daily-robots/daily/bridge_manager.py

"""Windows 兼容的 Cursor SDK bridge 管理。"""

from __future__ import annotations

import codecs
import json
import logging
import os
import subprocess
import threading
import time
from pathlib import Path
from typing import Any, Mapping

from daily.config import ROOT, env

logger = logging.getLogger(__name__)

READY_LINE_PREFIX = "cursor-sdk-bridge ready "
_bridge_lock = threading.Lock()
_bridge_process: subprocess.Popen[bytes] | None = None


def _cursor_cwd() -> str:
    return env("DAILY_CURSOR_CWD") or env("CURSOR_CWD") or str(ROOT)


def _parse_discovery_line(line: str) -> Mapping[str, Any] | None:
    if not line.startswith(READY_LINE_PREFIX):
        return None
    payload = line[len(READY_LINE_PREFIX) :].strip()
    loaded = json.loads(payload)
    if not isinstance(loaded, dict):
        raise RuntimeError("Bridge discovery payload must be an object")
    return loaded


def _read_discovery_polling(process: subprocess.Popen[bytes], timeout: float = 60) -> Mapping[str, Any]:
    """不用 selectors，避免 Windows 上 WinError 10038。"""
    if process.stderr is None:
        raise RuntimeError("Bridge stderr unavailable")

    fd = process.stderr.fileno()
    was_blocking = os.get_blocking(fd)
    os.set_blocking(fd, False)
    decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
    pending = ""
    stderr_lines: list[str] = []
    deadline = time.monotonic() + timeout

    try:
        while time.monotonic() < deadline:
            try:
                chunk = os.read(fd, 8192)
            except BlockingIOError:
                chunk = b""

            if chunk:
                pending += decoder.decode(chunk)
                while "\n" in pending:
                    line, pending = pending.split("\n", 1)
                    stderr_lines.append(line)
                    discovery = _parse_discovery_line(line)
                    if discovery is not None:
                        return discovery
            else:
                code = process.poll()
                if code is not None:
                    pending += decoder.decode(b"", final=True)
                    if pending.strip():
                        stderr_lines.append(pending.strip())
                    joined = "\n".join(stderr_lines)[-2000:]
                    raise RuntimeError(
                        f"Bridge 启动失败 exit={code}: {joined or '无 stderr 输出'}"
                    )
                time.sleep(0.05)
    finally:
        os.set_blocking(fd, was_blocking)

    raise RuntimeError("等待 Cursor bridge 就绪超时")


def _auth_token_from_discovery(discovery: Mapping[str, Any]) -> str:
    token = str(discovery.get("authToken") or "").strip()
    if token:
        return token
    token_file = discovery.get("authTokenFile")
    if token_file:
        return Path(str(token_file)).read_text(encoding="utf-8").strip()
    raise RuntimeError("Bridge discovery 缺少 auth token")


def warm_cursor_bridge(force: bool = False) -> None:
    """启动 cursor-sdk-bridge 并写入 CURSOR_SDK_BRIDGE_* 环境变量。"""
    global _bridge_process

    with _bridge_lock:
        if (
            not force
            and _bridge_process is not None
            and _bridge_process.poll() is None
            and os.environ.get("CURSOR_SDK_BRIDGE_URL")
            and os.environ.get("CURSOR_SDK_BRIDGE_TOKEN")
        ):
            return

        if _bridge_process is not None and _bridge_process.poll() is None:
            _bridge_process.terminate()
            try:
                _bridge_process.wait(timeout=5)
            except subprocess.TimeoutExpired:
                _bridge_process.kill()

        from cursor_sdk._vendor import resolve_bridge_path

        cwd = _cursor_cwd()
        argv = [resolve_bridge_path(), "--workspace", cwd]
        logger.info("启动 Cursor bridge workspace=%s", cwd)

        process = subprocess.Popen(
            argv,
            stdout=subprocess.DEVNULL,
            stderr=subprocess.PIPE,
        )
        try:
            discovery = _read_discovery_polling(process)
        except (RuntimeError, OSError, ValueError) as exc:
            logger.warning("Cursor bridge discovery 失败，终止子进程：%s", exc)
            process.kill()
            process.wait(timeout=5)
            raise

        url = str(discovery.get("url") or "").strip()
        if not url:
            host = str(discovery.get("host") or "127.0.0.1")
            port = discovery.get("port")
            url = f"http://{host}:{port}"

        token = _auth_token_from_discovery(discovery)
        os.environ["CURSOR_SDK_BRIDGE_URL"] = url
        os.environ["CURSOR_SDK_BRIDGE_TOKEN"] = token
        _bridge_process = process
        logger.info("Cursor bridge 就绪: %s", url)


def shutdown_cursor_bridge() -> None:
    global _bridge_process
    with _bridge_lock:
        if _bridge_process is None:
            return
        if _bridge_process.poll() is None:
            _bridge_process.terminate()
            try:
                _bridge_process.wait(timeout=5)
            except subprocess.TimeoutExpired:
                _bridge_process.kill()
        _bridge_process = None