mslxdff 0.1.188 → 0.1.197

package/bin/mslxdff.js

@@ -1,4 +1,4 @@
 #!/usr/bin/env node
-// thin adapter — all logic lives in src/cli (deep module, single inlet run)
-import { run } from "../src/cli/index.js";
-await run(process.argv.slice(2));
+// thin adapter — all logic lives in src/cli (deep module, single inlet run)
+import { run } from "../src/cli/index.js";
+await run(process.argv.slice(2));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mslxdff",
-  "version": "0.1.188",
+  "version": "0.1.197",
   "description": "测试项目，请勿使用。",
   "type": "module",
   "bin": {

package/src/anthropic/count-tokens.js

@@ -13,7 +13,9 @@ const CHAR_CAP = 32_000_000;
 // （每条消息 content 整段一次 + 每个 tool_call 一次）。逐块各套会把「非空至少 1」下限乘上块数
 // —— 几十个 1 字符 text 块能被高报几十倍（评审 P3：方向虽偏高安全，但与生成侧口径分叉就是第二个真相源）。
 function add(acc, text) {
-  if (acc.chars > CHAR_CAP) return; // 超限后停止累计，只保留已计部分与图片面值
+  // 超限后停止累计（chars 与 tokens 一起停）：只护 chars 会让 tokens 无界增长，
+  // 客户端拿这个读数决定何时压实上下文 ⇒ 超长请求触发一次毫无必要的激进压缩。
+  if (acc.chars > CHAR_CAP) return;
   const s = text == null ? "" : String(text);
   const c = createTokenCounter();
   c.add(s);
@@ -36,8 +38,13 @@ function countMessage(m, acc) {
     else if (b.type === "thinking") put(String(b.thinking || ""));
     else if (b.type === "image") acc.images += 1;
   }
-  if (acc.chars <= CHAR_CAP) acc.chars += chars;
-  acc.tokens += c.tokens();
+  // tokens 与 chars 必须**同受 CHAR_CAP 约束**：只护 chars 不护 tokens 时，超长请求（readBody 无大小上限，
+  // 实测 10 条 4000 万字符消息 → 1.11 亿 tokens）会把读数推到天文数字，而客户端正是拿它决定何时压实上下文
+  // ⇒ 一次毫无必要的激进压缩。超限后不再累加，保留已计部分。
+  if (acc.chars <= CHAR_CAP) {
+    acc.chars += chars;
+    acc.tokens += c.tokens();
+  }
 }
 
 /** 字符估算版 token 计数（Anthropic 官方：端点缺失时客户端本来就退回字符估算，精度非目标）。 */

package/src/auto.js

@@ -1,254 +1,254 @@
-import { statSync } from "node:fs";
-import { loadModelErrors, saveModelErrors, loadModelLatencies, saveModelLatencies, loadPreferredModel, loadModelPicks, saveModelPicks, defaultStateFile } from "./state.js";
-
-// 出厂默认首选模型（state.json 的 preferredModel / env MSLXDFF_PREFERRED_MODEL 可覆盖）
-export const DEFAULT_PREFERRED_MODEL = "big-pickle";
-// 兼容旧导出名：语义为"出厂默认"，当前生效值请用 getPreferredModel()
-export const PREFERRED_MODEL = DEFAULT_PREFERRED_MODEL;
-
-// 当前生效的首选模型：state.json > env > 出厂默认；mtime 缓存保证 daemon 热生效
-const _prefCache = { mtimeMs: -1, file: null, value: null };
-export function getPreferredModel({ file = defaultStateFile() } = {}) {
-  try {
-    const st = statSync(file);
-    if (_prefCache.file !== file || st.mtimeMs !== _prefCache.mtimeMs) {
-      _prefCache.file = file;
-      _prefCache.mtimeMs = st.mtimeMs;
-      _prefCache.value = loadPreferredModel({ file });
-    }
-  } catch {
-    _prefCache.file = file;
-    _prefCache.mtimeMs = -1;
-    _prefCache.value = null;
-  }
-  const env = (process.env.MSLXDFF_PREFERRED_MODEL || "").trim();
-  return _prefCache.value || env || DEFAULT_PREFERRED_MODEL;
-}
-
-export const DEFAULT_AUTO_MODELS = [
-  PREFERRED_MODEL,
-  "mimo-v2.5-free",
-  "deepseek-v4-flash-free",
-  "ling-3.0-flash-fin-free",
-  "nemotron-3-ultra-free",
-  "nemotron-3.5-lightning-free",
-  "muse-spark-1.3-contributor-free",
-].filter((id, i, arr) => id && arr.indexOf(id) === i);
-
-export function isAutoModel(model) {
-  return !model || model === "auto";
-}
-
-export const MODEL_STATUS = Object.freeze({
-  NORMAL: "normal",
-  LIMIT: "limit",
-  ERROR: "error",
-});
-
-// Legacy modelErrors entries are bare timestamps ({id: ts}); newer ones are
-// objects ({id: {status, at, code}}). Normalize both to an entry object.
