(feature): Fix OneOf behavior on invalid discriminator

According to the spec, propertyName is required when using a discriminator. If it is missing, the schema is invalid and should throw.

jambo/parser/oneof_type_parser.py

@@ -31,33 +31,50 @@ class OneOfTypeParser(GenericTypeParser):
         if not kwargs.get("required", False):
             mapped_properties["default"] = mapped_properties.get("default")
 
-        field_types = [
+        subfield_types = [Annotated[t, Field(**v)] for t, v in sub_types]
-            Annotated[t, Field(**v)] if self._has_meaningful_constraints(v) else t
-            for t, v in sub_types
-        ]
-
-        union_type = Union[(*field_types,)]
 
+        # Added with the understanding of discriminator are not in the JsonSchema Spec,
+        # they were added by OpenAI and not all implementations may support them,
+        # and they do not always generate a model one-to-one to the Pydantic model
+        # TL;DR: Discriminators were added by OpenAI and not a Official JSON Schema feature
         discriminator = properties.get("discriminator")
-        if discriminator and isinstance(discriminator, dict):
+        if discriminator is not None:
-            property_name = discriminator.get("propertyName")
+            validated_type = self._build_type_one_of_with_discriminator(
-            if property_name:
+                subfield_types, discriminator
-                validated_type = Annotated[
+            )
-                    union_type, Field(discriminator=property_name)
+        else:
-                ]
+            validated_type = self._build_type_one_of_with_func(subfield_types)
+
         return validated_type, mapped_properties
 
+    @staticmethod
+    def _build_type_one_of_with_discriminator(
+        subfield_types: list[Annotated], discriminator_prop: dict
+    ) -> Annotated:
+        if not isinstance(discriminator_prop, dict):
+            raise ValueError("Discriminator must be a dictionary")
+
+        property_name = discriminator_prop.get("propertyName")
+        if property_name is None or not isinstance(property_name, str):
+            raise ValueError("Discriminator must have a 'propertyName' key")
+
+        return Annotated[Union[(*subfield_types,)], Field(discriminator=property_name)]
+
+    @staticmethod
+    def _build_type_one_of_with_func(subfield_types: list[Annotated]) -> Annotated:
+        """
+        Build a validation function for the oneOf constraint.
+        This function will validate that the value matches exactly one of the schemas.
+        """
+
         def validate_one_of(value: Any) -> Any:
             matched_count = 0
-            validation_errors = []
 
-            for field_type in field_types:
+            for field_type in subfield_types:
                 try:
-                    adapter = TypeAdapter(field_type)
+                    TypeAdapter(field_type).validate_python(value)
-                    adapter.validate_python(value)
                     matched_count += 1
-                except ValidationError as e:
+                except ValidationError:
-                    validation_errors.append(str(e))
                     continue
 
             if matched_count == 0:
@@ -69,8 +86,7 @@ class OneOfTypeParser(GenericTypeParser):
 
             return value
 
-        validated_type = Annotated[union_type, BeforeValidator(validate_one_of)]
+        return Annotated[Union[(*subfield_types,)], BeforeValidator(validate_one_of)]
-        return validated_type, mapped_properties
 
     @staticmethod
     def _has_meaningful_constraints(field_props):

tests/parser/test_oneof_type_parser.py

@@ -354,54 +354,9 @@ class TestOneOfTypeParser(TestCase):
             },
         }
 
-        Model = SchemaConverter.build(schema)
+        # Should throw because the spec determines propertyName is required for discriminator
-
+        with self.assertRaises(ValueError):
-        # Should succeed because input matches exactly one schema (the first one)
+            SchemaConverter.build(schema)
-        # The first schema matches: type="a" matches const("a"), value="test" is a string
-        # The second schema doesn't match: type="a" does not match const("b")
-        obj = Model(value={"type": "a", "value": "test", "extra": "invalid"})
-        self.assertEqual(obj.value.type, "a")
-        self.assertEqual(obj.value.value, "test")
-
-        # Test with input that matches the second schema
-        obj2 = Model(value={"type": "b", "value": 42})
-        self.assertEqual(obj2.value.type, "b")
-        self.assertEqual(obj2.value.value, 42)
-
-        # Test with input that matches neither schema (should fail)
-        with self.assertRaises(ValueError) as cm:
-            Model(value={"type": "c", "value": "test"})
-        self.assertIn("does not match any of the oneOf schemas", str(cm.exception))
-
-    def test_oneof_multiple_matches_without_discriminator(self):
-        """Test case where input genuinely matches multiple oneOf schemas"""
-        schema = {
-            "title": "Test",
-            "type": "object",
-            "properties": {
-                "value": {
-                    "oneOf": [
-                        {"type": "object", "properties": {"data": {"type": "string"}}},
-                        {
-                            "type": "object",
-                            "properties": {
-                                "data": {"type": "string"},
-                                "optional": {"type": "string"},
-                            },
-                        },
-                    ],
-                    "discriminator": {},  # discriminator without propertyName
-                }
-            },
-        }
-
-        Model = SchemaConverter.build(schema)
-
-        # This input matches both schemas since both accept data as string
-        # and neither requires specific additional properties
-        with self.assertRaises(ValueError) as cm:
-            Model(value={"data": "test"})
-        self.assertIn("matches multiple oneOf schemas", str(cm.exception))
 
     def test_oneof_overlapping_strings_from_docs(self):
         """Test the overlapping strings example from documentation"""