Sorry - it's not totally clear to me what this fixes.

Can you add a test? Maybe there's an existing test for this in the masked array tests that check the other dtypes you can extend.

It looks like if I just extend the existing tests I can trigger the bug this PR is fixing. See ngoldbaum@4343115. You can pull that commit and push or I can just push directly to your PR branch, whichever you feel most comfortable with. Generally I don't push to people's PR branches without explicit permission.

Once there are tests I'm happy to merge this.

Oh I see now, my apologies Nathan, as a first time contributor I totally miss the testing inclusion with numpy development standards. I think I'll do the pull and push, but let me double check this just to make sure I'm understanding this correctly. Thank again.

Not ready to merge yet, I'm double checking something I just noticed.

Not ready to merge yet, I'm double checking something I just noticed.

Still double-checking?

I was checking the behavior of MaskedArray when is fill_value is not set. For the other dtypes it uses a default fill_value defined in default_filler dictionary

I think using "N/A" is fine. It might make sense eventually to also do something sensible if someone wants to use a fill value that isn't a string, but we can handle that then.

You can use StringDType with non-string fill values, so you can do stuff like this:

In [10]: dt = np.dtypes.StringDType(na_object=None)

In [11]: arr = np.array(['hello', 'world', None], dtype=dt)

In [12]: arr
Out[12]: array(['hello', 'world', None], dtype=StringDType(na_object=None))

In [13]: arr[2] is None
Out[13]: True

but "N/A" is probably a fine default to use. If you wanted to make it so non-string fill values work, you'll need to set na_object on the StringDType instance on the MaskedArray.

OK, awesome. It’s borderline but IMO this deserves a release note too.

See https://github.com/numpy/numpy/blob/main/doc/release/upcoming_changes/README.rst and the other release note fragments in that directory for instructions and examples.

Oops, looks like I forgot to come back to this. I'll try to give this another once-over and hopefully hit the merge button later today.

Let me know if there anything else you need me to do @ngoldbaum

Thanks for your patience @phdparedes and for the contribution! Hope to see your github handle again :)

hi folks, cheers for this - @ngoldbaum this has a side-effect that I'm not sure if it's a bug or a feature: previously a fill_value that was a one-element ndarray eg [-900.] was accepted and worked, now that'll toss a ValueError: setting an array element with a sequence in np.ma.masked_equal(data, missing_value) - see #30591