\"B.H'J8OA&[&X<<G%4GJ;'TZMK+/-;<\\C'*8.\"6NH[03`35V>0R93I[-DB%I\nMP&1B.`XF8Z3(BN&T&\\?B.B)G,$KP6F^J`6*B!`^A=.\"(3C8_R4IQ11R)?76L\nM;'-B9+4+VLA$OC8!W&ETS#B[9[%D/J\"/FAW6_J66#:;OIQG,%]1GE^BRHC(E\nM!^@]Q+25DB$GEHG\"E[>U_LGI!'+=X2Z8X!E^*3IQI<AO;F>/0YW5RF[H5F@R\nMC>NI<;$Z;#7_#GQLJUW_L>$L4&E%A]OFB-DIQ^:`(<+PM$_?=*A32!0=(EJ@\nM#<@743\\G4[(^CM+WP%E<L84T-_V#4B_B3H3ZUR9F^T6.ATW0\"=$,\"$#M7E^E\nM?315AQZ@0K6/;\"0>9-6-C8=KM;)8(0Z-D32I]PP$+H5LT[0?C2OV1%-('.-K\nMC#3\"+:-!-A[;L`?6E&HE!@10\"SO8$E+#?GP9]7GZREBE'T<4\\QT[*QD$,:U@\nM;V)@``UD%PIZE,/\"8]%FK941&65Q_J%W#R/\"B\\.]LL41WA3\"#/ZZC3:K-FC=\nMR`2L,YO)(0W5;,NZP?M^@XW;&FQD&XR=!G/0?]70_<G@PWY&G56IPT\\/O*<?\nM@O\"`*SX^:L\\$N+;B9$\\9S8[7)WO5L;TQD>Z=G`T\\3_6#GYUYRC7Y5C=I;;%F\nMIAA9>?=N+LELM';J1PU=.T>>X9TQ,!K9\\&/Y9CZ9?9^\\IWUK#).`$^S$(TIC\nM0+?P,M*\"(#>=GC:.0^\".[)YS@R^<(7?>7QEH^F5];Z_]^OCPY)4)MZ`][1<:\nMB\\>8R/E&9:R[:\\'RI_3(1I=X^*E/_O$/0'S6@$;@R$;W)-;CZ@HY]-'O!\\[O\nM'U9TC$>EI.BGE<QU<F9JX*T)\"2A]UBS5,'$Z%Y#6_\"/9+(/>Y?D9GWU(KQI<\nMZ`U&?7**P5SDJ7I@T\\U3RA7T!,,:Z`5*TJQFSI$6%V]^+2F0SCMP`@R3'_=.\nM=E\\U3'7-6E-DF[*UE7'X0?SD?=%U.`A+N!)]\\S[E,%W3C,_J0PZ:H3DP(3S&\nM3877O.KF!YPQF@0YR:;GDS[=#G+:/#H6<A-\"L@8UZ##AYA#T!0Y?8-!U:%D`\nM+U92245DBIN;8$=.H/`FT*@IE94%;\"^L'.!&NF\"H(I1(!?B.2G``[V2N]QA]\nM]\\)W'DT2LH9\\+O)]8N4Z<CDOXGK*AD/3/)=QN_4(3R[L\"+/0(=B6EG'SSU\"\"\nMP\\)F*[TX-F4SB211Y=P;7L)RBX$XV=QTDL$@<N(2F;:;!R\\Q$M'//K=G87'(\nMGPQ(`\\>`L>(!4@\\R^]0W_?1)_JEY#\"Y-WV]V^C2]8;7/53P=]U)@D9Q(/3F9\nMWQI!(\"`].N\\0\"A]>+XZ-LCYP:MOIS24W^@Q\"_^=1;/P]BP/)B:H@#E)<3O:@\nMP`P?\\22RD?!*:IEU8V@9GY=%\\T;I\\^105N4NS99!@XMDV3%OX7G([H%L&(_P\nM2,])V_<BJ.W#>$L)I\\XAVP=']Q<<=][(WK.Q]T+5&&-[X%9>]2B<`I9\"ZHQ*\nM(/26`?J,#BLJ1N^B(46\\0;J(Q=:QN@+VO_,^YP5M3D=64L*;2)WWR'TV4N_[\nMT^XEYH`;L@U=!&=AU*/+P/M8E/0F#&P2O8_1$@6E#<-U616C^-^RU\"*NG.<4\nM6)NG#226MKC@LK\"&/O9I;TQ1(C`JFHHO+N+.A%6%T!5@Z_D($A<E.MN=2!D,\nMRMR6H/OU%#@%LFQ?8[]8'$#4P]M.J#3HPR1N$0-PG8PQ#EKO(_ISZ:F\"O@Z(\nMA,+$8\"\"J\"\"TDR=&%!\\/>371N\\:S1=>E'M-CH?8C79N*S7$+PFM,RAW%;')\"@\nMF%9V2\"`R=%E%+$/7;S]8L*E2,W%!,<9<=UX=VD)NB!9?T`]J;_D16`,KG&H>\nMRRAPNL\"%[email protected]^LN%$\\`U+A/*$P)Q/^EMODY%@HIHC.-#MMNOVS'BRC\nM,H],>N&K.TT&7U,WJS[$:%IEF\\O7V3J,\\?HL(HLP':$K(*+.DU!S`JH^4_**\nM:#$Q7TP3G:?P/\"=&&8GG.!(.=2=MI$\\&R6AYAK_0+'<A;P\\9`RF4?R(F=Z9;\nMK/5`/6S:CU+4?-P7I/16V&MAWLYS\"YI+BP@U)UVRO0K3O1^<727JSCNI0GW5\nM:2^H)5URS)DPZ&1K#P[^ZK-GM4=B5>CK-F7\">%5(0ZQS?\")KA>[Y[#]OHR@B\nMB\"^J;^5+7LXQ\\&[%V8BW:0<\\$P>\"*P/^U0V,Z[5E[9/\\XI[M3:Z.9VTT6\\C\\\nM-]>[F-%:*>0.6B5;VRB\\_]?5V9(1'/%#Z62IQ@G7\\/DV/^5%Z!!`I&/=MU6(\nMRS%L/_T;J<7UKA#K>MO'N4KRW\"J25AR9`-ZMR.F@DI>FL5+AX\"(<:,.<31D(\nMOX5UG0NV7W7UG/9U+?SZ8?CUH_#KQ^'7WX5??Q]^_23\\^JG9Y&@7UH&M_3_\\\nM$9,<K^@,+WJEK3&3BT*.R9*3-?V3CR24*]7N(X-#PA)A<&8GNJ&+6L9&RD4L\nM)QKIR$0C=0I+--(]K[.S*G'GI<JQ7X5<VU!(=WM\"F5=MS<Q@3.[WH+8@)T@W\nM6B<@<VEK42%#CNG13-W_:E;O[![C7T2!ZZ##7%4NL;JKFM6U;.),[>3O4T[.\nMU`O.5`O.UDS.5$Q:6UP9H4.>/GE<_2SE5U[W]2E$N6?JP/0.FJ,!6UP!EE-9\nMW:[^FJ']8D@+ZKY\"JJ\\%-%]:\\>7,N=&`?3+H-5,#-E-$RL+**,%NUX'-5(']\nMEL<98@*SO!B(]B7L^WU5/)_2O:IZ=[_$AG56%ZO-LZKFOD\"_V0CK;A@!K`3C\nM7)/0IZ6`=!)R^0QHIUQ)!'<FN;YJ:4,CY-)LG97#_ELS8H[405COF5?F5BZ_\nM.\\A:2>/[[LF1Y%');$;^%Z^8T3#:>I+1J29K)NW/$\"]U[YY+C$N_R=P6TPM$\nMY68IV(BG=99)W@AVL*H\"`[PMA75JH;`0>L7H+W6`,<,B+V(\\!@J'4^_&1%[#\nM2'B;Y`A`]HLZ#$`)[_J-EU^9X0TYP1!\"B\"38G-QJS=/Z:-9]#E8X:5E@M4PG\nM1JXJAKWM-7KX)N#:\"R(=52EJTB<0'F!G;^)?S\\RO4_Z+`J.X=6JY.L^#=;+V\nMF+MXSPKSV!Y/^S>JU<.HDR@AMS#D6M2O[.!Y]B.IR%(OGHIX^PB[)$;ECXT[\nMK&\"Z\\8,8=M#!B0MBUG=@-[XC]F3#=?29\"RX'9?GMYK/-T]/E#\"@M#?H:JC=6\nM3&/WT^\\\\6WZS8;[,\\&I?9GC/%QQ>8^'A>0R.UN4)7@+9$*I(@8PQ4Q#J2@5I\nM<==$I-MV+J0)S*D&L_,:0U%@R`E\\MY7E-XHKE4#)RDI`U9^YXSE=\\<Z\\\"B-T\nM21N,2\\@8\"D%A51:WMF@,F$Y7G)/Q1W-\"S;)&!F*!\\8?C-'4\\/\"ZF8XJ(Y9UW\nM&4A9A0.Y-SJDWM4;Z5.FIV%8)PR!I)TG*L!08][$;3]Q(K9I,LST[-$C80MR\nM/!E\"=N+-&(Z,#GY*V-@S%VUV,CL9H8!*I^A!T9\\`9;[G^@,5)'*(LTY&,<M(\nM%IZ++S!ZW?.=UR\"JR;N\\2L<.J^>P;;=JBBQ03,+IU)PS$1)W)3,5^*].`VCK\nM:J\\D,@7^^/&CU?-A(Q@OAL%@P6I.L3V:]`8X4<S.3A*,?#8QG*4K'>E@-0Q-\nM*](E4DYVT<RT&3W8X4M**RU!<AP@>;:!]JMG*TLOZ!;K;AQ$*%A4-A3-C$/=\nM)1KA<SV3:>V3>^9JW;!]9IV]H!*'A<EJ@^70I9A3Z*EU3?