failing tests with pandas 2.1

[dcherian]

What happened?

See #8101

FAILED xarray/tests/test_missing.py::test_interpolate_pd_compat - ValueError: 'fill_value' is not a valid keyword for DataFrame.interpolate
FAILED xarray/tests/test_missing.py::test_interpolate_pd_compat_non_uniform_index - ValueError: 'fill_value' is not a valid keyword for DataFrame.interpolate

and this doctest

FAILED xarray/core/dataarray.py::xarray.core.dataarray.DataArray.to_unstacked_dataset

@pydata/xarray can someone take a look please?

[kmuehlbauer]

FAILED xarray/tests/test_missing.py::test_interpolate_pd_compat - ValueError: 'fill_value' is not a valid keyword for DataFrame.interpolate
FAILED xarray/tests/test_missing.py::test_interpolate_pd_compat_non_uniform_index - ValueError: 'fill_value' is not a valid keyword for DataFrame.interpolate

fill_value - kwarg was removed upstream and raises error now -> pandas-dev/pandas#53962

To me it looks like fill_value should have only been removed for ["ffill", "bfill", "pad", "backfill"] - methods.

[kmuehlbauer]

FAILED xarray/core/dataarray.py::xarray.core.dataarray.DataArray.to_unstacked_dataset

The doctest works for me locally, investigating package differences in the environment now.

Update:

OK, it does run, but I get a different result (after to_stacked_array):

MultiIndex([('a',   0),
            ('a',   1),
            ('a',   2),
            ('b', nan)],
           name='z')

instead of

MultiIndex([('a', 0.0),
            ('a', 1.0),
            ('a', 2.0),
            ('b', nan)],
           name='z')

[mathause]

I think the doctest failure is unrelated to the removal of fill_value. I think this is pandas avoiding the cast of int to float when the nan value is added. The cheap solution is to convert the ints to floats

-         ...     coords=[("x", ["a", "b"]), ("y", [0, 1, 2])], 
+         ...     coords=[("x", ["a", "b"]), ("y", [0., 1, 2])],

in

xarray/xarray/core/dataarray.py

Line 2893 in 1043a9e

... coords=[("x", ["a", "b"]), ("y", [0, 1, 2])],

[kmuehlbauer]

I've opened pandas-dev/pandas#54920 upstream for the fill_value-issue.

[kmuehlbauer]

Upstream is already working on a fix for the fill_value-issue: pandas-dev/pandas#54927

[kmuehlbauer]

Reopening until doc-test is fixed.

[benbovy]

There are also two pandas-related warnings when building the docs, which cause a "failure" for PR builds on RTD:

/home/docs/checkouts/readthedocs.org/user_builds/xray/conda/8142/lib/python3.10/site-packages/pandas/core/indexes/base.py:docstring of pandas.core.indexes.base.Index.join:14: WARNING: Inline literal start-string without end-string.
/home/docs/checkouts/readthedocs.org/user_builds/xray/conda/8142/lib/python3.10/site-packages/pandas/core/indexes/base.py:docstring of pandas.core.indexes.base.Index.join:15: WARNING: Inline literal start-string without end-string.

Not sure how to (temporarily) fix those, though.