|
{ |
|
"number": 1, |
|
"title": "WIP: repo consolidation scouting kick-off - make clr build locally on Windows", |
|
"body": "This is my first contribution to the scouting to let other people\r\nsee the first problems I'm hitting as I believe them to be mostly\r\ngeneral. With this small change I now have at least the native part\r\nof CoreCLR build running - I'm not yet past that so then I'll see\r\nwhat other problems I hit when building the managed components,\r\nbuilding and running tests.\r\n\r\nAs you can easily see, the first (and so far the only) problem is\r\nthe split of the two roots (subrepo root vs. repo root) making the\r\nbuild script fail when trying to locate \"msbuild.ps1\" which is now\r\nunder a repo-global eng folder. I can imagine it may be undesirable\r\nto query GIT in the build script; in this particular case we have\r\nmultiple alternate ways to proceed (keep the duplicates of\r\nmsbuild.ps1? autodetecting the root based on some well-known file?).\r\n\r\nThanks\r\n\r\nTomas", |
|
"createdAt": "2019-10-15T17:51:16Z", |
|
"closedAt": "2019-10-29T16:41:31Z", |
|
"mergedAt": null, |
|
"state": "CLOSED", |
|
"author": { |
|
"login": "trylek" |
|
}, |
|
"labels": { |
|
"nodes": [] |
|
}, |
|
"headRefName": "BuildCoreCLR", |
|
"additions": 2, |
|
"deletions": 2, |
|
"changedFiles": 2, |
|
"comments": { |
|
"totalCount": 19, |
|
"nodes": [ |
|
{ |
|
"body": "I'm seeing a similar issue in core-setup.", |
|
"createdAt": "2019-10-15T17:53:02Z" |
|
}, |
|
{ |
|
"body": "Can you use the `RepositoryEngineeringDir` variable?", |
|
"createdAt": "2019-10-15T17:58:16Z" |
|
}, |
|
{ |
|
"body": "@ViktorHofer this is just to find the `eng/common` folder before even invoking the scripts.", |
|
"createdAt": "2019-10-15T18:00:35Z" |
|
}, |
|
{ |
|
"body": "Hmm, where is this supposed to be coming from? I'm normally building the repo from the Windows prompt - who is going to set the variable there? Or are you saying we should go through some central script that would set the variable? That would probably solve the problem but we're back to chicken-egg problem w.r.t. how to locate such a global script? ", |
|
"createdAt": "2019-10-15T18:01:22Z" |
|
}, |
|
{ |
|
"body": "Arcade defines that property, so you would get it if you import the Arcade SDK. Are you saying that at this point you don't have Arcade imported yet?\r\n\r\nIf that's the case we can probably manually define it in a runtime root's Directory.Build.props file which is then imported by the repos.", |
|
"createdAt": "2019-10-15T18:02:27Z" |
|
}, |
|
{ |
|
"body": "In fact, that might be one of the solutions - we would check in a special script at the root of each of the repos being merged, that would set the path to the repo root. In the source repos that would equal the repo root while we would massage it during the port to point to the global root.", |
|
"createdAt": "2019-10-15T18:03:08Z" |
|
}, |
|
{ |
|
"body": "https://github.com/dotnet/arcade/blob/e0112fe2f1fd7f6db08b228db98f22225f82267b/src/Microsoft.DotNet.Arcade.Sdk/tools/RepoLayout.props#L22-L65", |
|
"createdAt": "2019-10-15T18:04:18Z" |
|
}, |
|
{ |
|
"body": "@ViktorHofer we haven't even found MSBuild yet, let alone invoked a build with the Arcade SDK at this point.\r\n\r\nI encountered the same issue in core-setup's `build.cmd` just to find Arcade's `build.ps1`.", |
|
"createdAt": "2019-10-15T18:04:22Z" |
|
}, |
|
{ |
|
"body": "OK got it.", |
|
"createdAt": "2019-10-15T18:04:49Z" |
|
}, |
|
{ |
|
"body": "I am afraid that we're still on the chicken-egg plan as in CoreCLR running \"msbuild.ps1\" is exactly the thing that requires locating the repo root.", |
|
"createdAt": "2019-10-15T18:06:06Z" |
|
}, |
|
{ |
|
