issue_comments

Since the 2022.06.0 release, the following tests have been failing in Debian and Ubuntu on big-endian systems: ``` =================================== FAILURES =================================== ____ TestPandasIndex.test_join _______

self = <xarray.tests.test_indexes.TestPandasIndex object at 0x3ff828b6ef0>

def test_join(self) -> None:
    index1 = PandasIndex(["a", "aa", "aaa"], "x", coord_dtype="<U3")
    index2 = PandasIndex(["aa", "aaa", "aaaa"], "x", coord_dtype="<U4")

    expected = PandasIndex(["aa", "aaa"], "x")
    actual = index1.join(index2)
    print(actual.index)
    assert actual.equals(expected)

  assert actual.coord_dtype == "<U4"

E AssertionError: assert dtype('>U4') == '<U4' E + where dtype('>U4') = <xarray.core.indexes.PandasIndex object at 0x3ff716d5300>.coord_dtype

/usr/lib/python3/dist-packages/xarray/tests/test_indexes.py:267: AssertionError ----------------------------- Captured stdout call ----------------------------- Index(['aa', 'aaa'], dtype='object', name='x') ___ TestPandasMultiIndex.test_concat ___

self = <xarray.tests.test_indexes.TestPandasMultiIndex object at 0x3ff82e70370>

def test_concat(self) -> None:
    pd_midx = pd.MultiIndex.from_product(
        [[0, 1, 2], ["a", "b"]], names=("foo", "bar")
    )
    level_coords_dtype = {"foo": np.int32, "bar": "<U1"}

    midx1 = PandasMultiIndex(
        pd_midx[:2], "x", level_coords_dtype=level_coords_dtype
    )
    midx2 = PandasMultiIndex(
        pd_midx[2:], "x", level_coords_dtype=level_coords_dtype
    )
    expected = PandasMultiIndex(pd_midx, "x", level_coords_dtype=level_coords_dtype)

    actual = PandasMultiIndex.concat([midx1, midx2], "x")
    assert actual.equals(expected)

  assert actual.level_coords_dtype == expected.level_coords_dtype

E AssertionError: assert {'bar': dtype...type('int32')} == {'bar': '<U1'...numpy.int32'>} E Omitting 1 identical items, use -vv to show E Differing items: E {'bar': dtype('>U1')} != {'bar': '<U1'} E Full diff: E - {'bar': '<U1', 'foo': \<class 'numpy.int32'>} E + {'bar': dtype('>U1'), 'foo': dtype('int32')}

/usr/lib/python3/dist-packages/xarray/tests/test_indexes.py:395: AssertionError ___ TestPandasMultiIndex.testjoin ____

self = <xarray.tests.test_indexes.TestPandasMultiIndex object at 0x3ff828b6590>

def test_join(self):
    midx = pd.MultiIndex.from_product([["a", "aa"], [1, 2]], names=("one", "two"))
    level_coords_dtype = {"one": "<U2", "two": "i"}
    index1 = PandasMultiIndex(midx, "x", level_coords_dtype=level_coords_dtype)
    index2 = PandasMultiIndex(midx[0:2], "x", level_coords_dtype=level_coords_dtype)

    actual = index1.join(index2)
    assert actual.equals(index2)

  assert actual.level_coords_dtype == level_coords_dtype

E AssertionError: assert {'one': dtype...type('int32')} == {'one': '<U2', 'two': 'i'} E Omitting 1 identical items, use -vv to show E Differing items: E {'one': dtype('>U2')} != {'one': '<U2'} E Full diff: E - {'one': '<U2', 'two': 'i'} E + {'one': dtype('>U2'), 'two': dtype('int32')}

/usr/lib/python3/dist-packages/xarray/tests/test_indexes.py:493: AssertionError ``` Full test log on IBM s390x with Ubuntu 22.10: https://autopkgtest.ubuntu.com/results/autopkgtest-kinetic/kinetic/s390x/p/python-xarray/20220911_200538_24725@/log.gz

@max-sixty done in #3135

@max-sixty

Yes, #2334 is still relevant, I recently rebased and added another XFAIL for a new test.

This issue has gone away. Weirdly, it seems to have happened after NumPy was upgraded to 1.14, and Dask was still at 0.16.

Same results with dask: 0.17.2. Thanks for your help.

I have dask: 0.16.0 installed and there is a 0.17.2 out. I'll try upgrading that and see if it helps.

Thanks!

After adding: import dask.array as da I get: numpy 1.07926 /usr/lib/python3/dist-packages/dask/array/reductions.py:297: RuntimeWarning: invalid value encountered in true_divide u = total / n /usr/lib/python3/dist-packages/dask/array/numpy_compat.py:46: RuntimeWarning: invalid value encountered in true_divide x = np.divide(x1, x2, out) dask 1.09559 pandas 1.0792573690414429 bottleneck 1.0792573690414429

Hi, any ideas, please?

Thanks, here you go: ```

xarray.show_versions()

INSTALLED VERSIONS

commit: None python: 3.6.5.candidate.1 python-bits: 32 OS: Linux OS-release: 4.15.0-12-generic machine: i686 processor: athlon byteorder: little LC_ALL: None LANG: en_ZA.UTF-8 LOCALE: en_ZA.UTF-8

xarray: 0.10.3 pandas: 0.22.0 numpy: 1.13.3 ... bottleneck: 1.2.0 ... ```