[BUG] Sum of grouped bool has inconsistent dtype

[rhshadrach]

• closes Sum of grouped bool column has inconsistent type #7001
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Would appreciate any feedback on this attempt.

The strategy is to modify the dtype after the aggregation is computed in certain cases when casting. In order for this to work, the cast functions need to be made aware of how the data was aggregated. I've added an optional "how" argument to maybe_downcast_numeric and _try_cast. Because this dtype change is needed in two places, I've added the function groupby_result_dtype to dtypes/common.py to handle the logic.

I wasn't sure where the mapping information needed by groupby_result_dtype should be stored. Currently it is in the function itself, but maybe there is a better place for it.

If this is a good approach, it could potentially be expanded for other aggregations and datatypes. One thought is that perhaps groupby(-).mean() should always return a float for numeric types.

[WillAyd]

Thanks for taking a stab at this - the code paths here are tricky but for general feedback on your approach I think this should all be self-contained within groupby, i.e. not leaking into the dtype / inference modules

[jreback]

@rhshadrach if you can update

[rhshadrach]

@jreback Failure is unrelated, ready for another review:

##[error]./pandas/tests/indexes/timedeltas/test_constructors.py:173:String has a space at the beginning instead of the end of the previous string.

[jreback]

looks pretty good. small comments, ping on green.

cc @TomAugspurger @jorisvandenbossche @WillAyd @jbrockmendel if any comments

[WillAyd]

lgtm - certainly a more logical placement of things

[rhshadrach]

@jreback ping

[jreback]

lgtm. small follow request.

[jreback]

actually need a whatsnew note for this. if you can fix the out-of-context comment and change the name of the try_cast_to_ea as suggested. ping on green.

[rhshadrach]

@jreback ping

[jreback]

changes look good. one more item. ping on green and can get this in.

[rhshadrach]

@jreback ping

[jreback]

very nice @rhshadrach keep em coming!

[rhshadrach]

Thanks for the feedback @jbrockmendel. I'll clean these type hitns up in a subsequent PR.

[simonjayhawkins]

I'll clean these type hitns up in a subsequent PR.

FYI mypy error: pandas\core\dtypes\cast.py:333:18: error: "type" has no attribute "_from_sequence"

so maybe worth adding type hints to maybe_cast_to_extension_array and add a subclass check to the instance check