Support request: unexpected new analyzer failure when no relevant change was made

[Smaug123]

Hi - I've got a new analyser failure on https://github.com/Smaug123/fsharp-prattparser/pull/9 which is a change completely unrelated to the analysers. I can't reproduce it locally on macOS. The failing run is https://github.com/Smaug123/fsharp-prattparser/actions/runs/8123334984/job/22203652378?pr=9#step:4:1 and the preceding successful run (two weeks ago) was https://github.com/Smaug123/fsharp-prattparser/actions/runs/7965613558/job/21745483323 .

Do your eyes see what might be going wrong here? The .NET SDK and fsharp-analyzers are both binary-identical between the two runs (as validated by Nix); I haven't validated that the version of G-Research.FSharp.Analyzers is binary-identical.

[nojaf]

Hi Patrick, thanks for reaching out! I was able to run the analyzers locally on my Windows machine.

One thing I'm not entirely sure is if we support wildcards in --analyzers-path ./.analyzerpackages/g-research.fsharp.analyzers/*/

https://github.com/ionide/FSharp.Analyzers.SDK/blob/0719b7296843dc552e68bf6143d565227f2c1b33/src/FSharp.Analyzers.Cli/Program.fs#L543-L554

The aborted output text does make me wonder if type-checking succeeded:

https://github.com/ionide/FSharp.Analyzers.SDK/blob/25ab689f9285a81ff1c782e9ac62abed0eab785c/src/FSharp.Analyzers.SDK/FSharp.Analyzers.SDK.fs#L252

I don't quite see any good reason it shouldn't. Would you mind re-running that build sometime? I'm curious if the behaviour is consistent or not.

[Smaug123]

The wildcard should be interpreted by the shell, and the analysers should never even see it. (I'll verify that when I get home.) The error is consistent in the sense that I have several open PRs on that repo, all failing in the same way. Note that the analysers are the only failing step; in particular the dotnet build ran successfully in another step.

Actually now you mention it I can boot my Macbook into NixOS and get a testing environment that's pretty close to what the Action is running. I'll do that tonight.

[Smaug123]

Yeah, I'm confident the * was interpreted by the shell as a glob and was expanded before fsharp-analyzers got to see it:

Loading analyzers from /home/runner/work/fsharp-prattparser/fsharp-prattparser/.analyzerpackages/g-research.fsharp.analyzers/0.9.3/

[Smaug123]

Huh. When I added the .NET SDK to $PATH before running the analysers, it worked. I have no idea what changed underneath me, because it used to work fine without that, but that seems to reliably fix the problem. (Before that fix, the analysers reliably failed.)

[nojaf]

Hmm, this could explain why the design time build failed and thus type-checking no longer worked. I'm happy you found a solution. Was running out of ideas 😌.