API: Should NumericIndex([1 2, 3]).dtype be int64 or int32 on 32-bit systems?

[topper-123]

I've had some 32-bit issues while working on removing Int64Index and friends from the code base. One issue that I would appreciate input on the the desired dtype for numeric indexes on 32-bit systems. Currently we have (for 32-bit systems):

>>> from pandas.core.api import Series, NumericIndex
>>> Series([1, 2, 3]).dtype
dtype('int64')
>>> NumericIndex([1, 2, 3]).dtype
dtype('int32')

IMO these two should have the same dtype and because changing Series would break backwards compatability, I'd say that NumericIndex([1, 2, 3]).dtype should be dtype('int64')

Do you agree with me on this, or is there a need for discussion?

Added: BTW, NumericIndex will of course be removed also for pandas 2.0 and numerical functionality included in Index (as we previously have discussed), so the above question is the same as asking if in pandas 2.0 Index([1, 2, 3]).dtype should be int32 or int64 on 32-bit systems.

[jreback]

yes these should match and be 64bit