Fix imshow to work with subclasses of ndarray.

[l-johnston]

PR Summary

Closes #18077

In the internals of _ImageBase._make_image, the clip bounds arithmetic fails with ndarray subclasses that support masked arrays such as unyt.

PR Checklist

• Code is Flake 8 compliant
• Code has test coverage

[jklymak]

Seems right - OTOH do you have a test?

[l-johnston]

I took a look at matplotlib's Quantity defined in test_units.py, but it doesn't have all the ndarray features unyt has that are needed to test the fix. And I didn't want to import unyt.

[jklymak]

(Please don't put the issue in title, but yes, please do put issue in description!)

[l-johnston]

I might be able to implement a skeleton ndarray subclass with just enough features to reproduce the bug and then test the fix. Is this of interest?

[dopplershift]

If it's a huge amount of effort, I think it would be worthwhile. That should have said: "If it's NOT a huge amount of effort, I think it would be worthwhile."

[l-johnston]

@dopplershift I implemented the skeleton ndarray subclass that reproduces the bug and validates the fix. Is this sufficient?

[dopplershift]

That seems fine to me.

There is a slight question as to whether this should go into test_units.py (see #18278), but I'm not holding this up for that.

[dstansby]

Thoughts on backporting to 3.3.x as a bugfix?

[neutrinoceros]

Thanks for fixing this !