issue_comments: 904642129

I agree that this kind of "boolean indexing" could be clarified a bit in the docs.

I also had to double check here since I use isel() / sel() much more often than [] and .loc[].

Even now, it's not really clear to me why "boolean"-masking should only work with integer-indexing and not label-indexing. Given that the highlighted maybe_cast_to_coords_dtype is the only reason it does not work for label-based indexing.

Hmm you're right. The primary reason of maybe_cast_to_coords_dtype is because pandas indexes don't support floating point values with lower precision than float64 (e.g., #5700). But here https://github.com/pydata/xarray/blob/a6b44d72010c5311fad4f60194b527194e26094b/xarray/core/indexes.py#L236-L237 we actually bypass the underlying pandas index when the labels are a boolean array-like so that boolean indexing should normally work with both integer and label based indexing (#182).

So we should probably fix maybe_cast_to_coords_dtype with something like

python def maybe_cast_to_coords_dtype(label, coords_dtype): label_dtype = np.dtype(getattr(label, "dtype", None)) if coords_dtype.kind == "f" and not isinstance(label, slice) and label_dtype.kind != "b": label = np.asarray(label, dtype=coords_dtype) return label