series created with DatetimeIndex when index is [None]

[dbew]

This looks like a change from 0.13.1 to HEAD. If you create a series like this:

pd.Series(index=np.array([None]))

Then in 0.13.1 the result is:

Out[16]: 
NaN   NaN
dtype: float64

In [19]: x.index.dtype
Out[19]: dtype('O')

In [20]: type(x.index)
Out[20]: pandas.core.index.Index

but in HEAD the result is

Out[191]: 
NaT   NaN
dtype: float64

In[192]: x.index.dtype
Out[192]: dtype('<M8[ns]')

In[193]: type(x.index)
Out[193]: pandas.tseries.index.DatetimeIndex

So the index has changed from Index/object to DatetimeIndex/datetime64.

[dbew]

Running through the construction of a pd.Series object in the debugger, it looks like this is happening in Series._set_axis because is_all_dates is True for an index with just None in it.

    def _set_axis(self, axis, labels, fastpath=False):
        """ override generic, we want to set the _typ here """

        if not fastpath:
            labels = _ensure_index(labels)

        is_all_dates = labels.is_all_dates
        if is_all_dates:
            from pandas.tseries.index import DatetimeIndex
            from pandas.tseries.period import PeriodIndex
            if not isinstance(labels, (DatetimeIndex, PeriodIndex)):
                labels = DatetimeIndex(labels)

        .....

Constructing an Index manually confirms this:

i = pd.Index(np.array([None]))
i.is_all_dates    
Out[208]: True

But is_all_dates just defers to is_datetime_array which I think is cimported from numpy

[dbew]

Looks like it's PR 6341 which made the change:

[jreback]

no is_datetime_array is defined in src/inference.pyx; a single element None in this context should be simply nan. hmm.

cc @immerrr