DCU_DesktopController/scripts/guard_probe.py

"""tx 冻结守卫真机验证的共用底座：一条通路 + 一份前置判据。

为什么要有这个文件：两个真机脚本（check_guard_idle_silence.py / check_guard_fires_hw.py）
必须用**同一份**前置条件。前置没满足就跑，等于什么都没测——比如 loop=0 或 pm_hold=1
时判据按设计是静音的，这种"跑绿了其实没测"正是本次加固要根除的缺陷类别，所以
判据只放一个执行点，两个脚本都从这里取。

通路优先级：先试页面代理 :2625/ecat/api/*（用户点发送走的就是这条，代理自己那套
"不给 connect 设 timeout"的约束在 esp32_server/ecat_client 里），失败再直连 native
8090..8095（APP 没起时也能跑门槛）。
"""
import json
import os
import sys
import time
import urllib.error
import urllib.request

PAGE_BASE = os.environ.get("DCU_PAGE_BASE", "http://127.0.0.1:2625")
NATIVE_PORTS = list(range(8090, 8096))
GUARD_KEYS = ("alive_ticks", "armed", "tx_stale_s", "since_send_s", "since_full_s",
              "cooldown_left_s", "tx_count", "tx_base", "motor_send_hits", "fires",
              "last_fire_age_s")
# 注：这里的 timeout 只作用于门槛工具自己的请求。esp32_server 的 ECAT 代理严禁设
# socket timeout（Windows 上会让 connect 走 select 轮询，每请求白加约 10ms），
# 那是另一条路径，见 ecat_client.py 的同名注释。
TIMEOUT = float(os.environ.get("DCU_GUARD_TIMEOUT", "30"))


class Blocked(Exception):
    """前置不满足：调用方必须以退出码 2 结束，绝不折算成 PASS。"""


def _post(url, obj):
    data = json.dumps(obj).encode("utf-8") if obj is not None else b"{}"
    req = urllib.request.Request(url, data=data, method="POST",
                                 headers={"Content-Type": "application/json"})
    with urllib.request.urlopen(req, timeout=TIMEOUT) as r:
        return json.loads(r.read().decode("utf-8", "replace"))


class Link:
    def __init__(self, base, how):
        self.base = base
        self.how = how

    def api(self, path, obj=None):
        return _post(self.base + path, obj if obj is not None else {})

    def health(self):
        return self.api("/api/native/health")

    def state(self):
        return self.api("/api/motor/state")

    def send(self, ch=2, ident=0x00030200, dlc=32, payload=None):
        body = {"frames": [{"ch": ch, "id": ident, "dlc": dlc,
                            "data": payload or "03" * dlc}]}
        t0 = time.perf_counter()
        r = self.api("/api/motor/send", body)
        return r, (time.perf_counter() - t0) * 1000.0


def open_link(verbose=True):
    last = None
    for url in [PAGE_BASE + "/ecat/api"]:
        try:
            lk = Link(url, "页面代理")
            h = lk.health()
            if not h.get("ok"):
                raise ValueError("health.ok 非真")
            if verbose:
                print("通路：%s（%s）" % (url, lk.how))
            return lk
        except Exception as e:
            last = e
    for p in NATIVE_PORTS:
        try:
            lk = Link("http://127.0.0.1:%d" % p, "直连 native")
            h = lk.health()
            if h.get("ok"):
                if verbose:
                    print("通路：http://127.0.0.1:%d（%s；页面代理不可用：%s）" % (p, lk.how, last))
                return lk
        except Exception as e:
            last = e
    raise Blocked("native 连不上（试过 %s 与 %s）：%s"
                  % (PAGE_BASE + "/ecat/api", NATIVE_PORTS, last))


def guard_of(health):
    g = health.get("guard")
    if not isinstance(g, dict):
        raise Blocked("health 里没有 guard 块 ⇒ 跑的是旧 exe（不含可观测面），"
                      "先跑 tools/ecat/build_native.bat 换二进制")
    missing = [k for k in GUARD_KEYS if k not in g]
    if missing:
        raise Blocked("guard 缺字段 %s ⇒ 二进制与脚本不同代" % missing)
    return g


def preflight(lk, want_tx_alive=True):
    """总线必须真的在交换数据、且守卫没被静音，否则跑了也测不到东西。"""
    h = lk.health()
    g = guard_of(h)
    need = {"bus_open": 1, "op": 1, "loop": 1}
    bad = ["%s=%s（需 1）" % (k, h.get(k)) for k, v in need.items() if h.get(k) != v]
    if h.get("dc") != 1 or h.get("dc_cycle_us") != 250:
        bad.append("dc=%s/%sus（需 1/250）" % (h.get("dc"), h.get("dc_cycle_us")))
    if (h.get("wkc") or 0) < 3:
        bad.append("wkc=%s（需 >=3）" % h.get("wkc"))
    if h.get("pm_hold"):
        bad.append("pm_hold=1（休眠豁免期判据按设计静音，测不到）")
    if h.get("recovering"):
        bad.append("recovering=1（正在重拉，此刻的绿灯不算数）")
    t0 = g["alive_ticks"]
    time.sleep(1.0)
    g1 = guard_of(lk.health())
    if g1["alive_ticks"] - t0 < 3:
        bad.append("guard.alive_ticks +%d/1s（需 >=3）⇒ 巡检线程没在跑"
                   % (g1["alive_ticks"] - t0))
    if bad:
        raise Blocked("；".join(bad))
    if want_tx_alive:
        s = lk.state()
        tx = ((s.get("bridge") or {}).get("tx_count"), (s.get("head") or {}).get("global_seq"))
        print("前置 OK：bus/op/loop=1 dc=250us wkc=%s ticks=+%d/1s 当前 tx_count=%s gseq=%s"
              % (h.get("wkc"), g1["alive_ticks"] - t0, tx[0], tx[1]))
    else:
        print("前置 OK（未要求出口活动）")
    return h, g1


def bridge_of(state):
    return state.get("bridge") or {}


def verdict(ok, why=""):
    """末行恒为 VERDICT，包装脚本 grep 它；退出码 0/1/2 是真门槛。"""
    if ok:
        print("VERDICT PASS")
        return 0
    print("VERDICT FAIL " + why)
    return 1


__all__ = ["Link", "Blocked", "open_link", "guard_of", "preflight", "bridge_of",
           "verdict", "GUARD_KEYS", "PAGE_BASE"]


if __name__ == "__main__":
    try:
        lk = open_link()
        preflight(lk)
    except Blocked as e:
        print("BLOCKED %s" % e)
        sys.exit(2)
    sys.exit(0)