core: default_tool_parser crashes on malformed function dicts, silently drops all valid tool calls

[Mustafa-Shoukat1]

Checked other resources

• I searched existing issues and this has not been reported before

Example Code

from langchain_core.messages.tool import default_tool_parser, default_tool_chunk_parser

# Case 1: function value is None
raw = [{"function": None, "id": "call_1"}]
default_tool_parser(raw)  # TypeError

# Case 2: function dict missing keys
raw = [{"function": {}, "id": "call_1"}]
default_tool_parser(raw)  # KeyError: name

# Case 3: One malformed call drops ALL valid calls
raw = [
    {"function": None, "id": "bad"},
    {"function": {"name": "good_tool", "arguments": "{\"a\": 1}"}, "id": "good"},
]
default_tool_parser(raw)  # Crashes on first item, good call never parsed

Error Message and Stack Trace

TypeError: NoneType object is not subscriptable
KeyError: name

Description

default_tool_parser and default_tool_chunk_parser in langchain_core/messages/tool.py crash with unhandled KeyError/TypeError when a raw tool call has a malformed function field.

Root cause: Both functions check if "function" not in raw_tool_call but do not validate that the value is actually a dict with the expected keys. They use direct indexing (raw_tool_call["function"]["name"]) instead of .get().

Impact: When ANY single tool call in a list is malformed, the exception kills the entire parsing loop. The caller in ai.py catches Exception broadly, so ALL tool calls (including valid ones) are silently dropped. This is problematic because:

1. The function is documented as Best-effort parsing but is actually all-or-nothing
2. default_tool_parser already has an invalid_tool_calls return list for exactly this purpose, but malformed function dicts never reach it

Affected functions:

• default_tool_parser (line 361): function_name = raw_tool_call["function"]["name"]
• default_tool_chunk_parser (lines 405-406): tool_call["function"]["arguments"] and tool_call["function"]["name"]

System Info

langchain-core latest (main branch as of 2025-07-12)

[Mustafa-Shoukat1]

I would like to work on this issue. I have already identified the root cause and prepared a fix with 16 unit tests (PR #36680 was auto-closed pending assignment). Happy to resubmit once assigned.