-// `slow` flags a model whose last request was slow (over the wall-clock
-// threshold) — those get a longer cooldown so they lie low until they recover.
-function normEntry(e) {
-  if (typeof e === "number") return { status: MODEL_STATUS.ERROR, at: e, code: null, slow: false };
-  if (e && typeof e === "object") {
-    return {
-      status: e.status || MODEL_STATUS.ERROR,
-      at: typeof e.at === "number" ? e.at : 0,
-      code: e.code ?? null,
-      slow: Boolean(e.slow),
-    };
-  }
-  return null;
-}
-
-export function classifyErrorEvent(evt = {}) {
-  if (evt.slow) return MODEL_STATUS.ERROR;
-  const code = Number(evt.status);
-  if (code === 429) return MODEL_STATUS.LIMIT;
-  const msg = String(evt.message || evt.note || "").toLowerCase();
-  if (msg.includes("rate limit") || msg.includes("limit exceeded") || msg.includes("429")) {
-    return MODEL_STATUS.LIMIT;
-  }
-  return MODEL_STATUS.ERROR;
-}
-
-export const DEFAULT_COOLDOWN_MS = 60_000;
-export const DEFAULT_SLOW_COOLDOWN_MS = 5 * 60_000;
-export const DEFAULT_LATENCY_ALPHA = 0.3;
-
-function effectiveCooldown(entry, slowCooldownMs, cooldownMs) {
-  if (entry && entry.slow) return slowCooldownMs || 0;
-  return cooldownMs || 0;
-}
-
-function inCooldown(id, errors, now, cooldownMs, slowCooldownMs) {
-  const e = normEntry(errors[id]);
-  if (!e || !(e.at > 0)) return false;
-  if (e.status === MODEL_STATUS.NORMAL) return false;
-  const cd = effectiveCooldown(e, slowCooldownMs, cooldownMs);
-  return cd > 0 && now - e.at < cd;
-}
-
-// Latency EMA helpers
-function normLatency(e) {
-  if (!e || typeof e !== "object") return null;
-  const ema = Number(e.emaMs);
-  return Number.isFinite(ema) && ema > 0 ? ema : null;
-}
-
-export function rankModels(ids, errors = {}, { now = Date.now(), cooldownMs = 0, slowCooldownMs = 0, latencies = {}, preferred } = {}) {
-  const pref = preferred ?? getPreferredModel();
-  // 最近一次成功的模型（NORMAL 且非慢且 at 最大），用于“上次成功优先”——慢模型即使刚成功也不应钉死
-  let lastSuccessId = null;
-  let lastSuccessAt = 0;
-  for (const [id, e] of Object.entries(errors)) {
-    const ne = normEntry(e);
-    if (ne && ne.status === MODEL_STATUS.NORMAL && !ne.slow && ne.at > lastSuccessAt) {
-      lastSuccessAt = ne.at;
-      lastSuccessId = id;
-    }
-  }
-  return [...new Set(ids)]
-    .filter(Boolean)
-    .map((id) => ({
-      id,
-      e: normEntry(errors[id]),
-      err: normEntry(errors[id])?.at ?? 0,
-      isPreferred: id === pref,
-      isLastSuccess: id === lastSuccessId,
-      cooling: inCooldown(id, errors, now, cooldownMs, slowCooldownMs),
-      latency: normLatency(latencies[id]) ?? Number.MAX_SAFE_INTEGER,
-    }))
-    .sort(
-      (a, b) =>
-        (a.cooling ? 1 : 0) - (b.cooling ? 1 : 0) ||
-        (b.isLastSuccess ? 1 : 0) - (a.isLastSuccess ? 1 : 0) ||
-        (b.isPreferred ? 1 : 0) - (a.isPreferred ? 1 : 0) ||
-        a.latency - b.latency ||
-        a.err - b.err
-    )
-    .map((x) => x.id);
-}
-
-export function createAutoSelector({
-  loadCandidates,
-  file,
