Allow the same scheme name, if description matches

Fixes #929

Author: hyanwong

tests/test_inference.py

@@ -4350,11 +4350,39 @@ def test_is_copy(self):
         assert schema is not other
 
     def test_name_collision(self):
+        schema = MetadataSchema.permissive_json().schema
+        schema = tsinfer.add_to_schema(
+            schema, "name", definition={"type": "number", "description": "something"}
+        )
+        with pytest.raises(ValueError):
+            tsinfer.add_to_schema(
+                schema, "name", definition={"type": "number", "description": "alt"}
+            )
+
+    def test_name_collision_no_definition(self):
         schema = MetadataSchema.permissive_json().schema
         schema = tsinfer.add_to_schema(schema, "name")
         with pytest.raises(ValueError):
             tsinfer.add_to_schema(schema, "name")
 
+    def test_name_collision_same_description(self, caplog):
+        schema = MetadataSchema.permissive_json().schema
+        with caplog.at_level(logging.WARNING):
+            schema1 = tsinfer.add_to_schema(
+                schema,
+                "name",
+                definition={"description": "a unique description", "type": "number"},
+            )
+            assert caplog.text == ""
+        with caplog.at_level(logging.WARNING):
+            schema2 = tsinfer.add_to_schema(
+                schema1,
+                "name",
+                definition={"type": "number", "description": "a unique description"},
+            )
+            assert "already in schema" in caplog.text
+        assert schema1 == schema2
+
     def test_defaults(self):
         schema = MetadataSchema.permissive_json().schema
         schema = tsinfer.add_to_schema(schema, "name")

tsinfer/inference.py

@@ -72,11 +72,29 @@
 
 
 def add_to_schema(schema, name, definition=None, required=False):
+    """
+    Adds the specified metadata name to the schema, with the specified definition.
+    If the metadata name is already in the schema then either will warn about
+    potential overwriting (if the definition is the same and there is a description),
+    or will raise an error otherwise (to avoid conflicting metadata definitions).
+    """
     schema = copy.deepcopy(schema)
     if definition is None:
         definition = {}
     try:
         if name in schema["properties"]:
+            try:
+                if (
+                    schema["properties"][name] == definition
+                    and definition["description"] != ""
+                ):
+                    logger.warning(
+                        f"Metadata {name} with identical description already in schema."
+                        " Schema left unchanged: existing metadata may be overwritten."
+                    )
+                    return schema
+            except KeyError:
+                pass
             raise ValueError(f"The metadata {name} is reserved for use by tsinfer")
     except KeyError:
         schema["properties"] = {}