<-\\<J'6%TFFE?6\nM\"DRMOE4_:%*X:9,XN'I-HV2D5+ER<:RKU4R]65)W):PG.JX?O#)2@F,9I(W:\nMCQOMD^,#(Z$X8'WYN^.EU@W(V[,5MIO&*LBB`LV=:55\"3=PF[?SH6JLY-_2.\nMO$GG3,5Z#MA#8:?]LS5W<T*-.3MBD::S>D%AS4JA5G)KM&UF`C4*(+\"@Z]__\nMQEP<[NW5V]YLQ%]D-K8SL^&W$T`NF0X.PE8.7.;@.S6.AI>Q3M9`M)-NM%&*\nME,M@/%>(SC(C/9E>7)@9U5>J3K0WSPZ(++6!FP@$;*D9SB5SSN)ECNU43F`/\nM<\"Z&FBA:%5W0.;?L]U!/\\]HZ4]SD_I#)(7K$^9`=&3\\EK;-K@>7JYTB$9(BR\nM>)9R^#=`[*J]\\[K?X8HT0_0H0;3$=9M\\MV<3(_;>=BR5+6_`XU$]VR4G84WH\nM?,^0H\\!)Q9O=D7`-@E7X\"T7,ILBI)A=P;]9A%0Q2N.AQY9&=NQ]8%-YT9'V`\nM:=`\"0P:Y*N[OF?@.!9,59FH?^?A;\"C'N#IOIZ?\\<1!='>1&%*B`QT6M:2(H*\nMP#W\"A$,=+(5S+\"P4OI$X0!YVB?4=262P'!G93$-!X:R,SL\\L4A&)DT]OX>TI\nM7W/R(>L'2_A]D\"RR6HPC%E&/+N]Z;JP2Q&.!<L-9C9J_0=SC(R#R37%*H2G:\nMSJZX1<YSJU.84#]'E,/AE^S2:##,F&Y!15UK2TTQAI.$,YA2SD.&M+^#D`AK\nMRVHZBR&,,]MLZ!YTG['3@G$]\\SMM:9YI56^(85F`9X[7D#K.V9KQYVW-S$;J\nM\"`&WL4+QCG-;)(KLB@0FD2E_B%;A6?AV;>T4*+HY[(2?0G,8LBWBVOY\\S@]V\nM&M\"M6G98#CM6KF+(9NZ<J%F=8RH\\7QEFF\\#HB;J-O?\"4[:016UO)Q&[Q+0>7\nMA\"_/S(5L-1Y\":&W69&VRW,S:'&Y&]\\`&7`R=.KH9=S%U[RHQ(ZK=$)2PL8=A\nMI&8LWZSPL]GHL]Y:AM?13I+NAEE2;W((9GAMC:E;1TL4-!)XP59(^*6,2A]K\nMCN,G)YMUC'$6:SG%N!4YPX('E6-GUG$U2AG;%]?<]U.`]KJ,K7^KP;S7[8<=\nMI=7X_]]19VX0\\.1)NC%[\"W!^<R?A.#9)<Z35%Z8BK+3V')#;N(P!9/84U!M-\nM@LOW')7<W/LFH_BB4&E4$*\\2Z`W=?\\3C$;*ZJ*$8*N)MSF/8(X'M%HS9/'M?\nMB9NC;:.L.Z!3D5N`G7`Z,&U:8-FNSI43E<>)1=JYRI#!2<(Q2$T%HZ]U`I0&\nM:EEPIA;-IAK%G6F_%X'TA9>!';$T0[?.[$V@S$QF(4P\\:_A-3BV1ZDZC/BT$\nM):-!&SA4$TVNDPHOCVM#G.$$9D7']B-=:S;@17)3-L$-,NP`I_X38V$0A68P\nM`[;!$:5)FT?_G!7TJ%O2[^J0/PZ)RQ8Q\\7PP-X)_5HH&[>U#DQ!.,,3#)(IW\nM`!.`BV9NN$KJFI9L#,LX-C$W;!`B*;<E;[D3^A;P@6)K2-W^!EY)\\2C-NRJ^\nM`TKKO:R=NCJBG*F<-9WA7G@MEK;\\'LKL^3WD^0KM>ZTVT-@6,@L-(-:,$.I9\nMO,HCB&>-LC!^4!@?G976ZL!-S\"\"]SA33;N9%0T8VZW`6^A!2T$;-XK*[T_/A\nM[$0/C+&*RJYYIU3\\2T9?+.4RM@2FL#E#)'_M<U04<!0D.GVN4,)WTS/\"O+SM\nM7)V:>S*</_U*6VG9^7(NR#,(88T^-4HX^GZ7`J'&.X`8,P/EC[XTJ?',4Q?A\nMH;\\H;7ED:,O\"Q.%AGCBL_&,E0!W>#4/D(4,S'GX^S7CX^V@&6_O`#QUNFS/Q\nM(3L7P0$P)2[6=9QE8MK[$$M4`U(\\!G`GF#=AH;0)(?RP1DE2DVM!C3\"RN&XD\nMFBTG*XEAUS$QYYQN.@7/R'5=P4*NO3K%]+')W5P[*JXP28RK\\23Q\"J%A1P\\S\nMS14'T<VY47,6R/];&V^5\\CXP1K'I1?VRR6^@?LU]`:`*ZJ'[YKAQ5+R`62H5\nM.+[HZA.TMIEXL8K5?IVB`D,!_.<9LJT8+WN(-RWT:GO;Z>4/^!N^L\"VRB;`T\nMNG&I=NB>F7R6DHL+[:H1J_%T&%U'-QJCD@Q3E3GN:>ZA=X(DU+K<M>`P<118\nM!*14_`FE/!O`B\"+6'A4WT+)@$WD[S^880T-?NGX#XALC2U!AK!-.O#M.1APM\nM0Q`J<&OD-ECE!J-4??Q4+5=_,V:6;J$#K]`P7`BFOL2&U%+N-\\^A_(OY3\\MH\nM\\[&:7+5Y9DNRU6VU;)*?9/RK!:9\\(:=G=U(1<L;AV?>D=BMF')_O[!F-TUDU\nM\"P-;_T,OZK-\\Y$RG$QQ@)@AGV7Z@A:M4?_LCE^1.$QN:QC_3(1W)`I`O;:/M\nMH(GL*U2K2EEM*$!5MFV!1_;25J,;EA%,HQ\"N\\Q?9WS,/9D50\",ZO&\"I[WT+-\nM''B(\\/'3H)+!A-!,E+UL*_D!$L&K+#;.@^PX/WZ\"]LOP[LOUX?;KRV4.O\\@2\nM\\K)I.!A-D2J0;7)OZ.^\\N9=8G!B!?UCO&Y-LU^=#R(TOE'?)2[LTTYH63@9T\nMS\"79'VW(@,&]&<7LRXS'!8Y3G!9MOH9XOI;0,++>,;EMC\\GW<3RB.,C`V$^R\nMPS-1Z&R8VSC#_6$\"%,VW3,8WY.L\\EL-M2''U@:^[BOH3*PTSAT@5EWS5DIX)\nMLJ##$T^^L?KT.N[W*Y(U,^B[J<U&]<^(C5]C3I!L?/2)V7Q[ZJ_=W\"18`54]\nMEX=GQU9UCAPA%LZ.]QC98(DHI:\\,)=Z@>3R_F;!X6LA<@)@BG13M94VD9]^\\\nMRU'D&@,:,GI*4\"1T0A5GS9XHIV$TE\"R527\\Z(*F-HN0Z.&$B'3\\0<?)%L\\T;\nME>)..B7XK=!<^/B-*<^X*+5HO(@;';5N2ZS\"9/.H'8LRCE[LV86Y']!\\#!?(\nM1+4&D#:N+WS2AX:]V`]7!^+`4=5-C'%33%LK\"6^M9C7A=9I#,8?[+6&:B^B\\\nM[_2<9B7;W4_SZ\\[N-A;T>U[P6BB(-0,+URG%MZ6P-#`,W0[*&LDUC<&0T6R4\nMZV'B60_8:7O;.UT;3<9F)VB@>,N'&:4`B%U\\UYTBN_JBYR?%C`<I8RZ8B?B<\nM;54F&(@&M%_24P-[;:-,;\\(33UM9PH%F';B.'`\\1XSF,X7WB=)0,01CI]3DM\nM.(J@&-.5;>J-+&\"<1QQQ)#_NC$5C@2SGOW7F>9B<\"NVH/)\\[X4`02KDY)Q/*\nM*,4(IE7U[)F4^B90ZBHBJ7'#O$BUZ:7SSKS28<%=TB\\WRAER+L:3\\@LH.>!<\nMA3.<XHPMY74`X<2$BYB(^(::84(>NH1VC8[-=PR4XB^%2_-/[T;`L7!V;R$]\nMM)LK9Y;+UK,[email protected]&<WCGH\"`5,P10'+C/D]K-2F;$F_]*KTAN\"%#,:\nM)Q3X`XY\\#/R</V[S/-*L9)\"S+0PP90,&9D%\\U+[0@WA\\2:D:NA@/!C89F>92\nM<!4.>#)-IQ0E)L*(\"1,=<(65(YAC>'*58$H$G`A\"85)$#8`_18M?#K&`TX6*\nMA)+<R``'>-T;,J3SWB721*,AQB;/8^P18NB:3<%3QEM9A'8=L=Y2Q]?$5AD6\nMO\"3I;TKS&'6%_1I'=`)/KJ#=CX`#%6)L5+VUTVRJMU']E*V1\\=,D84@4&`)O\nM!XEGNI=%=Y??Q*EV]4I9\\^/%=P$<.@`2B8\"F!UG6XVZ;PVFC=O?=0B'/D_>L\nM4QP*<SLF0^GKJQZ<CQ)].