fix: handle empty or missing tool-call arguments in from_openai_dict_format

[chuenchen309]

Problem

ChatMessage.from_openai_dict_format parses tool-call arguments with:

arguments=json.loads(tc["function"]["arguments"])

This assumes arguments is always present and always a valid JSON string. A
zero-argument tool call from an OpenAI-compatible server (vLLM, llama.cpp,
Ollama, LM Studio, ...) commonly sends an empty string, null, or omits the key:

# empty string  -> json.JSONDecodeError
{"id": "1", "function": {"name": "now", "arguments": ""}}
# missing key   -> KeyError('arguments')
{"id": "1", "function": {"name": "now"}}

_validate_openai_message only checks that a function field exists, so nothing
guards this — and the method's docstring explicitly courts these "shallow
OpenAI-compatible APIs".

Fix

Treat empty/null/missing arguments as an empty dict, leaving valid-JSON parsing
unchanged:

raw_arguments = tc["function"].get("arguments")
arguments = json.loads(raw_arguments) if raw_arguments else {}

Tests

Added two tests in test/dataclasses/test_chat_message.py (empty-string and
missing-key arguments). Both fail before the change (JSONDecodeError /
KeyError) and pass after. All 15 from_openai_dict_format tests pass.

$ python -m pytest test/dataclasses/test_chat_message.py -k from_openai_dict_format
15 passed

ruff check / ruff format --check pass. Added a release note.

---

Disclosure: I used AI assistance (Claude) to find this crash-on-edge-input and
draft the tests. I reviewed the change, ran the suite, and take responsibility
for its correctness.

[vercel]

@chuenchen309 is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[bogdankostic]

Thanks for the PR @chuenchen309. It already looks good in principle, we just need to update the release notes.

[bogdankostic]

This release note doesn't seem to be related to the changes made in this PR.

[bogdankostic]

We need to use double-backticks for in-line code in release notes.

Suggested change

	Fixed `ChatMessage.from_openai_dict_format` crashing on zero-argument tool
	calls. OpenAI-compatible servers (vLLM, llama.cpp, Ollama, LM Studio, ...) may
	send an empty string or omit the `arguments` field for a tool call that takes
	no arguments; this previously raised `json.JSONDecodeError` or `KeyError`.
	Such tool calls are now parsed into a `ToolCall` with empty `arguments`.
	Fixed ``ChatMessage.from_openai_dict_format`` crashing on zero-argument tool
	calls. OpenAI-compatible servers (vLLM, llama.cpp, Ollama, LM Studio, ...) may
	send an empty string or omit the ``arguments`` field for a tool call that takes
	no arguments; this previously raised ``json.JSONDecodeError`` or ``KeyError``.
	Such tool calls are now parsed into a ``ToolCall`` with empty ``arguments``.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
haystack/dataclasses
chat_message.py
Project Total

This report was generated by python-coverage-comment-action

[chuenchen309]

Thanks @bogdankostic — added the release note for the fix and rebased onto latest main, which also dropped a stray release note the branch had picked up from a rebased-over PR. The diff is now just the from_openai_dict_format tool-call fix + its test + release note, and the local suite is green. Ready for another look whenever you have time.