Fix test_melt with pyarrow strings active

[j-bennet]

Part of #10029.

Follow-up for #10000.

• Tests added / passed
• Passes pre-commit run --all-files

[jrbourbeau]

Thanks @j-bennet

It's not totally clear to me whether this is a case where we should convert to string[pyarrow] or leave as object when combining mixed dtype columns (e.g. combining strings and floats). What's your thinking behind the current approach?

[j-bennet]

It's not totally clear to me whether this is a case where we should convert to string[pyarrow] or leave as object when combining mixed dtype columns (e.g. combining strings and floats). What's your thinking behind the current approach?

For some reason, I had an impression that pandas would lose string[pyarrow] dtypes when melting them together, but this turns out to not be the case (perhaps I've seen this in the past, but it was fixed? not sure). pandas has reasonable upcasting behavior:

obj + str = obj
obj + int = obj
str + str = str
str + int = obj

so the best fix seems to be to turn off convert_string and delegate to upstream. This PR is now doing just that.

[jrbourbeau]

I've not checked locally -- is pre-commit happy with this formatting? Not a big deal if we need what's current in this PR, but just thought I'd check

Suggested change

	(
	a,
	b,
	) = _maybe_convert_string(a, b)
	a, b = _maybe_convert_string(a, b)

[j-bennet]

This is what pre-commit did, for whatever reason! I didn't do that. :)

[jrbourbeau]

Hrm, weird. When I make this change, pre-commit seems to be happy locally

[j-bennet]

Yeah, it's happy now.

[jrbourbeau]

Thanks @j-bennet -- will merge after CI finishes

[j-bennet]

Test failure is unrelated (upstream): FAILED dask/dataframe/tests/test_pyarrow_compat.py::test_pickle_roundtrip[uint8].