Fix modulo operation between floatingpoints and integers (#207)

Author: cbourjau

CHANGELOG.rst

@@ -7,6 +7,14 @@
 Changelog
 =========
 
+0.18.1 (2026-04-22)
+-------------------
+
+**Bug fix**
+
+- Fixed a regression that forbade using the modulo operator between integers and floats.
+
+
 0.18.0 (2026-04-07)
 -------------------
 

ndonnx/_typed_array/onnx.py

@@ -1525,6 +1525,12 @@ def __mul__(self, other: TyArrayBase | PyScalar) -> TyArrayBase:
     def __rmul__(self, other: TyArrayBase | PyScalar) -> TyArrayBase:
         return self._apply(other, op.mul, forward=False, result_type=TyArrayNumber)
 
+    def __rmod__(self, other) -> TyArrayBase:
+        if isinstance(other, int | float):
+            b, a = promote(self, other)
+            return a.__mod__(b)
+        return NotImplemented
+
     def __pow__(self, other: TyArrayBase | PyScalar) -> TyArrayBase:
         return self._apply(other, op.pow, forward=True, result_type=TyArrayNumber)
 
@@ -1697,15 +1703,16 @@ def __or__(self, other) -> TyArrayBase:
     def __ror__(self, other) -> TyArrayBase:
         return self._apply_int_only(other, op.bitwise_or, forward=False)
 
-    def __mod__(self, other) -> TyArrayInteger:
-        return self._apply_int_only(
-            other, lambda a, b: op.mod(a, b, fmod=0), forward=True
-        )
+    def __mod__(self, other) -> TyArrayBase:
+        if isinstance(other, type(self)):
+            var = op.mod(self._var, other._var, fmod=0)
+            return safe_cast(type(self), _var_to_tyarray(var))
 
-    def __rmod__(self, other) -> TyArrayInteger:
-        return self._apply_int_only(
-            other, lambda a, b: op.mod(a, b, fmod=0), forward=False
-        )
+        if isinstance(other, TyArrayNumber | int | float):
+            a, b = promote(self, other)
+            return a.__mod__(b)
+
+        return NotImplemented
 
     def __xor__(self, other) -> TyArrayBase:
         return self._apply_int_only(other, op.bitwise_xor, forward=True)
@@ -1884,33 +1891,30 @@ def __floordiv__(self, other: TyArrayBase | PyScalar) -> TyArrayBase:
         return NotImplemented
 
     def __mod__(self, other) -> TyArrayBase:
-        if isinstance(other, TyArrayFloating | float):
+        if isinstance(other, type(self)):
             # This function is complicated for two reasons:
             # 1. The ONNX standard is undefined if dividend is 0, but the array-api is not.
             # 2. The array-api follows the Python semantics, which are rather odd.
-            a, b = promote(self, other)
-            var = op.mod(a._var, b._var, fmod=1)
+            var = op.mod(self._var, other._var, fmod=1)
             mod = safe_cast(TyArrayFloating, _var_to_tyarray(var))
             # NOTE: onnxruntime appears to have a bug where the sign
             # of zeros is only preserved if they are on the
             # false-branch!
             # TODO: File a bug!
-            fixed_mod = where((b < 0) == (mod < 0), mod, mod + b)
-            fixed_zeros = where(b > 0, const(0.0, mod.dtype), const(-0.0, mod.dtype))
+            fixed_mod = where((other < 0) == (mod < 0), mod, mod + other)
+            fixed_zeros = where(
+                other > 0, const(0.0, mod.dtype), const(-0.0, mod.dtype)
+            )
             return where(
-                safe_cast(TyArrayBool, ~((mod == 0.0) & (b != 0.0))),
+                safe_cast(TyArrayBool, ~((mod == 0.0) & (other != 0.0))),
                 fixed_mod,
                 fixed_zeros,
             )
+        if isinstance(other, TyArrayNumber | int | float):
+            a, b = promote(self, other)
+            return a.__mod__(b)
 
-        return super().__mod__(other)
-
-    def __rmod__(self, other) -> TyArrayBase:
-        if isinstance(other, TyArrayFloating | float):
-            b, a = promote(self, other)
-            var = op.mod(a._var, b._var, fmod=1)
-            return safe_cast(TyArrayFloating, _var_to_tyarray(var))
-        return super().__mod__(other)
+        return NotImplemented
 
     def __ndx_logaddexp__(self, x2: TyArrayBase | int | float, /) -> TyArrayFloating:
         if isinstance(x2, TyArrayNumber | int | float):

pyproject.toml

@@ -49,7 +49,6 @@ ignore = [
     "N803",  # https://docs.astral.sh/ruff/rules/invalid-argument-name
     "N806",  # https://docs.astral.sh/ruff/rules/non-lowercase-variable-in-function
     "E501",  # https://docs.astral.sh/ruff/faq/#is-the-ruff-linter-compatible-with-black
-    "UP038", # https://github.com/astral-sh/ruff/issues/7871
     "N807",  # Free functions may start/end with dunders __array_namespace_info__
     "UP007",
 ]

tests/test_elementwise.py

@@ -17,4 +17,15 @@ def do(npx):
         return x_arr % y_arr
 
     np.testing.assert_array_equal(do(ndx).unwrap_numpy(), do(np))
-    assert do(ndx).unwrap_numpy() == x % y
+    assert do(ndx).unwrap_numpy() == x % y  # Compare to Python result
+
+
+@pytest.mark.parametrize("a,b", [(12.3, 3), (12, 3.3)])
+def test_mod_float_int(a, b):
+    a1 = np.asarray(a)
+    a2 = np.asarray(b)
+
+    def do(npx):
+        return npx.asarray(a1) % npx.asarray(a2)
+
+    np.testing.assert_array_equal(do(ndx).unwrap_numpy(), do(np))