GUI v0.2.9: fix chronic-lag root cause (log trim crash on PySide6 6.11) + 5 hardening fixes

- log trim used instance-level enum access (removed in PySide6 6.11), crashed

AGENTS.md

@@ -160,9 +160,40 @@
     （app+window 两级 setWindowIcon，_icon_path() 兼容源码态与 _MEIPASS 打包态）、
     Windows 任务栏独立标识 AppUserModelID "hulk.FilmForgeGUI"。README"无应用
     图标"遗留项关闭。
+  - v0.2.8：**慢性卡顿修复**——`_log()` 手工 cursor 裁剪（`cursor.Start`/
+    `BlockUnderCursor` 实例级枚举访问）在 PySide6 6.11 上必然抛 AttributeError，
+    日志超 500 行（100ms 上报 ≈50s）后每个 100ms tick 异常中止 `_tick` 后半段：
+    应答超时清理失效（pending 永不清 → "停止"点击被幂等去重永久拦截，实拍状态栏
+    "M2S0 应答中,已忽略重复发送"）、RX 队列积压 ratchet（帧处理滞后 54ms→均值
+    1037ms/峰值 2701ms，日志/chip/曲线显示滞后数秒，"停止 5s 后才见命令发出"实为
+    日志显示积压，TX 本身 5~166ms 即出）、日志无限增长、CPU +30%。改原生
+    `setMaximumBlockCount(500)` 环形裁剪；selftest 新增"500 行封顶/封顶后可追加/
+    最新行在末尾"3 条回归。离屏 A/B 实测（GUI/_diag_latency_buggy.py，假固件
+    100ms 上报+定时点击）：buggy 组 5.5 分钟异常 2241 次、修复组全程漂移 ≤15ms/
+    零异常/零积压。v0.2.7 exe（2026-10-10 打包）内嵌同版本 PySide6 6.11.2，
+    同带此 bug，需重新打包生效。
+  - v0.2.9：**整体审查修复 5 项**——①CSV 写失败（磁盘满/记录盘被拔）自动停记录
+    并提示，不再沿 _tick 帧处理路径抛 OSError 中止主循环（与 v0.2.8 日志裁剪同款
+    失效管道）；②故障标记 faults 封顶 1000 条（曾是唯一无上限缓冲：NTC 抖动时
+    10 条/s 无限增长，且 _draw_faults 每 paint 全量扫描）；③串口写超时单独捕获
+    SerialTimeoutException——只丢该条命令、链路保留，不再按 OSError 当"设备拔出"
+    关端口断链（适配器停滞曾迫使手动重连）；④日志自动滚动仅在贴底时跟随（上报
+    运行时可上翻历史，不再被拽回底部）；⑤冻结视图切片右缘同样二分（暂停时长不再
+    增加 render 成本）。selftest 新增 7 项回归（封顶/滚动跟随/CSV 失败自停/冻结
+    切片）共 59 项全过。整体审查报告（2026-10-12）其余 P3 遗留：selftest 未覆盖
+    _tick 主循环（超时清理/丢失判定）、"全部"窗口全量切片有 MAX_SAMPLES 兜底、
+    _last_set 单槽位罕见误补偿（8s 自愈）、build_exe remove_stale 无提示强杀
+    运行实例、M888 后固定 2s 重查可能早于固件就绪、_danger_reset 把去重忽略记成
+    "已下发"。协议建议的 300ms×3 自动重发维持不做（设计决策，用户拍板）。
   - selftest 全过（累计新增 15 项回归用例，含窗口图标非空检查）；exe 已本地
     打包冒烟，Gitea 发布待用户确认后执行。
   遗留见其 README"已知事项"：USB-RS485 适配器更换。
+- [x] **GUI 慢性卡顿修复（根因已定位并修复 v0.2.8，待真机确认）**：
+      根因/机理/实测数据见上方 v0.2.8 条目（`cursor.Start` 实例级枚举访问在
+      PySide6 6.11 必抛 → tick 异常中止 → 应答超时清理失效 + RX 积压 ratchet +
+      日志无限增长）。复现/验证脚本：`GUI/_diag_latency.py`（修复基线）/
+      `GUI/_diag_latency_buggy.py`（带 bug 对照，假固件 100ms 上报+定时点击）。
+      遗留：待真机长时间运行确认 + 重新打包 exe。
 - [ ] **LED 引脚确认**：FilmForge 自制板 LED 引脚未定（当前沿用模板默认 PA15，
       仅为构建占位），按原理图修改。
 - [ ] **delay_10us 校准**：M4@168MHz 系数为估算值（/300000），需实测修正。

