The parsing is actually implemented twice: once in numpy/lib/function_base.py for np.vectorize in python (which you fixed and test), and again in '_parse_signature from 'numpy/core/src/umath/ufunc_object.c for ufuncs. The latter is tested in numpy/core/test/test_ufunc.py via various test_signature functions, but I don't see any that test white space. Could you add some tests there too?

The parsing is actually implemented twice: once in numpy/lib/function_base.py for np.vectorize in python (which you fixed and test), and again in '_parse_signature from 'numpy/core/src/umath/ufunc_object.c for ufuncs. The latter is tested in numpy/core/test/test_ufunc.py via various test_signature functions, but I don't see any that test white space. Could you add some tests there too?

Ah, thanks for pointing that out! Sure, I will take a look and update the PR.

Update:

There is already one signature parsing test in numpy/core/test/test_ufunc.py that tests for whitespace. I have added a couple more.

The C-code apparently accepts both \t and space, maybe we should do the same? Also I am wondering if a sig = sig.replace(" ", "") is not just as well, regex is always a bit tricky.

The C-code apparently accepts both \t and space, maybe we should do the same? Also I am wondering if a sig = sig.replace(" ", "") is not just as well, regex is always a bit tricky.

The regex I have used should take care of all whitespace characters, not just space (unless I'm missing something).

Replacing all whitespaces in the input signature would also involve using regex: sig = re.sub(r'\s+', "", sig). But I agree, even that is more clear and concise than the current change.

Thanks @ankitdwivedi23 .