Accept PandasArray (with correct dtype) in DatetimeArray constructor

[jorisvandenbossche]

I think it make sense that eg in the following case, we extract the numpy array out of a PandasArray, and thus accept it as an argument:

In [43]: pd.array([1, 2], dtype='datetime64[ns]')                       
Out[43]: 
<PandasArray>
[numpy.datetime64('1970-01-01T00:00:00.000000001'), numpy.datetime64('1970-01-01T00:00:00.000000002')]
Length: 2, dtype: datetime64[ns]

In [44]: pd.arrays.DatetimeArray(pd.array([1, 2], dtype='datetime64[ns]'))
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-44-4cc26981aff2> in <module>
----> 1 pd.arrays.DatetimeArray(pd.array([1, 2], dtype='datetime64[ns]'))

~/scipy/pandas/pandas/core/arrays/datetimes.py in __init__(self, values, dtype, freq, copy)
    265                 "ndarray, or Series or Index containing one of those."
    266             )
--> 267             raise ValueError(msg.format(type(values).__name__))
    268 
    269         if values.dtype == 'i8':

ValueError: Unexpected type 'PandasArray'. 'values' must be a DatetimeArray ndarray, or Series or Index containing one of those.

There might be more similar cases like this.

[mroeschke]

This looks supported on master now. Could use a test?

In [128]: In [43]: pd.array([1, 2], dtype='datetime64[ns]')
     ...:
Out[128]:
<DatetimeArray>
['1970-01-01 00:00:00.000000001', '1970-01-01 00:00:00.000000002']
Length: 2, dtype: datetime64[ns]

In [129]: In [44]: pd.arrays.DatetimeArray(pd.array([1, 2], dtype='datetime64[ns]'))
     ...:
Out[129]:
<DatetimeArray>
['1970-01-01 00:00:00.000000001', '1970-01-01 00:00:00.000000002']
Length: 2, dtype: datetime64[ns]

[jorisvandenbossche]

The initial example isn't really valid anymore, as pd.array(.., dtype="datetime64[ns]") is apparently not creating a PandasArray anymore, but a DatetimeArray.

So the original issue still fails:

In [16]: a = pd.arrays.PandasArray(np.array([1, 2, 3], dtype="datetime64[ns]"))                                                                                                                                    

In [17]: a                                                                                                                                                                                                         
Out[17]: 
<PandasArray>
[numpy.datetime64('1970-01-01T00:00:00.000000001'),
 numpy.datetime64('1970-01-01T00:00:00.000000002'),
 numpy.datetime64('1970-01-01T00:00:00.000000003')]
Length: 3, dtype: datetime64[ns]

In [18]: pd.arrays.DatetimeArray(a)                                                                                                                                                                                
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-18-98567534d92b> in <module>
----> 1 pd.arrays.DatetimeArray(a)

~/scipy/pandas/pandas/core/arrays/datetimes.py in __init__(self, values, dtype, freq, copy)
    237         if not isinstance(values, np.ndarray):
    238             raise ValueError(
--> 239                 f"Unexpected type '{type(values).__name__}'. 'values' must be "
    240                 "a DatetimeArray ndarray, or Series or Index containing one of those."
    241             )

ValueError: Unexpected type 'PandasArray'. 'values' must be a DatetimeArray ndarray, or Series or Index containing one of those.

(I am not sure it is the most important issue to support though)