API/BUG: get_indexer_non_unique(object_dtype)?

[jbrockmendel]

For get_indexer we do type inference on object dtype, but we dont do this for get_indexer_non_unique. Is this intentional?

dti = pd.date_range("2016-01-01", periods=3)
target = dti.astype(object)

>>> dti.get_indexer_non_unique(target)
(array([-1, -1, -1]), array([-1, -1, -1]))

>>> dti.get_indexer(target)
array([0, 1, 2])

We could also conceivably fall back to object-vs-object instead of returning no-matches.

There's also Cateegorical with matching categories dtype:

ci = pd.CategoricalIndex(dti)

>>> dti.get_indexer_non_unique(ci)
(array([-1, -1, -1]), array([-1, -1, -1]))

>>> dti.get_indexer(ci)
array([0, 1, 2])

[jbrockmendel]

@jreback can you weigh in on the desired behavior or (and possibly who else to ping)?

[jbrockmendel]

Fixed in or near #38105