Run tests on ndonnx

array_api_compat/common/_helpers.py

@@ -627,17 +627,13 @@ def device(x: Array, /) -> Device:
     to_device : Move array data to a different device.
 
     """
-    if is_numpy_array(x):
+    if is_numpy_array(x) or is_ndonnx_array(x):
         return "cpu"
     elif is_dask_array(x):
         # Peek at the metadata of the jax array to determine type
-        try:
-            import numpy as np
-            if isinstance(x._meta, np.ndarray):
-                # Must be on CPU since backed by numpy
-                return "cpu"
-        except ImportError:
-            pass
+        if is_numpy_array(x._meta):
+            # Must be on CPU since backed by numpy
+            return "cpu"
         return _DASK_DEVICE
     elif is_jax_array(x):
         # JAX has .device() as a method, but it is being deprecated so that it
@@ -758,7 +754,7 @@ def to_device(x: Array, device: Device, /, *, stream: Optional[Union[int, Any]]
     device : Hardware device the array data resides on.
 
     """
-    if is_numpy_array(x):
+    if is_numpy_array(x) or is_ndonnx_array(x):
         if stream is not None:
             raise ValueError("The stream argument to to_device() is not supported")
         if device == 'cpu':
@@ -780,7 +776,6 @@ def to_device(x: Array, device: Device, /, *, stream: Optional[Union[int, Any]]
         if not hasattr(x, "__array_namespace__"):
             # In JAX v0.4.31 and older, this import adds to_device method to x.
             import jax.experimental.array_api # noqa: F401
-        return x.to_device(device, stream=stream)
     elif is_pydata_sparse_array(x) and device == _device(x):
         # Perform trivial check to return the same array if
         # device is same instead of err-ing.

docs/supported-array-libraries.md

@@ -137,3 +137,14 @@ The minimum supported Dask version is 2023.12.0.
 ## [Sparse](https://sparse.pydata.org/en/stable/)
 
 Similar to JAX, `sparse` Array API support is contained directly in `sparse`.
+
+(ndonnx-support)=
+## [ndonnx](https://github.com/quantco/ndonnx)
+
+Similar to JAX, `ndonnx` Array API support is contained directly in `ndonnx`.
+
+(array-api-strict-support)=
+## [array-api-strict](https://data-apis.org/array-api-strict/)
+
+array-api-strict exists only to test support for the Array API, so obviously it
+does not need any wrappers.

tests/_helpers.py

@@ -1,19 +1,14 @@
 from importlib import import_module
-import sys
 
 import pytest
 
 wrapped_libraries = ["numpy", "cupy", "torch", "dask.array"]
 all_libraries = wrapped_libraries + [
-    "array_api_strict", "jax.numpy", "sparse"
+    "array_api_strict", "jax.numpy", "ndonnx", "sparse"
 ]
 
-# `sparse` added array API support as of Python 3.10.
-if sys.version_info >= (3, 10):
-    all_libraries.append('sparse')
-
 def import_(library, wrapper=False):
-    if library == 'cupy':
+    if library in ('cupy', 'ndonnx'):
         pytest.importorskip(library)
     if wrapper:
         if 'jax' in library:

tests/test_array_namespace.py

@@ -14,14 +14,17 @@
 
 @pytest.mark.parametrize("use_compat", [True, False, None])
 @pytest.mark.parametrize("api_version", [None, "2021.12", "2022.12", "2023.12"])
-@pytest.mark.parametrize("library", all_libraries + ['array_api_strict'])
+@pytest.mark.parametrize("library", all_libraries)
 def test_array_namespace(library, api_version, use_compat):
     xp = import_(library)
 
     array = xp.asarray([1.0, 2.0, 3.0])
-    if use_compat is True and library in {'array_api_strict', 'jax.numpy', 'sparse'}:
+    if use_compat and library not in wrapped_libraries:
         pytest.raises(ValueError, lambda: array_namespace(array, use_compat=use_compat))
         return
+    if library == "ndonnx" and api_version in ("2021.12", "2022.12"):
+        pytest.skip("Unsupported API version")
+
     namespace = array_api_compat.array_namespace(array, api_version=api_version, use_compat=use_compat)
 
