CI complain about missing standard module interface for Windows build

[filipsajdak]

Some tests are failing due to the following error:

cpp2util.h(48): fatal error C1011: cannot locate standard module interface. Did you install the library part of the C++ modules feature in VS setup?

[alexreinking]

Thanks for opening this issue... I have not had a chance to look at this.

[alexreinking]

@JohelEGP -- do you have any idea why this is happening? It used to work, then became intermittent, and now is consistently failing.

[JohelEGP]

I suppose it's the pure2 tests that are failing because they're trying to use C++ modules. I can't find platform-specific flags in the CMake sources, so perhaps it's missing one of the advertised ones at https://github.com/hsutter/cppfront#how-do-i-build-cppfront.

[alexreinking]

so perhaps it's missing one of the advertised ones

No, unfortunately it's not so simple. Here's one of the command lines from the tests:

C:\PROGRA~1\MICROS~2\2022\ENTERP~1\VC\Tools\MSVC\1434~1.319\bin\HostX64\x64\cl.exe  /nologo /TP  -external:ID:\a\cppfront\cppfront\_local\include -external:W0 /DWIN32 /D_WINDOWS /EHsc /O2 /Ob2 /DNDEBUG -std:c++20 -MD /experimental:module /showIncludes /FoCMakeFiles\test-case.dir\_cppfront\pure2-bounds-safety-span.cpp.obj /FdCMakeFiles\test-case.dir\ /FS -c D:\a\cppfront\cppfront\build\regression-tests\pure2-bounds-safety-span\build\_cppfront\pure2-bounds-safety-span.cpp

I can clearly see both -std:c++20 and /EHsc in there.

[JohelEGP]

Then it might be -experimental:module from https://github.com/hsutter/cppfront/blob/main/regression-tests/test-results/msvc-2022/run-tests-msvc-2022.bat.

[alexreinking]

Ugh, I wish I knew why that wasn't always needed. We can try adding that.

[JohelEGP]

It seems like it's already added at https://github.com/modern-cmake/cppfront/blob/main/regression-tests/cmake/CMakeLists.txt.in#L14.

[alexreinking]

It seems like it's already added

Oh, and sure enough, it's in the command line above... maybe removing it would help, if the MSVC version is new enough?

[JohelEGP]

Another difference is that https://github.com/hsutter/cppfront/blob/main/regression-tests/test-results/msvc-2022/run-tests-msvc-2022.bat uses -std:c++latest.

[alexreinking]

We could try this

target_compile_features(
    cppfront_cpp2util
    INTERFACE
    $<IF:$<CXX_COMPILER_ID:MSVC>,cxx_std_23,cxx_std_20>
)

instead of this: https://github.com/modern-cmake/cppfront/blob/43340d35bbc34d8cb788e627dee2638fa3dcf218/CMakeLists.txt#L57

That should give -std:c++latest for MSVC and will turn into -std:c++23 in future versions.

[JohelEGP]

I don't think that's it. It still fails: https://compiler-explorer.com/z/aGaj5z58q.

[JohelEGP]

Maybe it's just as the error message says, and there's something that needs to be installed in the environment.

[JohelEGP]

See https://learn.microsoft.com/en-us/cpp/error-messages/compiler-errors-1/fatal-error-c1011?view=msvc-170.

[alexreinking]

See https://learn.microsoft.com/en-us/cpp/error-messages/compiler-errors-1/fatal-error-c1011?view=msvc-170.

Yeah, that's not going to work on GitHub Actions, though.

[JohelEGP]

Maybe you can downgrade the version of MSVC to one that works: https://github.com/ilammy/msvc-dev-cmd#inputs.

[alexreinking]

This is frustrating... I cannot reproduce the error locally (with VS2022, v14.34 toolset, same as GHA). Everything "just works" here. The documentation suggests that the C++ modules are indeed installed (component Microsoft.VisualStudio.Component.VC.Modules.x86.x64)

[alexreinking]

I've put in the time I have to look at this during the week. I'll try again on the weekend.

[alexreinking]

This is no longer the issue with MSVC. Now we have issues in the build/check tests