Series.__sub__(NaT) vs DatetimeIndex.__sub__(NaT)

[jbrockmendel]

One treats NaT result as datetime, other treats it as timedelta.

>>> dti  = pd.DatetimeIndex([pd.NaT, pd.Timestamp('19900315')])
>>> ser = pd.Series(dti)

>>> dti - pd.NaT
TimedeltaIndex([NaT, NaT], dtype='timedelta64[ns]', freq=None)

>>> ser - pd.NaT
0   NaT
1   NaT
dtype: datetime64[ns]

[gfyoung]

Hmmm...that does seem a little inconsistent. Have a look as to why that's happening.

[jreback]

I would argue that the series is wrong here. a sub operation with a DTI-like should always yield a TDI-like.

[jbrockmendel]

I would argue that the series is wrong here. a sub operation with a DTI-like should always yield a TDI-like.

So I can change that and consider it a bugfix as opposed to a Breaking Change?

[jreback]

no its a breaking change, but you can still change it

[jbrockmendel]

Got a fix for this about ready. Two questions.

One: there are a couple of tests in series.test_operators test_datetime64_ops_nat that specifically check for the return type to be datetime64. Sure the behavior isn't intentional?

Two: also in test_datetime64_ops_nat are tests for -pd.NaT + Series[datetime64]. Should that also be changed to return Series[timedelta64]?