[67J'R;G-8R/=LXW5+X]AEN)/JI]QQOL9HF?MOJ\nMBO3H4!;O`%+/OY,8VIOS>.91Z6]G<7>\\)P;X-?(_`*CFC6N2;^)E<\\1!-QA`\nM#[N%!Y@4-9&UH>A-[!TVEFC(6+,,X04J9I6CGP@8MO,!I%=,@]4D'N]S2T26\nM*@X+%31Q%TMZ>YG,X6B\\1.=,9XJ\\GJ1:HX@<)KXK=+44(#S!A+)SSP=WPY@>\nM;\"'^2$[%1\"UC0)EEUQ'2]P7XO-/D\"QX087<.LS)\\>]3S5/^4-\\TWN7'DLV@,\nM#,08K_E\"8MB0UL[)M&XE+[UR^:7))^Z];5G@8?AGS;&6WUS?U,^9\\Z$(?'I.\nMAR'\\IYS$*,/\"WU'`8HA$ZDZ8`PKE,_;2&9M0:UY>XQ!+Q*#$]+`S2[K-3+&Y\nM\".M8V5?/LKD;<WA$/9DNVV@M(:%5U-DY3$W.K4<HC\\-@BMCD0/0O7_W26E\"R\nM+\\LJ5W5.VTSC9PL``7-2&%QG=.,\"?\"#W9:SZTAFQ;(&W[O1`&??V+H<\\F/ND\nM0JI>W'5!#\"(UVQP<8GXRF`%;NP,$T$5SH8PKNK[&\"@\\E!/LND!48=-`V5]?D\nM69/TS%Z6KXN1N8@Q=S0C*PS1>Y#U!L\"SP8>R@H^4J(DTP/JX+!0*_EP6W6(D\nM`^I)+YECS2F_K1R3.(,7J46,#6V&+]FO''@B$8;%H4+!PQ6#+)^!K0$G.'>6\nM$++V@;-8T\\/TG13F*X\\N1+5Q/7_(TVVLN3BN,-F6.HMA\"D]*T<>2!09$!M\\5\nMMON>.RA_3-P!T]J\"X^+/-#2WFPMN@U$.H4)$SDR%@^]T!HV$I!J[\\I&U*Q\\Y\nMU'3DV)4;4T6V5\"RK4<FU_K\"-C3PZ5LQ/NHM\"Y.I<T3[/R&.C[V'0[C6=M0!\"\nM@L3%\\PXFZ-<V,!N<D)(,.];R*-\\\"H^*E%Z.0V/3,1I:,%!9?Q.,QA_\\/'XJ>\nM[^GB;,>BCJDY<BYVE<@,9`TJ4YU&X;-GF2R7_OWFV'%T_L-GN)=BEN9)RN2/\nM(V)XDL;PAJ3L`.W357.I(2_%DFV^.#&\".>C1)06;=6AHE^BD8,?M)YEU(BM:\nMT9MF(NGGA&[A6N$3\\;<V[=\\#533J==0/N>H_JJE-RP\"^RD>M,4X7E@?&&SS,\nM`FCUBZ(&QL:)+^82#ZB(YJT=83KIOYUVO)!A?M2;^2;(*,O+*D8<,@IP97*=\nM9(2/^\"/\\%N47B?\\S5E:#F[FR5?FW=J<E-F!QB9'3J]VVSE7G=^U/7/1J:-6K\nMV5*U4*G:'X4<U5/.,H:_:Z<9)-GPD:0:0A*;D!`Q)D&+]*FK5[`FC@%BZ60S\nMS-N895=PIE\"5S8D8)F]!C*\"5=VU9Z<(G9\\D:BKW**AFNB_\\XFC*.Y3R(Z(:9\nM+WG3WF0JT7VN$VO'G#>T<A@#;\\V19&B#TRW5<:U-X=%3H<[email protected].$3P,]\nM+1H4Y>'F+5'I(\\-1,CJM92,4[9`0\"9B9Z4G-5[DQD)KMC;==:;<:]LB6M2T&\nMS`?0^,&UQ1_UF1IPK`:=QR]-1AF,8Z-%,IV8;]#(#XY1A8=T8BXGZMZ)^\nM;E#9L('&#,0,F#EF(NAZQH3&,D`L%,F^P:@163O`Y3FQD9N3C-^S\".C;Q6ED\nMTU(@?WR^[1K..&XQ'J*J(OPK\"H%A/QI?QA+.D7,>DRC(#`-U>(#'_W08?QS%\nM'0I;**IYK3ZUMW-<G^_54Y8G\"126`K$4)7YS!6ES)C_0Q>7B46$B@T`I'EG9\nMZJBPA3D)FA\\8034W[BR*)H2=!DFY18[PE82L][E\"$\"\\IP++([K!\\78^!;(*P\nM2!?):1P/U'D\\H72<B37@YAFG^.&#J#-.V$DU4A>P3SBK:,J08'ZZ[+Z\"<3&G\nM([Z<Z4U64K;PP=1C>#N#MI(8:`A7L,\\!AQ!-@4%@.&DR(`^8(8;GFG!J\\&XT\nMB03SIV/B8O=)=X^IS,:S.%:9$NA(0O_#R&?N']@0\"?_6:K$1&TI_^F,H,AH6\nM&H]?BGF->\"JFI*DJ^D94935(8(HPR2Q,6,G'<NPIG-IT'7QXU'K=?-DN!9M!\nM*VYJI^L:.IMLX8B@LE'L.>_LQJ*S':O(I=1@L1Z*#=>#QQO?Y#<?@GQF(3J]\nMHEO*/EZ\\(7[@]1GP9-,.9S((78*ZL6`1[H,'2.5;AT=%L[C!;80GEXB_]\".B\nM'0$G,#=#,(/;B<LOM*$TP==A2</(*``==.1`IG?'23\\TNTNOTZ&FTE`Y%[C]\nMCT'E=.@?$61I:A=%SIINPB9S:5D\"M'MV2X(SDBP`UV)HS4)3MI'@Y4^'I^Y0\nMND1:8.(`TT8PQ212^'B84'1(Q^._1_%ZYR1]U_&232AAW!:)?WWH?M<,CV</\nMHZ,MSZC+7[V:6KFHU:7FR/G6(G^\"`D/UM*S\"'T]%I:IGLH)MDU@QZU@R-5&E\nM&]Q!F.BVHG38`*3\\R?@::#WLV`L@W,-.'-P_4&T6ET0197$?67_Q\\)9!&#H\"\nM+9>=RP#)AF#L-][U?Q`%%_3B^\\4`^<W-KA?0&EZ5U*_*=#(SZT)VG5_SB)4F\nMTK.84DU[0Y/L$'A-]>=PF'+QIDE.05S7]9SHHT0S7?8<R9\",$5WL`08X1G%Z\nM<K@C(7PEY>CH]USTF$5!;!^99:(&N;LA+0962(?32]D&>'\\S;SE,^=\\CDEH@\nMGR614@3W2<2&FW)D;LD'/0'9!7!M\\60E]\"$^8S$LL#LLB=,#W9!9GYS-P;!+\nM9@3X$WB$WI\"M\"HB%Q'X17]-/D%5$Z^QA%[GC&]7OP3F`ID>4$\"RP1!KPXBNT\nM9-6M41_EW,G50'S!X6B[B(:=&VU<W>^]QQ`64?_F7VQR[3!V(.C]RJ`NR*0;\nM*#WFA3I/XW].D92FE#AW$'>C&VTS1090\\*\\DEB'?M1[\"&B0Z]94HG.GX@YF@\nMK$J49LKI.@4=IF1`_A1BD#X0U3!W%[&&W2D9BH.$/1W-XJS-]'T6;EK\"`4SN\nMEO]&HI-M^*^'\\77N\"](C=!:-\\RR/Y_NIXS]QC3^0#V*NMLNJ?T%)\\@E6>G$I\nM!YC#>7/W[:#0D(W5<2)?LFJOF]AMA=N9YDUS#MK[#4Z7-*N:26/N`/FDF$X`\nM>^$E/B3;<^D)#HTLS:]M>5.PX\"P#];5BTA[(4(SFQW4/D8Z0A\\VFSBI`3@B(\nMC!QXEHJ[EDQ2B5-#;+JOCMU7@78X>T\"^(>WIE[[O@4S5S32T\\_I,@\\2AL94E\nM/O'D%YQ<ES3T!WB>'^P6[?VQ71KSAKQPB$LH\\=Y$?@DP\"V0R=/8:C5$,F4Z<\nMI)\"T&+24B&:2[P\"57LZ[G=?P0I<OV`^'QS6C\"#.=+7B9NU[4=TWGO'OM@LV]\nM^9N2F#*W=L,,'3$?CI#)U9A,-_5H'+^@S%KP<D<!''.0\\3GN!\"11>E2?I\"T*\nMP^_BBHUJJ#'4C$I`\\\"=-*YS19K:@CU',E>?GHW&P:V19.08'9OOHT;(E&B*W\nMT]4<Q2`E4&_\"`JD;W$IR6>;\"D].(II0'SD0D@#-;S&N,(LKT\":6H53F\\300#\nM/WR!A9B%X@+A*CJ(@1XVU7,/_X*Y90&V\\FJ+53]PUG$V=ST/=EDVPGP:Q1\"2\nM!ESB:%?7B'(\"BR(FNP&3>0\\XBV=))R\\CDD8=5R$9PJG)K+BUU!@!7Z\\'^<!@\nM(Y=9Q3!&[B*4/()`@[!PMK=G`=)PR#!`9H[F#/@\"=L;O$Z'H].