BUG: Array.__setitem__ failing with nullable boolean mask

[charlesdong1991]

• closes REGR: Array.__setitem__ failing with nullable boolean mask #31446
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[WillAyd]

Looks generally good

[WillAyd]

Suggested change

	- Bug in ``__setitem__`` in ``Array`` which fails with nullable boolean mask (:issue:`31446`)
	- Bug where assigning to a Series using a IntegerArray / BooleanArray as a mask would raise ``TypeError`` (:issue:`31446`)

[jorisvandenbossche]

Added some comments.
Those are not strictly needed to fix the regression (only nullable integer already existed before), but I think it could be good to fix this more generally

[TomAugspurger]

@charlesdong1991 I haven't looked closely, but you may need to skip that setitem test for PandasArray. There are some issues with setting sized values in an ndarray.

[charlesdong1991]

ahh, thanks a lot for letting me know!! @TomAugspurger

i am indeed struggling with one failed test which is caused by PandasArray
seeing the error message, looks like original array is pd.array([0, 0, 0, (3,), (4,)]), while somehow expected became: pd.array([(0,), (0,), (0,), (3,), (4,)]).

I will skip this for now, and pls let me know if you think a new issue needs to be created or not.

just FYI, the only failed test

self = <pandas.tests.extension.test_numpy.TestSetitem object at 0x14d3eea20>
data = <PandasArray>
[    0,     0,     0,  (3,),  (4,),  (5,),  (6,),  (7,),  (8,),  (9,), (10,),
 (11,), (12,), (13,), (14,...(87,),
 (88,), (89,), (90,), (91,), (92,), (93,), (94,), (95,), (96,), (97,), (98,),
 (99,)]
Length: 100, dtype: object

    def test_setitem_nullable_mask(self, data):
        # GH 31446
        arr = data[:5]
        expected = data.take([0, 0, 0, 3, 4])
        mask = pd.array([True, True, True, False, False])
        arr[mask] = data[0]
>       self.assert_extension_array_equal(expected, arr)

pandas/tests/extension/base/setitem.py:205: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pandas/_testing.py:1057: in assert_extension_array_equal
    obj="ExtensionArray",
pandas/_libs/testing.pyx:65: in pandas._libs.testing.assert_almost_equal
    cpdef assert_almost_equal(a, b,
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

>   raise_assert_detail(obj, msg, lobj, robj)
E   AssertionError: ExtensionArray are different
E   
E   ExtensionArray values are different (60.0 %)
E   [left]:  [(0,), (0,), (0,), (3,), (4,)]
E   [right]: [0, 0, 0, (3,), (4,)]

[MarcoGorelli]

:class:`Series`

instead of

``Series``

so that there's a link in the docs?

[charlesdong1991]

yeah, tbh i wasn't quite sure about the rule here, since i see three different representations used in whatsnew

:class:`Series`, ``Series``, Series

but seems having :class:Series is more consistent, will change

[jreback]

can you create an issue for this? (e.g. PandasArray support for setitem)

[charlesdong1991]

will do

[martinfleis]

@charlesdong1991 is there an opened issue for this yet? I haven't found it and it causes CI of geopandas to fail as contrary to the comment this test is not skipped.

[jorisvandenbossche]

@martinfleis The issue for geopandas is not due to PandasArray not working, but rather because we need to update our __setitem__ implementation (similar to how I updated the __getitem__), will do a PR for that.

[jreback]

thanks @charlesdong1991