fix(core): raise ValueError when explicit tool_outputs length mismatches tool_calls in tool_example_to_messages

libs/core/langchain_core/utils/function_calling.py

@@ -713,6 +713,13 @@ class Person(BaseModel):
     messages.append(
         AIMessage(content="", additional_kwargs={"tool_calls": openai_tool_calls})
     )
+    if tool_outputs is not None and len(tool_outputs) != len(openai_tool_calls):
+        msg = (
+            f"The number of tool_outputs ({len(tool_outputs)}) must match the number "
+            f"of tool_calls ({len(openai_tool_calls)}). Got {len(tool_outputs)} "
+            f"output(s) for {len(openai_tool_calls)} tool call(s)."
+        )
+        raise ValueError(msg)
     tool_outputs = tool_outputs or ["You have correctly called this tool."] * len(
         openai_tool_calls
     )

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

@@ -795,6 +795,24 @@ def test_tool_outputs() -> None:
     assert not response.tool_calls
 
 
+def test_tool_outputs_fewer_than_tool_calls_raises() -> None:
+    with pytest.raises(ValueError, match="must match"):
+        tool_example_to_messages(
+            input="Extract both values",
+            tool_calls=[FakeCall(data="a"), FakeCall(data="b")],
+            tool_outputs=["only one output"],
+        )
+
+
+def test_tool_outputs_more_than_tool_calls_raises() -> None:
+    with pytest.raises(ValueError, match="must match"):
+        tool_example_to_messages(
+            input="Extract one value",
+            tool_calls=[FakeCall(data="a")],
+            tool_outputs=["output1", "extra output"],
+        )
+
+
 @pytest.mark.parametrize(
     "typed_dict",
     [ExtensionsTypedDict, TypingTypedDict],