Pandas Bug: groupby(..., observed=True) doesn't respect sort key

df.groupby(..., observed=True) behaves inconsistently for unordered Categoricals:

>>> df = pd.DataFrame({'A': pd.Categorical(['b', 'a']), 'B': [1,2]}) 
>>> df.groupby('A', observed=False).sum()  # ok
   B
A
a  2
b  1
>>> df.groupby('A', observed=True).sum()  # not ok
   B
A
b  1
a  2

Because the sort parameter in the groupby is implicitly True, the second result should be the same as if sort_index had been called:

>>> df.groupby('A', observed=True).sum().sort_index()  # ok result, but shouldn't be needed
   B
A
a  2
b  1

My guess is that somewhere there is missing a if sort: obj.sort_index() block or similar.

Comment From: rhshadrach

This is fixed on main and is tested in test_category_order_reducer.

Comment From: topper-123

That’s really great. 👍