Sorting within groupby causes duplication of index level

[tdhopper]

When I run this code on 0.17.0 on Python 3.5.0

import pandas as pd
d = pd.DataFrame(data=[[2, 1]])
print(d.index)
print((d.
 groupby(level=0).
 apply(lambda d: 
       d.sort_values(0)
      )
 ).index)

I get

Int64Index([0], dtype='int64')
MultiIndex(levels=[[0], [0]],
           labels=[[0], [0]])

The index gets duplicated (and turned into a multiindex) as a result of the sort within the groupby.

Appears to be a bug to me.

[jreback]

this is the same issue as in #9946 (same because the same underlying cython routine is used)

you will notice that using lambda x: x.copy() is the same result.

pandas is trying to infer whether you have mutated things (or not), by comparing indexes. This is prob a bug, but this is a bit of a rabbit hole as I think the best thing to do would actually to ban this entirely (mutation). So the outputs can be more predictible. You are welcome to dig-in!.

[rhshadrach]

This can now be controlled by specifying group_keys=False in groupby.