GUI/README.md

@@ -75,6 +75,19 @@ python publish_gitea.py release/FilmForge-GUI-v<版本>-win64.zip --replace --in
   才返回（Windows 串口"字节间隔超时"150ms 因帧间隔更短永不触发，100ms 上报时
   ≈2.5s 一批），表现为曲线/命令全部 3 秒一卡、一股脑一大坨。已改为
   `read(in_waiting or 1)`，延迟回到 ≤150ms 量级（v0.2.2，真机实测确认）。
+- **日志裁剪走原生 `setMaximumBlockCount(500)`**：手工 `textCursor` 裁剪用了
+  `cursor.Start`/`cursor.BlockUnderCursor` 实例级枚举访问，PySide6 6.11 已移除
+  （类上仍在），必然抛 AttributeError——日志超 500 行（100ms 上报 ≈50s）后每个
+  100ms tick 在 `_log` 处异常中止：`_tick` 后半段（应答超时清理/连接状态机/统计
+  节流）永不执行（pending 永不清 → "停止"点击被幂等去重永久拦截）、RX 队列积压
+  ratchet（帧处理滞后涨到 1~2.7s，日志/chip/曲线显示全面滞后）、日志无限增长。
+  v0.2.8 改原生环形裁剪修复（离屏 A/B 实测：修复后 6 分钟漂移 ≤15ms、零异常、
+  零积压；selftest 新增 3 条回归用例）。
+- **日志自动滚动仅在贴底时跟随**：上报运行时可上翻查看历史（每行强制滚底会抢
+  滚动条）；CSV 写失败（磁盘满/记录盘被拔）自动停止记录并提示——`_csv_write` 在
+  `_tick` 帧处理路径上，未捕获异常会中止主循环（v0.2.8 日志裁剪同款教训）；故障
+  标记封顶 1000 条；冻结视图切片右缘二分（暂停时长不增加渲染成本）；串口写超时
+  只丢该条命令、链路保留（不按设备拔出断链）。均 v0.2.9。
 - Fluent 控件自定义样式必须走 `setCustomStyleSheet`（覆盖 hover/pressed/disabled），
   直接 `setStyleSheet` 会在切主题时被库重刷（DLPilot 同款坑）；标题字号走 QFont。
 - 图表刷新双节流：数据到达即重绘（≤10Hz）+ 0.5s 时间轴推进；“全部”窗口超 4000 点

