DatetimeIndex.to_frame() inconsistent with dti.to_series().to_frame()

[jbrockmendel]

For TZ-aware DatetimeIndex, to_series() drops the timezone unless we pass keep_tz=True

>>> dti = pd.date_range('1977-04-15', periods=3, freq='MS', tz='US/Hawaii')

>>> dti.to_series().to_frame().dtypes
0    datetime64[ns]

>>> dti.to_frame().dtypes
0    datetime64[ns, US/Hawaii]

IIRC there has been discussion about deprecating the need for keep_tz; this should be another reason to do so.

[jbrockmendel]

Never mind, this was true in 0.23 but not in master.

[jbrockmendel]

Actually, it looks like the keep_tz kwarg is not allowed for to_frame, so there is an inconsistency, just not the one I first thought.

[mroeschke]

Would the idea then be for to_series to always preserve the timezone?

[jbrockmendel]

I think that's the most reasonable behavior yes. IIRC there was a historical reason why it wasnt the default. @jorisvandenbossche I think you've opened an issue about changing this behavior?

[mroeschke]

xref #17832

[jorisvandenbossche]

Yes, we should deprecate keep_tz in to_series, that should make to_series and to_frame consistent.
Since that is covered in #17832, I think this can be closed?

[jorisvandenbossche]

Opened a PR for it: #23739