Files
aiturk-hermes-ide/apps/desktop/assets/codex-bridge/test_model_switch.py
T
yilsemandClaude Opus 5.5 adf73eeedd fix(codex-link): summarize with a subscription model, reject tiny summaries, describe added models
Model switch lost the conversation: on 10 Oct a 1 MB thread was compacted by the
newly selected small model into 2201 characters. The compaction summary for a bridge
model is now written by a subscription model (ozet_kopru.ozet_modeli, default
gpt-6.1-sol); if that request is refused before it starts, the target model is used
as before. A summary far shorter than the conversation warrants is not accepted:
Codex gets response.failed and keeps the history.

The line under each added model no longer repeats its name. Teams show their
members, fallback chains their order, single models their provider and the traits
readable from the name; the text fits one phone row (30 characters).

Also normalizes tool call ids in the streamed response so Codex persists ids the
subscription backend accepts (fc_ -> ctc_), and bumps the bridge protocol to 6.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 15:31:53 +03:00

90 lines
5.3 KiB
Python

"""Provider-switch history regression tests, without startup or network access."""
import ast
import copy
import json
from pathlib import Path
import unittest
tree = ast.parse(Path(__file__).with_name("bridge.py").read_text(encoding="utf-8"))
names = {"flatten_request", "fix_tool_ids", "drop_foreign_items", "SseRewriter"}
nodes = [n for n in tree.body if isinstance(n, (ast.FunctionDef, ast.ClassDef)) and n.name in names]
env = {"json": json, "log": lambda message: None}
for node in tree.body:
if isinstance(node, ast.Assign) and any(isinstance(t, ast.Name) and t.id in
{"ID_PREFIX", "FOREIGN_ID_MARK", "REFERENCE_TYPES", "INLINE_TYPES"} for t in node.targets):
nodes.insert(0, node)
exec(compile(ast.Module(body=nodes, type_ignores=[]), "bridge.py", "exec"), env)
class ModelSwitch(unittest.TestCase):
def test_external_to_chatgpt_preserves_messages_and_paired_tool_calls(self):
req = {"input": [
{"type": "reasoning", "id": "rs_native", "encrypted_content": "native-reasoning"},
{"type": "reasoning", "id": "rs_chatcmpl-foreign"},
{"type": "item_reference", "id": "chatcmpl-foreign"},
{"type": "message", "id": "msg_chatcmpl-foreign", "role": "assistant", "content": "remember-4827"},
{"type": "custom_tool_call", "id": "fc_chatcmpl-call", "call_id": "call_4827", "name": "probe", "input": "hello"},
{"type": "custom_tool_call_output", "id": "fc_chatcmpl-output", "call_id": "call_4827", "output": "result-4827"},
]}
self.assertEqual(env["fix_tool_ids"](req), 2)
self.assertEqual(env["drop_foreign_items"](req), 2)
self.assertEqual(req["input"][0]["encrypted_content"], "native-reasoning")
self.assertEqual(req["input"][1]["content"], "remember-4827")
self.assertEqual(req["input"][2]["id"], "ctc_chatcmpl-call")
self.assertEqual(req["input"][3]["id"], "ctco_chatcmpl-output")
self.assertEqual(req["input"][2]["call_id"], req["input"][3]["call_id"])
self.assertEqual(req["input"][3]["output"], "result-4827")
def test_chatgpt_to_external_preserves_custom_tool_history(self):
original = {"access_programs": {"cyber": "standard"}, "input": [
{"type": "custom_tool_call", "id": "ctc_4827", "call_id": "call_4827", "name": "probe", "input": "remember-4827"},
{"type": "custom_tool_call_output", "id": "ctco_4827", "call_id": "call_4827", "output": "result-4827"},
], "tools": [{"type": "namespace", "name": "functions", "tools": [{"type": "custom", "name": "probe"}]}]}
req = copy.deepcopy(original)
self.assertEqual(env["flatten_request"](req), {"probe"})
env["fix_tool_ids"](req)
self.assertEqual(json.loads(req["input"][0]["arguments"]), {"input": "remember-4827"})
self.assertEqual(req["input"][1]["output"], "result-4827")
self.assertEqual(req["input"][0]["call_id"], req["input"][1]["call_id"])
self.assertEqual(req["input"][0]["type"], "function_call")
self.assertEqual(req["input"][1]["type"], "function_call_output")
self.assertIn("access_programs", original)
def test_streamed_external_custom_tool_can_return_to_codex(self):
event = {"type": "response.output_item.done", "item": {
"type": "function_call", "id": "fc_chatcmpl-probe", "call_id": "call_4827",
"name": "probe", "arguments": json.dumps({"input": "remember-4827"})}}
raw = ("event: response.output_item.done\ndata: " + json.dumps(event) + "\n\n").encode()
adapter = env["SseRewriter"]({"probe"})
out = adapter.feed(raw[:25]) + adapter.feed(raw[25:]) + adapter.flush()
rewritten = json.loads(next(line[6:] for line in out.decode().splitlines() if line.startswith("data: ")))
self.assertEqual(rewritten["item"]["type"], "custom_tool_call")
self.assertEqual(rewritten["item"]["input"], "remember-4827")
self.assertEqual(rewritten["item"]["call_id"], "call_4827")
self.assertEqual(rewritten["item"]["id"], "ctc_chatcmpl-probe")
def test_custom_stream_ids_are_valid_before_history_is_saved(self):
# Real failure after disabling the bridge and returning to OpenAI:
# input[32].id fc_call_... was saved as a custom_tool_call by Codex.
item = {"type": "function_call", "id": "fc_call_3f427ecc539d4c779c1252077701216f",
"call_id": "call_original", "name": "probe",
"arguments": json.dumps({"input": "keep this exact tool input"})}
adapter = env["SseRewriter"]({"probe"})
converted = []
for kind in ("response.output_item.added", "response.output_item.done"):
converted.append(adapter._fix({"type": kind, "item": dict(item)})["item"])
final = adapter._fix({"type": "response.completed", "response": {"output": [dict(item)]}})
converted.append(final["response"]["output"][0])
for result in converted:
self.assertEqual(result["type"], "custom_tool_call")
self.assertEqual(result["id"], "ctc_call_3f427ecc539d4c779c1252077701216f")
self.assertEqual(result["call_id"], "call_original")
self.assertEqual(result["input"], "keep this exact tool input")
history = {"input": converted}
self.assertEqual(env["fix_tool_ids"](history), 0)
if __name__ == "__main__":
unittest.main()