Pandas BUG: adding DatetimeIndex with series of timedeltas results in strange series

In [21]: dtidx = pd.date_range('2012-01-01', periods=2)

In [22]: dtidx
Out[22]:
<class 'pandas.tseries.index.DatetimeIndex'>
[2012-01-01, 2012-01-02]
Length: 2, Freq: D, Timezone: None

In [23]: dtidx + pd.Series([pd.Timedelta('10 days'), pd.Timedelta('20 days')])
Out[23]:
2012-01-01 00:00:00   NaT
2012-01-02 00:00:00   NaT
0                     NaT
1                     NaT
dtype: datetime64[ns]

Comment From: jreback

this is correct, its just 'misuse'. The dt index is aligning with the index of the series (which in an integer index) while odd, it is allowed. Then the addition happens, but since nothing aligns you get all NaT.

you can do this

In [14]: Series(dtidx) + s
Out[14]: 
0   2012-01-11
1   2012-01-22
dtype: datetime64[ns]

 [13]: dtidx + s.values
Out[13]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2012-01-11, 2012-01-22]
Length: 2, Freq: D, Timezone: None

Comment From: jorisvandenbossche

ah, yes, of course, stupid of me, I forgot the basic pandas feature, alignment, ..

I tried to use Series because an array or list did not work:

dtidx + [pd.Timedelta('10 days'), pd.Timedelta('20 days')]
dtidx + np.array([pd.Timedelta('10 days'), pd.Timedelta('20 days')])

But indeed, the s.values approach works.