issue_comments

If everyone is happy with this I'm going to merge it.

I agree with @max-sixty A > C > B but only weakly.

Same opinion.

This looks v good @benbovy!

I'm just fixing int32 vs int64 issue with tests run on appveyor, then to me it's ready to merge unless someone else has objections.

If you want to add that example as a test, that could be a good way of documenting the function. But I don't think it's strictly needed

Yes I could use those examples for the tests. I think it might be good to write shallow tests at least for diff_array_repr and diff_dataset_repr (in formatting.py).

I'm wondering which is the best among the options below:

A.

Differing data variables: L var1 (x, y) int64 1 2 3 4 5 7 R var1 (x) int64 1 2 L var2 (x, y) int64 4 5 6 7 8 9 description: variable 2 R var2 (x, y) int64 4 5 6 7 8 9

B.

Differing data variables: L var1 (x, y) int64 1 2 3 4 5 7 R var1 (x) int64 1 2 L var2 (x, y) int64 4 5 6 7 8 9 description: variable 2 R var2 (x, y) int64 4 5 6 7 8 9

C.

Differing data variables: L var1 (x, y) int64 1 2 3 4 5 7 var2 (x, y) int64 4 5 6 7 8 9 description: variable 2 R var1 (x) int64 1 2 var2 (x, y) int64 4 5 6 7 8 9

Thanks for reminding me this PR @max-sixty ! It deserves more than no commits since 1.5 years!

The assertion error messages now look like pytest's failure reports (see test_eq_dict). I've tried to reuse as much as possible the repr formatting that we already have for the xarray objects.

As examples:

```python ds_a = xr.Dataset(data_vars={'var1': (('x', 'y'), [[1, 2, 3], [4, 5, 7]]), 'var2': ('x', [3, 4])}, coords={'x': ['a', 'b'], 'y': [1, 2, 3]}, attrs={'units': 'm', 'description': 'desc'})

ds_b = xr.Dataset(data_vars={'var1': ('x', [1, 2])}, coords={'x': ('x', ['a', 'c'], {'source': 0}), 'label': ('x', [1, 2])}, attrs={'units': 'kg'})

xr.testing.assert_identical(ds_a, ds_b) AssertionError: Left and right Dataset objects are not identical Differing dimensions: (x: 2, y: 3) != (x: 2) Differing coordinates: L * x (x) <U1 'a' 'b' R * x (x) <U1 'a' 'c' source: 0 Left contains more coordinates: * y (y) int64 1 2 3 Right contains more coordinates: label (x) int64 1 2 Differing data variables: L var1 (x, y) int64 1 2 3 4 5 7 R var1 (x) int64 1 2 Left contains more data variables: var2 (x) int64 3 4 Differing attributes: L units: m R units: kg Left contains more attributes: description: desc ```

```python da_a = xr.DataArray([[1, 2, 3], [4, 5, 7]], dims=('x', 'y'), coords={'x': ['a', 'b'], 'y': [1, 2, 3]}, attrs={'units': 'm', 'description': 'desc'})

da_b = xr.DataArray([1, 2], dims='x', coords={'x': ['a', 'c'], 'label': ('x', [1, 2])}, attrs={'units': 'kg'})

xr.testing.assert_equal(da_a, da_b) AssertionError: Left and right DataArray objects are not equal Differing values: L array([[1, 2, 3], [4, 5, 7]]) R array([1, 2]) Differing coordinates: L * x (x) <U1 'a' 'b' R * x (x) <U1 'a' 'c' Left contains more coordinates: * y (y) int64 1 2 3 Right contains more coordinates: label (x) int64 1 2 ```

Those examples are probably not the most readable ones but it's for a full showcase.

A downside of this approach (i.e., report the exact differing) is that when a and b differs, full comparison is done once again for formatting the report. I don't think there is an easy way to avoid that, but I don't think it's a big deal either.

Indeed, exact differing variables/attributes would be much better here. I'll look into this next week.

pytest's failure reports and reports of numpy.testing functions may be good sources of inspiration.