jorisvandenbossche edited a comment on pull request #7169: URL: https://github.com/apache/arrow/pull/7169#issuecomment-642463399
> The new tests are failing for older pandas. So this was actually pointing to something that still needs to be fixed: the check you do in `pandas_compat` for non-nanosecond datetime64 to cast those to object dtype, a similar check will need to be done in `array.pxi::_array_like_to_pandas`. In addition, in that function we will need to pass `dtype=object` to the Series creation in this case (and dtype=None for all others), to ensure pandas doesn't convert the non-ns datetime64 back to ns resolution (otherwise the resulting dtype can depend on the actual dates you have, which is something we should avoid I think) ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
