TST: Refactor to consistently use CompilerChecker

[HaoZeke]

Closes #27045, by using the CompilerChecker object to make sure tests which would need a compiler are skipped correctly.

TODO

• Documentation addition detailing the no-compiler and CompilerChecker bits in f2py-testing.rst.

Implementation note

• CompilerChecker basically runs meson, so if there are environments where the tests may be run without meson being installed at all then this needs to be reworked.
  • Like all the failing Windows builds In these scenarios the compiled tests can be skipped anyway.
• Soft reverts 17f529a

[HaoZeke]

The remaining test failures on Windows are real errors, not related to the compiler presence, which were previously being suppressed. (though I might still end up skipping them for now since they seem to be linking errors, i.e. system configuration related issues). These will be handled eventually with #25134 anyway.

2024-07-26T14:35:17.6418897Z FAILED: test_array_from_pyobj_ext.cp310-win32.pyd 
2024-07-26T14:35:17.6419172Z 
2024-07-26T14:35:17.6423831Z "link"  /MACHINE:x86 /OUT:test_array_from_pyobj_ext.cp310-win32.pyd test_array_from_pyobj_ext.cp310-win32.pyd.p/112443d308cc40139cb5e53d97c4d31678bcb07b_f2py_tests_src_array_from_pyobj_wrapmodule.c.obj test_array_from_pyobj_ext.cp310-win32.pyd.p/26ebc5753ea73c490d2b7fd0b1c7c485b8e26306_.._.._f2py_src_fortranobject.c.obj "/LIBPATH:C:/mingw64/lib/gcc/x86_64-w64-mingw32/8.1.0" "/LIBPATH:C:/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/8.1.0" "/LIBPATH:C:/mingw64/lib/gcc" "/LIBPATH:C:/mingw64/bin/../lib/gcc" "/LIBPATH:C:/mingw64/x86_64-w64-mingw32/lib" "/LIBPATH:C:/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/8.1.0/../../../../x86_64-w64-mingw32/lib/../lib" "/LIBPATH:C:/mingw64/lib" "/LIBPATH:C:/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/8.1.0/../../../../lib" "/LIBPATH:C:/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/8.1.0/../../../../x86_64-w64-mingw32/lib" "/LIBPATH:C:/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/8.1.0/../../.." "/release" "/nologo" "/OPT:REF" "/DLL" "/IMPLIB:test_array_from_pyobj_ext.cp310-win32.lib" "C:\hostedtoolcache\windows\Python\3.10.11\x86\libs\python310.lib" "quadmath.lib" "kernel32.lib" "user32.lib" "gdi32.lib" "winspool.lib" "shell32.lib" "ole32.lib" "oleaut32.lib" "uuid.lib" "comdlg32.lib" "advapi32.lib" "gfortran.lib"
2024-07-26T14:35:17.6428982Z 
2024-07-26T14:35:17.6429200Z LINK : fatal error LNK1181: cannot open input file 'quadmath.lib'

[mattip]

Is skipping windows f2py CI critical enough that this should wait for #25134?

[HaoZeke]

Is skipping windows f2py CI critical enough that this should wait for #25134?

I think skipping here is OK since this PR is mostly a regression fix to unblock conda-forge, reported in #27045.

However, I'll try to get #25134 ready ASAP, just need to find some time to test on a clean Windows machine.

[HaoZeke]

Should be good to go for now.

[rgommers]

Thanks @HaoZeke! Changes LGTM; I'm testing this the lazy way to be sure: pushed a commit to trigger wheel builds.

[rgommers]

CompilerChecker basically runs meson, so if there are environments where the tests may be run without meson being installed at all then this needs to be reworked.

It's a required test dependency also outside of f2py, so should be fine. Only ninja may be missing in a Pyodide environment, but CI wasn't complaining.

[rgommers]

All happy, so let's give it a go - thanks Rohit!