CI: failing timedelta64 test on Linux py37_locale build

This is seen on a few PRs (eg https://dev.azure.com/pandas-dev/pandas/_build/results?buildId=34508&view=logs&jobId=c4100758-acc4-5cd0-72d8-90828968cf00&j=c4100758-acc4-5cd0-72d8-90828968cf00)

=================================== FAILURES ===================================
___ TestSeriesDtypes.test_astype_generic_timestamp_no_frequency[timedelta64] ___
[gw0] linux -- Python 3.7.6 /home/vsts/miniconda3/envs/pandas-dev/bin/python

self = <pandas.tests.series.test_dtypes.TestSeriesDtypes object at 0x7f3d21a6c050>
dtype = <class 'numpy.timedelta64'>

    @pytest.mark.parametrize("dtype", [np.datetime64, np.timedelta64])
    def test_astype_generic_timestamp_no_frequency(self, dtype):
        # see gh-15524, gh-15987
        data = [1]
        s = Series(data)
    
        msg = (
            fr"The '{dtype.__name__}' dtype has no unit\. "
            fr"Please pass in '{dtype.__name__}\[ns\]' instead."
        )
        with pytest.raises(ValueError, match=msg):
>           s.astype(dtype)
E           Failed: DID NOT RAISE <class 'ValueError'>

pandas/tests/series/test_dtypes.py:395: Failed

[jbrockmendel]

Just to be sure, is the failure newer, or just something that came up now randomly?

I’ve been seeing it for a few days, but its not in the npdev build. Its py37, Linux, numpy 1.18.1

[jbrockmendel]

This just stopped failing somehow? Or did we actually do something?

xfailed the affected test in the relevant conditions