BUG: astype of datetimelike to non-NS

[xhochy]

The current pandas master includes a behaviour change date conversion.

Before #19176

In [1]: import pandas as pd

In [2]: pd.DataFrame({'x': [3, 2, 1]}).astype('M8[us]')
Out[2]:
                           x
0 1970-01-01 00:00:00.000003
1 1970-01-01 00:00:00.000002
2 1970-01-01 00:00:00.000001

After #19176

In [1]: import pandas as pd

In [2]: pd.DataFrame({'x': [3, 2, 1]}).astype('M8[us]')
Out[2]:
                              x
0 1970-01-01 00:00:00.000000003
1 1970-01-01 00:00:00.000000002
2 1970-01-01 00:00:00.000000001

[jorisvandenbossche]

Hmm, we apparently didn't have any tests that catched this.
Anyhow, this is a regression that we should fix. By using plainly pd.to_datetime for any datetime64 dtype, we ignore the resolution.

[jreback]

yes, PR shortly