API/INT: why is to_pydatetime handled differently in Series.dt accessor?

[jorisvandenbossche]

While reviewing: #20198, I notices this was defined in the DatetimeProperties itself:

pandas/pandas/core/indexes/accessors.py

Lines 114 to 143 in 78ded25

	class DatetimeProperties(Properties):
	"""
	Accessor object for datetimelike properties of the Series values.
	Examples
	--------
	>>> s.dt.hour
	>>> s.dt.second
	>>> s.dt.quarter
	Returns a Series indexed like the original Series.
	Raises TypeError if the Series does not contain datetimelike values.
	"""
	def to_pydatetime(self):
	return self._get_values().to_pydatetime()
	@property
	def freq(self):
	return self._get_values().inferred_freq
	DatetimeProperties._add_delegate_accessors(
	delegate=DatetimeIndex,
	accessors=DatetimeIndex._datetimelike_ops,
	typ='property')
	DatetimeProperties._add_delegate_accessors(
	delegate=DatetimeIndex,
	accessors=DatetimeIndex._datetimelike_methods,
	typ='method')

Is there any reason that we have it like that, while all others are just wrapped DatetimeIndex attributes/methods?

Because that means this returns (a bit strangly) an ndarray instead of a Series of the datetime.datetime values (making it inconsistent with other methods)

cc @jbrockmendel @jreback

[jorisvandenbossche]

Ah, now I opened this issue, I thought of a probable reason: because otherwise the datetime.datetime values would quickly get inferred / coerced back to datetime64 ?

[jreback]

yes that is the reason
though we could return an object dtype Index

[jbrockmendel]

Could change this for 2.0. Hard to deprecate since there isn't a good "do X instead."