Use to_numpy in ImplicitToExplicitIndexingAdapter.__array__ method

[weiji14]

Description

Fixes TypeError: Implicit conversion to a NumPy array is not allowed. Please use `.get()` to construct a NumPy array explicitly when using a backend engine that loads data onto CuPy, but indexes should stay as NumPy (following convention at xarray-contrib/cupy-xarray#90)

Full traceback

TypeError: Implicit conversion to a NumPy array is not allowed. Please use `.get()` to construct a NumPy array explicitly.
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[10], line 20, in test_xarray_backend_open_mfdataset()
     13         engine="cog3pio",
     14         concat_dim="band",
     15         combine="nested",
     16     )
---> 17     ds.raster.load()
     18     assert ds.sizes == {"band": 2, "y": 183, "x": 183}
     19     assert ds.x.min() == 700260.0
     20     assert ds.x.max() == 809460.0
File ~/mambaforge/envs/cupy-xarray/lib/python3.14/site-packages/xarray/core/dataarray.py:1173, in DataArray.load(self, **kwargs)
   1143 def load(self, **kwargs) -> Self:
   1144     """Trigger loading data into memory and return this dataarray.
   1145
   1146     Data will be computed and/or loaded from disk or a remote source.
   (...)   1171     Variable.load
   1172     """
-> 1173     ds = self._to_temp_dataset().load(**kwargs)
   1174     new = self._from_temp_dataset(ds)
   1175     self._variable = new._variable
File ~/mambaforge/envs/cupy-xarray/lib/python3.14/site-packages/xarray/core/dataset.py:569, in Dataset.load(self, **kwargs)
    565         if chunked_data:
    566             chunkmanager = get_chunked_array_type(*chunked_data.values())
    567
    568             # evaluate all the chunked arrays simultaneously
--> 569             evaluated_data: tuple[np.ndarray[Any, Any], ...] = chunkmanager.compute(
    570                 *chunked_data.values(), **kwargs
    571             )
    572
File ~/mambaforge/envs/cupy-xarray/lib/python3.14/site-packages/xarray/namedarray/daskmanager.py:85, in DaskManager.compute(self
, *data, **kwargs)
     80 def compute(
     81     self, *data: Any, **kwargs: Any
     82 ) -> tuple[np.ndarray[Any, _DType_co], ...]:
     83     from dask.array import compute
---> 85     return compute(*data, **kwargs)
File ~/mambaforge/envs/cupy-xarray/lib/python3.14/site-packages/dask/base.py:685, in compute(traverse, optimize_graph, scheduler
, get, *args, **kwargs)
    682     expr = expr.optimize()
    683     keys = list(flatten(expr.__dask_keys__()))
--> 685     results = schedule(expr, keys, **kwargs)
    687 return repack(results)
File ~/mambaforge/envs/cupy-xarray/lib/python3.14/site-packages/xarray/core/indexing.py:691, in ImplicitToExplicitIndexingAdapte
r.__array__(self, dtype, copy)
    687 def __array__(
    688     self, dtype: DTypeLike | None = None, /, *, copy: bool | None = None
    689 ) -> np.ndarray:
    690     if Version(np.__version__) >= Version("2.0.0"):
--> 691         return np.asarray(self.get_duck_array(), dtype=dtype, copy=copy)
    692     else:
    693         return np.asarray(self.get_duck_array(), dtype=dtype)
File cupy/_core/core.pyx:1760, in cupy._core.core._ndarray_base.__array__()
-> 1760 'Could not get source, probably due dynamically evaluated source code.'
TypeError: Implicit conversion to a NumPy array is not allowed. Please use `.get()` to construct a NumPy array explicitly.

Somewhat similar to #10078. Helps with some cupy-xarray work that is making TIFF to GPU possible at xarray-contrib/cupy-xarray#81 (comment) (though not fully working yet 🙂).

Checklist

• Closes #xxxx
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

AI Disclosure

[weiji14]

Should probably add a test for this somehow.

[weiji14]

Please don't merge yet, as I don't think the current patch is correct (even though the new test passes).

Edit: should be ok now after 0c405d9.

[weiji14]

This fixes the TypeError, but I don't think it is the correct solution (to_numpy doesn't handle the dtype/copy arguments). Probably should use duck_array_ops.as_array instead? Also, best if we can drop NumPy<2.0 to make the code cleaner 🙂

[weiji14]

Ok, using np.asarray(to_numpy(self.get_duck_array()), dtype=dtype, copy=copy) now, which should be the correct way.