     if use_compat is False or use_compat is None and library not in wrapped_libraries:

tests/test_common.py

@@ -1,9 +1,10 @@
 from array_api_compat import (  # noqa: F401
     is_numpy_array, is_cupy_array, is_torch_array,
     is_dask_array, is_jax_array, is_pydata_sparse_array,
+    is_ndonnx_array,
     is_numpy_namespace, is_cupy_namespace, is_torch_namespace,
     is_dask_namespace, is_jax_namespace, is_pydata_sparse_namespace,
-    is_array_api_strict_namespace,
+    is_array_api_strict_namespace, is_ndonnx_namespace,
 )
 
 from array_api_compat import device, is_array_api_obj, is_writeable_array, to_device
@@ -22,6 +23,7 @@
     'dask.array': 'is_dask_array',
     'jax.numpy': 'is_jax_array',
     'sparse': 'is_pydata_sparse_array',
+    'ndonnx': 'is_ndonnx_array',
 }
 
 is_namespace_functions = {
@@ -32,6 +34,7 @@
     'jax.numpy': 'is_jax_namespace',
     'sparse': 'is_pydata_sparse_namespace',
     'array_api_strict': 'is_array_api_strict_namespace',
+    'ndonnx': 'is_ndonnx_namespace',
 }
 
 
@@ -135,26 +138,40 @@ def test_to_device_host(library):
 @pytest.mark.parametrize("target_library", is_array_functions.keys())
 @pytest.mark.parametrize("source_library", is_array_functions.keys())
 def test_asarray_cross_library(source_library, target_library, request):
-    if source_library == "dask.array" and target_library == "torch":
+    def _xfail(reason: str) -> None:
         # Allow rest of test to execute instead of immediately xfailing
         # xref https://github.com/pandas-dev/pandas/issues/38902
+        request.node.add_marker(pytest.mark.xfail(reason=reason))
 
+    if source_library == "dask.array" and target_library == "torch":
         # TODO: remove xfail once
         # https://github.com/dask/dask/issues/8260 is resolved
-        request.node.add_marker(pytest.mark.xfail(reason="Bug in dask raising error on conversion"))
-    if source_library == "cupy" and target_library != "cupy":
+        _xfail(reason="Bug in dask raising error on conversion")
+    elif (
+        source_library == "ndonnx" 
+        and target_library not in ("array_api_strict", "ndonnx", "numpy")
+    ):
+        _xfail(reason="The truth value of lazy Array Array(dtype=Boolean) is unknown")
+    elif source_library == "ndonnx" and target_library == "numpy":
+        _xfail(reason="produces numpy array of ndonnx scalar arrays")
+    elif source_library == "jax.numpy" and target_library == "torch":
+        _xfail(reason="casts int to float")
+    elif source_library == "cupy" and target_library != "cupy":
         # cupy explicitly disallows implicit conversions to CPU
         pytest.skip(reason="cupy does not support implicit conversion to CPU")
     elif source_library == "sparse" and target_library != "sparse":
         pytest.skip(reason="`sparse` does not allow implicit densification")
+
     src_lib = import_(source_library, wrapper=True)
     tgt_lib = import_(target_library, wrapper=True)
     is_tgt_type = globals()[is_array_functions[target_library]]
 
-    a = src_lib.asarray([1, 2, 3])
+    a = src_lib.asarray([1, 2, 3], dtype=src_lib.int32)
     b = tgt_lib.asarray(a)
 
     assert is_tgt_type(b), f"Expected {b} to be a {tgt_lib.ndarray}, but was {type(b)}"
+    assert b.dtype == tgt_lib.int32
+
 
 @pytest.mark.parametrize("library", wrapped_libraries)
 def test_asarray_copy(library):