⚠️ Nightly upstream-dev CI failed ⚠️

[github-actions]

Workflow Run URL

Python 3.8 Test Summary Info

FAILED xarray/tests/test_backends.py::TestNetCDF4Data::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestNetCDF4ViaDaskData::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestZarrDictStore::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestZarrDirectoryStore::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestScipyInMemoryData::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestScipyFileObject::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestScipyFilePath::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestNetCDF3ViaNetCDF4Data::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestNetCDF4ClassicViaNetCDF4Data::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestGenericNetCDFData::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestH5NetCDFData::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestH5NetCDFFileObject::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_backends.py::TestH5NetCDFViaDaskData::test_roundtrip_string_with_fill_value_nchar
FAILED xarray/tests/test_conventions.py::TestCFEncodedDataStore::test_roundtrip_string_with_fill_value_nchar

[mathause]

• I wonder if that has something to do with coords: retain str dtype #4759. But why would this only fail upstream?
• It is also similar to fix compatibility with h5py version 3 and unpin tests #4570 but h5py is still pinned...
• Some of also seems to be correctly failing: we have encoding = {"dtype": "S1"} but expected has dtype U1.
• The tests are enclosed with with pytest.raises(NotImplementedError): should these not fail if no error is raised? So why do we even get a result...
• Why are some dask arrays not evaluated in assert_identical?

[keewis]

this is unrelated to #4759, the errors appeared before that (see https://github.com/pydata/xarray/runs/1684617075). The issue was opened today because the scheduled CI was broken until the merge of #4806.

[mathause]

I guess the failure could easily be fixed - replace "abc" by b"abc" but the question is - why does this raise a NotImplementedError and still manage to write the file (for all backends)? When I try it locally I only get an empty netCDF.

[keewis]

interestingly, this didn't fail in the run yesterday. As far as I can tell, only the versions of identify, bottleneck, numpy, and pandas changed. Not sure if any of these could have fixed this. Maybe we did without noticing it?

[mathause]

As it touches all backends my bet would be on numpy or pandas - maybe numpy/numpy#18152? I think it's not really worth digging - so I suggest to close this.

[keewis]

that makes sense, I agree we don't really have to figure out what the actual cause was. If this pops up again we can just wait on the CI to open a new issue.