Is the min 3.12 requirement new?

Not really. It is already in the docstring and I am just making the implementation consistent. This line was added in #5486 by @taldcroft .

Oh, the the 32bit failure is related

Re: 32-bit -- I think the failure is not so much 32-bit but the parallel part. Maybe __doctest_requires__ = {'*': ['yaml']} doesn't work well with pytest-xdist. I'll have to ponder this.

ugly workaround just for this, but what if we run the 32bit tests with alldeps? I would generarly feel a bit better to have alldeps the default to make sure we run as many tests as possible for a setting, and only have one run with the mandatory only to make sure the bare minimum works as well, rather the other way around.

pinging @astrofrog as he may recall if there was any such reason to do the 32bit this way.

I am a little afraid that adding alldeps to 32-bit will open a can of worms but why not try it?

yeah, I wouldn't be surprised by the tiny monster 🐛 crawling out from their hides

Yup... Can o' worms:

    src/checkdep_freetype2.c:9:22: fatal error: ft2build.h: No such file or directory
     #include <ft2build.h>
                          ^
    compilation terminated.
    error: command 'gcc' failed with exit status 1

I'll think of a way to just do a custom skip on that doctest...

I think I addressed all the comments. Had to move the docstring to RST though to use the doctest directive that we know would work. I successfully ran all the tests locally without yaml installed with this patch. Let's see what the CI says... 🤞 CI passed!

I don't think this needs a changelog, nothing really changed on for the users.

Thanks @pllim for taking care all these import issues.

I hope it works. I think I must have missed something but I think @mhvk fixed it somewhere else. I am sorry!

Yes, fixed elsewhere. And, overall, very nice to have those imports be truly optional! It would be even nicer if there was some way of bailing from module loading (with __all__ empty), but I don't think that's possible.