Fix Enum collision with same choices & varying labels #790 #1104

drf_spectacular/hooks.py

@@ -45,6 +45,16 @@ def create_enum_component(name, schema):
         generator.registry.register_on_missing(component)
         return component
 
+    def extract_hash(schema):
+        if 'x-spec-enum-id' in schema:
+            # try to use the injected enum hash first as it generated from (name, value) tuples,
+            # which prevents collisions on choice sets only differing in labels not values.
+            return schema['x-spec-enum-id']
+        else:
+            # fall back to actual list hashing when we encounter enums not generated by us.
+            # remove blank/null entry for hashing. will be reconstructed in the last step
+            return list_hash([(i, i) for i in schema['enum'] if i not in ('', None)])
+
     schemas = result.get('components', {}).get('schemas', {})
 
     overrides = load_enum_name_overrides()
@@ -58,8 +68,8 @@ def create_enum_component(name, schema):
                 prop_schema = prop_schema.get('items', {})
             if 'enum' not in prop_schema:
                 continue
-            # remove blank/null entry for hashing. will be reconstructed in the last step
-            prop_enum_cleaned_hash = list_hash([i for i in prop_schema['enum'] if i not in ['', None]])
+
+            prop_enum_cleaned_hash = extract_hash(prop_schema)
             prop_hash_mapping[prop_name].add(prop_enum_cleaned_hash)
             hash_name_mapping[prop_enum_cleaned_hash].add((component_name, prop_name))
 
@@ -110,14 +120,14 @@ def create_enum_component(name, schema):
 
             prop_enum_original_list = prop_schema['enum']
             prop_schema['enum'] = [i for i in prop_schema['enum'] if i not in ['', None]]
-            prop_hash = list_hash(prop_schema['enum'])
+            prop_hash = extract_hash(prop_schema)
             # when choice sets are reused under multiple names, the generated name cannot be
             # resolved from the hash alone. fall back to prop_name and hash for resolution.
             enum_name = enum_name_mapping.get(prop_hash) or enum_name_mapping[prop_hash, prop_name]
 
             # split property into remaining property and enum component parts
             enum_schema = {k: v for k, v in prop_schema.items() if k in ['type', 'enum']}
-            prop_schema = {k: v for k, v in prop_schema.items() if k not in ['type', 'enum']}
+            prop_schema = {k: v for k, v in prop_schema.items() if k not in ['type', 'enum', 'x-spec-enum-id']}
 
             # separate actual description from name-value tuples
             if spectacular_settings.ENUM_GENERATE_CHOICE_DESCRIPTION:
@@ -148,6 +158,30 @@ def create_enum_component(name, schema):
 
     # sort again with additional components
     result['components'] = generator.registry.build(spectacular_settings.APPEND_COMPONENTS)
+
+    postprocess_schema_enum_id_removal(result, generator)
+
+    return result
+
+
+def postprocess_schema_enum_id_removal(result, generator, **kwargs):
+    """
+    Iterative modifying approach to scanning the whole schema and removing the
+    temporary helper ids that allowed us to distinguish similar enums.
+    """
+    def clean(sub_result):
+        if isinstance(sub_result, dict):
+            for key in list(sub_result):
+                if key == 'x-spec-enum-id':
+                    del sub_result['x-spec-enum-id']
+                else:
+                    clean(sub_result[key])
+        elif isinstance(sub_result, (list, tuple)):
+            for item in sub_result:
+                clean(item)
+
+    clean(result)
+
     return result
 
 

drf_spectacular/plumbing.py

@@ -431,6 +431,8 @@ def build_choice_field(field):
     if spectacular_settings.ENUM_GENERATE_CHOICE_DESCRIPTION:
         schema['description'] = build_choice_description_list(field.choices.items())
 
+    schema['x-spec-enum-id'] = list_hash([(k, v) for k, v in field.choices.items() if k not in ('', None)])
+
     return schema
 
 
@@ -499,10 +501,12 @@ def build_root_object(paths, components, version):
 def safe_ref(schema):
     """
     ensure that $ref has its own context and does not remove potential sibling
-    entries when $ref is substituted.
+    entries when $ref is substituted. also remove useless singular "allOf" .
     """
     if '$ref' in schema and len(schema) > 1:
         return {'allOf': [{'$ref': schema.pop('$ref')}], **schema}
+    if 'allOf' in schema and len(schema) == 1 and len(schema['allOf']) == 1:
+        return schema['allOf']
     return schema
 
 
@@ -815,11 +819,12 @@ def load_enum_name_overrides():
         if inspect.isclass(choices) and issubclass(choices, Choices):
             choices = choices.choices
         if inspect.isclass(choices) and issubclass(choices, Enum):
-            choices = [c.value for c in choices]
+            choices = [(c.value, c.name) for c in choices]
         normalized_choices = []
         for choice in choices:
             # Allow None values in the simple values list case
             if isinstance(choice, str) or choice is None:
+                # TODO warning
                 normalized_choices.append((choice, choice))  # simple choice list
             elif isinstance(choice[1], (list, tuple)):
                 normalized_choices.extend(choice[1])  # categorized nested choices
@@ -828,7 +833,9 @@ def load_enum_name_overrides():
 