"body": "OK there are multiple options here. \r\n\r\n1. Importing a cmd/sh script that defines those paths that lives either in the parent (src) directory or parent-parent (repo root) dir?\r\n2. Hardcoding the relative path to the repo root (../../)\r\n3. Using git to resolve to the repo-root\r\n\r\nOther ideas?", |
|
"createdAt": "2019-10-15T18:08:39Z" |
|
}, |
|
{ |
|
"body": "Circling back to my suggestion to \"check in a script at the root of each repo being merged\", that may be also usable for other things like exporting an environment variable declaring whether we're in the split or merged repo.", |
|
"createdAt": "2019-10-15T18:08:50Z" |
|
}, |
|
{ |
|
"body": "[not sure what kind of script that should be, ideally something that doesn't require two versions for Windows vs. non-Windows]", |
|
"createdAt": "2019-10-15T18:09:27Z" |
|
}, |
|
{ |
|
"body": "I think I like Jared's suggestion to ideally make most of the changes in the original repos so that we can arbitrarily repeat this process. So I think hard-coding ..\\.. is not ideal unless it comes from a script the porting process will create. I leave it to others to chime in as to whether it's legal to query git or not. I don't think it's strictly illegal, I think that the BuildVersion script is doing that already, and both in the lab and locally we always have a legal GIT repository available whenever we run the build script. Considering the dichotomy is supposed to be temporary, using the GIT magic may be a reasonable workaround.", |
|
"createdAt": "2019-10-15T18:12:48Z" |
|
}, |
|
{ |
|
"body": "I agree with you. We should make most changes in the original repository. Just as an FYI, in cases where this isn't possible, you can create a GIT patch and place it here: https://github.com/dotnet/consolidation/tree/master/patches/coreclr", |
|
"createdAt": "2019-10-15T18:16:22Z" |
|
}, |
|
{ |
|
"body": "Conceptually, if we decide to emit / special case some scripts in the subrepos to interface the subrepos to their \"embedding\" within either the split or consolidated repo, we might also want to introduce similar per-subrepo interfacing YAML files under eng somewhere, going back to Jared's proposal for encoding the script / source location details via variables.", |
|
"createdAt": "2019-10-15T18:16:59Z" |
|
}, |
|
{ |
|
"body": "Ahh, thanks Viktor for making me aware of the patch list, that's very nice!", |
|
"createdAt": "2019-10-15T18:19:54Z" |
|
}, |
|
{ |
|
"body": "OK, so if I may try to divide the problem space, I believe that we're basically in agreement that we need some environment variable representing the \"main root repo\" that translates to the normal thing for the existing repos like CoreCLR and to the root of the consolidated repo in the merged case. In such case I can continue making further progress using my temporary git query hack while we discuss how to tackle this cleanly as there seem to be several available options.", |
|
"createdAt": "2019-10-15T18:24:10Z" |
|
}, |
|
{ |
|
"body": "Source-build does have a requirement to be able to build without a valid Git repo since we package sources and tools into a tarball without the accompanying Git repo. We could work with an environment variable override or similar though.", |
|
"createdAt": "2019-10-15T18:29:33Z" |
|
} |
|
] |
|
}, |
|
"reviewThreads": { |
|
"totalCount": 4, |
|
"nodes": [ |
|
{ |
|
"comments": { |
|
"nodes": [ |
|
{ |
|
"path": "NuGet.config", |
|
"diffHunk": "@@ -14,6 +14,7 @@\n <add key=\"dotnet-core\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-core/index.json\" />\n <add key=\"dotnet-coreclr\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-coreclr/index.json\" />\n <add key=\"dotnet-windowsdesktop\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-windowsdesktop/index.json\" />\n+ <add key=\"myget.org dotnet-core\" value=\"https://dotnet.myget.org/F/dotnet-core/api/v3/index.json\" />", |
