API should empty ._get_bool_data have RangeIndex?

[MarcoGorelli]

This came up in #50636

In [9]: Series([], dtype='bool').index
Out[9]: RangeIndex(start=0, stop=0, step=1)

In [10]: Series([], dtype='bool')._get_bool_data().index
Out[10]: NumericIndex([], dtype='int64')

cc @topper-123

[jbrockmendel]

I'd expect it to always have the same index type as the original DataFrame

[topper-123]

I'd expect it to always have the same index type as the original DataFrame

+1.

[topper-123]

I've looked a bit into this. The underlying issue is that _get_bool_data() does an implicit slice and slicing a RangeIndex always returns a NumericIndex[dtype=int64].

>>> import pandas as pd
>>> ri = RangeIndex(0)
>>> ri[[]]
NumericIndex([], dtype='int64')  # soon Index([], dtype='int64')

I think the above is correct, i.e. slicing a RangeIndex should always return the same index type, i.e. a NumericIndex, even if its has length 0. SO IMO RangeIndex behaves correctly.

So the issue here seems to be in BaseBlockManager._combine where the use of slicing the index is forced, which does seems to be not needed in this case. So the question is maybe if slicing there should be not done in some case (especially where there is only 1 block).