Intermittent MSB6006 compiler failures using CodeQL in GitHub Actions

[martincostello]

For some time period of time (I don't remember exactly, but it's been months) I've been getting intermittent failures in the C# compiler in CodeQL analysis legs in GitHub Actions workflows.

Seemingly at random, the code will fail to compile with an error like this:

Running /home/runner/work/Polly/Polly/.dotnet/dotnet build --no-incremental /home/runner/work/Polly/Polly/Polly.sln
MSBuild version 17.7.3+8ec440e68 for .NET
  Determining projects to restore...
  All projects are up-to-date for restore.
  Polly.Core -> /home/runner/work/Polly/Polly/src/Polly.Core/bin/Debug/net6.0/Polly.Core.dll
  Polly.Core -> /home/runner/work/Polly/Polly/src/Polly.Core/bin/Debug/net462/Polly.Core.dll
/home/runner/work/Polly/Polly/.dotnet/sdk/7.0.402/Roslyn/Microsoft.CSharp.Core.targets(80,5): error MSB6006: "csc.dll" exited with code 1. [/home/runner/work/Polly/Polly/src/Polly/Polly.csproj::TargetFramework=net6.0]

This only affects compilation for CodeQL so I imagine that something there is part of the problem, however I've attempted to get more information out of the build using approaches like --verbosity diagnostic and /p:ErrorLog=./msbuild.log (e.g. martincostello/alexa-london-travel-site#1910), but to no avail. I don't seem to be able to get any further detail out of the compiler as to why it is failing.

In this case today, it failed 5 times in a row.

Maybe this is related to #42616 and some custom analysis for CodeQL is the root cause, but unless I'm incorrectly configuring something, the C# compiler should emit some more detailed information about what is wrong to aid resolution of the problem.

Version Used:

At least the following versions of the .NET SDK have manifested this issue:

• 7.0.400
• 7.0.401
• 7.0.402

Steps to Reproduce:

Run either of these two workflows in GitHub Actions to perform CodeQL code analysis:

• .github/workflows/codeql-analysis.yml in App-vNext/Polly
• .github/workflows/codeql-analysis.yml in martincostello/alexa-london-travel-site

Diagnostic Id:

MSB6006

Expected Behavior:

The code compiles.

Actual Behavior:

Errors such as the below occur:

/home/runner/work/Polly/Polly/.dotnet/sdk/7.0.402/Roslyn/Microsoft.CSharp.Core.targets(80,5): error MSB6006: "csc.dll" exited with code 1. [/home/runner/work/Polly/Polly/src/Polly/Polly.csproj::TargetFramework=net6.0]

[jaredpar]

Sorry, this bug somehow slipped through my triage board. I just came across it again today.

Is this still happening? If so can you try to get a binary log of the build (use the -bl argument)? That will give us a full stack and we can start tracking down where this is coming from.

[martincostello]

I'm not sure if it's would still happen to be honest.

In Polly this workaround seemed to avoid the problem App-vNext/Polly@160efa5, and this seems to have done a similar job in the other repo: martincostello/alexa-london-travel-site@e372a22

When I get a moment I'll undo the workaround in a branch and see if it comes back. If it does I'll try and grab a binary log.

[martincostello]

No errors after 5 re-runs in Polly and 10 in the other repo.

I'll close this for now on the assumption something somewhere fixed this, and merge the removal of my workaround. I'll open a new issue referencing this one if it comes back again.