#P3X-WI:/^\nM--6I)?C?!X#MQ`8L!?37IOAGL:>>\\EI:`7HVFLB^M]\"_)&\\@![N6\\Z`-;#K+\nM#@RBC^[K6T]N8SUI-NP@8@<03@R*&]W8L/PQQS.?BD)Z]+CF':/DC*HKT.CT\nM9#Q7,GYS\"MCYT(6V,BW9<^X.9,]OUDYJ\\*C1,ZW[EI.OM,*7O\"JU4CEESAY-\nM>\"M1VNGU*D3G+J?1N)O%2,J38A.0\\!6_X).7^4`,EXL=BJS.D8\"?;:L3-^:*\nMAM+Q::2IF_LN!%&&E.V2!*;_G5V2V#8SNY3Y/K]+'.;J]W9)@F7-[%+F^_PN\nM<4BNW]LE\">PULTN9[Z9+\\.^%JG14Y4!5QM\"A])_]-9#O+M?3<6<=KS*@Z/HX\nMOHP_TM](2=8Z7&YND:75U=5%P17:5U/U7T!?5%5M/-W<V(#_5/7I]QM+E4IE\nM@;:H_C[LD.I357VX^?#Q9JT*OY\\^P4ZX?ZA3&XI^(NAJN59[JN!7A39DA7?D\nM3C*Z8?=\"F\"^$4ROCWP_I[T?J=3P$(;LU0M9-;E]NJ4)E\"@5T&3J&3@^MD=3)\nML$>W'YSZ9@>$4#C8A[T(I-\\ZW@H@2/+1B\\<?M''2DI7%M7#<C<<`1V*SI,G%\nMY!H-3O!8'_?.IQ.^5W@1C]_'_?A&G=\\PB,!`^,-QW.VE7)6NN$\"LFZ8LOR73\nM<8>E\\'..J07]':1EN<GA)%W`U#.<0=*E##\\<-`>[-,+,QQ/LT&B<?.A1[F9D\nMB7$N.&LM2>_:UUONUK#F(*;P\\Y@N:2W3P93OJZEG[\"^%'`1@/KEG8?Z@\\^1#\nM3-[\\-*&\\'`J/NEXG+LLML#%7TVWKG-E.QZ#53C_J#<RZU_)]@3:=J=%]@>'B\nM!<\\?U1W.$!YK4-VD,[7AIZ#B>B)A#.&4!/ZZ%_53NP*T=G3/Z`Q$!OAPC1`Q\nMZJ*!88\\\\P2P$;`**XLN+.**[2<0!Q!6=R$DCH^X930>T,^I'-YG1P.8>)B\"/\nM=\"_IYG73[)H>];6+/),D&[=P`?4_Q/UDQ&%&<UM*8-B-5;;;(*+@RZG=)LE8\nM5$;JT9HZD%\"3\")(B6^9W[#\"QGPD'L]#,H&&L'&6G*^&62+L\"U6%<`]1,R?@R\nM.SDX@[+#5(KYG\"XP0.^XAQMOC/MJR#N,<H![U.)ULZ5:AR_;;^K'#06_CXX/\nM?VKN-G;5BY\\Q;`EZ##8.VBU5!YYOY_\"@?=Q\\<=(^/&ZI7WZIMZ#\"R@I^8F#U\nM@Y]5X^]'QXU62QT>J^;^T5X3(`'HX_I!N]EH82\".G;V3W>;!*YCODS:FEU![\nMS?UF&XJU#RE0\"H/*UU6'+]5^[email protected]_45SK]G^F3KULMD^P`9?0HMU=50_\nM;C=W3O;JQ^KHY/CHL\"7P<'2[S=;.7KVYW]A=*S0/H&W5^`G&IEJOZWM[WF`!\nME#?6%PWH9?W%G@\"CIF\"LN\\WCQDX;!V5_[<#D00?WRJIUU-AIXH_&WQLPFOKQ\nMSV4!W&K\\[00*P4>&MUO?K[^\"$19OF1U8FYV3X\\8^=AIFHW7RHM5NMD_:#?7J\nM\\'\"WQ<\"@B5;C^*?F3J.UI?8.6S1Q)^C8N5MOUZD+``=F#3[#[Q<GK2;-7_,`\nM-=,G1^WF(86OPG/@\\`W,$'2Y#O5W:;8/#VCD,%F'QS\\C9)P76HRR>O.:8MP`\nM))Z].LY(\"V9QI\\WPG++0,LQLVQFR.FB\\VFO\"_.\\T\\.LA@GK3Q%B&]>-F\"PLT\nMN>TW]9]EI\"<T#;APT#_^Z6!SF997-5^J^NY/31R`%`:L:#4%@PY?,JP6<$VR\nM#L[V*!3^6OQ+R>$GGJP]4D5-*$KJX7IM8UV.<9&'@$-3$O:CN-=\\L7/6VMEI\nM<0AB_;J/WH66W^-L%)U.VNM29K?EA=I<%E&..3:K7C!M$AE#H0V;<[K'A%(]\nM2V_2=10NTK6KY_Z'2;>7Y%^BWB7[EJ23[,M^#T[R$-1^[SS[EE4!_-;[8#FY\nMC]DZ_`G.HGZN$:=6S0\"%2=G6?Q3=35.X-2#A.JH?'1%KZX/W5TY)5\\5T=M9!\nMP0[_AY?Q1SCGAH7\"\\LZR9LN=5'O<G')[email protected]#E@0(EHDO2*<#`0)P\"=/H.%\nM&L$O[<*!86'ZY],+$]XNU)W?7$3(CA?]M!8>+4N>VSKZ78&\\P@\\.]^OMG=>%\nM@JH&OH(\\>U1O2[\"][$<=I)D#[^6_8B(#J/HH])&3R\\/7QZ&OG'88OGX7^DI1\nMNN'C]Z&/I\">'CT]FU<1&GX:'^N*X4*ANA\"I2UA+X&)HCON:#CZ$YXBS'\\#$X\nM1?M'[9_A6VB&.%84?`Q-4//@I_HQS$\\U-#_U]F&S4*@]?@P$8QWY$G*`(5X&\nMF!!QA2W>*X7@M@_KA<+&HXT-KRI7P<H(1*IZZC&MV1J/ES*I5;NL73074PC\"\nM?Q-_!*:P1VJOWTA!/QZG2\"D9[PD8*X\\<=\"VK9>=I&1YI@\\6=8DE=1+T^,US`\nML7:N<!,7?BO[8!BO!0H_()#>\\`-PC&CC<3GMPPZ%KHUC8JJ\"4/0&$#CZT864\nM\\X*9`0CWB@&##QX0/V=6&`3O*`V#GQ#(9`SS@1U`*3KM8XB`XKMWI3`0WG@:\nM\"#^Y/9'<WVP`(X@1AD2;5`.B!X0CH5M357RK3DNL=(SZ$0#KAL'0=M9@Z`'!\nMD.43^56D\"\\\"@7>]VI:&[LDAUH@L64UX<^XCBYS(NIC,FE@B([@,]A!>8,DK-\nM6)LC9QCT@\"!0)``19!`/DO'-K`$`\";(C@`?>,+KK%6U@:X)Z2K_\"_4\"BI;N!\nMOQ%8/!B!8%1,I^>E6_8,TS6ISP\\(X)UG^OEN&<.+W\"13\"?:`]NF707!,\"04<\nM/[@3&XTOI]I/G2B$&L.$`;W+`=N`6E@3-473(6<[email protected]!#XRC616&TM\nM>1<2FB6`G\\8LX0*-_5#THP^\\7]B\\7M@,;5\"I*Z,^$/U//)Y!\\PO:_-)S_`3^\nM^56+&'BFR%0\"?HPNGSX^,V#A;QQ`63'_`8_`=^A_S[3%D@3JX[+82(!YH=>&\nM=E/]K25KK>E`M-<;]G@`OHCO+C*)J22.<$0W9MM*NJ\"^5?^CSR7_T$CM(UYC\nM?8`VWS[>,$:2J*O7,(#]T<<B:^S9>9FY,V''&(+OF\\]X01=N8ZH!Y]*6&G,\"\nM#+H`@\"?CM8$MZF\\8[(A&XH<[67(<:^T`]?B\\:Q(7FN.%9MSUN;]E[`T>J/I.\nMVLD,E](]^X4M2CMDX^,W'P'+I7=22R[L=)P#F0OUS,2/DS>Z/(56X7?.U8G3\nM.$UPY;ESL!LK3\\^+/O4C*CBH0[:K=A'R7^T%N4R)QNA9,=(H$XI;SD-^-\\8H\nMOWZ;^>H%*?<M5_5EC[[=S$49)U0C8VS<\\E%?$R'?XT`CZ5+`!_RNPD0N2PE)\nMO#[2FPH+['-==LZ67M*WT[=ME@4SX*`S^%AGK<$.59Z/XS,0>$:!=-RN!_GM\nME6:E<'/V;^;&='ECV=@^Z)VEYP2VUC?=Y;(>K._[YZW*G6]=@*&]Y=*%2BQV\nMYT)%/__*Q5;_S!N7C<=?;UR^WKA\\O7'Y>N/R]<;EZXW+UQN7_]LW+LQ./%E[\nM^.==N,QK\\C/O6[0Q%WD[3,F\"]2H&*@XDRB@DLZS$!1J8&?(:#R]A$,#EK4ZN\nMX5`\"&3^Z0#B#:><JP<+=+FGO]-$AYSWZ)(M9,1W$TS0VX6;[J&`S447H,VMX\nM8C2OQ$`DR(:B#D^X!SPT@?&14XQCTI,;#4>Q-<OQ^;=+H<NA\\)U3^'HI>!/U\nM9]PN.7%4AC\")9&0IMM]R_F)8'NH(R#OI%1IV$H9,1PZ2Z/6B%/71L#>:ZNRT\nMY*P$TBC:()ND9#S9DJ.(BQ30MR+WMJJ_FAZ1:[CNE]6)=^$$RX/>V6O4CXL?