issue_comments

in any case, I think any changes to the signature of _repr_inline_ should be a separate PR

I agree, let's discuss this in https://github.com/pydata/xarray/issues/5372

in any case, I think any changes to the signature of _repr_inline_ should be a separate PR

I see your points. I chose the name as _repr_inline_ because that seemed the most straight-forward but didn't think too deeply about this. If we do change the name I would like to keep the name as _repr_*_. Maybe _repr_array_inline_?

However, as I said I don't think it matters if the name is reused in a different context, as long as that is not incompatible with our use: we only care about arrays. Which is why I would slightly prefer the second alternative (a generic name and a parameter which is interpreted depending on the context). The only issue I have is that it's usually not a good idea to extend the API unless there's a specific need so I'm somewhat reluctant to add that parameter.

It's more that _repr_inline_ sounds very generic, neither really specific to xarray or even multi-dimensional arrays. So if I saw that without knowing the context of how it is used in xarray, it would surprise me that it's supposed to omit shape and dtype, which are otherwise essential parts of the repr.

Other possibilities: - Give the protocol a namespace, like _xarray_repr_inline_ to make it clearer that it's for Xarray's use. - Maybe add an omit argument, e.g., obj._repr_inline_(max_width, skip=['dtype', 'shape']). Attributes in skip are already represented elsewhere and can be skipped if desired.

hmm... adding shape and dtype as parameters would mean that we expect cases where we actually want to include either of those and where the normal repr is not enough. Do you have anything in mind for that?

I'm also not quite sure I get your point about the name: xarray currently does not allow wrapping anything other than duck arrays so I'm inclined to just not worry about this until that changes or someone else wants to use that name for a different purpose.

I wonder if _repr_inline_ should include something explicit about excluded fields in the protocol, e.g., obj._repr_inline_(shape=False, dtype=False) and/or if the name should make it clearer that it's for array objects, e.g., _ndarray_repr_inline_?

I don't think we need tests (or a whats-new) for this so this should be ready for merging. The only issue notable change I think users can see is that when (if?) dask accepts the _repr_inline_, they will probably use chunktype instead of meta (because that's what their repr uses).