Slicing bug gh 1135

[oleksandr-pavlyk]

Closes gh-1135.

Handle x[True] and x[False] as NumPy does, even though the behavior
may be undocumented. NumPy treats True as None (insert axis with size 1),
and treats False as None followed by empty slicing (insert axis with size 0).

Changed the logic of _basic_slice_meta utility function to correctly handle
boolean scalars (surprisingly, insinstance(True, int) evaluates to True).
0d arrays are handled on the same footing as the Python scalars.

Introduced _is_integral and _is_boolean utilty functions and used them in
_basic_slice_meta utility.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?

[github-actions]

View rendered docs @ https://intelpython.github.io/dpctl/pulls/1136/index.html

[github-actions]

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_21 ran successfully.
Passed: 46
Failed: 788
Skipped: 280

[antonwolfy]

The below example works differently with numpy:

import numpy
import dpctl.tensor as dpt

numpy.ones((3, 2))[True, False].shape
Out: (0, 3, 2)

dpt.ones((3, 2))[True, False].shape
Out: (1, 0, 3, 2)

[oleksandr-pavlyk]

Actually this behavior in NumPy is undocumented and we should not rely on it: https://stackoverflow.com/questions/75828008/docs-about-boolean-scalars-in-indexing-of-numpy-array

I will change the implementation to raise IndexError for boolean scalar indexing.

[antonwolfy]

Looks it was answered here, referred from numpy issue #9515, and stated as expected behavior intended for 0-dimensional boolean index in numpy.

[oleksandr-pavlyk]

Ok, in that case, I am inclined to leave things as is. The behavior of np.nonzero on non-zero array is slated to change, so this behavior will change in NumPy too.

[github-actions]

Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞

[github-actions]

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_21 ran successfully.
Passed: 46
Failed: 788
Skipped: 280