BUG: 0th power of sparse array/matrix always returns the identity of float64

[takagi]

Describe your issue.

The power of sparse array/matrix is expected to return one with the same dtype. However, in SciPy 1.8.0rc1, its 0th power always returns the identity of float64. I guess it was unintentionally missing in c3e4ec5.

Reproducing Code Example

>>> import numpy as np
>>> import scipy.sparse as sp
>>> a = sp.coo_array(np.ones((2, 2), dtype='D'))
>>> a.dtype
dtype('complex128')
>>> (a**0).dtype
dtype('float64')

Error message

N/A

SciPy/NumPy/Python version information

1.8.0rc1 1.22.0rc1 sys.version_info(major=3, minor=9, micro=7, releaselevel='final', serial=0)

[takagi]

xref: cupy/cupy#6233

[tylerjereddy]

@stefanv @WarrenWeckesser should this be on the radar for 1.8.0 as well?

[WarrenWeckesser]

@tylerjereddy, I think so. Ideally any PRs that fix problems discovered in the release candidate that were missed in the review of #14822 should be targeted for 1.8.0. I'm not singling out #14822 for special treatment; this applies to any PRs that fix problems with new 1.8 features that are discovered in the release candidate, such as #15220, which fixes a small issue that is the result of the big namespace clean up. (That particular fix, however, is definitely not a high priority, so bumping it to 1.8.1 would be fine.)

[tylerjereddy]

Right, I just meant to assess priority.

Regardless, this seems to be fixed by the now-merged gh-15225