BUG(?): Non nano + DatetOffset(component=0) re-infers resolution

[mroeschke]

In [1]: import pandas as pd; from io import StringIO; import numpy as np

In [2]: ser = pd.Series(pd.DatetimeIndex(["2000-01-01 00:00:00.012"], dtype="datetime64[ms]"))

In [3]: ser
Out[3]: 
0   2000-01-01 00:00:00.012
dtype: datetime64[ms]

In [4]: ser + pd.DateOffset(microseconds=0)  # 2.1.4
Out[4]: 
0   2000-01-01 00:00:00.012
dtype: datetime64[ms]

In [3]: ser + pd.DateOffset(microseconds=0). # main
Out[3]: 
0   2000-01-01 00:00:00.012
dtype: datetime64[us]

Possibly from #55595. Should the resolution be re-inferred here? cc @jbrockmendel

[jbrockmendel]

i think bc you passed microseconds=0, offset._pd_timedelta comes back with unit="us", which is why this gets upcast

[mroeschke]

Would it be worth no-oping these cases though?

[jbrockmendel]

i dont think so, no. This is analogous to pd.Timedelta(0).as_unit("us") in that the "us" unit is informative despite the value being zero.