CI prints Closing remaining open files:...test_hdf_empty_dataframe0/data.h5

[graingert]

example eg https://github.com/dask/dask/actions/runs/4727453904/jobs/8388080950#step:8:24759

prints: Closing remaining open files:/tmp/pytest-of-runner/pytest-0/popen-gw0/test_hdf_empty_dataframe0/data.h5...done

we're using with pd.HDFStore(...): correctly so this could be a pandas or pytables bug

[graingert]

I can reproduce with pure pandas with:

def test_hdf_empty_dataframe(tmp_path):
    pytest.importorskip("tables")
    # https://github.com/dask/dask/issues/8707

    df = pd.DataFrame({"A": [], "B": []}, index=[])
    df.to_hdf(tmp_path / "data.h5", format="fixed", key="df", mode="w")
    try:
        pd.read_hdf(tmp_path / "data.h5", "/df", mode="r", stop=0)
    except IndexError:
        pass

[graingert]

raised upstream as pandas-dev/pandas#52781

[jrbourbeau]

Thanks @graingert! Should we close this issue in favor of the upstream pandas issue then?

[graingert]

@jrbourbeau I have a performance enhancement that side-steps the pandas bug: #10205