-  now = () => Date.now(),
-  cooldownMs = DEFAULT_COOLDOWN_MS,
-  slowCooldownMs = DEFAULT_SLOW_COOLDOWN_MS,
-  latencyAlpha = DEFAULT_LATENCY_ALPHA,
-  errors: seedErrors,
-  latencies: seedLatencies,
-  persist = (errors, f = file) => saveModelErrors(errors, f ? { file: f } : {}),
-  persistLatencies = (latencies, f = file) => saveModelLatencies(latencies, f ? { file: f } : {}),
-  loadPicks = () => (file ? loadModelPicks({ file }) : []),
-  persistPicks = (picks) => (file ? saveModelPicks(picks, { file }) : picks),
-} = {}) {
-  const lastErrorAt = { ...(seedErrors ?? loadModelErrors(file ? { file } : {})) };
-  const latencies = { ...(seedLatencies ?? loadModelLatencies(file ? { file } : {})) };
-
-  async function loadList() {
-    let list;
-    try {
-      const loaded = await loadCandidates?.();
-      list = Array.isArray(loaded) && loaded.length ? loaded : DEFAULT_AUTO_MODELS;
-    } catch {
-      list = DEFAULT_AUTO_MODELS;
-    }
-    return [...new Set(list)].filter(Boolean);
-  }
-
-  // 勾选集 = auto 候选池白名单：只在勾选的模型里择优；空勾选或勾选中无可用模型时回退全量
-  async function pickedPool(list) {
-    const picks = loadPicks();
-    if (!picks.length) return list;
-    const pickedSet = new Set(picks);
-    const filtered = list.filter((id) => pickedSet.has(id));
-    return filtered.length ? filtered : list;
-  }
-
-  async function candidates() {
-    const list = await loadList();
-    const pool = await pickedPool(list);
-    return rankModels(pool, lastErrorAt, { now: now(), cooldownMs, slowCooldownMs, latencies, preferred: getPreferredModel({ file: file ?? undefined }) });
-  }
-
-  async function candidatesFor(requested) {
-    if (!requested) return candidates();
-    const list = await loadList();
-    // 显式指定某模型 = 认可它，自动加入勾选集（仅当它是真实上游 free 模型时，避免垃圾 id 污染）
-    const picks = loadPicks();
-    if (list.includes(requested) && !picks.includes(requested) && persistPicks) {
-      await persistPicks([...picks, requested]);
-    }
-    const all = list.includes(requested) ? list : [requested, ...list];
-    const others = rankModels(all.filter((id) => id !== requested), lastErrorAt, {
-      now: now(),
-      cooldownMs,
-      slowCooldownMs,
-      latencies,
-      preferred: getPreferredModel({ file: file ?? undefined }),
-    });
-    // 显式指定模型：严格优先，永不因冷却被挤到最后（原设计：A deepseek 失败 → B/D deepseek 并发 → 都失败才 fallback）
-    // 冷却仅影响 auto 的择优，不影响指定模型的“很难被更改”语义
-    return [requested, ...others];
-  }
-
-  function isCooling(id) {
-    return inCooldown(id, lastErrorAt, now(), cooldownMs, slowCooldownMs);
-  }
-
-  async function recordError(id, evt = {}) {
-    if (!id) return;
-    lastErrorAt[id] = {
-      status: classifyErrorEvent(evt),
-      at: now(),
-      code: Number.isInteger(Number(evt.status)) ? Number(evt.status) : null,
-      slow: Boolean(evt.slow),
-    };
-    await persist({ ...lastErrorAt });
-  }
-
-  async function recordOk(id, evt = {}) {
-    if (!id) return;
-    lastErrorAt[id] = { status: MODEL_STATUS.NORMAL, at: now(), code: 200, slow: false };
