fix(imperative): fix error message when applying custom function with non-tensor arguments

GitOrigin-RevId: 387d6fda4a

imperative/python/megengine/core/autodiff/grad.py

@@ -123,6 +123,11 @@ class Function(ops.PyOpBase):
 
             This method should return a tuple of Tensor or a single Tensor representing the output
             of the function.
+
+        .. note::
+
+            positional arguments should all be Tensor
+
         """
         raise NotImplementedError
 

imperative/src/impl/op_trait.cpp

@@ -98,6 +98,12 @@ OpTraitRegistry& OpTraitRegistry::fallback() {
     if (!trait->decide_dispatch_mode) {
         trait->decide_dispatch_mode = fallback_decide_dispatch_mode;
     }
+    if (!trait->make_name) {
+        static auto make_name = [](const OpDef& def) -> std::string {
+            return def.trait()->name;
+        };
+        trait->make_name = make_name;
+    }
     return *this;
 }
 

imperative/src/impl/ops/utility.cpp

@@ -18,6 +18,7 @@
 namespace mgb::imperative {
 
 MGB_DYN_TYPE_OBJ_FINAL_IMPL(GenericPyOp);
+OP_TRAIT_REG(GenericPyOp, GenericPyOp).fallback();
 
 namespace { namespace fastpathcopy {
     auto apply_on_var_node(