muse-agent-example/tests/集成/test_作者审校入口.py
zizi cf0f4fb985 W20–W24:保存知识方法、迁移框架、审校修订与案例行为的集成成果
接续 88dd570,保存 W20–W24 已实现的共享接口、业务入口、迁移、工作台、测试与文档。
W20/W22/W23 保持 in_progress,W21/W24 保持 verified;此提交不宣称方法或规则正式启用、多轮返修、真实角色评测完成。

W25 新增实验、标定、逐调用交付与角色执行及其迁移/测试/索引留在实施工作树,原有私人和旧实现保留项不纳入。

验证:离线 571、前端 33 通过;PG 469 项通过、2 项浏览器未启用,2 项误带入的 W25 用例已移出本提交;最终任务与交付边界 37 项通过。make 检查、最终类型、84 项资源及 diff 检查通过。未重跑浏览器或 Pi 宿主,不以合成调用认证外部模型效果。
独立整体审查四维通过;证据保存在 R2-20260909/提交W20-W24。
2026-09-13 16:24:42 +08:00

240 lines
9.5 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""审校、规则与声音的真实作者浏览器旅程,使用隔离PG和合成模型。"""
import json
from dataclasses import replace
import pytest
import test_完整文学审校 as 文学测试
from muse.任务运行.接口 import 任务状态
pytestmark = pytest.mark.数据库
文学环境 = 文学测试.文学环境
生成环境 = 文学测试.生成环境
@pytest.mark.浏览器
@pytest.mark.timeout(240)
def test_作者审校与声音浏览器往返__229004(文学环境, tmp_path):
import os
import secrets
import socket
import subprocess
from contextlib import asynccontextmanager
from pathlib import Path
from threading import Event, Thread
import uvicorn
from muse.启动 import 装配审校修订
from muse.接入.http.应用 import 创建应用
from muse.配置 import 服务配置
if os.environ.get("MUSE_RUN_BROWSER") != "1":
pytest.skip("浏览器驱动需显式启用;不将未执行记为通过")
环境 = 文学环境
task_id, pieces = 文学测试.发起(环境)
环境["剧本"].extend({"类型": "文本", "文本": 文学测试.片结果(p, 有声音=False)} for p in pieces)
assert 文学测试.跑(环境, task_id).状态 == 任务状态.已完成
from muse.正文写作.接口 import 文本节点, 正文草稿, 段落
环境["正文"].保存人工(
环境["作者"],
"review-patch-seed",
"ch-4",
0,
正文草稿((段落("patch-p1", (文本节点("值得注意的是,雨已经停了。"),)),)),
)
口令 = tmp_path / "browser-password"
口令.write_text(secrets.token_hex(24))
口令.chmod(0o600)
sock = socket.socket()
sock.bind(("127.0.0.1", 0))
port = sock.getsockname()[1]
base = f"http://127.0.0.1:{port}"
配置 = replace(
环境["装配"].配置,
HTTP=服务配置(
str(口令),
作者ID="gen-author",
端口=port,
公开地址=base,
允许来源=(base,),
),
)
app = 创建应用(配置)
原生命周期 = app.router.lifespan_context
就绪 = Event()
@asynccontextmanager
async def 生命周期(app_):
async with 原生命周期(app_):
app_.state.装配 = 环境["装配"]
就绪.set()
yield
app.router.lifespan_context = 生命周期
server = uvicorn.Server(
uvicorn.Config(app, log_level="warning", access_log=False, timeout_graceful_shutdown=3)
)
thread = Thread(target=server.run, kwargs={"sockets": [sock]}, daemon=True)
root = Path(__file__).resolve().parents[2]
装配审校修订(环境["装配"], 文学测试.写手测试.合成计价(), root / "配置/流程模板/审校修订.yaml")
worker_stop = Event()
worker_errors = []
def 运行合成修订():
from datetime import UTC, datetime, timedelta
from decimal import Decimal
from muse.任务运行.接口 import 任务预算计划, 角色预算, 预算管理, 额度策略
run = 环境["装配"].任务运行
done = set()
try:
while not worker_stop.wait(0.1):
for task in run.列出任务(环境["作者"].作者, 作品ID="gen-work"):
if task.流程.流程ID != "审校修订" or task.任务ID in done:
continue
done.add(task.任务ID)
budget = 预算管理(环境["库"], "synthetic")
budget.登记策略(额度策略("synthetic", "1", Decimal("20"), 40))
budget.登记任务预算(
task.任务ID,
任务预算计划(
Decimal("4"),
(角色预算("writer", 2, 2, Decimal("1")),),
"synthetic-browser-approval",
datetime.now(UTC) + timedelta(minutes=5),
),
)
auth = task.冻结输入["输入"]["修订授权"]
