issue_comments

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

With da.loc[{'x':~mask}], you're actually doing label-based indexing, so ~mask will select repeatedly the items for which x=1.0 (True) and x=0.0 (False):

```python

da.loc[{'x':~mask}] <xarray.DataArray (x: 10)> array([1., 0., 1., 0., 1., 0., 1., 0., 1., 0.]) Coordinates: * x (x) float64 1.0 0.0 1.0 0.0 1.0 0.0 1.0 0.0 1.0 0.0 ```

What you want is integer-based indexing (i.e., not using .loc):

```python

da[{'x':~mask}] <xarray.DataArray (x: 5)> array([0., 2., 4., 6., 8.]) Coordinates: * x (x) float64 0.0 2.0 4.0 6.0 8.0 ```

doc link.