BUG: numpy.datetime('NaT') printing inconsistency

[kawochen]

Whether np.datetime64('NaT') is printed as NaT or NaN depends on other types present in the column. This is consistent with pd.NaT, whose display also depends on other types. But when all the other scalars are None, pd.NaT and None are both printed as NaT, whereas in a column of np.datetime64('NaT') and None, it depends on the order of the entries (last two lines).

In [44]: pd.DataFrame([[np.datetime64('NaT')], [None]])
Out[44]:
      0
0   NaN
1  None

In [45]: pd.DataFrame([[pd.NaT], [None]])
Out[45]:
    0
0 NaT
1 NaT

In [46]: pd.DataFrame([[np.datetime64('NaT')], [pd.NaT]])
Out[46]:
    0
0 NaT
1 NaT

In [47]: pd.DataFrame([[pd.NaT], [np.datetime64('NaT')]])
Out[47]:
    0
0 NaT
1 NaT

In [81]: pd.DataFrame([[None],[np.datetime64('NaT')]])
Out[81]:
    0
0 NaT
1 NaT

In [82]: pd.DataFrame([[np.datetime64('NaT')],[None]])
Out[82]:
      0
0   NaN
1  None

In [86]: pd.DataFrame([[None],[np.datetime64('NaT')]]).dtypes
Out[86]:
0    datetime64[ns]
dtype: object

In [87]: pd.DataFrame([[np.datetime64('NaT')],[None]]).dtypes
Out[87]:
0    object
dtype: object

[jreback]

yeh these are coercion issues actually, e.g. in infer.pyx, the np.datetime64('NaT') is not recognized like NaT,None,NaN (IOW it should trigger the conversions like what NaT does

[mroeschke]

Looks like np.datetime('NaT') is getting correctly recognized as datetime64ns now. Could use a test.

In [137]: In [44]: pd.DataFrame([[np.datetime64('NaT')], [None]])
     ...:
Out[137]:
    0
0 NaT
1 NaT

In [138]: pd.__version__
Out[138]: '0.26.0.dev0+576.gde67bb72e'