mirror of
https://github.com/andrewyng/openworker.git
synced 2026-11-20 20:37:44 +00:00
time.monotonic() on Windows is GetTickCount64, which ticks every 15.625 ms, so the transcript's timing sidecar (model_ms, tool_ms, waited_ms) read 0 or 15.6 for anything short. perf_counter() is monotonic too and ticks in fractions of a microsecond on every OS. The new test fails on Windows without the change.
63 lines
1.9 KiB
Python
63 lines
1.9 KiB
Python
"""OPE-245: the `timing` sidecar measures with a millisecond clock on every OS.
|
|
|
|
On Windows `time.monotonic()` ticks every 15.625 ms (GetTickCount64), so a tool that took
|
|
4 ms was saved as `tool_ms` 0 or 15.6. `time.perf_counter()` is monotonic too and ticks in
|
|
fractions of a microsecond everywhere."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import time
|
|
|
|
from coworker.engine import TurnEngine
|
|
from coworker.permissions import PermissionEngine
|
|
from coworker.providers import AssistantTurn, ModelCapabilities, ProviderClient, ToolCall
|
|
from coworker.tools import ToolRegistry
|
|
|
|
|
|
class _Provider(ProviderClient):
|
|
def __init__(self, turns):
|
|
self._turns = list(turns)
|
|
|
|
def complete(self, *, model, messages, tools=None, **settings):
|
|
return self._turns.pop(0)
|
|
|
|
def capabilities(self, model):
|
|
return ModelCapabilities()
|
|
|
|
|
|
class _ReadOnly:
|
|
requires_approval = False
|
|
risk_level = "low"
|
|
|
|
|
|
def four_ms() -> dict:
|
|
"""Busy for about 4 ms."""
|
|
end = time.perf_counter() + 0.004
|
|
while time.perf_counter() < end:
|
|
pass
|
|
return {"ok": True}
|
|
|
|
|
|
def test_tool_ms_has_millisecond_resolution(tmp_path):
|
|
registry = ToolRegistry()
|
|
registry.register(four_ms, metadata=_ReadOnly())
|
|
engine = TurnEngine(
|
|
provider=_Provider([
|
|
AssistantTurn(tool_calls=[ToolCall(id="c1", name="four_ms", arguments={})],
|
|
finish_reason="tool_calls"),
|
|
AssistantTurn(text="done", finish_reason="stop"),
|
|
]),
|
|
registry=registry,
|
|
permissions=PermissionEngine(workspace_root=tmp_path),
|
|
model="gpt-5.5",
|
|
)
|
|
|
|
async def go():
|
|
return [ev async for ev in engine.run("go")]
|
|
|
|
asyncio.run(go())
|
|
tool = next(m for m in engine.messages if m.get("role") == "tool")
|
|
# 15.6 ms ticks would give 0 or ~15.6 here.
|
|
assert 3.5 <= tool["timing"]["tool_ms"] < 15
|