https://issues.apache.org/jira/browse/ARROW-14052

@lidavidm When I test the hash aggregate variation of this this from the R bindings on arrays of integer data that have integral medians within groups, I'm seeing errors like this:

Error: Invalid: Referenced field x was double but should have been int32

Do these kernels always return float64 output or are there some cases where integer input will give integer output?

@ianmcook hmm, no, it should always give float64 regardless of the input type - I'll take a look tomorrow (if you push it somewhere I can dig into it too)

@ianmcook hmm, no, it should always give float64 regardless of the input type - I'll take a look tomorrow (if you push it somewhere I can dig into it too)

I built Arrow C++ from this PR and I'm using that to test the changes to the R bindings in #11018. The test that fails with this error is here: https://github.com/apache/arrow/pull/11018/files#diff-a18898eb61bfc3d3dda292cf594b26d5f08a8cd7b455b72ed9c58f6e9c1311edR234-R249

Thanks, that should be fixed now - I had declared the wrong output type, leading to expressions getting bound with the wrong types. The tests pass locally for me now.

Thanks, that should be fixed now - I had declared the wrong output type, leading to expressions getting bound with the wrong types. The tests pass locally for me now.

Yep that fixes it, thanks!

Is this ready to merge? The CI failure appears to be unrelated.

I think we should be good here? We can tackle anything in a followup if needed.