BUG: handle NaNs in FloatingArray.equals

[jbrockmendel]

• closes BUG: FloatingArray.equals should consider NaNs in same location as equal #44382
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

[jorisvandenbossche]

Thanks! One small comment

[jorisvandenbossche]

Could you do something like left2 = FloatingArray(data.copy(), mask.copy()) to create the other array to test equality with? I would just avoid the special case of "identical" objects (in case we might add a fastpath for that in the future)

[jbrockmendel]

updated; is this what you had in mind?

[jorisvandenbossche]

That's fine!
Can you add a small whatsnew note?

[jreback]

do we have reference docs for these anywhere? (i don't remember), if so can you add otherwise lgtm

[jreback]

actually would be great to tests Series equals with FloatDtype (does it dispatch to this?)

[jorisvandenbossche]

do we have reference docs for these anywhere? (i don't remember), if so can you add otherwise lgtm

Yes. The relevant one for users is https://pandas.pydata.org/docs/dev/reference/api/pandas.Series.equals.html#, and the ExtensionArray version is also listed: https://pandas.pydata.org/docs/dev/reference/api/pandas.api.extensions.ExtensionArray.equals.html

actually would be great to tests Series equals with FloatDtype (does it dispatch to this?)

Yes, it does dispatch, and that's already tested.

[jreback]

where is the series test explicitly added here?

[jorisvandenbossche]

It is not added here because we already have a general extension test (test_equals) that ensures both DataFrame and Series dispatch to the EA method.

[jreback]

sure but do we have an example that was failing with Float dtypes that check nans - we couldn't have as it would have failed - so need this case

[jorisvandenbossche]

We have a lot of specific tests in the /tests/arrays that are not all repeated each time for Series or DataFrame. And I think that is fine.

[jbrockmendel]

i think this wasnt ready, had failures on older numpys

[jorisvandenbossche]

It seems the equal_nan keyword is new in numpy 1.19 (https://numpy.org/doc/stable/reference/generated/numpy.array_equal.html)

[jbrockmendel]

shimmed a fix for this into #44428

[jorisvandenbossche]

Can you quickly put it in its own PR? That seems easier to merge it quickly, while the other PR still needs to be reviewed?