Fix true division integer casting. (#15)

CHANGELOG.rst

@@ -6,12 +6,19 @@
 Changelog
 =========
 
+0.6.1 (unreleased)
+------------------
+
+**Bug fixes**
+
+- Division now complies more strictly with the Array API standard by returning a floating-point result regardless of input data types.
+
 0.6.0 (2024-07-11)
 ------------------
 
 **Other changes**
 
-- :func:`ndonnx.promote_nullable` is now publicly exported
+- ``ndonnx.promote_nullable`` is now publicly exported.
 
 0.5.0 (2024-07-01)
 ------------------

ndonnx/_core/_impl.py

@@ -112,7 +112,20 @@ def cosh(self, x):
         return _unary_op(x, opx.cosh, dtypes.float32)
 
     def divide(self, x, y):
-        return _variadic_op([x, y], opx.div, via_dtype=dtypes.float64)
+        x, y = promote(x, y)
+        if not isinstance(x.dtype, (dtypes.Numerical, dtypes.NullableNumerical)):
+            raise TypeError(f"Unsupported dtype for divide: {x.dtype}")
+        bits = (
+            ndx.iinfo(x.dtype).bits
+            if isinstance(x.dtype, (dtypes.Integral, dtypes.NullableIntegral))
+            else ndx.finfo(x.dtype).bits
+        )
+        via_dtype = (
+            dtypes.float64
+            if bits > 32 or x.dtype in (dtypes.nuint32, dtypes.uint32)
+            else dtypes.float32
+        )
+        return _variadic_op([x, y], opx.div, via_dtype=via_dtype, cast_return=False)
 
     def equal(self, x, y) -> Array:
         x, y = promote(x, y)
@@ -134,7 +147,20 @@ def floor(self, x):
         return x
 
     def floor_divide(self, x, y):
-        return self.floor(self.divide(x, y))
+        x, y = promote(x, y)
+        dtype = x.dtype
+        out = self.floor(self.divide(x, y))
+        if isinstance(
+            dtype,
+            (
+                dtypes.Integral,
+                dtypes.NullableIntegral,
+                dtypes.Unsigned,
+                dtypes.NullableUnsigned,
+            ),
+        ):
+            out = out.astype(dtype)
+        return out
 
     def greater(self, x, y):
         return _via_i64_f64(opx.greater, [x, y], cast_return=False)

skips.txt

@@ -32,9 +32,6 @@ array_api_tests/test_has_names.py::test_has_names[linalg-cross]
 array_api_tests/test_has_names.py::test_has_names[linalg-cholesky]
 array_api_tests/test_linalg.py
 array_api_tests/test_operators_and_elementwise_functions.py::test_atan2
-array_api_tests/test_operators_and_elementwise_functions.py::test_divide[__itruediv__(x, s)]
-array_api_tests/test_operators_and_elementwise_functions.py::test_divide[__itruediv__(x1, x2)]
-array_api_tests/test_operators_and_elementwise_functions.py::test_floor_divide
 array_api_tests/test_operators_and_elementwise_functions.py::test_sign
 array_api_tests/test_operators_and_elementwise_functions.py::test_sinh
 array_api_tests/test_set_functions.py::test_unique_all
@@ -68,6 +65,8 @@ array_api_tests/test_special_cases.py::test_binary[atan2(x1_i is -0 and x2_i is
 array_api_tests/test_special_cases.py::test_binary[atan2(x1_i is -infinity and x2_i is +infinity) -> roughly -pi/4]
 array_api_tests/test_special_cases.py::test_binary[atan2(x1_i is -infinity and x2_i is -infinity) -> roughly -3pi/4]
 array_api_tests/test_special_cases.py::test_binary[floor_divide(copysign(1, x1_i) != copysign(1, x2_i) and isfinite(x1_i) and x1_i != 0 and isfinite(x2_i) and x2_i != 0) -> negative sign]
+array_api_tests/test_special_cases.py::test_binary[__floordiv__(copysign(1, x1_i) != copysign(1, x2_i) and isfinite(x1_i) and x1_i != 0 and isfinite(x2_i) and x2_i != 0) -> negative sign]
+array_api_tests/test_special_cases.py::test_binary[__floordiv__((x1_i is +infinity or x1_i == -infinity) and (x2_i is +infinity or x2_i == -infinity)) -> NaN]
 array_api_tests/test_special_cases.py::test_binary[remainder(isfinite(x1_i) and x1_i < 0 and x2_i is +infinity) -> x2_i]
 array_api_tests/test_special_cases.py::test_binary[remainder(isfinite(x1_i) and x1_i > 0 and x2_i is -infinity) -> x2_i]
 array_api_tests/test_special_cases.py::test_binary[remainder(x1_i is +0 and x2_i < 0) -> -0]
@@ -77,6 +76,7 @@ array_api_tests/test_special_cases.py::test_iop[__imod__(isfinite(x1_i) and x1_i
 array_api_tests/test_special_cases.py::test_iop[__imod__(isfinite(x1_i) and x1_i > 0 and x2_i is -infinity) -> x2_i]
 array_api_tests/test_special_cases.py::test_iop[__imod__(x1_i is +0 and x2_i < 0) -> -0]
 array_api_tests/test_special_cases.py::test_iop[__imod__(x1_i is -0 and x2_i > 0) -> +0]
+array_api_tests/test_special_cases.py::test_iop[__ifloordiv__(copysign(1, x1_i) != copysign(1, x2_i) and isfinite(x1_i) and x1_i != 0 and isfinite(x2_i) and x2_i != 0) -> negative sign]
 array_api_tests/test_special_cases.py::test_nan_propagation[prod]
 array_api_tests/test_special_cases.py::test_unary[acos(x_i < -1) -> NaN]
 array_api_tests/test_special_cases.py::test_unary[acos(x_i > 1) -> NaN]

tests/ndonnx/test_core.py

@@ -531,3 +531,11 @@ def test_searchsorted_raises():
         b = ndx.array(shape=(3,), dtype=ndx.int64)
 
         ndx.searchsorted(a, b, side="middle")  # type: ignore[arg-type]
+
+
+def test_truediv():
+    x = ndx.asarray([1, 2, 3], dtype=ndx.int64)
+    y = ndx.asarray([2, 3, 3], dtype=ndx.int64)
+    z = x / y
+    assert isinstance(z.dtype, ndx.Floating)
+    np.testing.assert_array_equal(z.to_numpy(), np.array([0.5, 2 / 3, 1.0]))