FindPython: Regression on GitHub Actions macOS 11/12 with Python 3.11.3 and CMake 3.26.3

[dalcinl]

I originally reported this issue to CMake: https://gitlab.kitware.com/cmake/cmake/-/issues/24811
After doing some local testing on a macOS machine that allowed me to reproduce the failure, maybe the problem is related to scikit-build-core and/or build environment isolation, and not CMake.
For your convenience, I paste below my original issue report to CMake:

FindPython cannot find the Interpreter Development.Module components [logs].

• GitHub-hosted runners
• macOS 11 and 12
• Python 3.11.3 (from actions/setup-python@v4 with python_version: 3)
• CMake 3.26.3 (AFAICT, Hombrew installation in /usr/local)

This is mpi4py CI infrastructure trying to build with the new scikit-build-core Python build backend at version 0.3.0. The build uses Python from an isolated build environment. Previous builds from 3 days ago where using Python 3.11.2 and CMake 3.26.1, and were all good.

I don't have tons of experience with scikit-build-core or CMake. Let me know what else I can do on my side to further debug the issue.

[henryiii]

This might be a side effect of adding the python lib location in 0.3.0. I thought this was fixed starting in CMake 3.24 (and we backport before 3.26.1). Could you test with scikit-build-core<0.3? We really only need the lib location on Windows, so for now we could revert this on non-Windows systems.

[dalcinl]

This might be a side effect of adding the python lib location in 0.3.0.

Can you point me to the commit/PR that added this change?

Could you test with scikit-build-core<0.3?

Looks like you may be right, Henry. With scikit-build-core < 0.3, all builds are OK.

[dalcinl]

BTW, why aren't you testing the latest CMake version in CI? something like cmake-version: ["3.15.x", "latest"] in the build matrix should do (according to docs from jwlawson/actions-setup-cmake).

[henryiii]

#263 (initial PR and reasoning in scikit-build)

We do test on a matrix. I think this is a fairly specific issue.

[dalcinl]

We do test on a matrix. I think this is a fairly specific issue.

Yes, I know. I noticed you are not testing the latest CMake on all platforms (only Windows), and somehow assumed that this issue is related to latest CMake 3.26.3, not I understand that may not be the case.

Anyway, many thanks for your prompt action and the new patch release. I confirm #285 fixed the issue in mpi4py CI runs.