Newer system gets an extra skip on exit in testing

[HinTak]

Since I upgraded to fedora 39 (and upgraded to python 3.12), my pytest statistics seem to get an extra skip. I first noticed it as my number seems to differ from a few days ago (with only a re-compile between) and CI but I thought it might be just some extra local code I have. Now I sync'ed up everything, and it is still happening. I tried using xrvb-run like CI, still the extra skip.

Somehow, while I have a pretty normal mesa-based openGL , but I get this line executed on exit:

skia-python/tests/conftest.py

Line 52 in a942b21

pytest.skip('OpenGL is not available')

While the ubuntu CI machines don't.

This isn't Ubuntu vs Fedora, bit fedora 38 vs fedora 39.

So my numbers don't add up. The sum of all the numbers at the end is 1 larger than the total showed at the beginning of pytest.

Strange - it is probably a pytest version or something else.

[HinTak]

The pytest statistics is a bit more complicated lately - it used to be just differing by openGL (linux + newer python on mac os x, vs older python on mac os x and Windows) on 87.x . Now with the upcoming m120, there are 4:

Linux: have both opengl and vulkan
Windows: vulkan only
Old python on mac os: neither
New python on mac os: opengl only

I suspect the vulkan api exists in 87 but not always functional.

[HinTak]

Ha, interesting this extra skip on exit seems to affect mac os too:

Linux
2072 passed, 108 skipped, 13 xfailed

Windows
1848 passed, 339 skipped, 6 xfailed, 1 warning

Mac
2061 passed, 120 skipped, 13 xfailed

So we have fedora 39 and mac os newer python, vs everybody else (windows, ubuntu, fedora 38, mac os with python 3.7/3.8)