Fix bug when querying unnamed dataarray

[TomNicholas]

There might be a slightly neater way to do this, but this works.

• Closes query not possible on unnamed DataArray #5492
• Tests added
• Passes pre-commit run --all-files
• User visible changes (including notable bug fixes) are documented in whats-new.rst

[keewis]

is there a reason why we don't just use _to_temp_dataset / _from_temp_dataset regardless of the name?

[TomNicholas]

_to_dataset_whole is just the same thing as _to_temp_dataset except it accepts the shallow_copy argument, which is was already being used here.

Using _to_dataset_whole with name=None will fail (so I can't just pass self.name because that will be None for an unnamed dataarray). Using _to_dataset_whole without passing a name fails in the same way.

Then trying to use _from_temp_dataset fails with

    def _from_temp_dataset(
        self, dataset: Dataset, name: Union[Hashable, None, Default] = _default
    ) -> "DataArray":
>       variable = dataset._variables.pop(_THIS_ARRAY)
E       KeyError: <this-array>

so I think it will fail for named dataarrays?

[keewis]

I don't think we need the shallow_copy=True as this is a temporary dataset (and it's the usual pattern for delegating to the Dataset implementation). No need to use the name parameter for _from_temp_dataset, either, as it's being restored from self.

Edit: actually, the failure for _from_temp_dataset fails because of the name parameter to _to_dataset_whole. In total, I'd suggest to use the same pattern as in sel

[TomNicholas]

I'd suggest to use the same pattern as in sel

That doesn't work for query because query needs the name of the dataarray to be a variable on the temporary dataset, otherwise it can't evaluate queries involving the dataarray name (such as x="a > 5" where da.name = 'a'). The tests will fail with this error if you try that:

>               raise UndefinedVariableError(key, is_local) from err
E               pandas.core.computation.ops.UndefinedVariableError: name 'a' is not defined

[TomNicholas]

Basically in order for query to work you have to provide a name to construct the temporary dataset with, which rules out using _to_temp_dataset.

[keewis]

indeed. However, if I had to choose between "self" for unnamed and the name for named DataArray objects or always "self" I'd choose the latter because it's easier to use (and the implementation would be easier)? Not sure if we should try to support both (i.e. allow referencing by name and using "self")?

[TomNicholas]

I see what you mean. What I just wrote supports both individually: You use
the name when it's named, and self when it isn't. But if it's named you
can't use self. I suppose you could support using self even when it's named
by examining the query and replacing parts of it before giving it to
ds.query?

[TomNicholas]

So I added support for using 'self' even when the dataarray has a name, by using .replace(). It works, but makes the implementation more complicated again.

[TomNicholas]

One way to slightly simplify this would be to always rename all dataarray names to 'self' before querying.

[dcherian]

This will break if there's a non-dimensional coordinate variable named self won't it? I'm not sure there's good solution though, perhaps ... :D

[TomNicholas]

This line causes a mypy error

xarray/core/dataarray.py:4510: error: Incompatible types in assignment (expression has type "Hashable", variable has type "str")  [assignment]
Found 1 error in 1 file (checked 142 source files)