-    await persist({ ...lastErrorAt });
-    const ms = Number(evt.latencyMs ?? evt.totalMs ?? evt.elapsedMs);
-    if (Number.isFinite(ms) && ms > 0) {
-      const prev = latencies[id]?.emaMs;
-      const ema = prev ? Math.round(prev * (1 - latencyAlpha) + ms * latencyAlpha) : Math.round(ms);
-      latencies[id] = { emaMs: ema, lastMs: Math.round(ms), at: now(), count: (latencies[id]?.count ?? 0) + 1 };
-      await persistLatencies({ ...latencies });
-    }
-  }
-
-  async function recordLatency(id, ms) {
-    if (!id || !Number.isFinite(ms) || ms <= 0) return;
-    const prev = latencies[id]?.emaMs;
-    const ema = prev ? Math.round(prev * (1 - latencyAlpha) + ms * latencyAlpha) : Math.round(ms);
-    latencies[id] = { emaMs: ema, lastMs: Math.round(ms), at: now(), count: (latencies[id]?.count ?? 0) + 1 };
-    await persistLatencies({ ...latencies });
-  }
-
-  function statuses() {
-    return { ...lastErrorAt };
-  }
-
-  function latencyStatuses() {
-    return { ...latencies };
-  }
-
-  return {
-    candidates,
-    candidatesFor,
-    recordError,
-    recordOk,
-    recordLatency,
-    statuses,
-    latencyStatuses,
-    isCooling,
-    errors: () => ({ ...lastErrorAt }),
-    latencies: () => ({ ...latencies }),
-  };
-}
+import { statSync } from "node:fs";
+import { loadModelErrors, saveModelErrors, loadModelLatencies, saveModelLatencies, loadPreferredModel, loadModelPicks, saveModelPicks, defaultStateFile } from "./state.js";
+
+// 出厂默认首选模型（state.json 的 preferredModel / env MSLXDFF_PREFERRED_MODEL 可覆盖）
+export const DEFAULT_PREFERRED_MODEL = "big-pickle";
+// 兼容旧导出名：语义为"出厂默认"，当前生效值请用 getPreferredModel()
+export const PREFERRED_MODEL = DEFAULT_PREFERRED_MODEL;
+
+// 当前生效的首选模型：state.json > env > 出厂默认；mtime 缓存保证 daemon 热生效
+const _prefCache = { mtimeMs: -1, file: null, value: null };
+export function getPreferredModel({ file = defaultStateFile() } = {}) {
+  try {
+    const st = statSync(file);
+    if (_prefCache.file !== file || st.mtimeMs !== _prefCache.mtimeMs) {
+      _prefCache.file = file;
+      _prefCache.mtimeMs = st.mtimeMs;
+      _prefCache.value = loadPreferredModel({ file });
+    }
+  } catch {
+    _prefCache.file = file;
+    _prefCache.mtimeMs = -1;
+    _prefCache.value = null;
+  }
+  const env = (process.env.MSLXDFF_PREFERRED_MODEL || "").trim();
+  return _prefCache.value || env || DEFAULT_PREFERRED_MODEL;
+}
+
+export const DEFAULT_AUTO_MODELS = [
+  PREFERRED_MODEL,
+  "mimo-v2.5-free",
+  "deepseek-v4-flash-free",
+  "ling-3.0-flash-fin-free",
+  "nemotron-3-ultra-free",
+  "nemotron-3.5-lightning-free",
+  "muse-spark-1.3-contributor-free",
+].filter((id, i, arr) => id && arr.indexOf(id) === i);
+
+export function isAutoModel(model) {
+  return !model || model === "auto";
+}
+
+export const MODEL_STATUS = Object.freeze({
+  NORMAL: "normal",
+  LIMIT: "limit",
+  ERROR: "error",
+});
+
+// Legacy modelErrors entries are bare timestamps ({id: ts}); newer ones are
+// objects ({id: {status, at, code}}). Normalize both to an entry object.
