Christian's comments

src/spox/_attributes.py

@@ -48,13 +48,12 @@ def value(self) -> T:
     def _validate(self):
         try:
             type_in_onnx = self._to_onnx_deref("dummy").type
-        except TypeError as e:
-            # TypeError: 'a' has type str, but expected one of: int -- this is raised from within protobuf
+        except Exception as e:
+            # Likely an error from within onnx/protobuf, such as:
+            # 1) AttributeError: 'int' object has no attribute 'encode'
+            # 2) TypeError: 'a' has type str, but expected one of: int -- this is raised from within protobuf
             # when .extend()-ing with the wrong type.
-            if "but expected one of" in str(e):
-                raise self._get_pretty_type_exception() from e
-            else:
-                raise
+            raise self._get_pretty_type_exception() from e
 
         if type_in_onnx != self._attribute_proto_type:
             raise self._get_pretty_type_exception()
@@ -207,13 +206,6 @@ def _to_onnx_deref(self, key: str) -> AttributeProto:
 class AttrFloat32s(_AttrIterable[float]):
     _attribute_proto_type = AttributeProto.FLOATS
 
-    def _to_onnx_deref(self, key: str) -> AttributeProto:
-        try:
-            transformed = [float(v) for v in self.value]
-        except (ValueError, TypeError) as e:
-            raise AttributeTypeError("Attribute values don't seem to be floats.") from e
-        return make_attribute(key, transformed, attr_type=self._attribute_proto_type)
-
 
 class AttrInt64s(_AttrIterable[int]):
     _attribute_proto_type = AttributeProto.INTS
@@ -222,29 +214,10 @@ class AttrInt64s(_AttrIterable[int]):
 class AttrStrings(_AttrIterable[str]):
     _attribute_proto_type = AttributeProto.STRINGS
 
-    def _to_onnx_deref(self, key: str) -> AttributeProto:
-        try:
-            transformed = [v.encode() for v in self.value]
-        except AttributeError as e:
-            raise AttributeTypeError(
-                "Attribute values don't seem to be strings."
-            ) from e
-        return make_attribute(key, transformed, attr_type=self._attribute_proto_type)
-
 
 class AttrTensors(_AttrIterable[np.ndarray]):
     _attribute_proto_type = AttributeProto.TENSORS
 
-    def _to_onnx_deref(self, key: str) -> AttributeProto:
-        try:
-            transformed = [from_array(v) for v in self.value]
-        except AttributeError as e:
-            raise AttributeTypeError(
-                "Attribute values don't seem to be numpy arrays."
-            ) from e
-
-        return make_attribute(key, transformed, attr_type=self._attribute_proto_type)
-
 
 def _deref(ref: _Ref[T]) -> T:
     if isinstance(ref._concrete._value, _Ref):

tests/test_standard_op.py

@@ -75,8 +75,16 @@ def test_multiple_outputs():
             ["a"],
             "Unable to instantiate `AttrInt64s` with value of type `tuple[str, ...]`",
         ),
-        ("value_strings", [1], "Attribute values don't seem to be strings."),
-        ("value_floats", ["a"], "Attribute values don't seem to be floats."),
+        (
+            "value_strings",
+            [1],
+            "Unable to instantiate `AttrStrings` with value of type `tuple[int, ...]",
+        ),
+        (
+            "value_floats",
+            ["a"],
+            "Unable to instantiate `AttrFloat32s` with value of type `tuple[str, ...]",
+        ),
         (
             "value_int",
             "a",
@@ -92,6 +100,8 @@ def test_passing_wrong_type(key: str, values: Any, match: str):
 def test_passing_wrong_type_tensors():
     with pytest.raises(
         AttributeTypeError,
-        match=re.escape("Attribute values don't seem to be numpy arrays."),
+        match=re.escape(
+            "Unable to instantiate `AttrTensors` with value of type `tuple[int, ...]"
+        ),
     ):
         AttrTensors([1])  # type: ignore