Raise NotImplementedError for groupby.agg if duplicate columns would be created

[mroeschke]

Description

xref #17649

For cudf.pandas, we will dispatch to pandas instead of silently dropping the duplicate column

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[vyasr]

I'm approving, deferring to you on to add a warning in the non-pandas_compatible case as well.

Do we need to add similar logic for duplicate columns in a dataframe itself? i.e. to prevent something like

>>> df = pd.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6]})
>>> df.rename({'b': 'a'}, axis=1)

or any other way that duplicate names could manifest? That could be done in another PR of course.

[vyasr]

Are you asking if we should start issuing a warning in non-pandas_compatible mode, i.e. in an else clause? I would support that.

[mroeschke]

Yup, exactly. Thanks added a UserWarning in non-pandas_compatible mode.

[mroeschke]

any other way that duplicate names could manifest?

Yeah we have some checks like this in ColumnAccessor already that would check if duplicate columns would be created (e.g. your example raises in cudf).

The harder-to-catch cases are when we're preprocessing column-related operations with mappings (dicts) first which won't alert us if duplicates are introduced.

[mroeschke]

/merge