GUI/_diag_latency.py

@@ -0,0 +1,232 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""诊断 v2: 干净基线 — 旁路 app_fluent._log 的坏裁剪(cursor.Start 在 PySide6
+6.11.2 上抛 AttributeError), 用 Qt 原生 setMaximumBlockCount 替代, 其余不动源码。
+目的: 验证 100ms 持续上报数分钟下, 修复裁剪后 点击->TX 是否仍瞬时、事件循环
+是否仍流畅(对照 buggy 版的 tick 中止 + 队列积压)。"""
+import ctypes
+import gc
+import os
+import sys
+import threading
+import time
+
+os.environ.setdefault("QT_QPA_PLATFORM", "offscreen")
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import filmforge_rs485 as ff
+
+
+class FakeSerial:
+    """模拟 CH34x + 固件: M2S<ms> 控制周期温度帧, M2S0 回 ok 并停流。"""
+
+    def __init__(self, *a, **kw):
+        self._buf = bytearray()
+        self._lock = threading.Lock()
+        self._alive = True
+        self.write_log = []              # [(time, cmd_bytes)]
+        self.report_ms = 0
+        self._next_tx = 0.0
+        self._th = threading.Thread(target=self._fw, daemon=True)
+        self._th.start()
+        FakeSerial.last = self
+
+    def _fw(self):
+        while self._alive:
+            now = time.time()
+            if self.report_ms and now >= self._next_tx:
+                self._push(self._frame("34.6"))
+                self._next_tx = now + self.report_ms / 1000.0
+            time.sleep(0.002)
+
+    @staticmethod
+    def _frame(payload, type_code=0xF0):
+        body = bytes([type_code, len(payload)]) + payload.encode("ascii")
+        return bytes([0xB5]) + body + bytes([ff.sum_crc(body)])
+
+    def _push(self, b):
+        with self._lock:
+            self._buf += b
+
+    @property
+    def in_waiting(self):
+        with self._lock:
+            return len(self._buf)
+
+    def read(self, n):
+        deadline = time.time() + 0.15
+        while True:
+            with self._lock:
+                if self._buf:
+                    out = bytes(self._buf[:n])
+                    del self._buf[:n]
+                    return out
+            if time.time() >= deadline:
+                return b""
+            time.sleep(0.004)
+
+    def write(self, data):
+        self.write_log.append((time.time(), bytes(data)))
+        cmd = bytes(data[3:-1]).decode("ascii", "replace").strip()
+        threading.Timer(0.05, lambda: self._on_cmd(cmd)).start()
+        return len(data)
+
+    def _on_cmd(self, cmd):
+        if cmd.startswith("M2S"):
+            try:
+                ms = int(cmd[3:])
+            except ValueError:
+                ms = 0
+            self.report_ms = ms
+            self._next_tx = time.time() + ms / 1000.0
+            self._push(self._frame("ok"))
+        elif cmd in ("M1", "M2", "M3", "M4", "M999"):
+            self._push(self._frame({"M1": "34.6", "M2": "OFF", "M3": "OFF",
+                                    "M4": "35.0", "M999": "v41.0.0.1"}[cmd]))
+        else:
+            self._push(self._frame("ok"))
+
+    def close(self):
+        self._alive = False
+        try:
+            self._th.join(timeout=0.5)
+        except Exception:
+            pass
+
+
+ff.serial.Serial = FakeSerial
+
+from PySide6.QtCore import QTimer
+from PySide6.QtWidgets import QApplication
+
+app = QApplication([])
+import app_fluent
+
+# ---- 测试侧修复: 旁路坏裁剪(原生上限 500 行, 语义等同原意图) ----
+window = app_fluent.App(0)
+window.txt_log.setMaximumBlockCount(500)   # blockCount 永不 >500, 坏分支不触发
+window.show()
+fake = None
+exception_count = [0]
+
+# ---- 指标 ----
+stats = {"paint": [0, 0.0, 0.0], "tick": [0, 0.0, 0.0],
+         "gc": [0, 0.0, 0.0], "drift": [], "qmax": [0], "click": []}
+
+_orig_paint = app_fluent.TemperatureChart.paintEvent
+def timed_paint(self, ev):
+    t0 = time.perf_counter()
+    _orig_paint(self, ev)
+    dt = time.perf_counter() - t0
+    s = stats["paint"]; s[0] += 1; s[1] += dt; s[2] = max(s[2], dt)
+app_fluent.TemperatureChart.paintEvent = timed_paint
+
+_orig_tick = app_fluent.App._tick
+def timed_tick(self):
+    t0 = time.perf_counter()
+    try:
+        _orig_tick(self)
+    except Exception as e:
+        exception_count[0] += 1
+        if exception_count[0] <= 3:
+            print("TICK EXCEPTION:", e, flush=True)
+    dt = time.perf_counter() - t0
+    s = stats["tick"]; s[0] += 1; s[1] += dt; s[2] = max(s[2], dt)
+app_fluent.App._tick = timed_tick
+
