Fix "🐍 3 • windows-latest • mingw64" CI job; test_gil_scoped.py xfail macOS free-threading

[rwgk]

Description

Apparently msys2/setup-msys2@v2 cannot be run twice anymore:

• https://github.com/pybind/pybind11/actions/runs/17394902023/job/49417376616?pr=5796

Run msys2/setup-msys2@v2
  with:
    msystem: mingw64
    install: mingw-w64-x86_64-python-numpy mingw-w64-x86_64-python-scipy mingw-w64-x86_64-eigen3
    path-type: minimal
    update: false
    pacboy: false
    release: true
    location: RUNNER_TEMP
    platform-check-severity: fatal
    cache: true
  env:
    PYTHONDEVMODE: 1
    PIP_BREAK_SYSTEM_PACKAGES: 1
    PIP_ONLY_BINARY: numpy
    FORCE_COLOR: 3
    PYTEST_TIMEOUT: 300
    VERBOSE: 1
    CMAKE_COLOR_DIAGNOSTICS: 1
    MSYSTEM: MINGW64
Error: Trying to install MSYS2 to D:\a\_temp\msys64 but that already exists, cannot continue.

Suggested changelog entry:

• Placeholder.

[rwgk]

Hi @henryiii, could you please review? This should stop the frequent CI / 🐍 (macos-13, 3.13t, -DCMAKE_CXX_STANDARD=11) / 🧪 (pull_request) failures.

[henryiii]

This doesn't always fail, does it? So strict=False?

[rwgk]

The strict flag only applies to @pytest.mark.xfail(...) (the decorator/mark form). Inline pytest.xfail() doesn't take strict at all — it immediately aborts the test and reports XFAIL. There’s no concept of XPASS in this case, so strict isn't relevant here.

To be sure I looked it up the good-old way:

https://docs.pytest.org/en/stable/how-to/skipping.html#xfail-mark-test-functions-as-expected-to-fail

Note that no other code is executed after the pytest.xfail() call, differently from the marker.

[henryiii]

This didn't run before, now it will. Guessing that's fine?

[rwgk]

See other comment: Note that no other code is executed after the pytest.xfail() call, differently from the marker.

[henryiii]

I'd name this "extra_install", I think, since it's there to install extra stuff. And you can leave it off on the other job, variables default to empty.

[rwgk]

Done: commit bb8a794

[rwgk]

Thanks @henryiii!

WDYT about disabling the two failing CIBW tests in this PR, too, until someone gets a chance to fix those?

CIBW / iOS wheel macos-13 (pull_request)
CIBW / iOS wheel macos-13 (pull_request)Failing after 19s

CIBW / iOS wheel macos-14 (pull_request)
CIBW / iOS wheel macos-14 (pull_request)Failing after 11s

[rwgk]

Ugh...

It is not working as intended:

• 🐍 (macos-13, 3.13t, -DCMAKE_CXX_STANDARD=11) / 🧪

            if process.exitcode is None:
                assert t_delta > 0.9 * timeout
                msg = "DEADLOCK, most likely, exactly what this test is meant to detect."
                if env.WIN and env.PYPY:
                    pytest.xfail("[TEST-GIL-SCOPED] Windows PyPy: " + msg)
                elif env.MACOS and not env.sys_is_gil_enabled():
                    pytest.xfail("[TEST-GIL-SCOPED] macOS free-threading: " + msg)
>               raise RuntimeError(msg)
E               RuntimeError: DEADLOCK, most likely, exactly what this test is meant to detect.

@henryiii I'm a bit lost, should I maybe checking for env.PY_GIL_DISABLED here?

IIUC, it's failing even though the GIL is enabled? This has been happening with macOS ever since I added this test, but only fairly infrequently.