core[patch]: Fix handling of title when tool schema is specified manually via JSONSchema (#30479)

Fix issue: #30456

Author: eyurtsev

libs/core/langchain_core/utils/function_calling.py

@@ -62,17 +62,36 @@ class ToolDescription(TypedDict):
 
 
 def _rm_titles(kv: dict, prev_key: str = "") -> dict:
+    """Recursively removes "title" fields from a JSON schema dictionary.
+
+    Remove "title" fields from the input JSON schema dictionary,
+    except when a "title" appears within a property definition under "properties".
+
+    Args:
+        kv (dict): The input JSON schema as a dictionary.
+        prev_key (str): The key from the parent dictionary, used to identify context.
+
+    Returns:
+        dict: A new dictionary with appropriate "title" fields removed.
+    """
     new_kv = {}
+
     for k, v in kv.items():
         if k == "title":
-            if isinstance(v, dict) and prev_key == "properties" and "title" in v:
+            # If the value is a nested dict and part of a property under "properties",
+            # preserve the title but continue recursion
+            if isinstance(v, dict) and prev_key == "properties":
                 new_kv[k] = _rm_titles(v, k)
             else:
+                # Otherwise, remove this "title" key
                 continue
         elif isinstance(v, dict):
+            # Recurse into nested dictionaries
             new_kv[k] = _rm_titles(v, k)
         else:
+            # Leave non-dict values untouched
             new_kv[k] = v
+
     return new_kv
 
 

libs/core/tests/unit_tests/test_tools.py

@@ -60,7 +60,10 @@
     _is_message_content_type,
     get_all_basemodel_annotations,
 )
-from langchain_core.utils.function_calling import convert_to_openai_function
+from langchain_core.utils.function_calling import (
+    convert_to_openai_function,
+    convert_to_openai_tool,
+)
 from langchain_core.utils.pydantic import (
     PYDANTIC_MAJOR_VERSION,
     _create_subset_model,
@@ -2560,3 +2563,44 @@ def foo_args_jsons_schema_with_description(x: int) -> str:
         ]
         == "description"
     )
+
+
+def test_title_property_preserved() -> None:
+    """Test that the title property is preserved when generating schema.
+
+    https://github.com/langchain-ai/langchain/issues/30456
+    """
+    from typing import Any
+
+    from langchain_core.tools import tool
+
+    schema_to_be_extracted = {
+        "type": "object",
+        "required": [],
+        "properties": {
+            "title": {"type": "string", "description": "item title"},
+            "due_date": {"type": "string", "description": "item due date"},
+        },
+        "description": "foo",
+    }
+
+    @tool(args_schema=schema_to_be_extracted)
+    def extract_data(extracted_data: dict[str, Any]) -> dict[str, Any]:
+        """Some documentation."""
+        return extracted_data
+
+    assert convert_to_openai_tool(extract_data) == {
+        "function": {
+            "description": "Some documentation.",
+            "name": "extract_data",
+            "parameters": {
+                "properties": {
+                    "due_date": {"description": "item due date", "type": "string"},
+                    "title": {"description": "item title", "type": "string"},
+                },
+                "required": [],
+                "type": "object",
+            },
+        },
+        "type": "function",
+    }

libs/core/tests/unit_tests/utils/test_rm_titles.py

@@ -190,10 +190,44 @@
     "required": ["properties"],
 }
 
+schema5 = {
+    "description": "A list of data.",
+    "items": {
+        "description": "foo",
+        "properties": {
+            "title": {"type": "string", "description": "item title"},
+            "due_date": {"type": "string", "description": "item due date"},
+        },
+        "required": [],
+        "type": "object",
+    },
+    "type": "array",
+}
+
+output5 = {
+    "description": "A list of data.",
+    "items": {
+        "description": "foo",
+        "properties": {
+            "title": {"type": "string", "description": "item title"},
+            "due_date": {"type": "string", "description": "item due date"},
+        },
+        "required": [],
+        "type": "object",
+    },
+    "type": "array",
+}
+
 
 @pytest.mark.parametrize(
     "schema, output",
-    [(schema1, output1), (schema2, output2), (schema3, output3), (schema4, output4)],
+    [
+        (schema1, output1),
+        (schema2, output2),
+        (schema3, output3),
+        (schema4, output4),
+        (schema5, output5),
+    ],
 )
 def test_rm_titles(schema: dict, output: dict) -> None:
     assert _rm_titles(schema) == output