runtime root build fails on Windows

[ManickaP]

Fresh main, repo cleaned with git clean -dfx and then running .\build.cmd clr+libs -rc release ends with:

C:\Users\mapichov\Repositories\runtime\src\tools\illink\src\ILLink.RoslynAnalyzer\DataFlow\LocalDataFlowState.cs(30,13): error IDE0031: Null check can be simplified (https://learn.microsoft.com/dotnet/fundamentals/code-analysis/style-rules/ide0031) [C:\Users\mapichov\Repositories\runtime\src\tools\illink\src\ILLink.RoslynAnalyzer\ILLink.RoslynAnalyzer.csproj]

Build FAILED.

C:\Users\mapichov\Repositories\runtime\src\tools\illink\src\ILLink.RoslynAnalyzer\DataFlow\LocalDataFlowState.cs(30,13): error IDE0031: Null check can be simplified (https://learn.microsoft.com/dotnet/fundamentals/code-analysis/style-rules/ide0031) [C:\Users\mapichov\Repositories\runtime\src\tools\illink\src\ILLink.RoslynAnalyzer\ILLink.RoslynAnalyzer.csproj]
    0 Warning(s)
    1 Error(s)

Time Elapsed 00:00:49.89
Build failed with exit code 1. Check errors above.
Some builds failed:
        Configuration: Debug, Architecture: x64

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/illink
See info in area-owners.md if you want to be subscribed.

[ManickaP]

I just retested this on a completely new and clean DevBox and it reproduces. Any ideas? Any possible workarounds for this?

[filipnavara]

It is happening on my machine too (both Windows and macOS), so it's not an isolated incident. I locally disabled the warning in Directory.Build.props to get any work done at all.

[jkotas]

Any possible workarounds for this?

Set RunAnalyzers=false in your environment.

[akoeplinger]

I looked at this with @filipnavara and we found what looks like a Roslyn bug that caused this warning to not be reported on some machines even though it should. Tracked at dotnet/roslyn#80301

[filipnavara]

For posterity, the rule for this specific warning is specified on two places with conflicting values:

runtime/.editorconfig

Line 102 in 838491d

dotnet_style_null_propagation = true:suggestion

runtime/eng/CodeAnalysis.src.globalconfig

Lines 1669 to 1670 in 838491d

	# IDE0031: Use null propagation
	dotnet_diagnostic.IDE0031.severity = warning

It seems that the order of /analyzerconfig parameter processing in CSC is not deterministic. On some of my machines the first one wins, on others the second one. Depending on that the compilation either passes or fails.

To add insult to the injury, binlog/complog doesn't capture this in reproducible way since the non-determinism happens in CSC itself. Tested with multiple different CSC versions (eg. 4.14.0-3.25413.5, 5.0.0-2.25420.111). The same CSC version on different machine behaved differently.