Pandas API: pd.Series([py_date, py_datetime]) vs pd.Index([py_date, py_datetime])

ts = pd.Timestamp.now()

vals= [ts.date(), ts.to_pydatetime()]

ser = pd.Series(vals)  # <- dt64 dtype
idx = pd.Index(vals)  # <- object dtype

Index uses lib.maybe_convert_objects for inference while Series uses infer_datetimelike_array in maybe_infer_to_datetimelike.

Ideally these would match.

Comment From: jorisvandenbossche

Personally, if I would have to choose, I prefer the Index behaviour. I don't think we should automatically convert datetime.date objects to datetime dtype.

For the case of only datetime.date objects, we also don't infer it as datetime dtype.

Comment From: jbrockmendel

if I would have to choose, I prefer the Index behaviour

agreed

Comment From: jreback

ok i'll agree with you folks here.