While probably right (don't know this stuff well enough to say right away). This requires a test to be merged (the tests are usually in the same folder where the code is in the "tests" subfolder.

Also, please try to format your git commit according to our guidelines, such as BUG: .... Writing the text also in the git commit itself may be nice too, since it will ease understanding if anyone has to look at it again.

Otherwise, thanks for creating a PR :)

Why does MaskedArray._optinfo even exist? Is it documented? Can we deprecate it?

@seberg: I added a simple test and sorry for ignoring your commit message format.
@eric-wieser: I have of course no idea why _optinfo was introduced, but it is a convient way to add attributes to a MaskedArray subclass, I would miss it! I once read about it here.

...allright, I try to bring this issue back to the agenda by providing more details (maybe I should have done this earlier). At the moment I can add items to the _optdict attribute and the dictionary's content is usually propagated to masked arrays created during operations on the original one:

x = np.ma.ones(10)
x._optinfo['test'] = True
(x > 4)._optinfo  # {'test': True}
(x * 4)._optinfo  # {'test': True}

Only the __eq__/__ne__ methods behave differently:

(x != 4)._optinfo  # {}
(x == 4)._optinfo  # {}

I understand, that this is not a big issue for you, in the end it only affects a hardly documented feature. But it is still an inconsistency with a rather simple fix, that bit me, so I would be happy if this could be merged!
Anyhow, if you do not think that is worth it, I simply close the PR.
Cheers!

Ok, if comparison preserves it, then I think I can accept that as an argument to make equality tests do so too

Done...

Closing in favor of #9540

Why did you close this?

He had created it against his own master, which is tricky stuff. Possibly you can change the backing branch nowadays, not sure.

@seberg I haven't found a solution to do so...
@eric-wieser ...that's why I opened #9540.