Wow, cool! I have to admit, I do not walk into the depth of distutils, but it does seem great to me. If you want style nitpicks, I think textwrap.dedent() would make for nicer inline code than '\n'.join(...) ;)

EDIT: to be clear, I don't see why not to use this test. Although I am not sure if there are any skip needed potentially?

Wow, cool! I have to admit, I do not walk into the depth of distutils, but it does seem great to me. If you want style nitpicks, I think textwrap.dedent() would make for nicer inline code than '\n'.join(...) ;)

EDIT: to be clear, I don't see why not to use this test. Although I am not sure if there are any skip needed potentially?

I do like dedent better. I had to use a combination of indent/dedent to do the right thing for Fortran 77 (first 6 columns are reserved). The only skips in the numpy random C API tests are for slow (which I have) and if cython/numba/cffi are not present (which aren't used here), so I think we should be good on skips.