if task.冻结输入["输入"].get("source_kind") == "selection":
from pydantic import TypeAdapter
from muse.正文写作.接口 import 选段写作授权
selected = TypeAdapter(选段写作授权).validate_python(auth)
edits = [
{
"selection_id": s.selection_id,
"action": "replace",
"replacement": "细雨已经停了。",
"reason": "合成通用润色",
}
for s in selected.selections
]
else:
edits = [
{
"finding_id": auth["approved_findings"][0],
"action": "replace",
"replacement": "",
"reason": "合成删除元话语",
}
]
环境["剧本"].append(
{
"类型": "文本",
"文本": {"edits": edits},
}
)
while run.读取任务(task.任务ID).状态 != 任务状态.已完成:
claim = run.领取步骤(
"browser-revision", ["revision.freeze", "revision.generate"]
)
assert claim is not None
run.执行一步(claim)
except Exception as exc:
worker_errors.append(exc)
worker = Thread(target=运行合成修订, daemon=True)
输出 = Path(os.environ.get("MUSE_BROWSER_ARTIFACTS", str(tmp_path / "browser"))).resolve()
输出.mkdir(parents=True, exist_ok=True)
env = {
**os.environ,
"MUSE_WORKBENCH_URL": base,
"MUSE_AUTHOR_PASSWORD_FILE": str(口令),
"MUSE_REVIEW_TASK_ID": task_id,
}
chrome = Path("/Applications/Google Chrome.app/Contents/MacOS/Google Chrome")
if chrome.exists():
env.setdefault("MUSE_BROWSER_EXECUTABLE", str(chrome))
thread.start()
worker.start()
try:
assert 就绪.wait(10), "隔离 HTTP 服务未就绪"
报告路径 = 输出.parent / "作者审校-浏览器执行.json"
with 报告路径.open("w") as 报告:
过程 = subprocess.run(
[
"pnpm",
"exec",
"playwright",
"test",
"作者审校.spec.ts",
"--reporter=json",
"--output",
str(输出),
],
cwd=root / "web",
env=env,
stdout=报告,
stderr=subprocess.STDOUT,
timeout=90,
)
assert 过程.returncode == 0, f"浏览器失败,见 {报告路径}"
文件 = list(输出.rglob("审校回执.json"))
assert len(文件) == 1
新任务 = json.loads(文件[0].read_text())["created_task_id"]
assert 新任务 != task_id
with 环境["库"].连接(只读=True) as 连:
assert (
连.execute("SELECT COUNT(*) FROM muse_task WHERE task_id=%s", (新任务,)).fetchone()[
0
]
== 1
)
assert (
连.execute(
"SELECT COUNT(*) FROM muse_method_usage WHERE task_id=%s", (新任务,)
).fetchone()[0]
== 0
)
声音 = 环境["装配"].要求审校().当前声音(环境["作者"], "gen-work")
assert 声音["revision"] == 1
assert 声音["narrator"]["句式画像"] == "短句推进,保留动作间停顿"
assert 环境["正文"].读取正文(环境["审校作者"], "ch-2") == 环境["审校正文"]
修订候选 = 环境["正文"].列出候选(环境["作者"], "ch-4")
assert not worker_errors
assert len(修订候选) == 4 and {c["origin"] for c in 修订候选} == {"author", "model"}
assert all(c["decision"] == "undecided" for c in 修订候选)
assert sorted(c["visible_text"] for c in 修订候选) == sorted(
[
"雨已经停了。",
"雨已经停了。",
"值得注意的是,雨已经停了。屋檐开始滴水。",
"值得注意的是,细雨已经停了。",
]
)
assert 环境["正文"].读取正文(环境["作者"], "ch-4", 分支="雨后试写")["revision"] == 2
assert (
环境["正文"].读取正文(环境["作者"], "ch-4")["visible_text"]
== "值得注意的是,雨已经停了。"
)
if os.environ.get("MUSE_BROWSER_INSPECT") == "1":
(输出 / "检查入口.json").write_text(json.dumps({"origin": base, "task_id": task_id}))
import time
deadline = time.monotonic() + 150
while not (输出 / "检查结束").exists() and time.monotonic() < deadline:
Event().wait(0.25)
finally:
worker_stop.set()
worker.join(10)
assert not worker.is_alive(), "测试修订worker未停止"
server.should_exit = True
thread.join(10)
sock.close()
assert not thread.is_alive(), "测试 HTTP 服务未关闭"