BUG: DatetimeArray._from_sequence(arg, dtype='tz') raises TypeError

[mroeschke]

Discovered in #25308 (comment), is this supposed to work @jbrockmendel, @TomAugspurger?

cc @jreback

In [3]: arg = np.array([np.datetime64('2018-01-01')], dtype=object)

In [4]: pd.arrays.DatetimeArray._from_sequence(arg, dtype='UTC')
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
 in pandas_dtype(dtype)
   2036     try:
-> 2037         npdtype = np.dtype(dtype)
   2038     except Exception:

TypeError: data type "UTC" not understood

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
<ipython-input-4-506f2a230c35> in <module>
----> 1 pd.arrays.DatetimeArray._from_sequence(arg, dtype='UTC')

 in _from_sequence(cls, data, dtype, copy, tz, freq, dayfirst, yearfirst, ambiguous, int_as_wall_time)
    374             data, dtype=dtype, copy=copy, tz=tz,
    375             dayfirst=dayfirst, yearfirst=yearfirst,
--> 376             ambiguous=ambiguous, int_as_wall_time=int_as_wall_time)
    377
    378         freq, freq_infer = dtl.validate_inferred_freq(freq, inferred_freq,

 in sequence_to_dt64ns(data, dtype, copy, tz, dayfirst, yearfirst, ambiguous, int_as_wall_time)
   1714     inferred_freq = None
   1715
-> 1716     dtype = _validate_dt64_dtype(dtype)
   1717
   1718     if not hasattr(data, "dtype"):

 in _validate_dt64_dtype(dtype)
   1996     """
   1997     if dtype is not None:
-> 1998         dtype = pandas_dtype(dtype)
   1999         if is_dtype_equal(dtype, np.dtype("M8")):
   2000             # no precision, warn

 in pandas_dtype(dtype)
   2041             raise TypeError("data type not understood")
   2042         raise TypeError("data type '{}' not understood".format(
-> 2043             dtype))
   2044
   2045     # Any invalid dtype (such as pd.Timestamp) should raise an error.

TypeError: data type 'UTC' not understood

In [5]: pd.__version__
Out[5]: '0.25.0.dev0+170.g5bf07a99d.dirty'

[TomAugspurger]

No, I don't think so. 'UTC' isn't a dtype or an alias (and I don't think it should be inferred as one).

[TomAugspurger]

dtype='datetime64[ns, UTC]' would work I think.

[mroeschke]

Gotcha. dtype='datetime64[ns, UTC]' works, and agreed that 'UTC' alone shouldn't work here. Closing