Alternative implementation for .where on EA-backed Indexes

[TomAugspurger]

Followup to #24114.

Right now Categorical.where and IntervalIndex.where involve converting to objects.

We could use an alternative implementation that copies the underlying array and sets the values.
#24114 (comment) should give a rough idea of the kinds of speedups you'd get.

[charlesdong1991]

I am sorry to bother, could I ask you a question, each time when i tried to do the pytest, even with no change at all after git rebase, I got ImportError while loading conftest '/Users/kaiqi/pandas/pandas/tests/indexes/conftest.py'. pandas/tests/indexes/conftest.py:21: in <module> tm.makeCategoricalIndex(100), pandas/util/testing.py:1902: in makeCategoricalIndex return CategoricalIndex(np.random.choice(x, k), name=name, **kwargs) pandas/core/indexes/category.py:129: in __new__ dtype) pandas/core/indexes/category.py:192: in _create_categorical dtype=dtype) pandas/core/arrays/categorical.py:413: in __init__ raise NotImplementedError("> 1 ndim Categorical are not " E NotImplementedError: > 1 ndim Categorical are not supported at this time error, but i tested on pandas-dev/ci, there is no such problem. Do you have any idea where this problem might come from and how can I fix it locally? thank you so much!!

[TomAugspurger]

Did you rebuild the C extensions? python setup.py build_ext -i

[charlesdong1991]

ahh, it works... @TomAugspurger thanks so much!! i will keep this in mind!!!

[jbrockmendel]

Index.where now dispatches to Index.putmask, which has a non-object-casting implementation on ExtensionIndex. closing.