fix: where for cupy.array

[hoxbro]

• 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

Noticed the following will fail; I copied the logic from the asarray function just below, and added a simple test.

import cupy as cp
import numpy as np
import xarray as xr

xp = cp
arr = xp.arange(6.0).reshape(3, 2)
arr[0] = xp.nan
mask = xp.isnan(arr)

da = xr.DataArray(arr, dims=("x", "y"), name="example")
da.where(mask, 0)

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[1], line 11
      8 mask = xp.isnan(arr)
     10 da = xr.DataArray(arr, dims=("x", "y"), name="example")
---> 11 da.where(mask, 0)

File [~/projects/xarray/xarray/core/common.py:1251](http://localhost:8888/home/shh/projects/xarray/xarray/core/common.py#line=1250), in DataWithCoords.where(self, cond, other, drop)
   1247     cond = cond.isel(**indexers)
   1249 from xarray.computation import ops
-> 1251 return ops.where_method(self, cond, other)

File [~/projects/xarray/xarray/computation/ops.py:184](http://localhost:8888/home/shh/projects/xarray/xarray/computation/ops.py#line=183), in where_method(self, cond, other)
    182 # alignment for three arguments is complicated, so don't support it yet
    183 join: Literal["inner", "exact"] = "inner" if other is dtypes.NA else "exact"
--> 184 return apply_ufunc(
    185     duck_array_ops.where_method,
    186     self,
    187     cond,
    188     other,
    189     join=join,
    190     dataset_join=join,
    191     dask="allowed",
    192     keep_attrs=True,
    193 )

File [~/projects/xarray/xarray/computation/apply_ufunc.py:1269](http://localhost:8888/home/shh/projects/xarray/xarray/computation/apply_ufunc.py#line=1268), in apply_ufunc(func, input_core_dims, output_core_dims, exclude_dims, vectorize, join, dataset_join, dataset_fill_value, keep_attrs, kwargs, dask, output_dtypes, output_sizes, meta, dask_gufunc_kwargs, on_missing_core_dim, *args)
   1267 # feed DataArray apply_variable_ufunc through apply_dataarray_vfunc
   1268 elif any(isinstance(a, DataArray) for a in args):
-> 1269     return apply_dataarray_vfunc(
   1270         variables_vfunc,
   1271         *args,
   1272         signature=signature,
   1273         join=join,
   1274         exclude_dims=exclude_dims,
   1275         keep_attrs=keep_attrs,
   1276     )
   1277 # feed Variables directly through apply_variable_ufunc
   1278 elif any(isinstance(a, Variable) for a in args):

File [~/projects/xarray/xarray/computation/apply_ufunc.py:312](http://localhost:8888/home/shh/projects/xarray/xarray/computation/apply_ufunc.py#line=311), in apply_dataarray_vfunc(func, signature, join, exclude_dims, keep_attrs, *args)
    307 result_coords, result_indexes = build_output_coords_and_indexes(
    308     args, signature, exclude_dims, combine_attrs=keep_attrs
    309 )
    311 data_vars = [getattr(a, "variable", a) for a in args]
--> 312 result_var = func(*data_vars)
    314 out: tuple[DataArray, ...] | DataArray
    315 if signature.num_outputs > 1:

File [~/projects/xarray/xarray/computation/apply_ufunc.py:820](http://localhost:8888/home/shh/projects/xarray/xarray/computation/apply_ufunc.py#line=819), in apply_variable_ufunc(func, signature, exclude_dims, dask, output_dtypes, vectorize, keep_attrs, dask_gufunc_kwargs, *args)
    815 elif vectorize:
    816     func = _vectorize(
    817         func, signature, output_dtypes=output_dtypes, exclude_dims=exclude_dims
    818     )
--> 820 result_data = func(*input_data)
    822 if signature.num_outputs == 1:
    823     result_data = (result_data,)

File [~/projects/xarray/xarray/core/duck_array_ops.py:414](http://localhost:8888/home/shh/projects/xarray/xarray/core/duck_array_ops.py#line=413), in where_method(data, cond, other)
    412 if other is dtypes.NA:
    413     other = dtypes.get_fill_value(data.dtype)
--> 414 return where(cond, data, other)

File [~/projects/xarray/xarray/core/duck_array_ops.py:404](http://localhost:8888/home/shh/projects/xarray/xarray/core/duck_array_ops.py#line=403), in where(condition, x, y)
    402     condition = asarray(condition, dtype=dtype, xp=xp)
    403 else:
--> 404     condition = astype(condition, dtype=dtype, xp=xp)
    406 promoted_x, promoted_y = as_shared_dtype([x, y], xp=xp)
    408 return xp.where(condition, promoted_x, promoted_y)

File [~/projects/xarray/xarray/core/duck_array_ops.py:258](http://localhost:8888/home/shh/projects/xarray/xarray/core/duck_array_ops.py#line=257), in astype(data, dtype, xp, **kwargs)
    256 if xp is np:
    257     return data.astype(dtype, **kwargs)
--> 258 return xp.astype(data, dtype, **kwargs)

File [~/.local/conda/envs/holoviz/lib/python3.13/site-packages/cupy/__init__.py:1087](http://localhost:8888/home/shh/.local/conda/envs/holoviz/lib/python3.13/site-packages/cupy/__init__.py#line=1086), in __getattr__(name)
   1084 if name in _deprecated_apis:
   1085     return getattr(_numpy, name)
-> 1087 raise AttributeError(f"module 'cupy' has no attribute {name!r}")

AttributeError: module 'cupy' has no attribute 'astype'

[max-sixty]

Suggested change

	if xp is np or not hasattr(xp, "astype"):
	if not hasattr(xp, "astype"):

wdyt?

[hoxbro]

No objections, what I originally did until I saw this below:

xarray/xarray/core/duck_array_ops.py

Line 275 in 3c6b050

if xp is np or not hasattr(xp, "astype"):

[hoxbro]

Done in abd935a

[max-sixty]

(I don't know this code at all, looks very reasonable, if anyone knows it better...)

[hoxbro]

Reverted abd935a as it made a lot of tests fail.

[keewis]

looks good to me.

It appears cupy>=14 will support the array api (similarly to numpy>=2), which means we might be able to drop this again once we can require that (provided there are no other array types that need it)

[dcherian]

Thanks @hoxbro !