bug in Series.replace with timezone-aware datetime columns

[multiloc]

Seeing the following behavior in 0.17.1 with the patch from #11715 applied (would raise without the patch):

In [29]: ser = pd.Series([pd.NaT, pd.Timestamp('2015/01/01', tz='UTC')])

# works
In [30]: ser.replace(pd.NaT, pd.Timestamp.min)
Out[30]: 
0    1677-09-22 00:12:43.145225
1     2015-01-01 00:00:00+00:00
dtype: object

# doesn't work
In [31]: ser.replace([np.nan, pd.NaT], pd.Timestamp.min)
Out[31]: 
0                         NaT
1   2015-01-01 00:00:00+00:00
dtype: datetime64[ns, UTC]

# works without timezone-aware datetimes
In [32]: ser = pd.Series([pd.NaT, pd.Timestamp('2015/01/01')])
In [33]: ser.replace([np.nan, pd.NaT], pd.Timestamp.min)
Out[33]: 
0   1677-09-22 00:12:43.145225
1   2015-01-01 00:00:00.000000
dtype: datetime64[ns]

[jreback]

yep, looks like a buggie! (thanks for the multiple reports BTW). interested in a pull-request to fix?

[multiloc]

sure, I'll look into it over the weekend