|
| 1 | +"""Regression: _handle_chat single-shot dict path used `payload = result` and |
| 2 | +then `setdefault`'d defaults onto the handler-supplied dict, mutating a |
| 3 | +caller's template / cached response (silently adding `text: ""` and |
| 4 | +`finish_reason: "stop"`). A handler returning a class-attribute or a memoised |
| 5 | +response dict therefore saw the SAME dict grow extra keys across calls. |
| 6 | +
|
| 7 | +Fix copies first (`payload = dict(result)`); these tests pin that behavior. |
| 8 | +""" |
| 9 | + |
| 10 | +import json |
| 11 | + |
| 12 | +import pytest |
| 13 | + |
| 14 | +from zhub.client import _handle_chat |
| 15 | +from zhub.protocol import Envelope |
| 16 | + |
| 17 | + |
| 18 | +class _FakeWS: |
| 19 | + def __init__(self) -> None: |
| 20 | + self.sent: list[dict] = [] |
| 21 | + |
| 22 | + async def send(self, text: str) -> None: |
| 23 | + self.sent.append(json.loads(text)) |
| 24 | + |
| 25 | + |
| 26 | +class _Pub: |
| 27 | + def __init__(self, handler) -> None: |
| 28 | + self.chat_handler = handler |
| 29 | + |
| 30 | + |
| 31 | +@pytest.mark.asyncio |
| 32 | +async def test_dict_handler_result_is_not_mutated_with_defaults(): |
| 33 | + template = {"tool_calls": [{"id": "c1", "function": {"name": "lookup"}}]} |
| 34 | + |
| 35 | + def handler(messages, options): |
| 36 | + return template |
| 37 | + |
| 38 | + ws = _FakeWS() |
| 39 | + env = Envelope(type="chat-request", payload={"messages": []}) |
| 40 | + await _handle_chat(_Pub(handler), ws, env) |
| 41 | + |
| 42 | + # Wire payload still carries the defaults the hub expects. |
| 43 | + payload = ws.sent[0]["payload"] |
| 44 | + assert payload["text"] == "" |
| 45 | + assert payload["finish_reason"] == "stop" |
| 46 | + assert payload["tool_calls"] == [{"id": "c1", "function": {"name": "lookup"}}] |
| 47 | + |
| 48 | + # But the caller's template is untouched — same keys it walked in with. |
| 49 | + assert set(template.keys()) == {"tool_calls"} |
| 50 | + |
| 51 | + |
| 52 | +@pytest.mark.asyncio |
| 53 | +async def test_dict_handler_text_present_no_default_inserted(): |
| 54 | + template = {"text": "hi", "usage": {"prompt_tokens": 7}} |
| 55 | + |
| 56 | + def handler(messages, options): |
| 57 | + return template |
| 58 | + |
| 59 | + ws = _FakeWS() |
| 60 | + env = Envelope(type="chat-request", payload={"messages": []}) |
| 61 | + await _handle_chat(_Pub(handler), ws, env) |
| 62 | + |
| 63 | + payload = ws.sent[0]["payload"] |
| 64 | + assert payload["text"] == "hi" |
| 65 | + assert payload["finish_reason"] == "stop" |
| 66 | + assert payload["usage"] == {"prompt_tokens": 7} |
| 67 | + |
| 68 | + # Caller's template unchanged across the call. |
| 69 | + assert set(template.keys()) == {"text", "usage"} |
| 70 | + |
| 71 | + |
| 72 | +@pytest.mark.asyncio |
| 73 | +async def test_dict_handler_reused_across_calls_keeps_clean_keys(): |
| 74 | + """Two calls reusing the same dict — second call must not see the first |
| 75 | + call's defaults baked into the template.""" |
| 76 | + template = {"tool_calls": []} |
| 77 | + |
| 78 | + def handler(messages, options): |
| 79 | + return template |
| 80 | + |
| 81 | + pub = _Pub(handler) |
| 82 | + ws1 = _FakeWS() |
| 83 | + ws2 = _FakeWS() |
| 84 | + await _handle_chat(pub, ws1, Envelope(type="chat-request", payload={"messages": []})) |
| 85 | + snapshot_after_first = set(template.keys()) |
| 86 | + await _handle_chat(pub, ws2, Envelope(type="chat-request", payload={"messages": []})) |
| 87 | + |
| 88 | + assert snapshot_after_first == {"tool_calls"} |
| 89 | + assert set(template.keys()) == {"tool_calls"} |
| 90 | + # Both wire payloads still carry the defaults. |
| 91 | + assert ws1.sent[0]["payload"]["text"] == "" |
| 92 | + assert ws2.sent[0]["payload"]["text"] == "" |
0 commit comments