html_url,issue_url,id,node_id,user,created_at,updated_at,author_association,body,reactions,performed_via_github_app,issue
https://github.com/pydata/xarray/pull/4163#issuecomment-653099470,https://api.github.com/repos/pydata/xarray/issues/4163,653099470,MDEyOklzc3VlQ29tbWVudDY1MzA5OTQ3MA==,2448579,2020-07-02T16:14:03Z,2020-07-02T16:14:03Z,MEMBER,Thanks @keewis. I'm really excited to see this go in.,"{""total_count"": 1, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 1, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029
https://github.com/pydata/xarray/pull/4163#issuecomment-650430723,https://api.github.com/repos/pydata/xarray/issues/4163,650430723,MDEyOklzc3VlQ29tbWVudDY1MDQzMDcyMw==,14808389,2020-06-26T22:25:46Z,2020-06-26T22:25:46Z,MEMBER,Once the CI passes this should be ready for review and merging.,"{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029
https://github.com/pydata/xarray/pull/4163#issuecomment-649860850,https://api.github.com/repos/pydata/xarray/issues/4163,649860850,MDEyOklzc3VlQ29tbWVudDY0OTg2MDg1MA==,14808389,2020-06-25T23:06:28Z,2020-06-25T23:16:04Z,MEMBER,"I tried that by writing some sort of `assert_duckarray_equal` and `assert_duckarray_allclose` function. That should work for most duck arrays, but not `pint`: the tests often compare the data with the variable (or the variable's data), and since the data is mostly a bare numpy array this fails for `pint`: bare numpy arrays are by definition dimensionless and so the comparison either returns `False` (the assertion fails) or a `DimensionalityError` is raised.

So to make this work I think we'd need to rewrite `VariableSubtests` to have another static method (`array_type`? `data_type`?) which is called on all data before it is passed to `cls`. That, however, has its own set of issues and in the end I concluded that `VariableSubtests` was just not written with duck arrays in mind, and that pytest doesn't encourage reusing tests by using inheritance (marking or removing tests from base classes is really clunky). With that in mind I simply removed the use of `VariableSubtests` for now (this shouldn't decrease the coverage too much) and we can add it back if / when we notice that we actually need these tests.

Edit: Right now the only failing tests are due to hgrecco/pint#1123. I guess I'll just remove the xfail condition and add it back once that PR is merged.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029
https://github.com/pydata/xarray/pull/4163#issuecomment-649854399,https://api.github.com/repos/pydata/xarray/issues/4163,649854399,MDEyOklzc3VlQ29tbWVudDY0OTg1NDM5OQ==,5635139,2020-06-25T22:44:43Z,2020-06-25T22:44:43Z,MEMBER,"I just saw your message from a couple of days ago @keewis . 
To the extent you've solved that, great. 
If you haven't — one option would be to change `assert_array_equal` in those tests to xarray's `assert_equal`, and ensure xarray's functions handle the duck arrays reasonably.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029
https://github.com/pydata/xarray/pull/4163#issuecomment-648266355,https://api.github.com/repos/pydata/xarray/issues/4163,648266355,MDEyOklzc3VlQ29tbWVudDY0ODI2NjM1NQ==,14808389,2020-06-23T16:14:44Z,2020-06-23T21:50:51Z,MEMBER,"unfortunately, this doesn't seem to really work: `VariableSubtests` was written to use the same tests for `Variable` and `IndexVariable`, and using it to test duckarrays (`pint` in particular) is difficult since the data that is often used to compare the result is not converted. I'm not sure if rewriting `VariableSubtests` to optionally allow converting the data or adding specific Variable-with-units tests instead of inheriting from `VariableSubtests` is easier...","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029
https://github.com/pydata/xarray/pull/4163#issuecomment-647576284,https://api.github.com/repos/pydata/xarray/issues/4163,647576284,MDEyOklzc3VlQ29tbWVudDY0NzU3NjI4NA==,2448579,2020-06-22T15:04:58Z,2020-06-22T15:04:58Z,MEMBER,">  I think we could also write our own version of assert_array_equal / assert_allclose that works on duck arrays. 

This sounds right to me.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029
https://github.com/pydata/xarray/pull/4163#issuecomment-647525098,https://api.github.com/repos/pydata/xarray/issues/4163,647525098,MDEyOklzc3VlQ29tbWVudDY0NzUyNTA5OA==,14808389,2020-06-22T13:38:14Z,2020-06-22T13:38:14Z,MEMBER,"except from the failing `operator.eq` tests (which are due to `pint` upstream and will be fixed soon) and the `pad` tests that need to be rewritten only the tests from the `VariableSubclassobjects` are left.

Since they use `assert_array_equal` and `np.testing.assert_allclose` (which don't dispatch), I'm not sure how to deal with them: If I wanted to fix them in `TestVariable` I would probably end up rewriting all of these. Other than that, I think we could also write our own version of `assert_array_equal` / `assert_allclose` that works on duck arrays. We already have the functionality in `duck_array_ops.py`, we'd just need to write the summary code. That would probably have to be a new PR, though.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,640740029