[BUG] loc behavior differs from pandas when a duplicated index is requested

[brandon-b-miller]

Describe the bug
When we have a series with a duplicated index such as

x = cudf.Series([1,2,3,4], index=[0,1,1,2])

And we try and fetch the items corresponding to that duplicated index, we get just the first element rather than a series of all the elements corresponding to that index:

x.loc[1]
# 2

Steps/Code to reproduce bug
See above

Expected behavior
We should get the same thing as pandas, basically all the rows corresponding to that index gathered.

x.to_pandas().loc[1]
# 1    2
# 1    3
# dtype: int64

Environment overview (please complete the following information)

• Environment location: [Bare-metal]
• Method of cuDF install: [source]

Environment details

Additional context
Add any other context about the problem here.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[shwina]

This is because of an "optimization" for scalar inputs to loc (

cudf/python/cudf/cudf/core/series.py

Line 292 in d49e412

if _is_scalar_or_zero_d_array(arg):

), that we should probably get rid of.

A workaround is to pass the index in a list:

s.loc[[0]]