d19963782b
`_response_messages_turn_start_index` located the current turn by matching `result["messages"]` against `conversation_history + [user]` row by row. The loop repairs host-fed history before its first call (consecutive assistant or user rows merge, orphan tool results drop) and compaction rewrites it, so the transcript legitimately stops sharing a prefix with the client history; the match then returned 0 and the WHOLE transcript was treated as the current turn: earlier turns' function_call / function_call_output items were replayed as this turn's `output` / `run.completed` turn_messages, and the stored `previous_response_id` chain grew by another copy of the history every turn. Anchor on the loop's canonical current-turn user index instead (`agent.turn_context.reanchor_current_turn_user_idx`: last user row that says this turn's text, else the last user-originated row), and keep the semantic prefix match only for suffix-only results without a user row (mocked/legacy paths). The boundary logic moves to the topical sibling `api_server_turn_boundary.py`; the routes mixin delegates. The dedupe test that asserted "divergent transcript => append history + transcript" encoded the duplication itself; it now asserts the fallback for the case it actually exists for (a suffix-only result). Fixes #89891 Co-authored-by: heyf <tonyheyifan@gmail.com>
46 lines
2.1 KiB
Python
46 lines
2.1 KiB
Python
"""Stored /v1/responses transcripts must not duplicate history across chained turns.
|
|
|
|
The agent's ``result["messages"]`` copies of the prior history carry ``timestamp`` /
|
|
``_db_persisted`` / ``reasoning`` / ``finish_reason`` that the API layer's bare
|
|
``{"role", "content"}`` dicts never have, so a whole-dict prefix check failed every turn and
|
|
the stored history grew 3 -> 8 -> 17 instead of 2 -> 4 -> 6 (#95137, #101644, #82513).
|
|
"""
|
|
|
|
from agent.agent_runtime_helpers import repair_message_sequence
|
|
from agent.message_metadata import append_message
|
|
from gateway.platforms.api_server import APIServerAdapter
|
|
|
|
|
|
class _Agent:
|
|
api_mode = "chat_completions"
|
|
session_id = "s"
|
|
|
|
|
|
def _agent_turn(history, user_message, answer):
|
|
"""Real producers: turn_context appends the user row via append_message (stamps timestamp)."""
|
|
messages = list(history)
|
|
append_message(messages, {"role": "user", "content": user_message})
|
|
repair_message_sequence(_Agent(), messages)
|
|
append_message(messages, {"role": "assistant", "content": answer, "reasoning": "r", "finish_reason": "stop"})
|
|
return {"messages": messages}
|
|
|
|
|
|
def test_chained_responses_turns_store_each_message_once():
|
|
build = APIServerAdapter._build_response_conversation_history
|
|
history, counts = [], []
|
|
for turn, prompt in enumerate(["Remember TOKEN-ABC.", "What token?", "Third"]):
|
|
answer = f"answer {turn}"
|
|
history = build(history, prompt, _agent_turn(history, prompt, answer), answer)
|
|
counts.append(len(history))
|
|
assert counts == [2, 4, 6]
|
|
assert [m["role"] for m in history] == ["user", "assistant"] * 3
|
|
|
|
|
|
def test_suffix_only_result_still_appends_to_history():
|
|
build = APIServerAdapter._build_response_conversation_history
|
|
prior = [{"role": "user", "content": "a"}, {"role": "assistant", "content": "b"}]
|
|
# Mocked/legacy paths return only this turn's rows (no user row): appended after the input.
|
|
result = {"messages": [{"role": "assistant", "content": "d"}]}
|
|
stored = build(prior, "c", result, "d")
|
|
assert [m["content"] for m in stored] == ["a", "b", "c", "d"]
|