+_gc_t = [0.0]
+def gc_cb(phase, info):
+    if phase == "start":
+        _gc_t[0] = time.perf_counter()
+    else:
+        dt = time.perf_counter() - _gc_t[0]
+        s = stats["gc"]; s[0] += 1; s[1] += dt; s[2] = max(s[2], dt)
+gc.callbacks.append(gc_cb)
+
+probe = {"expect": None}
+def probe_fire():
+    now = time.time()
+    if probe["expect"] is None:
+        probe["expect"] = now + 0.02
+        return
+    stats["drift"].append((now, now - probe["expect"]))
+    probe["expect"] += 0.02
+    if now - probe["expect"] > 0.5:
+        probe["expect"] = now + 0.02
+pt = QTimer(); pt.timeout.connect(probe_fire); pt.start(20)
+
+CLICK_PERIOD = 30
+click = {"expect": None, "n": 0}
+
+def do_click_stop():
+    now = time.time()
+    drift_ms = (now - click["expect"]) * 1000 if click["expect"] else 0.0
+    q0 = window._rx_q.qsize()
+    window._send_set("M2S0", "report", ("OFF", None))     # "停止"按钮真实路径
+    click["n"] += 1
+    stats["click"].append(dict(n=click["n"], t=now, drift_ms=drift_ms,
+                               qsize=q0, pending=len(window._pending),
+                               samples=len(window.samples_t),
+                               logblk=window.txt_log.blockCount()))
+    QTimer.singleShot(3000, do_click_restart)
+
+def do_click_restart():
+    window._send_set("M2S100", "report", ("ON", 100))
+    click["expect"] = time.time() + CLICK_PERIOD
+    QTimer.singleShot(CLICK_PERIOD * 1000, do_click_stop)
+
+def start():
+    global fake
+    window.link.open("COMFAKE", 115200)
+    def wait_conn():
+        if window.link.is_connected():
+            fake = FakeSerial.last
+            click["expect"] = time.time() + 15
+            QTimer.singleShot(15000, do_click_stop)
+            print("[t0] connected; stop-clicks every %ds from +15s" % CLICK_PERIOD,
+                  flush=True)
+        else:
+            QTimer.singleShot(200, wait_conn)
+    QTimer.singleShot(200, wait_conn)
+QTimer.singleShot(100, start)
+
+t_start = time.time()
+DURATION = float(sys.argv[1]) if len(sys.argv) > 1 else 320.0
+INTERVAL = float(sys.argv[2]) if len(sys.argv) > 2 else 40.0
+
+def monitor():
+    while time.time() - t_start < DURATION:
+        time.sleep(INTERVAL)
+        el = time.time() - t_start
+        recent = [d for t, d in stats["drift"] if t >= time.time() - INTERVAL]
+        rmax = max(recent) * 1000 if recent else 0
+        stats["qmax"][0] = max(stats["qmax"][0], window._rx_q.qsize())
+        print(f"[{el:6.1f}s] cpu={time.process_time():7.2f}s "
+              f"samples={len(window.samples_t):6d} q={window._rx_q.qsize():4d} "
+              f"(max {stats['qmax'][0]:4d}) logblk={window.txt_log.blockCount():4d} "
+              f"paint(n={stats['paint'][0]:5d} max={stats['paint'][2]*1000:6.1f}ms) "
+              f"tick_max={stats['tick'][2]*1000:6.1f}ms "
+              f"gc_max={stats['gc'][2]*1000:5.1f}ms drift_max={rmax:7.1f}ms "
+              f"exc={exception_count[0]}", flush=True)
+    print("==== CLICK 明细 ====", flush=True)
+    for c in stats["click"]:
+        tx = next((t for t, d in FakeSerial.last.write_log
+                   if d[3:-1].decode("ascii", "replace") == "M2S0"
+                   and c["t"] <= t <= c["t"] + 2.0), None)
+        total = (tx - c["t"]) * 1000 if tx else -1
+        print(f"click#{c['n']} t=+{c['t']-t_start:6.1f}s 调度漂移={c['drift_ms']:8.1f}ms "
+              f"click->TX={total:7.1f}ms q={c['qsize']:4d} pending={c['pending']} "
+              f"samples={c['samples']:6d} logblk={c['logblk']}", flush=True)
+    ds = sorted(d for _t, d in stats["drift"])
+    if ds:
+        print(f"drift: n={len(ds)} p50={ds[len(ds)//2]*1000:.1f}ms "
+              f"p99={ds[int(len(ds)*0.99)]*1000:.1f}ms max={ds[-1]*1000:.1f}ms "
+              f">500ms次数={sum(1 for d in ds if d>0.5)}", flush=True)
+    print(f"paint total={stats['paint'][1]:.2f}s max={stats['paint'][2]*1000:.1f}ms | "
+          f"tick total={stats['tick'][1]:.2f}s max={stats['tick'][2]*1000:.1f}ms | "
+          f"gc total={stats['gc'][1]:.2f}s max={stats['gc'][2]*1000:.1f}ms "
+          f"n={stats['gc'][0]} | exceptions={exception_count[0]}", flush=True)
+    os._exit(0)
+
+threading.Thread(target=monitor, daemon=True).start()
+app.exec()

