Issues with xarray v0.14

[zbruick]

Another xarray release, another round of broken tests. Main issues are:

• AttributeError: 'GiniFile' object has no attribute 'tell' (several test failures)
• AssertionError on test_narr_example_variable_without_grid_mapping
• AssertionError on test_dataset_parse_cf_varname_list

[jthielen]

The GiniFile errors appear to have arose from pydata/xarray#3389.

The assertion errors for test_narr_example_variable_without_grid_mapping and test_dataset_parse_cf_varname_list look to be the same error—a doubling of the 'x' coordinate's shape as pointed out in the comment in test_narr_example_variable_without_grid_mapping. This arose with the changes in pydata/xarray#3234, but what exactly in that PR, I'm not sure yet.

[kgoebber]

Is it worth starting to test on xarray master to get ahead of some of these changes due to our reliance on xarray for our data model?

[jthielen]

@kgoebber We already do: #1144.

[zbruick]

I haven't gone back to look at those builds since we started them...should have been doing this. I'll keep digging and see if I can get a PR up today to fix this so that we can get 0.11.1 out the door.

[zbruick]

@jthielen, looks like your fix in 2fe00b6 isn't having the correct effect anymore. You know this code better than the rest of us...any quick fix you see?

[jthielen]

@zbruick I found a quick fix that resolves the issue on MetPy's end (see #1206), but I still can't figure out what went wrong in the first place. I think there is an upstream issue with the way the Dataset constructor used in Dataset.apply combines coordinates/indexes after pydata/xarray#3389 that conflicts with the unit conversion MetPy does...but I can't quite track it down yet (or be totally confident it is all upstream and not a problem with MetPy's coordinate unit conversion).

[zbruick]

Thanks @jthielen for the quick fix...interesting bug here. Hopefully between #1204 and #1206 we have this taken care of now.

[zbruick]

With #1204 and #1206 being merged, this issue is resolved.