\nM2H5\"$?Y66CZ2KZU&>Z/XH:R&^ONWV^I_BAQ_>U@J98I6O:*_ZD#=0Z]<LP4E\nM?9@J#+$.%.7501&=C$M0LELBSRVO2.-O16`RS@E21.)=\\=SO5;O>;OQ4/VX5\nM4+VDNM/!X(:]<O%7>+:I#G3RY*@XX&Y2,%LRXP^4;,,$[AZ^[email protected]<+.F\nM.4$#45@UZJ0[5P?H4*RG*BFYX6U*P<D]V(%B7/;9,\\G`ZLP\\]K=Y@+,/!?3D\nM)P1B7<Q1HW,0B#_TF`QLX6C$>8J^,E\\&J$6QMCY$P)IB*BMBRNX9=60W4<M`\nMLYLK'(@#CG`*50U<;(*L&5D<+ZMXTC'\\U?DX&G:NM)FW[O'+-[O%;HIA1<8=\nM/0OP3+A5M#-1A,^E;XL(M50*S,J+^LZ/=X?S_'D0=0D8H:^90`N!3*FEHDRJ\nMB1;!C*N/%0?U_49+GS/F&,!HT:F<`5T4K/II_G9_61=?-A0&EV0*[*[V..&V\nM,/C!18&)A'VF?6\\?<:-[3U7[1,.VC[PM[7/C;TY%O=TRKPC],^_TOO&:=4OI\nM7>'T!/:#^[3C=9*1V[X\"[+$/N&R9(?FO>\"T6IM;LFSR/6AN56Y[42@H(P**-\nMLAIPN'&N-8_XOL7<S#GZ7,T7R>YZG]Y\"F9E45@+#=$DW.K-C#M7%\"H,1/68K\nM6&5;F!9_&&Z)6I+B(<PFP#ACGQ20U\\ISCIQ@W77@VRI]<-K=HIC\"PKYQ$E6G\nMJGCL:.-HUT?7JT=U/K\"?#FES9]/\\3Q2IWFFEE*F1I?W?#J3@VP&V\\>`!<G/N\nM&$X_]W\"8>1J(VG7.2?`V.?U/.P;>(H`',.A3(N1(O>G5Z4+$GXI6;JD=(OE4\nM\"D/U0M4%B/R>1^3136018KYD[A.!>]/W!G25R\"1ITKG27D[Z&A$WE6;W\\K>'\nM],*_0]31MPOF#U0@T%#G[2F%%.2KA)S!4/O%X5ZAL`%_0N8P\\+VAOX<L8@#[\nMCMN-@UTJ$+)\\:;.1SH8V0\\'0T931XT+A\\*8FUYE?;:]^C\"8Z&]4-K@93U8F%\nM5/L\"3J`RKO(Q5*YYE46])D8/)$DYPMR;6(VF$BB*L`(P^SHF+,>;KZ0W44;-\nMIG5:+%7EN[.JKJ]B3!2?)*3YDUP\"N!QL>8-Y>TA5G2:2P@-U=0EECJ6,)%32\nMQ*3Z$,,F0QW5.)DD*#LY`4Y-3$C\"R0V53CN8-*6L'(L:,J*9ZC!JYBH9L8LO\nMCUAX@VU!+9?52/Z-R15JWI62@XL,Q+T]9G#XQF)C@6&_I<M=N:&F5O*74=JQ\nM#GV_].7/I6]<Q]=.UH<4ZNH+:!A4ZGG)V%K'C=W&BY-7SLXF#='+O?JK5O&B\nM5%#%\"_<^:E8A^/O;HMU$O]K]\\JN[-4JE3\"_XEDR/:!!=@DAZ;UOMUU\\U=ZKJ\nMUU_)N=)[6_/OKZR9DY=9X5[1>J]]B\\YK?I0+[PMA\"]D08=84)$3#I,(NW6[6\nMGEDM\\I+!PM@I$5P)K4]**:*\"KKYR=4RH4WK@AOCQ'7TY<PF4DR#H*G3L+\\DU\nM;J&(41%*`&%P7B,@13^6*/1`4!XANI.D`XX\\VW:SQ.G`(2RFL[Y*!BQW_D2O\nMV#E2(>;AM;ALXR)OL)F[RS-3T)7[=ZG,3(L=F'MI\"9VVBZ*4M/-E1VS0Q!VR\nMN]RW$9A,&@5SW?I[P'VV=QS.YOQ[6BZQT#TM%_WL>UJG^N?=TS[Y>DW[]9KV\nMZS7MUVO:K]>T7Z]IOU[3_M^^IA5NXL^\\IIW;Y)_N%F<O+KT/7]1_S:A[*+6L\nM9)C%G'<WDYR>AX*W2=&B$:X=FT^)5:=%=BJ'DF.)96HJCQ:=6AX/\"]!>()D9\nM<B=+A(F3*UC\"&9DX.3TV?J<@JL!Z<&AL#)AMP\\OD1766,K2R=H9P:YN.+I&I\nMR06:*F1Z;>]M!]'X/<>>8A<!KFK:D((S2WIBJIV>431)4>NB<T%IF=,3M+RB\nM5OC,Q9?B2#M>YF@/CHG&X\\>D=/,QSZ@43])LG7R6ZE`U+.75)+XE7TV^!*1,\nM=7D7&UC:/^M1=[*^'[V/R60B+_3D\"\\V3L/*E[RIDS8#@R%G?;U:_7U#.>E33\nM<M9?*E_J#P(CJ5WWCU_AAC?32'8+>B\"F`IW>S9?-G3J>@[K2_=?D6[JIUJ?I\nM>)ULF=<['])QDDS6%YF>\\@<03JHXIT^_7Z_6@(8KF.3:$Y1&T\\XX&HUN5./C\nM2-W7_?B2,P'_M8YW@$.!/;VVMD[_+3W0'+-Y93J[=ME/SJ,^U^3CN7%V>-2&\nMZI4F%57\\;8?45P^V[Q>=4N0^8TF#+E/9W7\\!!?=?X'='HW?XXK]:`#CJ`**\"\nM.#>.;O`:(=4/?'PDZCQ)L$`G2J_6.'\\0_3,=OH=_NR!WP#]T<P6_IP-\\-UE+\nMR,((>7#\\V4\\B?'49)V=`+N57&O=U<W!,:AC`)G?6T,GS/,6TU&OL[XG_)).S\nMWG\"-G$=US^\"8T?7@IQ2$7P\"O9GX]PE]C&&1\\1B&$L2EVJN(?F*\\;?C$8IU/T\nM/*$6/D1C&3;\\PF&M)6R5Q..7CF+P6?@'GX\"]&`!XGN:HW]\\D?O'GEGZEGS;5\nM_2*N`A&O^\\6]W1*2HTIB\"CC?X;]NC,'+\\9]-KK\"S`Q7V]Y6/!6IUK:.><V&N\nM\"(./AIOL*#)0E8L9\\'L7\\3]5D2N6[Q>O>_UN!T]*?E,J.9AKH!M\\NB-)I=D*\nMZ:NR118CIUSV\\XFI4]\\CI;4GB[H6/'*#.7U!$J*E1>ZAO()3G6(=6T='X&F&\nM;*6'VA97S,RIP1Z5%]5X62@B&68I-/7$(<PD^`-9QC#02*7GSC939NS1X_6-\nM[]<?;JC:P\\W'CS<??J?BP?N/HXTJ$N9,'PHHF35:(B+4*:_)%32@AR);NT/3\nM<.\\>C-[JVT3'5BC\\B'H$=02\\5:J>25M_'4PZG;5N/``AO9.L3=\\_5\\7OUY\\^\nM+JUE>O!E%S=CTQBR1UP>)>GD$I.O7\"TS5ZA?*!`E)O'`[$HZ7'V0RSSK([JZ\nM6\"-/].RW\\VFO/^D-&?PZ70-RRCA[K=R-._UH+*:9ETFV#18JM*Z;6UGR#!=Q\nMQ1UE^`1$B;*RSR:5K1BS%NQUL_BII73]:#5*IFN#:!A=QF-7$&0@P/CV/L9=\nMH-=GV/HF?X,_$<>I@)Y1`?3CN\\1@YKBAI#5B)E5:*%1,O'F,)R[N_,SP`^^I\nMBBCF#6.\\NXS&O?X-G$[]?F6\"ZIXAQL%@KE1Q9$V&!2<02E'\\`4>/[HCZHW3%\nMW-1*$'77DPZ/XZ4'_MB$OTU-'%$YIV#+CB0Q'S>D@_SY@;I3[+#DKR\"Y#)\\=\nMEUC`M&E_8EXX&7_Y?@L%I'LC[U+E97VOU=\"E\\-(.1D.I1Q`TY8GW)RH@V'!1\nM_P9.C9R`^?.#[.:OT\\Y02W>&-W?Y>[4RSX&>)C>0H>U(IA-N'ZC%UF1\\L#.Z\nM*3(L)6ER@V%\\>5I(/H,UXD3#N\"F@&5F[,JD0(HU[?+&O,XT.(@QT;Q$$H8AV\nM,E8?^F>4M>FB%_>[#BX!8JQQV76[,#R6`/)1;QZP+A76RHD/P-(J>]&B.57S\nMOQN82[>\"#Z]WCUO_O95=RM&\\I?R\"RP@=0$4;]*9,399S:SF:L9;>.NHT,`Q#\nM.5!M\"K-5_BI9S33$DLE?G5EIU4V<X\\G8S\\A<\\@;#\"U3<S((^!-]X?(Y(D$W-\nM.WGA%])7D$4D$FACAF!=?V`A'MC5&K4UA0U=_0YC%%1#Q,/0\"TLKW,G[%NME\nM&_<I$VP\"($&UTJQ>#._4#=W&/3L\"KC.:U<`C`_]A[0\\?YJ.9O1C>J1O>,!\\M\nM,LPGMQT#.