Peter Sobolewski

Seems rather like an issue with whatever sdl2 is?

sdl2-feedstock is archived, but indeed the last recipe there is using the old cdt. Looks like sdl3 is what's maintained now on conda-forge under sdl-feedstock. And that one uses the...

https://github.com/libsdl-org/sdl2-compat

Looks like the current sdl-feedstock already has that? https://github.com/conda-forge/sdl-feedstock/pull/21

Yeah, looks like that was already used, if you look at the conda dependencies in this PR: + sdl2 2.32.54 h3f2d84a_0 conda-forge 587kB + sdl3 3.2.12 he3e324a_2 conda-forge 2MB So...

@brisvag I'm curious if using sdl2

I think you need to pin sdl2. Right now it's still the compatibility wrapper: Linking sdl2-2.32.54-h3f2d84a_0

Yeah :( I think it would be best to split the sdl2 into own job with old pyopengl and unpin for everything else?

Maybe I wasn't clear. My suggestion is to just add an extra job here: https://github.com/vispy/vispy/blob/4f1be84caf69e93156eed5a0a96aa1056adfb77a/.github/workflows/main.yml#L171-L192 that is like number 2 in the matrix (the failing one), but skips sdl2, so...

I still think it's split off sdl tests from other backends.