GUI/_diag_latency_buggy.py

@@ -0,0 +1,225 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""诊断 v2b: BUGGY 对照组 — 不加裁剪修复, 复现 PySide6 6.11.2 上
+_log 裁剪 AttributeError 触发后的行为: tick 中止 / RX 处理滞后 ratchet /
+事件循环漂移 / 点击->TX。指标同 v2, 另增帧处理年龄(frame_age)。"""
+import gc
+import os
+import sys
+import threading
+import time
+
+os.environ.setdefault("QT_QPA_PLATFORM", "offscreen")
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import filmforge_rs485 as ff
+
+
+class FakeSerial:
+    def __init__(self, *a, **kw):
+        self._buf = bytearray()
+        self._lock = threading.Lock()
+        self._alive = True
+        self.write_log = []
+        self.report_ms = 0
+        self._next_tx = 0.0
+        self._th = threading.Thread(target=self._fw, daemon=True)
+        self._th.start()
+        FakeSerial.last = self
+
+    def _fw(self):
+        while self._alive:
+            now = time.time()
+            if self.report_ms and now >= self._next_tx:
+                self._push(self._frame("34.6"))
+                self._next_tx = now + self.report_ms / 1000.0
+            time.sleep(0.002)
+
+    @staticmethod
+    def _frame(payload, type_code=0xF0):
+        body = bytes([type_code, len(payload)]) + payload.encode("ascii")
+        return bytes([0xB5]) + body + bytes([ff.sum_crc(body)])
+
+    def _push(self, b):
+        with self._lock:
+            self._buf += b
+
+    @property
+    def in_waiting(self):
+        with self._lock:
+            return len(self._buf)
+
+    def read(self, n):
+        deadline = time.time() + 0.15
+        while True:
+            with self._lock:
+                if self._buf:
+                    out = bytes(self._buf[:n])
+                    del self._buf[:n]
+                    return out
+            if time.time() >= deadline:
+                return b""
+            time.sleep(0.004)
+
+    def write(self, data):
+        self.write_log.append((time.time(), bytes(data)))
+        cmd = bytes(data[3:-1]).decode("ascii", "replace").strip()
+        threading.Timer(0.05, lambda: self._on_cmd(cmd)).start()
+        return len(data)
+
+    def _on_cmd(self, cmd):
+        if cmd.startswith("M2S"):
+            try:
+                ms = int(cmd[3:])
+            except ValueError:
+                ms = 0
+            self.report_ms = ms
+            self._next_tx = time.time() + ms / 1000.0
+            self._push(self._frame("ok"))
+        elif cmd in ("M1", "M2", "M3", "M4", "M999"):
+            self._push(self._frame({"M1": "34.6", "M2": "OFF", "M3": "OFF",
+                                    "M4": "35.0", "M999": "v41.0.0.1"}[cmd]))
+        else:
+            self._push(self._frame("ok"))
+
+    def close(self):
+        self._alive = False
+        try:
+            self._th.join(timeout=0.5)
+        except Exception:
+            pass
+
+
+ff.serial.Serial = FakeSerial
+
+from PySide6.QtCore import QTimer
+from PySide6.QtWidgets import QApplication
+
+app = QApplication([])
+import app_fluent
+
+# 注意: 不打 setMaximumBlockCount 补丁 —— 原样复现 bug
+window = app_fluent.App(0)
+window.show()
+fake = None
+exception_count = [0]
+
+stats = {"tick": [0, 0.0, 0.0], "gc": [0, 0.0, 0.0], "drift": [],
+         "qmax": [0], "click": [], "age": [0, 0.0, 0.0]}   # age: n,total,max
+
+_orig_tick = app_fluent.App._tick
+def timed_tick(self):
+    t0 = time.perf_counter()
+    try:
+        _orig_tick(self)
+    except Exception as e:
+        exception_count[0] += 1
+        if exception_count[0] <= 3:
+            print("TICK EXCEPTION:", type(e).__name__, e, flush=True)
+    dt = time.perf_counter() - t0
+    s = stats["tick"]; s[0] += 1; s[1] += dt; s[2] = max(s[2], dt)
+app_fluent.App._tick = timed_tick
+
+_orig_handle = app_fluent.App._handle_frame
+def timed_handle(self, fr, t_rx):
+    _orig_handle(self, fr, t_rx)
+    age = time.time() - t_rx                     # 帧从收到到被主线程处理的滞后
+    s = stats["age"]; s[0] += 1; s[1] += age; s[2] = max(s[2], age)
+app_fluent.App._handle_frame = timed_handle
+
+_gc_t = [0.0]
+def gc_cb(phase, info):
+    if phase == "start":