=BA_C^9\"7YX9_CW;+?2.1VO?O=%>E[];G8#G]EWZ=FLSJ.X204.\nMX,<NTMO5X5S@=%8.PZ?DC+$-*\\^1DM,(T8H/R=!>XV!6=X:9_BPX6#.264/%\nMHT`H4P;:_!:8.[C3B#6M34LT:'V\\I.[Q,JN+P\\_IHYD%.\\CL+\"PA1UFX?QQ_\nM(/^5391::J0O6ETM++,LLZSJ<-Z.,;T+L'>#$?X+!_1>\\\\>&R1U):J\\C8=V?\nM/EXC``SE$`0TSLMG[@R!SSU&`[S[4?]?937H80B5>V/HW!LX[PZ2#ZKV'8A.\nMFQL/4=V'22*J3Y]\\QS`+IB)Q6LFU>C9.X>FOY^<HVPR>2[$6WE9&?;SM'[ZG\nMC(][T3@%`7ARU4/;0/5L<@4R]W#XUV[O_72M\"R+1A5RO[M5?X!62EC7U.*P!\nMQ'4$TLD58A;PQC\"L=)+@&6N8E.71/_O+7!1&.KA!S0M>G#\"<M(\\\"*U1C:P48\nM\\6;):8;,,'0(*).J\\CH9=]4RYBE8!EYI3#:._:C#,L,R+$)'8E_=`@'U'B$(\nMN-++F5X0>X1W#.QS=]RP1WLZF5Y<T#RX4\"C#P/+9,ET2+7_CP4M54[*R$)\\7\nMD3F\"YL=:?]M#$6#818T,\"OLW4!J@8W%BK4%LP\\MQA@4#ZL<(1*PM5MZMJ#CM\nM1*/88>2TRF!-%9L``R^O=7XM;H=A43&8I.5OOEGF=*1H<@\"2ID[V)%/KKI`C\nM8:^I9^_Q*2]@.^7?PO\">UMAM$&,N\\44ZBWJY?L,TIM&-B$:K!=YEST2P>JXX\nM[A@(]IEZS]?07A;M$R)Q=N.=R?:O\"(B9;#:\\Q5[`+*+=!$S>&B92OU&H$#@U\nMW=8W>]K0&KIQUCX^:?\"58#7[B6@??]O(?J/MQ-\\JU24KQ4*O]DGD)'8<<0:H\nMHM`LTU[9@5]&\"P8+<FW)EQ[9A'\\W(:!S!?(E+X@>V[1UO2`K6YA30N2$!U`Y\nME\\EQE24:#EJ#][\"K=*.T\\LV*Q,UQ!T+=Y<@V.G7#ZBB?MV'EW;L5G80!+UT%\nM!5DN)Z,9QV1)K_R:35@SLBD>H#I,V5[[]?'AR:O7N>0Y4LH.Y-ZV[E\"FZ_90\nM*V03FDBGS]P^\\X)&0[Y:=?H6K/J-J2II(58?/*\"T)G8:\">I.,DS),OU#K#<D\nM;%78T$1Q2+F1#&.GM6QS/-*17B\\7=EO'[Q/`8NN=.C?$+N`L?FYE!H\"S:<\"S\nM@PRC%S#=&C4M6I4H!X:=:%-3M^/@9K9U!ZE\"X9(\\!,V>^+S[3M)X7.FC6E/K\nMAM:4.J;JJ<+!X7ZC?H6WVJV*+Y=9&KW=.)6EQ>TR>EL]]9=#^FTF57,O[JSA\nME&$M,_V+W'F**F_=,RIP-.;![R'5?+#@HGKYV94=I?S3S=KC!97RU6IM`4/2\nM.QF/+OT'F(V&>]1.Q@`D^6I3^M6F]*M-Z5>;TJ\\VI5]M2O_/VI027U%XLE9S\nM[3NKZP_S%J5#BH:/B/3WL]=GCMCFOEM:R+PS>TMJDP=0<6Z5'\"6AR>3B@D/R\nMT@^1N_1'UF?Y$IHV,R2^4/LRPLMA.CW?8EM(CJ$.],O$KJ:\\PU#`VA%H0TKV\nMD;2,JH2PW-)Y0-GQ,V:E%GI6'9F4?3E33C*KE-@D0R*P(%R#&#]-\\6(W59N5\nM4N#^SUQ1KHJUXA891,9X`98X=9R+EM^T>>G<^=*36A@/SM)$3#(YCRE&L2;I\nM3-_+V)*Q+HF#]\\O]YGB*FGML?(<&1Q@-GAS0/*=L=DUL-7?823GSI?%W=,%L\nMD']R-?.MN5-OD6LR>3:[7PX.@8+0ET?9+XTW>T`5\\%LUVQ;4`L*&GVK93[BD\nM[`2=^;![LG_$GLKN:#G*\\\\SA^ED>_(EP,CQX\\^!E=_\"_.)D=O`]>5@?OBY?1\nMP?OB9G/P/KB9''(U=!:'S%!,!@>O@IN]P>^5D[DA`\\EF;?!J.!D;W/=>M@8/\nM8]Q,#5X-)TO#ZIPL#?FU=),T.#5S21K<^%-.=H19\".)[]6>^>1[][K><-[_[\nMT??D7PUZ\\N>[XCORK\\YQY`]M:]>/?W6V'[^9'\";@VER>B,;,R`DZ)(*F\\'Z(\nM]84CM[-RPG.JSU9C=<J,P`V6X+EQ&O\"JWG8IX!]@Z+5UF[AG#E*<C[LJ*6JW\nM:2EJ\"ZLI:K]+3U'[78J*6O6[KQZO7SU>OVHGOFHGOFHGOFHGOFHG_H]K)Y\";\nMN#T1I([\\^Q(-3\\IR<X]&!W28P?Y.,02LG#ZP65.@LY7K9-SO>A;C2P_<W'RS\nM%1`9<5K)12]%_7)T#P633DCK'C\":U6+*!ZKO<IQ&[0`@9FD=N-&@S@$;GJ5R\nMP!%Y\"H/P^/0,**LM6,]H\"W+E8EW.U168]AQ-@3/OJR8U2VI\"-Y]-S')K\"8/]\nM7@O%8G%EO/(/%#)*SYX]*:E?U4J\\4O+QHJU30%&^%U]L(8MCBGXUPGC/*HW_\nM.:6$:G):&:NEE%Q\\=/1G#\"PV[*)7*C`?A6(+@_&Q%7?23RYO,)C>&L%F('!R\nM``C@O;#]/IP2*'9%XWX/EE!B5Z82:(R.I<&T/^F-^FSZ/DK7A#3LP*F/7%32\nM12XD3C5.9\\84C26H'QYOS..I\\70X%`?F50E3AN8Y,3!&*(O!D=9-MQ`>H\"4P\nM/IARKXQW\\WBB<_1!86=I?B8]<LY#:VM@%YB-.(_)2Q=Z#YON'-@;.H`=%K5'\nM!VHZ04MW/!`+.&L\"!9>4<#K==,[)BCH\\VCMIG5'UPS/\\38-;[0UQ#_-`^DDR\nMPA\\WJC..T0W/U`5ZWVKKRO3`M84$</7SFU&4IF$`.Z]U;:!3^:JT2I5K9\"<H\nM7J(+I*SHEE_8A,/#XRI#.CRN$218+S0S`NZB+]8]F%M->$I*G3H4\\9SWK0\"B\nMKQ*<<:UP8*)!\\G(CP@TIC5&O`VM.C5F.#H<C8\")JS\"M91=_O3DPHCN-U^97F\nM,!@M+T4,,1MD)05.>X)]P,CFB4*3=#1=BI2.'HM[>S\"`)N`11KUJ&EQ5RT8)\nML<RL'8Q)`[;A\\C1A,O;N&(D;4$S4L>RK;<P,O2JZY##1\\K<0DL.CX_UZZT>U\nM\\?'BR<9&1@EY>+0K'S>^OZ`_WL?6Z^;+MAMWN/9]R2M0'&+8SV'I6VG%_WJP\nM6Z10EO1]-_N]!=6341G&,^Q2\\-!1Z=<B/4BP2T.92(6@@+&-(T2:0D%R!])[\nM(:T%YT\\1=WQ975QW\"16!7),SND]@#U&35\"R*KJ3Z[)D,MR2Z7Z9]A4K!MF*J\nM8B@R6[?FUS762`7/;2$'!'5@!L9#'T8_OD`KL,Y5,J,'#;?R([\\R:P:XM@K6\nM!H;'UG[LUUY#T[09E0Y?VFK?^=7>KJVMG<+Q'&OU8!X`*LS.+(#O?0\"<3OD=\nMZI60<;!0?\"\"D=K-`GN36[7801'(MB*<^B`=()RYZ'PL%Q!_<Y\\!IX6.V(PC&\nM0:\"-+!A=#Y&1#BH!FX$C1-P\"RJ#B#XOVAP\\`\"Z>6A;-@A_9(&>X`RN!FL6\"Z\nM4LHO\\G&V=@8YL8KN0#&TLYR5J3X.H0B@-1KB2A?P#,B#`7+O@,F@ZJ]J-%E3\nM5=L/.AV2\\9*V-R-RL\\K'H\\KP9=2>!?U]\"'3-3!\"=4`'(9W@_DED_>&7A!O\":\nMQ^UU.D!3WCA`GH:F#VV?9]$4IW)M(]\\#JNK5=9C0%C'3:*\",//O;4TL'*T@4\nM_)-U3:D=8^9+P42TR@$-HV.T/L:S]@-P$`G:PE^.DRDJ79[@R\"/@:2:Q?*36\nMR.&_,T&.?\\UEB;N]#[WN%)A2[`&Q$%`:O>M29N`B(V>(8;@#6'2!E!