Creating DatetimeIndex from empty array with datetime64 dtype raises IndexError

[lesteve]

using pandas from master:

import numpy as np
import pandas as pd
pd.DatetimeIndex(np.array([], dtype='datetime64'))

I get the following stack trace:

/data/python/pandas/pandas/tseries/index.pyc in __new__(cls, data, freq, start, end, periods, copy, name, tz, verify_integrity, normalize, **kwds)
    229             else:
    230                 if data.dtype != _NS_DTYPE:
--> 231                     subarr = tslib.cast_to_nanoseconds(data)
    232                 else:
    233                     subarr = data

/data/python/pandas/pandas/tslib.so in pandas.tslib.cast_to_nanoseconds (pandas/tslib.c:14151)()

IndexError: invalid index

This happens because pandas.tslib.cast_to_nanoseconds expects the array to be not empty to get the datetime64 unit of the first element.

One way to fix it would be to check whether data is empty before trying the conversion.

It would be nice to have a better error inside pd.tslib.cast_to_nanoseconds saying that it expects a non-empty array as well.

[wesm]

thanks!