+        _gc_t[0] = time.perf_counter()
+    else:
+        dt = time.perf_counter() - _gc_t[0]
+        s = stats["gc"]; s[0] += 1; s[1] += dt; s[2] = max(s[2], dt)
+gc.callbacks.append(gc_cb)
+
+probe = {"expect": None}
+def probe_fire():
+    now = time.time()
+    if probe["expect"] is None:
+        probe["expect"] = now + 0.02
+        return
+    stats["drift"].append((now, now - probe["expect"]))
+    probe["expect"] += 0.02
+    if now - probe["expect"] > 0.5:
+        probe["expect"] = now + 0.02
+pt = QTimer(); pt.timeout.connect(probe_fire); pt.start(20)
+
+CLICK_PERIOD = 30
+click = {"expect": None, "n": 0}
+
+def do_click_stop():
+    now = time.time()
+    drift_ms = (now - click["expect"]) * 1000 if click["expect"] else 0.0
+    q0 = window._rx_q.qsize()
+    window._send_set("M2S0", "report", ("OFF", None))
+    click["n"] += 1
+    stats["click"].append(dict(n=click["n"], t=now, drift_ms=drift_ms,
+                               qsize=q0, pending=len(window._pending),
+                               samples=len(window.samples_t),
+                               logblk=window.txt_log.blockCount()))
+    QTimer.singleShot(3000, do_click_restart)
+
+def do_click_restart():
+    window._send_set("M2S100", "report", ("ON", 100))
+    click["expect"] = time.time() + CLICK_PERIOD
+    QTimer.singleShot(CLICK_PERIOD * 1000, do_click_stop)
+
+def start():
+    global fake
+    window.link.open("COMFAKE", 115200)
+    def wait_conn():
+        if window.link.is_connected():
+            fake = FakeSerial.last
+            click["expect"] = time.time() + 15
+            QTimer.singleShot(15000, do_click_stop)
+            print("[t0] connected; stop-clicks every %ds from +15s (BUGGY mode)"
+                  % CLICK_PERIOD, flush=True)
+        else:
+            QTimer.singleShot(200, wait_conn)
+    QTimer.singleShot(200, wait_conn)
+QTimer.singleShot(100, start)
+
+t_start = time.time()
+DURATION = float(sys.argv[1]) if len(sys.argv) > 1 else 320.0
+INTERVAL = float(sys.argv[2]) if len(sys.argv) > 2 else 40.0
+
+def monitor():
+    while time.time() - t_start < DURATION:
+        time.sleep(INTERVAL)
+        el = time.time() - t_start
+        recent = [d for t, d in stats["drift"] if t >= time.time() - INTERVAL]
+        rmax = max(recent) * 1000 if recent else 0
+        stats["qmax"][0] = max(stats["qmax"][0], window._rx_q.qsize())
+        n, tot, mx = stats["age"]
+        print(f"[{el:6.1f}s] cpu={time.process_time():7.2f}s "
+              f"samples={len(window.samples_t):6d} q={window._rx_q.qsize():5d} "
+              f"(max {stats['qmax'][0]:5d}) logblk={window.txt_log.blockCount():5d} "
+              f"frame_age avg={tot/max(n,1)*1000:7.1f}ms max={mx*1000:8.1f}ms "
+              f"tick_max={stats['tick'][2]*1000:6.1f}ms drift_max={rmax:8.1f}ms "
+              f"exc={exception_count[0]}", flush=True)
+    print("==== CLICK 明细 ====", flush=True)
+    for c in stats["click"]:
+        tx = next((t for t, d in FakeSerial.last.write_log
+                   if d[3:-1].decode("ascii", "replace") == "M2S0"
+                   and c["t"] <= t <= c["t"] + 2.0), None)
+        total = (tx - c["t"]) * 1000 if tx else -1
+        print(f"click#{c['n']} t=+{c['t']-t_start:6.1f}s 调度漂移={c['drift_ms']:8.1f}ms "
+              f"click->TX={total:7.1f}ms q={c['qsize']:5d} pending={c['pending']} "
+              f"samples={c['samples']:6d} logblk={c['logblk']}", flush=True)
+    ds = sorted(d for _t, d in stats["drift"])
+    if ds:
+        print(f"drift: n={len(ds)} p50={ds[len(ds)//2]*1000:.1f}ms "
+              f"p99={ds[int(len(ds)*0.99)]*1000:.1f}ms max={ds[-1]*1000:.1f}ms "
+              f">500ms次数={sum(1 for d in ds if d>0.5)}", flush=True)
+    n, tot, mx = stats["age"]
+    print(f"frame_age avg={tot/max(n,1)*1000:.1f}ms max={mx*1000:.1f}ms | "
+          f"exceptions={exception_count[0]} | tick n={stats['tick'][0]} "
+          f"max={stats['tick'][2]*1000:.1f}ms", flush=True)
+    os._exit(0)
+
+threading.Thread(target=monitor, daemon=True).start()
+app.exec()

