.groupby by should indicate it aligns the passed in Series

[kernc]

Code Sample, a copy-pastable example if possible

>>> series = pd.Series([1, 2, 3], index=range(3))

>>> key = pd.Series([0, 0, 1], index=range(3, 6))  # Note: disjunct index

>>> series.groupby(key).sum()
Series([], dtype: int64)

>>> series.groupby(key.values).sum()
0    3
1    3
dtype: int64

Problem description

The current behavior is a bug because the .groupby() documentation says about by parameter:

If a dict or Series is passed, the Series or dict VALUES will be used to determine the groups

Above example shows this not being the case.

Expected Output

With above setup:

>>> series.groupby(key).sum()
0    3
1    3
dtype: int64

Output of pd.show_versions()

pandas: 0.19.0+416.ge1390cd

[kernc]

Might be related to #15244.

I followed to problem to reindexing of the grouper Series which results in the new grouper containing all NaNs.

[jreback]

this is correct, if you pass a Series if will first .align it. So that's why you get an empty result set.
This is not a bug, though I suppose could be better documented.

This was brought up here: #15244

[jreback]

closing as dupe of #15244