gh-139935: do not skip test on real errors in os.getlogin

[picnixz]

This amends a PR I just merged. This is my fault here as I incorrectly assumed that all those errors were non-fatal.

• Issue: Some errors are not caught by test_os.LoginTests.test_getlogin #139935

[vstinner]

LGTM

[vstinner]

Thanks for the quickfix. I prefer to grow the list of ignored errno one by one, when we discover a clear failure in the wild, rather than ignoring them in advance without knowing if they can occur in practice.

[picnixz]

Yes, this makes sense. I don't think we'll actually see those EMFILE and ENFILE unless we're in a constrained environment. ENOMEM is only for getlogin() because it returns an allocated char * whereas getlogin_r takes a pre-allocated buffer. So the fix I merged was already too broad.

[vstinner]

ENOMEM is only for getlogin() because it returns an allocated char * whereas getlogin_r takes a pre-allocated buffer

os.getlogin() is implemented with getlogin() if getlogin_r() is not available :-)

[picnixz]

Oh! so indeed we should not catch this error even more.