BUG(?): can_hold_element(DTA, np.nan) returns True

[mroeschke]

In [12]: import pandas as pd

In [13]: import numpy as np

In [14]: arr = pd.arrays.DatetimeArray(np.array([np.datetime64("2020-01-01T00:01:01")]))

In [15]: pd.core.dtypes.cast.can_hold_element(arr, np.nan)
Out[15]: True

I think this is potentially the case for TDA, PeriodArray too.

Is this expected? cc @jbrockmendel

I think this is the crux behind #29024 where this looks buggy

In [17]: pd.Series([pd.NaT]).replace({np.nan: 0})
Out[17]: 
0    0
dtype: int64

[jbrockmendel]

This is expected. You can do DTA[0] = np.nan without raising.

[mroeschke]

Cool thanks closing