BUG: IntervalIndex.astype("category") doesn't preserve exact interval dtype in categories

[jorisvandenbossche]

Somewhere in the conversion, before factorizing, we convert the interval array/index to a object-dtype numpy array of Interval objects, and so afterwards infer the IntervalDtype again when creating the categories.

Example consequence is that if you have uint64 intervals, they get inferred as int64 afterwards:

In [29]: index = pd.IntervalIndex.from_breaks(np.arange(5, dtype="uint64"))

In [30]: index
Out[30]: 
IntervalIndex([(0, 1], (1, 2], (2, 3], (3, 4]],
              closed='right',
              dtype='interval[uint64]')  # <---- unsigned ints

In [31]: pd.CategoricalIndex(index)
Out[31]: CategoricalIndex([(0, 1], (1, 2], (2, 3], (3, 4]], categories=[(0, 1], (1, 2], (2, 3], (3, 4]], ordered=False, dtype='category')

In [33]: pd.CategoricalIndex(index).dtype.categories
Out[33]: 
IntervalIndex([(0, 1], (1, 2], (2, 3], (3, 4]],
              closed='right',
              dtype='interval[int64]')  # <---- no longer uint64

[Alvinwuzw]

Hi! I can reproduce the issue, could I look into this issue?

[Kyrpel]

I have confirmed in the latest main branch that this is not an issue anymore since both output the same
IntervalIndex([(0, 1], (1, 2], (2, 3], (3, 4]], dtype='interval[uint64, right]')