BUG: GroupBy.ffill()/bfill() do not return NaN values for NaN groups

[smithto1]

• closes BUG: fillna(method="ffill") and ffill() on DataFrameGroupBy gives different results #34725
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

See #34725 for a good description of the problem.

This fixes the issue in the tidiest way I could find.

One test is added that covers all relevant cases I could think of. The comments in the test can be removed, they are just there to make it easier for the reviewer to follow as the test is a bit complicated.

This also fixes the copy-pastable example from the #34725.

[rhshadrach]

Is there a way to fix group_fillna_indexer rather than correct the results with post-processing? Or is there a reason not to do this?

[WillAyd]

Agreed with @rhshadrach comment here - this should be fixed directly in the underlying function

[smithto1]

Is there a way to fix group_fillna_indexer rather than correct the results with post-processing? Or is there a reason not to do this?

@rhshadrach This is fixed in group_fillna_indexer now. It is much cleaner and faster.

[rhshadrach]

Looking really good, two minor requests.

[smithto1]

@rhshadrach @WillAyd Comments have been addressed. Can you please have another look?

[rhshadrach]

Looks good. Can you move tests to tests.groupby.test_missing - ref #36326

[smithto1]

Looks good. Can you move tests to tests.groupby.test_missing - ref #36326

@rhshadrach Done.

[rhshadrach]

@smithto1 CI failure due to import order in unrelated files; I think merging master will fix.

[smithto1]

@smithto1 CI failure due to import order in unrelated files; I think merging master will fix.

@rhshadrach I think that is all fixed now.

[rhshadrach]

@smithto1 Got a new conflict that needs resolved.

[jreback]

lgtm can merge on green.

[jreback]

thanks @smithto1