non-64-bit Indexes?

[jbrockmendel]

ser = pd.Series(np.arange(10, dtype=np.float128))
>>> ser.dtype
dtype('float128')

>>> pd.Index(ser).dtype
dtype('float64')

This makes dispatching Series ops for non-64 dtypes problematic. Is there a reason for enforcing only-64?

[jreback]

we don’t support this as it’s somewhat platform dependent

[jreback]

not necessarily objecting, but this would be tremendous work for very little gain.

[toobaz]

See #16404