         # Get all of choice values that should be used in the hash, blank and None values get excluded
         # in the post-processing hook for enum overrides, so we do the same here to ensure the hashes match
-        hashable_values = [value for value, _ in normalized_choices if value not in ['', None]]
+        hashable_values = [
+            (value, label) for value, label in normalized_choices if value not in ['', None]
+        ]
         overrides[list_hash(hashable_values)] = name
 
     if len(spectacular_settings.ENUM_NAME_OVERRIDES) != len(overrides):
@@ -840,7 +847,7 @@ def load_enum_name_overrides():
 
 
 def list_hash(lst):
-    return hashlib.sha256(json.dumps(list(lst), sort_keys=True, cls=JSONEncoder).encode()).hexdigest()
+    return hashlib.sha256(json.dumps(list(lst), sort_keys=True, cls=JSONEncoder).encode()).hexdigest()[:16]
 
 
 def anchor_pattern(pattern: str) -> str:

tests/test_postprocessing.py

@@ -9,9 +9,10 @@
 from rest_framework.views import APIView
 
 try:
-    from django.db.models.enums import TextChoices
+    from django.db.models.enums import IntegerChoices, TextChoices
 except ImportError:
     TextChoices = object  # type: ignore  # django < 3.0 handling
+    IntegerChoices = object  # type: ignore  # django < 3.0 handling
 
 from drf_spectacular.plumbing import list_hash, load_enum_name_overrides
 from drf_spectacular.utils import OpenApiParameter, extend_schema
@@ -244,35 +245,39 @@ def partial_update(self, request):
 
 
 def test_enum_override_variations(no_warnings):
-    enum_override_variations = ['language_list', 'LanguageEnum', 'LanguageStrEnum']
+    enum_override_variations = [
+        ('language_list', [('en', 'en')]),
+        ('LanguageEnum', [('en', 'EN')]),
+        ('LanguageStrEnum', [('en', 'EN')]),
+    ]
     if DJANGO_VERSION > '3':
-        enum_override_variations += ['LanguageChoices', 'LanguageChoices.choices']
+        enum_override_variations += [
+            ('LanguageChoices', [('en', 'En')]),
+            ('LanguageChoices.choices', [('en', 'En')])
+        ]
 
-    for variation in enum_override_variations:
+    for variation, expected_hashed_keys in enum_override_variations:
         with mock.patch(
             'drf_spectacular.settings.spectacular_settings.ENUM_NAME_OVERRIDES',
             {'LanguageEnum': f'tests.test_postprocessing.{variation}'}
         ):
             load_enum_name_overrides.cache_clear()
-            assert list_hash(['en']) in load_enum_name_overrides()
+            assert list_hash(expected_hashed_keys) in load_enum_name_overrides()
 
 
 def test_enum_override_variations_with_blank_and_null(no_warnings):
     enum_override_variations = [
-        'blank_null_language_list',
-        'BlankNullLanguageEnum',
-        ('BlankNullLanguageStrEnum', ['en', 'None'])
+        ('blank_null_language_list', [('en', 'en')]),
+        ('BlankNullLanguageEnum', [('en', 'EN')]),
+        ('BlankNullLanguageStrEnum', [('en', 'EN'), ('None', 'NULL')])
     ]
     if DJANGO_VERSION > '3':
         enum_override_variations += [
-            ('BlankNullLanguageChoices', ['en', 'None']),
-            ('BlankNullLanguageChoices.choices', ['en', 'None'])
+            ('BlankNullLanguageChoices', [('en', 'En'), ('None', 'Null')]),
+            ('BlankNullLanguageChoices.choices', [('en', 'En'), ('None', 'Null')])
         ]
 
-    for variation in enum_override_variations:
-        expected_hashed_keys = ['en']
-        if isinstance(variation, (list, tuple, )):
-            variation, expected_hashed_keys = variation
+    for variation, expected_hashed_keys in enum_override_variations:
         with mock.patch(
             'drf_spectacular.settings.spectacular_settings.ENUM_NAME_OVERRIDES',
             {'LanguageEnum': f'tests.test_postprocessing.{variation}'}
@@ -340,3 +345,28 @@ def get(self, request):
         uuid.UUID('93d7527f-de3c-4a76-9cc2-5578675630d4'),
         uuid.UUID('47a4b873-409e-4e43-81d5-fafc3faeb849')
     ]
+
+
+@pytest.mark.skipif(DJANGO_VERSION < '3', reason='Not available before Django 3.0')
+def test_equal_choices_different_semantics(no_warnings):
+
+    class Health(IntegerChoices):
+        OK = 0
+        FAIL = 1
+
+    class Status(IntegerChoices):
+        GREEN = 0
+        RED = 1
+
+    class XSerializer(serializers.Serializer):
+        some_health = serializers.ChoiceField(choices=Health.choices)
+        some_status = serializers.ChoiceField(choices=Status.choices)
+
+    class XAPIView(APIView):
+        @extend_schema(responses=XSerializer)
+        def get(self, request):
+            pass  # pragma: no cover
+
+    # This should not generate a warning even though the enum list is identical
+    # in both Enums. We now also differentiate the Enums by their labels.
+    generate_schema('x', view=XAPIView)