issue_comments

we discussed this in the meeting yesterday, and decided that since it makes the code cleaner and there's the dim parameter of to_array and to_dataset (which I didn't know about when I suggested the ugly to_unstacked_array hack) we would like to restrict .cross to accept only DataArray objects. Could you make that change?

To increase their visibility it would probably also be good to show how to use to_array and to_dataset to pass Dataset objects to .cross.

The use case is something similar to: python ds = xr.Dataset({"x": ("t", [0, 1]), "y": ("t", [2, 3]), "z": ("t", [4, 5])}) arr = xr.DataArray(data=[0, 2, 4], dims="cartesian", coords={"cartesian": ["x", "y", "z"]}) xr.cross(arr, ds, dim="cartesian") # result: Dataset with x, y, z variables This does return what I want it to (which is why I don't think it is a bug), but my worry was that since ds does not have a dimension called "cartesian" it might be surprising that this works (though good documentation might help).

In any case, I'd like to get the opinion of other maintainers: is the integrated conversion worth the trouble, or should we make combining / splitting variables easier?

I finally got around to testing this. This works really well for the use cases I tried, but the automatic conversion of Datasets with three variables to a DataArray is a bit surprising: python xr.cross(arr, ds, dim="cartesian") # ← ds does not have a "cartesian" dimension not sure what to do about that. @pydata/xarray, any ideas?

To be clear, the idea was to have a convenient way to convert a Dataset with three identically dimensioned variables to a DataArray with an additional additional dimension. If there's a concise way to do that I would lean towards not doing that in cross() (even though I was the one to propose it). For example: python xr.cross( arr, ds.to_array(concat_dim="cartesian"), dim="cartesian" ).to_dataset(cut_dim="cartesian")