fix: removed code modifying ndarray/torch tensor methods (caused an i…

…ssue with torch.compile recognising dunder methods). but added the same functionality using `__torch_function__`/`__array_ufunc__` which are the proper mechanisms to enable this behaviour. also simplified the test (ivy-llc#28394)

ivy/data_classes/array/array.py

@@ -349,7 +349,7 @@ def data(self, data):
     @classmethod
     def __torch_function__(cls, func, types, args=(), kwargs={}):
         args, kwargs = args_to_native(*args, **kwargs)
-        return func(*args, **kwargs)
+        return to_ivy(func(*args, **kwargs))
 
     def __ivy_array_function__(self, func, types, args, kwargs):
         # Cannot handle items that have __ivy_array_function__ other than those of
@@ -382,7 +382,7 @@ def __array_prepare__(self, *args, **kwargs):
 
     def __array_ufunc__(self, *args, **kwargs):
         args, kwargs = args_to_native(*args, **kwargs)
-        return self._data.__array_ufunc__(*args, **kwargs)
+        return to_ivy(self._data.__array_ufunc__(*args, **kwargs))
 
     def __array_wrap__(self, *args, **kwargs):
         args, kwargs = args_to_native(*args, **kwargs)

ivy/functional/backends/numpy/__init__.py

@@ -16,36 +16,6 @@
 
 use = ivy.utils.backend.ContextManager(_module_in_memory)
 
-# wrap __array_ufunc__ method of ivy.Array to prioritize Ivy array methods when using numpu backend
-
-
-def wrap__array_ufunc__(func):
-    def rep_method(self, ufunc, method, *inputs, **kwargs):
-        methods = {
-            "not_equal": "not_equal",
-            "greater": "greater",
-            "less": "less",
-            "greater_equal": "greater_equal",
-            "less_equal": "less_equal",
-            "multiply": "multiply",
-            "divide": "divide",
-            "remainder": "remainder",
-            "equal": "equal",
-            "bitwise_and": "bitwise_and",
-            "matmul": "matmul",
-            "power": "pow",
-            "subtract": "subtract",
-            "add": "add",
-        }
-        if ufunc.__name__ in methods:
-            return eval("ivy." + methods[ufunc.__name__] + "(*inputs, **kwargs)")
-        return func(self, ufunc, method, *inputs, **kwargs)
-
-    return rep_method
-
-
-ivy.Array.__array_ufunc__ = wrap__array_ufunc__(ivy.Array.__array_ufunc__)
-
 NativeArray = np.ndarray
 NativeDevice = str
 NativeDtype = np.dtype

ivy/functional/backends/torch/__init__.py

@@ -20,81 +20,6 @@
 
 use = ivy.utils.backend.ContextManager(_module_in_memory)
 
-
-# wrap dunder methods of native tensors to return NotImplemented to prioritize Ivy array methods.
-def dunder_wrapper(func):
-    def rep_method(*args, **kwargs):
-        for arg in args:
-            if ivy.is_ivy_array(arg):
-                return NotImplemented
-        return func(*args, **kwargs)
-
-    return rep_method
-
-
-# check for previously imported torch module
-modules_to_patch = []
-tensors_to_patch = []
-tmp_globals = dict(globals())
-for name, value in tmp_globals.items():
-    if value == "torch.Tensor":
-        tensors_to_patch.append(name)
-    try:
-        if value.__name__ == "torch":
-            modules_to_patch.append(name)
-    except AttributeError:
-        pass
-
-methods_to_patch = [
-    "__add__",
-    "__and__",
-    "__div__",
-    "__eq__",
-    "__floordiv__",
-    "__ge__",
-    "__gt__",
-    "__iadd__",
-    "__iand__",
-    "__idiv__",
-    "__ifloordiv__",
-    "__ilshift__",
-    "__imul__",
-    "__ior__",
-    "__ipow__",
-    "__irshift__",
-    "__isub__",
-    "__itruediv__",
-    "__ixor__",
-    "__le__",
-    "__lshift__",
-    "__lt__",
-    "__matmul__",
-    "__mul__",
-    "__or__",
-    "__pow__",
-    "__truediv__",
-    "__xor__",
-    "__ne__",
-    "__mod__",
-]
-
-for module in modules_to_patch:
-    for method in methods_to_patch:
-        exec(
-            module
-            + ".Tensor."
-            + method
-            + " = dunder_wrapper("
-            + module
-            + ".Tensor."
-            + method
-            + ")"
-        )
-
-for tensor in tensors_to_patch:
-    for method in methods_to_patch:
-        exec(tensor + "." + method + " = dunder_wrapper(" + tensor + "." + method + ")")
-
 NativeArray = torch.Tensor
 NativeDevice = torch.device
 NativeDtype = torch.dtype

ivy_tests/test_ivy/test_misc/test_array.py

@@ -2543,30 +2543,14 @@ def test_array_property_strides(dtype_x, backend_fw):
 
 @handle_test(
     fn_tree="functional.ivy.native_array",  # dummy fn_tree
-    dtype_x=helpers.dtype_and_values(
-        available_dtypes=helpers.get_dtypes("integer"),
-        min_dim_size=3,
-        max_dim_size=3,
-        min_num_dims=3,
-        max_num_dims=3,
-        num_arrays=2,
-        min_value=3.0,
-        max_value=10.0,
-    ),
     op=st.sampled_from(
-        ["!=", ">", "<", ">=", "<=", "*", "/", "%", "==", "&", "@", "**", "/"]
+        ["!=", ">", "<", ">=", "<=", "*", "/", "%", "==", "&", "@", "**", "/"],
     ),
 )
-def test_dunder_wrapping(
-    dtype_x,
-    backend_fw,
-    test_flags,
-    op,
-):
-    _, data = dtype_x
+def test_dunder_wrapping(backend_fw, op):
     ivy.set_backend(backend_fw)
-    x = ivy.to_native(ivy.array(data[0]))
-    y = ivy.array(data[1])
+    x = ivy.native_array([1])
+    y = ivy.array([1])
     assert ivy.is_ivy_array(y)
     assert ivy.is_native_array(x)
     res = eval(f"x {op} y")