%V$*6\\\nM8WKH@#R=\"%/:CS]$(!5A3XG?!0X)`:`V%=NXLE%\\@5E`5N0\">=H)L8B7VHZ;\nMDJ30%'22*4ALP&8\"`P8<*##3.0WL=:_?5YBV%#B!SON>\"$C$ISF)F[-YH#%;\nM2UV4FR/DSP>L,;RB^/\\(H1M'-O-+.B56#T/[E]%.FN)H=Z(QU(@D-0C.C/98\nM1+F'&$7;`]UNF>8/$X)T61=)&2Y9M<509(61I08.GH(JY-DA_UH7>J?=5\\=R\nM55MYKK#/;SLXC]#$J=SK4M$\"KIX4Q*7\",6(B%8Q],>\\Z&E=0JIG%U'#UP4YN\nMLITM/K*CKF>,S'YE9@>8:VSQ4*2^S6[#!NK1-T`IS;*3C!;G0-Z9:<`/;[GD\nM:2$Z?[?1Z;[;B\"_>;;S;T'?8.$GV]MH&7-]Y'7691=]Y#=(ZL4XPT.@B+LO;\nMY@'LSI@\\$6[(YW1CO9J?09.&X'6_4^P@9^=DONB4O@5&]`+9T-+SY]7O2EX%\nMZ$\"Q@^ZT6*N3EBK/88W?YJM?7'`L=2J\"RUN6WS2)#]SL);9*6<+N2]$^QE^2\nM/OJ]@,$OW`O*4Q/N1B7<#;^MYL'B33GC1:MA,QP['CL@MY7]'9Q7X+2II1&<\nMP(-.]HUGQZ!U\"=ZQ)4?7Q3!SH[N_@_/E0_??>(5[0^>+&G2\\9V=_W(X4G2N<\nMGML08='%U=!N6=`L.+M^,\\`Y??L/7A-K>D-B*.H>Z=RU9PZ(O)?)N)<U;_(]\nMMZ&48SG#)\\``%3XXR%X_[E:<0\">I9BC,>>#K_I8RCBNBES3QT#E\\FU*H0CMV\nM5&B_KC38RAZDY8+$10\"YVH1L0S_QF\"3'2+*3H=3-A)_;TL=,UMR*PIOAT4(T\nM5C,H;LTA596F;$5ZBS6G)K@YON+.V0I(WQ'P6P;CG7)2%+NPY9R)]DB42NLZ\nMC/:I/R2.<E\\0\"Q%V1[/8QY]MBGNQ_T)U`PR*XW!SL]O.L%(;KT^*DNLD%>?2\nMG+R^A^HV$L@QT,9X0/Z1&R6_:C_*U;Q`5TJNYXZEYX[%Z$4#QBBHX`<I7*D-\nMY*2Q-\"DG_Y$OU:!2-;?4_4RI%_7=@MIX)$4P>PFGY[H>)\\;_7Y#@/.GG4.`?\nM#-4K%P?*W0^4LYN/.)$KU%BB:G/V]BS03B2D<2LSWL`;SAI3(5S9;QYH7#%F\nM@LQZI!.S:=!&CK2D>\"')%^%XF<@<HMO9`485X5HV#@75E-76V1GTE8*+CG31\nMX($3.QOI.W(G&$41MQ')2=;U\\X<,SH[Z4XTB.%_=.!Z9ZI@(1STP5;`+,B5&\nM$XS8Z$UEP<Q:]916`29^2HC\\]F!'YN\\WRW9A_#,*C<=Z[CP3I50/?^*9OG^R\nMUSBCB[0#NMNC<'E::#MIP\\E3_>[[[[^O5;][4\"5933]BY=H_:H_PWX<DFNB6\nM,-2^@=\\XV3G[KR,,[F*?=P[\\YQ^/_>?VFV!'3EJOC]MG^_6_2T]JLUL].6CN\nM'.XV[@)F#;,?3@>(4B<[K4K-@I7SR3`/&7B,QP9<A)=%CKCR(>I/LVSW#*:V\nMV7IS>+S+;&U'/=]6&\\B*=3`_P8ENI>1Q9[TTZL/V14,=F$'\\!SV]SE9*(6['\nM`3^SGNWE'8RH)$;L3-LF^7Z[\"944_#P+*K>R;T#UW8(&5-\\]_&H_]=5^ZJO]\nMU%?[J:_V4U_MI[[:3_W?M9_2O,1\"\"0-T`$O8[A>]2XZOB9S@Z_I/C;/]QO[^\nMX4\\-UX>!.?,;YB*%H'OBWN[)$7!Z!?0`D3\\$[^/'CWC`$R6@T^('OUKSX\"6F\nMT_P9>%*&0,'H9K.;!SO8T:)FAU5%\\4^8=*F8Y2\"EAN9%H886XFP-87'#.IHI\nMY\\0350]'Z$B`S^0T;C0316#B*6AA?QQ'723<%R5%[W3R.B/0##,9[>3-03;%\nM';\\HR+(`(*>Y43\\V2=_^8EE]3R>%BIJDWY5`J2D?DN=X:!=+E`H`0.IP?]8)\nMS2:,<!'!58[I.IRI%M5\\#!0>NOB\\(#..=S[(`Z-N:_W5P<D@'VA^5I$02SZK\nM[*)<^=SZEC%_M+'Y:&'&_'\\QQ+R,1!4E\\J+6(.(P2P;([XN/_!<#Y_>$KY\\U\nM\\1PB^1'-]?I&=;WZ/<9(1L&H]B<&KY?8]4O9P/7!J/4ZZKRJ-.]C&I:COP&$\nMD@Y?KP4_=;_(!<]:>T+G]G;KN[M8;\\^KUQ_]$VFV_#&SO;M'X>N)J`+[>`/<\nM&]Z,]F\\J_21\"=C@Y_W\\8SHKZI^C>-X,!R,BEG&\"0Z#)E/^U-6-L3J9WC1AW.\nMZI<G!SNXGC#D-;57/WAU`H>.6ME9H;C0V!?NR;8&?[^XN]<Z>?FR^7=-OZD)\nMK(ZZ)6[ZGU,T0.P-1ZRO%@2\"0V\"O8<:$/4]5`'6YXXC@!`#[(56W%;9.$=V5\nMVS@U''\\<)6/`:8(KG_AH`2X0>VSA/+\"`-M>2[?M%*5'*9!`0Z\"`716QT27>U\nM%#P6NEGDM291$`/05Z8C22I/YHNH*8/R>B-2F'P,9$\\]($5\\I\\LCW(+W^_4?\nM@;F`0EL*7A,RXG=@H><4T&\\K.S;5@G#VZ^FH!P<0'+OM,^\":7@#ON5U%O6!O\nM.'&6D/4U$<@!,)@92^L.6E[\"*WQR!T1QAW0BABU>##C4)>OW?3C+&YBS94V3\nM@;7T2M6>?UM5OP)@@P-K+)47_AIWKA*U#$S;27T/QM$ZV6L3[PA,,7+TB+%M\nMM/M!5OK@\\`VR8]@/%XY$Z>;D``B4<P,X7<XTZDPG$\"ZNF'D/XX)=ASQ#X\"/'\nMHTV#WX+KPP47/T&=J9M!6VV)6\\]/6_2SCL],=7UZ/MJL?K>YL>CI6=VHFN/S\nMWOIY;[B>7A'.?<:Q8CO$I\\KW<JK4UM%-\\?'F1FVS^B1XJ@!#1.BV\\JZS`N@(\nM8$:J`LNHGF/`_W4*#XVHND3AHFEE&SNO#\\\\.MI>I7F6X;-[M;\"_C@[9V<,NY\nMA;\"I)8QROB29*=1;58%MO0YD79UNF78Z\\LXI>_2J_=_;RT>M]I.CW?;REM`]\nM>DU?0;\"\"C?'S7@,*23#.\\DG++6A*R\"Y_>0P\"4N-@=WN0`--/)D#FU?((MT)E\nMJ\"JQJOQ3ME3KYU:[L;_]RY14#I7T%W[->W;;_Z/0F#G%XV'NG^WLGV4#$+,!\nM#\"(,#<UZ&<U^G\\?:^MV$[<R<)\"G;#78Z<=Q=LP#QL$2+&_4OM+F2$*)D*I4H\nMF5=*+XXFU/T;DPH\\`]S\"@WW<P]O!\"PQ`!<?'FCIBUAT0\"6<<[Z0`Z<@O!'4H\nMYWUDU2>)`T$I.KS^JC,?T*Z]M`5:ED2NP\\&]N]]@\"R@TZ^H-487']F6VQO*\\\nM-4%S_7%R0V9,_:X[,HS'>PY]QP4+KHE4[9X[M>8U16;SV-`POI[5T(SE9XO[\nM;$NT5^[_`#@9`S&Q>X7^4!],/<RP'G?-MX_`_52=G13NL,8IIZ_(SO1\\))Z#\nMK\\OJN5-Y#<E[JOL=W=]_@>'^(NBXWN.#<\\+5[5^H.A3X=3)6RV_KE?\\^789_\nMH\\J_3I=_<6B*+D]+3&,A7@F5J[]TH@F>7409TU_4?2F+ZY;0B<0$22W?_]3[\nM#0XE0#U^M4-?];9W!_^,`&+Y-20%S_4QQZ]0QR>TL<!4[,(<DE2B<O\\3DPLN\nMB]%%\"WKD!>\"[&COM!I\"9.76(;IHTU#.JF((T(4J9_OQ\"Q^O]3[K>;_G^_ZJN\nMX?3M_Z)1RNMB4?'Y?+WS<\"X4870*@@-;RGL.\"R'+F4JJI)X'<4=*&'RVTT'O\nMD_=FY&A&BK@\"N+X(DE-8!8W?*DNE@T@^`TD(4HHHXH@<;+8Y1'^&6-J27`!B\nM(HIEV4BU1\\2Z2H:NE($1F2-.2X(\"D%`+(D#GT\\OM#7-P$B^^?)]>+\\,Q]4_E\nM'-0T^M;[WFBDV70\"R>:9_9@N'FQFG2QMLIMN!FW3\\&;2M;O.HE!7F<;;21W-\nM#==!UJ8WFI\"5+[JQX6S/)GWX8F.1\\V'.T((#FW$^W#X46_$VLJW[#B_^/Z`(\n&OF&I\\0$`\n`\nend\n",
