AttributeError: 'NoneType' object has no attribute 'get' when using with_structured_output(dict_schema, method='function_calling') with a titleless dict schema

[awg66]

Description

with_structured_output(schema, method="function_calling") raises AttributeError: 'NoneType' object has no attribute 'get' inside _prepare_request when schema is a dict that does not have a top-level "title" key.

Minimal Repro

from langchain_google_genai import ChatGoogleGenerativeAI

llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash-001")

# No "title" key — this crashes
schema = {
    "type": "object",
    "properties": {
        "answer": {"type": "string"},
        "score": {"anyOf": [{"type": "number"}, {"type": "null"}]},
    },
    "required": ["answer"],
}

structured = llm.with_structured_output(schema, method="function_calling")
structured.invoke("What weighs more, a pound of bricks or a pound of feathers?")

Error:

File ".../langchain_google_genai/_function_utils.py", line 404, in _format_to_genai_function_declaration
    elif "parameters" in tool and tool["parameters"].get("properties"):
                                  ^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'get'

A dict schema with "title" works fine:

schema = {"title": "MyOutput", "type": "object", "properties": {...}, "required": [...]}

Root Cause

Two problems compound in bind_tools + _prepare_request:

Problem 1 — bind_tools fallback produces a broken FunctionDeclaration

# ChatGoogleGenerativeAI.bind_tools (chat_models.py ~line 66)
try:
    formatted_tools = [convert_to_openai_tool(tool) for tool in tools]
except Exception:
    formatted_tools = [tool_to_dict(t) for t in convert_to_genai_function_declarations(tools)]

convert_to_openai_tool raises ValueError: must have a top-level 'title' key for titleless dicts. The except Exception fallback calls convert_to_genai_function_declarations([titleless_schema]).

Inside _format_to_genai_function_declaration, the schema has no "name", "description", or "parameters" keys, so it falls into the else branch, sets function["parameters"] = {}, and _format_dict_to_function_declaration can't find a name → FunctionDeclaration(name="MISSING_NAME", parameters=None).

tool_to_dict serialises this as:

{"function_declarations": [{"name": "MISSING_NAME", "parameters": null}]}

Problem 2 — Missing None guard in _format_to_genai_function_declaration

When _prepare_request calls convert_to_genai_function_declarations a second time on the stored {"function_declarations": [...]} tools, it iterates over the function declarations and calls _format_to_genai_function_declaration({"name": "MISSING_NAME", "parameters": null}).

Line 404 (_function_utils.py):

elif "parameters" in tool and tool["parameters"].get("properties"):

"parameters" in tool is True (the key exists, value is null), but tool["parameters"] is None — calling .get() on it crashes.

Fix

Either problem fixed independently would prevent the crash. The minimal correct fix is adding a None guard:

# _function_utils.py line ~404
elif "parameters" in tool and tool["parameters"] is not None and tool["parameters"].get("properties"):

A more complete fix would also make the fallback path in bind_tools raise a clearer error (or handle the missing-title case) instead of silently producing FunctionDeclaration(name="MISSING_NAME", parameters=None).

Environment

• langchain-google-genai version: checked against current main
• Python 3.12