SemanticSearch.ReferenceAssemblies is rebuilding the references every time, even if not needed

[jasonmalinowski]

This is more visible after #77038 is merged.

Problem: when Roslyn.VisualStudio.Setup is built, it's calling into targets in SemanticSearch.ReferenceAssemblies.csproj to rebuild the reference assemblies. This seems to rebuild every single time, even if the inputs or outputs are up to date. This means repeatedly pressing F6 in Visual Studio will never each a steady state, since each time Roslyn.VisualStudio.Setup builds, it'll rebuild the reference assemblies.

Running a few local builds I see this for the GenerateFilteredReferenceAssemblies task:

Input file "F:\Projects\Roslyn\artifacts\bin\Microsoft.CodeAnalysis.CSharp\Debug\net8.0\Microsoft.CodeAnalysis.CSharp.dll" is newer than output file "F:\Projects\Roslyn\src\Tools\SemanticSearch\ReferenceAssemblies\Apis\Microsoft.CodeAnalysis.CSharp.txt".

It seems odd to me that we have an output file pointing to our repo, but this will never get up to date properly.

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[jasonmalinowski]

@tmat can you chime in to what is the intent here?

[jasonmalinowski]

It appears to me that the task is intentionally doing its own version checking and avoiding the file writes if they're not changed, but this trips up MSBuild doing the same thing.

[tmat]

Running a few local builds I see this for the GenerateFilteredReferenceAssemblies task:

Input file "F:\Projects\Roslyn\artifacts\bin\Microsoft.CodeAnalysis.CSharp\Debug\net8.0\Microsoft.CodeAnalysis.CSharp.dll" is newer than output file "F:\Projects\Roslyn\src\Tools\SemanticSearch\ReferenceAssemblies\Apis\Microsoft.CodeAnalysis.CSharp.txt".

It seems odd to me that we have an output file pointing to our repo, but this will never get up to date properly.

The output file is a list of filtered APIs. This is checked in, so that we can review any changes in APIs exposed to semantic search.

[tmat]

It appears to me that the task is intentionally doing its own version checking and avoiding the file writes if they're not changed, but this trips up MSBuild doing the same thing.

Should we remove the Api files from the output set? Ie. the only output would be the rewritten assembly?
I think that'd work since the rewritten assembly file last write time is always updated by the task.

[jasonmalinowski]

@tmat I just removed the logic that writes only if the file is unchanged; that way the output timestamp is updated.

[tmat]

Would git consider that a changed file?

[jasonmalinowski]

@tmat It'll see the timestamp changed, but it'll rescan the contents and see it didn't change. No different than if you make a change to a file, decide you don't want it, and just edit the change out in your text editor.