"msg_date": "Thu, 19 Mar 1998 14:34:08 +0900",
"msg_from": "[email protected]",
"msg_from_op": false,
"msg_subject": "Re: [HACKERS] First mega-patch... "
}
] |
[
{
"msg_contents": "\n\n\n\nI have a table called files which is owned by the postgres user. I give\naccess to hostadm\nfor INSERT,UPDATE,DELETE to hostadm. Then hostadm does a copy from stdin\nusing a program\nthat I wrote in C using libpq. Once this copy occurs, the ACL on the table\nfiles is set so that hostadm\nno longer has insert permissions.\n\nPostGreSQL v6.3\nSlackware Linux 2.0.0\ngcc: 2.7.2\ngmake: 3.74\nflex: 2.5.4\n\nMatt\n\n\n\n",
"msg_date": "Mon, 16 Mar 1998 13:05:54 -0400",
"msg_from": "[email protected]",
"msg_from_op": true,
"msg_subject": "Problem with GRANT permissions"
}
] |
[
{
"msg_contents": "> > If I create a table with a datetime field with a default of 'now', \n> > every insert the value is the time of table creation instead of the \n> > time of insert, which is how it behaved in previous releases (I\n> think \n> > this was even documented).\n> \n> I can't recall it ever working that way, though before we discovered\n> that it didn't we all assumed that it _did_ work that way :)\n> \n> The workaround is to define it as\n> \n> ... default datetime('now'::text)\n> \n> which forces the string to be evaluated at runtime. The SQL symbol\n> CURRENT_TIMESTAMP also misbehaves in \"default\" clauses, and I'm\n> considering changing it a bit to get around the problem.\n> \n> - Tom\n> \ntest=> drop table tmp1;\nDROP\ntest=> create table tmp1 (c1 int4 primary key, estDate timestamp default\ntimestamp(now()) not null);\nNOTICE: CREATE TABLE/PRIMARY KEY will create implicit index tmp1_pkey\nfor table tmp1\nCREATE\ntest=> insert into tmp1(c1) values (0);\nINSERT 19653 1\ntest=> insert into tmp1(c1) select max(c1)+1 from tmp1;\nINSERT 19654 1\ntest=> insert into tmp1(c1) select max(c1)+1 from tmp1;\nINSERT 19655 1\ntest=> insert into tmp1(c1) select max(c1)+1 from tmp1;\nINSERT 19656 1\ntest=> select * from tmp1;\nc1|estdate \n--+----------------------\n 0|1998-03-16 13:26:32-05\n 1|1998-03-16 13:26:39-05\n 2|1998-03-16 13:26:40-05\n 3|1998-03-16 13:26:44-05\n(4 rows)\n\ntimestamp(now()) - works for me.\ndatetime(now()) - also works for datetimes.\n\t\t-DEJ\n",
"msg_date": "Mon, 16 Mar 1998 12:35:48 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] datetime default 'now' broken?"
}
] |
[
{
"msg_contents": "hello all,\n\nyep, i'm new to postgres so please bare with me. if this isn't the\ncorrect place to ask these questions, please point me in the right\ndirection.\n\nfirst and foremost,\n\ni need to have a field (say pin) that is the primary key of a given\nrelation. each time a record gets added to this relation, the pin\nincreases by 1. thus making an auto increasing primary key (i.e. a\nunique identifier for each record that isn't explicitly increased by 1,\nthe dbms does it for me.)\n\nsecond,\n\ni have a platform related installation question, which list should be\nused.\n\nthankx,\n\n-brian ([email protected])\n\n",
"msg_date": "Mon, 16 Mar 1998 16:09:24 -0500",
"msg_from": "\"PostgreSQL On Peanuts.roanoke.edu\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "AUTOINDEXING AND HOROLOGY REGRESSION ERROR ON PGSQL 6.3,\n\tLINUX-ELF i686"
}
] |
[
{
"msg_contents": "> *NOTE: This is not to be discussed in any other mailing list, period,\n> or\n> you will lose your secret decoder ring, and we'll have to cut\n> off\n> your hands so that you can't do the secret handshake anymore*\n> \nBut I don't have a decoder ring *sniff*. I can wait till Sunday. ;-p\n",
"msg_date": "Mon, 16 Mar 1998 15:11:06 -0600",
"msg_from": "\"Jackson, DeJuan\" <[email protected]>",
"msg_from_op": true,
"msg_subject": "RE: [HACKERS] First mega-patch..."
}
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.