Everything seems tested and passes.

I propose we merge this to main.

If we have more issues with masked arrays and new numpy we have this in our backpocket to backport if needed but we do not put speculative changes on the bug-fix branch.

I am weakly against merging this. In all of these cases we are grabbing the mask, but then passing it straight back into numpy. If there is no mask (which numpy internally stores as numpy's version of False), then there is no need to pay the cost to expand that to an array of bools to pass it back into numpy.

I do think we should take the simplifications where we in-line the getmask calls.

Sure, we can just let it hang around and see if anything shows up.

I realize that the case which broke is quite different from these, so it may very well be that it is OK to pass a scalar mask here.

I'll try to remind myself of it when 3.7 is closing in and create a PR with the inlining if not merged by then.

Since this Pull Request has not been updated in 60 days, it has been marked "inactive." This does not mean that it will be closed, though it may be moved to a "Draft" state. This helps maintainers prioritize their reviewing efforts. You can pick the PR back up anytime - please ping us if you need a review or guidance to move the PR forward! If you do not plan on continuing the work, please let us know so that we can either find someone to take the PR over, or close it.

I close this as not worth pursuing.

There's the larger idea that we should get rid of internally using masked arrays and instead use regular arrays with NaNs. That would simplify logic and increase performance a bit, see #31008.