GUI/app_fluent.py

@@ -60,7 +60,7 @@ import config
 import filmforge_rs485 as ff
 
 APP_NAME = "FilmForge"
-APP_VERSION = "0.2.7"
+APP_VERSION = "0.2.9"
 REPO_URL = "https://gitea.hulk.wang/hulk/FilmForge"
 MONO = QFont("Consolas", 10)
 CHIP_IDLE = "#666666"
@@ -69,6 +69,7 @@ CHIP_ERR = "#D9534F"
 CHIP_WARN = "#E8833A"
 PENDING_TIMEOUT = 1.2        # 应答超时(秒): 协议建议 300ms 重发, 监视台不自动重发只提示
 RESPONSE_FRESH = 0.5         # 应答新鲜度: 该窗口内的温度帧优先按“查询应答”派发
+MAX_FAULTS = 1000            # 故障标记内存上限(条): NTC 抖动场景防无限增长
 
 CHART_COLORS_DARK = dict(bg="#242424", grid="#3a3a3a", text="#c9c9c9",
                          measured="#00c8d7", target=CHIP_WARN, fault=CHIP_ERR)
@@ -133,7 +134,7 @@ class TemperatureChart(QWidget):
         ts, vs = self._samples_t, self._samples_v
         if not ts:
             return [], []
-        lo, hi = 0, len(ts)                    # 二分找左边界(留一点前文保证连线)
+        lo, hi = 0, len(ts)                    # 左边界二分(留一点前文保证连线)
         while lo < hi:
             m = (lo + hi) // 2
             if ts[m] < t0:
@@ -141,7 +142,14 @@ class TemperatureChart(QWidget):
             else:
                 hi = m
         lo = max(lo - 1, 0)
-        t_out, v_out = ts[lo:], vs[lo:]
+        end, hi = lo, len(ts)                  # 右边界二分: 暂停(冻结)后不切冻结尾后新样本
+        while end < hi:
+            m = (end + hi) // 2
+            if ts[m] <= t1:
+                end = m + 1
+            else:
+                hi = m
+        t_out, v_out = ts[lo:end], vs[lo:end]
         if len(t_out) > 4000:
             stride = math.ceil(len(t_out) / 4000)
             gt, gv = [], []
@@ -674,6 +682,9 @@ class App(QMainWindow):
         self.txt_log.setFont(MONO)
         self.txt_log.setLineWrapMode(QPlainTextEdit.NoWrap)
         self.txt_log.setMinimumHeight(200)
+        # 裁剪走原生环形上限: 手工 cursor 裁剪在 PySide6 6.11 上抛 AttributeError
+        # (实例级枚举访问已移除), 曾致每 tick 异常中止 + 日志无限增长 + 显示滞后
+        self.txt_log.setMaximumBlockCount(500)
         body.addWidget(self.txt_log)
         self._log(f"帧格式: D5 01 LEN <命令> CRC / B5 TYPE LEN <payload> CRC; "
                   f"响应超时建议 {ff.REPLY_TIMEOUT * 1000:.0f}ms 重发")
@@ -959,6 +970,8 @@ class App(QMainWindow):
         self.samples_t.append(t)
         self.samples_v.append(float("nan"))
         self.faults.append((t, code))
+        if len(self.faults) > MAX_FAULTS:   # 封顶: 抖动场景不无限增长(_draw_faults 按 paint 全量扫描)
+            del self.faults[:len(self.faults) - MAX_FAULTS]
         self.lbl_temp.setText("—.— ℃")
         self._csv_write(t, code, ok=False)
         self.chart.update()
@@ -1266,9 +1279,19 @@ class App(QMainWindow):
         target = f"{self.target_hist[-1][1]:g}" if self.target_hist else ""
         heater = self.heater_state or ""
         sensor = self.sensor_state or ("OK" if ok else "")
-        self._csv.write(f"{time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(t))},"
-                        f"{t - self._csv_t0:.3f},{temp_text},{target},{heater},{sensor}\n")
-        self._csv.flush()
+        try:
+            self._csv.write(f"{time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(t))},"
+                            f"{t - self._csv_t0:.3f},{temp_text},{target},{heater},{sensor}\n")
+            self._csv.flush()
+        except OSError as e:
+            # 写失败(磁盘满/记录盘被拔): 自动停记录。本函数在 _tick 帧处理路径上,
+            # 未捕获异常会中止整段消费(与 v0.2.8 日志裁剪同款失效管道, 勿回退)
+            self._csv.close()
+            self._csv = None
+            self.btn_csv.setText("开始记录")
+            self.lbl_csv.setText(f"记录失败已停止: {e}")
+            self._log(f"⚠ CSV 记录失败已停止: {e}")
+            return
         self._csv_count += 1
         self.lbl_csv.setText(f"记录中({self._csv_count} 行) → "
                              f"{os.path.basename(self._csv_path)}")
@@ -1279,15 +1302,11 @@ class App(QMainWindow):
 
     # ================= 日志 =================
     def _log(self, line):
-        self.txt_log.appendPlainText(time.strftime("%H:%M:%S  ") + line)
-        if self.txt_log.blockCount() > 500:
-            cursor = self.txt_log.textCursor()
-            cursor.movePosition(cursor.Start)
-            cursor.select(cursor.BlockUnderCursor)
-            cursor.removeSelectedText()
-            cursor.deleteChar()
         sb = self.txt_log.verticalScrollBar()
-        sb.setValue(sb.maximum())
+        at_bottom = sb.value() >= sb.maximum() - 4
+        self.txt_log.appendPlainText(time.strftime("%H:%M:%S  ") + line)
+        if at_bottom:                       # 仅原本贴底时跟随, 不抢用户上翻看历史
+            sb.setValue(sb.maximum())
 
     def _clear_log(self):
         self.txt_log.clear()