|
"body": "@trylek I think we should try to get rid of this feed in coreclr before we consolidate as myget is unreliable. What assets are you currently pushing there?", |
|
"createdAt": "2019-10-17T10:09:05Z" |
|
}, |
|
{ |
|
"path": "NuGet.config", |
|
"diffHunk": "@@ -14,6 +14,7 @@\n <add key=\"dotnet-core\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-core/index.json\" />\n <add key=\"dotnet-coreclr\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-coreclr/index.json\" />\n <add key=\"dotnet-windowsdesktop\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-windowsdesktop/index.json\" />\n+ <add key=\"myget.org dotnet-core\" value=\"https://dotnet.myget.org/F/dotnet-core/api/v3/index.json\" />", |
|
"body": "I believe this is only needed by crossgen2 and R2RDump - without the additional feed they cannot find the [old] System.CommandLine parser package. I have already switched SuperIlc over to use the new parser and it would be probably feasible to do the same for the two other apps if necessary.", |
|
"createdAt": "2019-10-17T14:12:39Z" |
|
}, |
|
{ |
|
"path": "NuGet.config", |
|
"diffHunk": "@@ -14,6 +14,7 @@\n <add key=\"dotnet-core\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-core/index.json\" />\n <add key=\"dotnet-coreclr\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-coreclr/index.json\" />\n <add key=\"dotnet-windowsdesktop\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-windowsdesktop/index.json\" />\n+ <add key=\"myget.org dotnet-core\" value=\"https://dotnet.myget.org/F/dotnet-core/api/v3/index.json\" />", |
|
"body": "IIRC there was some work in the dotnet/jitutils repo that was waiting on R2RDump to move off the old System.CommandLine, so there'd be a number of people happy about removing old System.CommandLine.", |
|
"createdAt": "2019-10-17T15:06:51Z" |
|
}, |
|
{ |
|
"path": "NuGet.config", |
|
"diffHunk": "@@ -14,6 +14,7 @@\n <add key=\"dotnet-core\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-core/index.json\" />\n <add key=\"dotnet-coreclr\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-coreclr/index.json\" />\n <add key=\"dotnet-windowsdesktop\" value=\"https://dotnetfeed.blob.core.windows.net/dotnet-windowsdesktop/index.json\" />\n+ <add key=\"myget.org dotnet-core\" value=\"https://dotnet.myget.org/F/dotnet-core/api/v3/index.json\" />", |
|
"body": "myget is unreliable but I wouldn't block consolidation on it. \r\n\r\nMy mental model for should or shouldn't we require this before consolidation is roughly this (assuming it's not a functional blocker);\r\n\r\n> Could we fix this in the consolidated repository in less than one business day. If so don't block consolidation on it because it's something we could iterate on pretty fast. \r\n\r\n", |
|
"createdAt": "2019-10-21T16:57:31Z" |
|
} |
|
] |
|
} |
|
}, |
|
{ |
|
"comments": { |
|
"nodes": [ |
|
{ |
|
"path": "eng/Tools.props", |
|
"diffHunk": "@@ -0,0 +1,15 @@\n+<?xml version=\"1.0\" encoding=\"utf-8\"?>\n+<Project>\n+ <PropertyGroup>\n+ <RestoreSources Condition=\"'$(AdditionalRestoreSources)' != ''\">", |
|
"body": "AFAIK, RestoreSources aren't honored by Arcade anymore.", |
|
"createdAt": "2019-10-17T10:09:29Z" |
|
}, |
|
{ |
|
"path": "eng/Tools.props", |
|
"diffHunk": "@@ -0,0 +1,15 @@\n+<?xml version=\"1.0\" encoding=\"utf-8\"?>\n+<Project>\n+ <PropertyGroup>\n+ <RestoreSources Condition=\"'$(AdditionalRestoreSources)' != ''\">", |
|
