issue_comments

I think this function mostly makes sense, but we would want to drop the stuff for scalar variables.

Unfortunately I don't know a good way to fix this stuff short of auditing a lot of code manually -- this is a hazard of data model changes. On the plus side, I've also started to do some of this for the explicit index refactor.

One thing that might turn up a few bugs is to try adding such a variable to the create_test_data() helper function which gets used in lots of places.

I had some spare time and started working on this again. I'm scared of how much internal refactoring it requires. For example, this function

https://github.com/pydata/xarray/blob/385b36cdd34431b4f6f14aad1f222f989e7e2de2/xarray/core/dataset.py#L92-L113

just doesn't make sense any more if dimensions are not guaranteed to be one-dimensional.

I've never touched this part of the code base before. I have no idea how many places there are that make such an assumption.

Hello @rabernat! Thanks for updating the PR.

• In the file xarray/core/formatting.py, following are the PEP8 issues :

Line 275:45: E225 missing whitespace around operator

• In the file xarray/core/indexes.py, following are the PEP8 issues :

Line 55:77: E225 missing whitespace around operator Line 55:80: E501 line too long (80 > 79 characters)

• In the file xarray/tests/test_dataset.py, following are the PEP8 issues :

Line 68:1: E302 expected 2 blank lines, found 1 Line 251:9: E265 block comment should start with '# '

With 68f170c there are basically no failing tests in the original test suite. I will now work on adding more tests along the lines outlined by @shoyer.

It would be good to add a tests for these variables in: - The data_vars argument to the Dataset constructor (these objects should become data_vars, not coords) - The DataArray constructor. - Explicitly adding these variables: - With Dataset.__setitem__, Dataset.coords.__setitem__ and DataArray.coords.__setitem__ - When there are no existing variables matching the dimension name, and when a 1D index variable matching the dimension name already exists (the existing index should be deleted/cleared) - Converting these objects to pandas with to_dataframe, to_series and to_pandas - Concatenating along an existing dimension whose name matches existing variables (e.g., concatenate along x when a variable x with dimensions ('x', 'y') exists) - Concatenating along a new dimension whose name matches existing variables (e.g., concatenate along x when a variable x with dimensions ('y',) exists) - More generally, test variables where the name matches a dimension but the variable does not include that dimension at all, e.g., a variable 'x' with dimensions ('y',) in a Dataset where x is a dimension.

I turned up a few lines that probably need to be fixed, just by grepping for in\ \w+\.dims: https://github.com/pydata/xarray/blob/66a8f8dd7f5a2997ff614f3966d1951587915e7e/xarray/backends/api.py#L125-L126 https://github.com/pydata/xarray/blob/66a8f8dd7f5a2997ff614f3966d1951587915e7e/xarray/core/coordinates.py#L111-L116 https://github.com/pydata/xarray/blob/66a8f8dd7f5a2997ff614f3966d1951587915e7e/xarray/core/dataarray.py#L502

It looks like this change broke align somehow -- possibly we have a bug where some indexes were not being created as IndexVariable objects?

Seeking suggestions about what else needs to be tested here.