BUG: datetime64[us] arrays with NaT cannot be cast to DatetimeIndex

[shoyer]

In [11]: t = pd.to_datetime(['2000-01-01T00:00', 'NaT']).values.astype('datetime64[us]')

In [12]: pd.DatetimeIndex(t)
OutOfBoundsDatetime: Out of bounds nanosecond timestamp: 294247-01-09 19:59:05

In [13]: pd.to_datetime(t)
Out[13]:
<class 'pandas.tseries.index.DatetimeIndex'>
[2000-01-01, NaT]
Length: 2, Freq: None, Timezone: None

The second line should give the same result as the last one.

[jreback]

this is more of a case of DatetimeIndex should have less code in its new method (and just defer to pd.to_datetime).