"body": "I must admit I actually don't know what exactly this file is needed for, I just see that in its absence CoreCLR build seems unable to install \"dotnet.exe\" (dotnet-install) and subsequently pull down some packages needed by the build like Microsoft.DotNet.Build.Tasks.Packaging. Having said that, I fully admit I've never been a Nuget expert so I may be easily misinterpreting the symptome.", |
|
"createdAt": "2019-10-17T14:14:34Z" |
|
} |
|
] |
|
} |
|
}, |
|
{ |
|
"comments": { |
|
"nodes": [ |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "I think we should make sure that coreclr works with the global nuget cache enabled. This is the default for corefx and it would be really sad if we need to go back to the old world with the repo packages dir.", |
|
"createdAt": "2019-10-17T10:10:44Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "Using the repo package level directory has helped us achieve higher fidelity in our CI. At the moment I believe arcade still uses the global nuget cache as well, but our goal pre-consolidation was to move as much as we could to the repo level nuget cache.", |
|
"createdAt": "2019-10-17T16:57:11Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "Do you have some data on that? We aren't hitting any issues with using the user global nuget cache locally and the repo package cache in CI (the default behavior in an arcade enabled repo) in corefx.\r\n\r\n> but our goal pre-consolidation was to move as much as we could to the repo level nuget cache\r\n\r\nCan you please elaborate on that? Not sure what you mean.", |
|
"createdAt": "2019-10-17T17:07:14Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "The issue is that the default global cache is on the home drive. Which for vms is the most expensive and smallest size disk. This equates to a performance problem, for our arm hardware our home disks are extremely small. Helix also consumes most of the home disk, so accidentally filling the disk with the global nuget cache on the main disk used to happen.", |
|
"createdAt": "2019-10-17T18:47:10Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "As for the comment. It is more of less whet I said. Coreclr’s expectation was to continue to use the non global nuget cache. I think there are still blockers keeping us from migrating away from that plan.", |
|
"createdAt": "2019-10-17T18:48:21Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "Got it, that makes sense. What I don't understand is why this causes a problem as on CI the user global cache isn't used by default (arcade convention).", |
|
"createdAt": "2019-10-17T18:48:52Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false\n+\n+# Always use the local repo packages directory instead of\n+# the user's NuGet cache\n+$script:useGlobalNuGetCache = $false", |
|
"body": "https://github.com/dotnet/arcade/blob/90d8a77c53d5f540ebda7a56d8a2e5661657d742/eng/common/tools.sh#L52-L57", |
|
"createdAt": "2019-10-17T18:51:09Z" |
|
} |
|
] |
|
} |
|
}, |
|
{ |
|
"comments": { |
|
"nodes": [ |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false", |
|
"body": "Same for the SDK here. CoreClr isn't using buildtools anymore, why is still required?", |
|
"createdAt": "2019-10-17T10:11:04Z" |
|
}, |
|
{ |
|
"path": "eng/configure-toolset.ps1", |
|
"diffHunk": "@@ -1,2 +1,12 @@\n+# We depend on a local cli for a number of our buildtool\n+# commands like init-tools so for now we need to disable\n+# using the globally installed dotnet\n+\n+$script:useInstalledDotNetCli = $false", |
|
"body": "This is possibly an artifact of our older hack to support building with arcade and buildtools. I am not familiar exactly with what can be removed.", |
|
"createdAt": "2019-10-17T16:58:35Z" |
|
} |
|
] |
|
} |
|
} |
|
] |
|
}, |
|
"commits": { |
|
"totalCount": 0, |
|
"nodes": [] |
|
} |
|
} |