GUI/filmforge_rs485.py

@@ -347,8 +347,17 @@ class FF485:
                         pass
                     if item is not None:
                         cmd, frame = item
-                        self.ser.write(frame)
-                        self.on_event(f'→ {cmd}  [{frame.hex(" ").upper()}]')
+                        sent = False
+                        try:
+                            self.ser.write(frame)
+                            sent = True
+                        except serial.SerialTimeoutException:
+                            # 写超时(适配器停滞): 只丢这条命令, 链路保留。
+                            # 勿并入下方 OSError——瞬时停滞曾被当成"设备拔出"
+                            # 关端口断链, 迫使用户手动重连(适配器换向本就不稳)
+                            self.on_event(f'✕ {cmd} 写超时未发出, 链路保留(可重发)')
+                        if sent:
+                            self.on_event(f'→ {cmd}  [{frame.hex(" ").upper()}]')
                     # 按到达量读取, 不攒大块: 固定 read(256) 在持续帧流下要攒满
                     # 256 字节才返回(win32 字节间隔超时 150ms 因帧间隔更短永不
                     # 触发, 100ms 上报 ≈ 2.5s 一批), 表现为曲线/命令全按 3s 一卡

GUI/selftest_gui.py

@@ -227,5 +227,57 @@ with open(window._csv_path, encoding='utf-8-sig') as f:
 check('CSV 表头+2 行', len(lines), 3)
 check('CSV 故障行', lines[2].split(',')[2], 'SHRT')
 
+# 6) 日志环形裁剪回归: 写满 500 行后封顶, 继续追加不抛异常、不无限增长。
+#    (手工 cursor 裁剪用 cursor.Start/BlockUnderCursor 实例级枚举访问,
+#    PySide6 6.11 上必然抛 AttributeError → 每 tick 异常中止, v0.2.8 改原生裁剪)
+for i in range(600):
+    window._log(f'日志压力行 {i}')
+check('日志 500 行封顶', window.txt_log.blockCount() <= 500, True)
+window._log('封顶后追加行')
+check('封顶后可继续追加', window.txt_log.blockCount() <= 500, True)
+check('封顶后最新行在末尾',
+      window.txt_log.toPlainText().splitlines()[-1].endswith('封顶后追加行'), True)
+
+# 7) v0.2.9 回归: 故障标记封顶 / 滚动条跟随 / CSV 写失败自停 / 冻结切片右边界
+t_end = time.time()
+for i in range(app_fluent.MAX_FAULTS + 50):
+    window._record_fault(t_end + i * 0.01, ff.SENSOR_OPEN)
+check('故障标记封顶', len(window.faults), app_fluent.MAX_FAULTS)
+
+sb = window.txt_log.verticalScrollBar()
+sb.setValue(0)
+window._log('上翻后新行')
+check('上翻时不抢滚动条', sb.value() <= 1, True)
+sb.setValue(sb.maximum())
+window._log('贴底跟随行')
+check('贴底时跟随滚动', abs(sb.value() - sb.maximum()) <= 4, True)
+
+
+class _BoomFile:
+    """模拟写坏盘: write/flush 必抛 OSError(磁盘满/记录盘被拔)"""
+    def write(self, *_a):
+        raise OSError(28, '模拟磁盘满')
+    def flush(self):
+        raise OSError(28, '模拟磁盘满')
+    def close(self):
+        pass
+
+
+window._csv = _BoomFile()
+window._csv_path = 'dummy.csv'
+window._add_sample(time.time(), 25.0)
+check('CSV 写失败自动停止记录', window._csv is None, True)
+check('CSV 失败后按钮复位', window.btn_csv.text(), '开始记录')
+
+c = window.chart
+c._samples_t, c._samples_v = [1.0, 2.0, 3.0, 4.0], [1, 2, 3, 4]
+c._frozen_at = 2.5
+ts, _vs = c._visible(0, 2.5)
+check('冻结切片不含冻结尾后样本', ts, [1.0, 2.0])
+c._frozen_at = None
+ts, _vs = c._visible(0, 99)
+check('实时切片完整', ts, [1.0, 2.0, 3.0, 4.0])
+c._samples_t, c._samples_v = window.samples_t, window.samples_v
+
 print('RESULT:', 'FAIL:' + ','.join(fails) if fails else 'ALL PASS')
 sys.exit(1 if fails else 0)