Coerce the labels passed to Index.query to array-like objects

[benbovy]

When looking at #5691 I noticed that the labels are sometimes coerced to arrays (i.e., #3153) but not always.

Later in PandasIndex.query those may again be coerced to arrays (i.e., _as_array_tuplesafe). In #5692 (a551c7f) they are always coerced to arrays before maybe be converted as scalars.

Shouldn't we therefore make things easier and ensure that the labels given to xarray.Index.query() always have an array interface? This would also yield a more predictable behavior to anyone who wants to implement custom xarray indexes.

[benbovy]

In #5692 (a551c7f) they are always coerced to arrays before maybe be converted as scalars.

This is not exactly true: slices are kept as-is.

#3153 is actually a workaround specific to Pandas index limitations (only float64 floating -point labels supported), so it would make sense to move this within PandasIndex.

Other use-cases like xoak point-wise indexing only support Variable or DataArray objects as indexers (to ensure that all indexers have the same dimensions).

So it would probably be better to leave indexers untouched and let Index objects do the validation and conversion.

[dcherian]

leave indexers untouched and let Index objects do the validation and conversion.

So does this mean custom indexes have to implement vectorized or pointwise indexing on their own?

[benbovy]

So does this mean custom indexes have to implement vectorized or pointwise indexing on their own?

No I don't think so (sorry my thoughts are still confusing), we would still use the current implementation in Xarray for positional (vectorized, pointwise, etc.) indexing.

Currently label-based indexers go through these normalization steps before passing them to Index.query:

1. cast floating-point labels to coordinate dtype (maybe_cast_to_coords_dtype())
2. convert DataArray objects to unlabelled data (via .variable.data)

The dimension (positional) indexers returned by Index.query may be also converted before passing them to .isel:

3. back to Variable or DataArray objects for advanced indexing

I think that PandasIndex should take care of 1. Casting floating-point labels (casting any object to a numpy array as a side effect) may be undesirable for other, custom indexes.

For 2 and 3, I'm not exactly sure what to do. We should probably avoid 2 since dimension labels are useful (e.g., it is used in .xoak.sel) and let the indexes handle that. This means that indexes should also handle 3. For pandas multi-indexes, this would allow us raising a more meaningful error message since they don't support DataArray or Variable objects as level indexers. It is also better to handle 2 and 3 in a per-case basis since a general solution for 3 is tricky with the new data model (non-dimensional indexes).

This means a lot of flexibility for custom indexes but also a great responsibility.