Timestamp unit is not respected

[david-cortes]

Timestamps always get returned in nanoseconds regardless of the unit I specify:

import numpy as np, pandas as pd
np.datetime64("2018-01-01").astype("datetime64[s]").astype(int)
>>> 1514764800
pd.Series(np.datetime64("2018-01-01")).astype("datetime64[s]")

0   2018-01-01
dtype: datetime64[ns]

pd.Series(np.datetime64("2018-01-01")).astype("datetime64[s]").astype(int)

0    1514764800000000000
dtype: int64

[TomAugspurger]

Pandas doesn't support non-ns resolution.

As of pandas 0.21, we raised on the Series[datetime64[ns]].astype('datetime64[s]'). It seems we don't raise any more, which may be a bug.

[marceloid]

Can confirm that as of pandas 1.2.3 the Series[datetime64[ns]].astype('datetime64[s]') doesn't raise any exception, and it still returns ns values.

[jbrockmendel]

In 2.0 pandas supports s, ms, us, and ns. Closing.