BUG: Fixed where keyword for np.mean & np.var methods

[charris]

Backport of #18560.

Addresses #18552

I've managed to fix where keyword issue by refactoring those if statements.

So the rcount is a scalar when where is not provided as we reduce equal number of elements for given axis (or axes). So the first and statement prevents from illegal comparison of array to 0 value that caused original error. If the where clause is provided and rcount is an array (as we might want to reduce different number of elements along axis due to masking) then we check second and statement if any of those reduce groups are zero. I also maintained this micro-optimization to avoid calling _any(rcount == 0) when where is True and not zero.

Is this correct?

Also I had to change umr_any to _any as I think there is another bug there (or umr_any shouldn't be used there). np.any without axis parameter evaluates to scalar value (that can be used in if statement). But umr_any does not behave that way - it only reduces one dimension. Here's code for reproducing:

import numpy as np
from numpy.core import umath as um
a = np.ones((3,4))

# used in numpy/core/_methods.py
umr_any = um.logical_or.reduce

# np.any works as expected
if np.any(a == 0):
    pass

# but umr_any can't be used in if statement as here in `_mean`
if umr_any(a == 0):
    pass

Here's a colab reproduction: https://colab.research.google.com/drive/1UJi6E50zzaLu2nZQWMEFHPLZnq2v6pzT?usp=sharing

Thank you for any help!

• Fixed keyword bug

• Added test case

• Reverted to original notation

• Added tests for var and std

Closes gh-18552