+// `slow` flags a model whose last request was slow (over the wall-clock
+// threshold) — those get a longer cooldown so they lie low until they recover.
+function normEntry(e) {
+  if (typeof e === "number") return { status: MODEL_STATUS.ERROR, at: e, code: null, slow: false };
+  if (e && typeof e === "object") {
+    return {
+      status: e.status || MODEL_STATUS.ERROR,
+      at: typeof e.at === "number" ? e.at : 0,
+      code: e.code ?? null,
+      slow: Boolean(e.slow),
+    };
+  }
+  return null;
+}
+
+export function classifyErrorEvent(evt = {}) {
+  if (evt.slow) return MODEL_STATUS.ERROR;
+  const code = Number(evt.status);
+  if (code === 429) return MODEL_STATUS.LIMIT;
+  const msg = String(evt.message || evt.note || "").toLowerCase();
+  if (msg.includes("rate limit") || msg.includes("limit exceeded") || msg.includes("429")) {
+    return MODEL_STATUS.LIMIT;
+  }
+  return MODEL_STATUS.ERROR;
+}
+
+export const DEFAULT_COOLDOWN_MS = 60_000;
+export const DEFAULT_SLOW_COOLDOWN_MS = 5 * 60_000;
+export const DEFAULT_LATENCY_ALPHA = 0.3;
+
+function effectiveCooldown(entry, slowCooldownMs, cooldownMs) {
+  if (entry && entry.slow) return slowCooldownMs || 0;
+  return cooldownMs || 0;
+}
+
+function inCooldown(id, errors, now, cooldownMs, slowCooldownMs) {
+  const e = normEntry(errors[id]);
+  if (!e || !(e.at > 0)) return false;
+  if (e.status === MODEL_STATUS.NORMAL) return false;
+  const cd = effectiveCooldown(e, slowCooldownMs, cooldownMs);
+  return cd > 0 && now - e.at < cd;
+}
+
+// Latency EMA helpers
+function normLatency(e) {
+  if (!e || typeof e !== "object") return null;
+  const ema = Number(e.emaMs);
+  return Number.isFinite(ema) && ema > 0 ? ema : null;
+}
+
+export function rankModels(ids, errors = {}, { now = Date.now(), cooldownMs = 0, slowCooldownMs = 0, latencies = {}, preferred } = {}) {
+  const pref = preferred ?? getPreferredModel();
+  // 最近一次成功的模型（NORMAL 且非慢且 at 最大），用于“上次成功优先”——慢模型即使刚成功也不应钉死
+  let lastSuccessId = null;
+  let lastSuccessAt = 0;
+  for (const [id, e] of Object.entries(errors)) {
+    const ne = normEntry(e);
+    if (ne && ne.status === MODEL_STATUS.NORMAL && !ne.slow && ne.at > lastSuccessAt) {
+      lastSuccessAt = ne.at;
+      lastSuccessId = id;
+    }
+  }
+  return [...new Set(ids)]
+    .filter(Boolean)
+    .map((id) => ({
+      id,
+      e: normEntry(errors[id]),
+      err: normEntry(errors[id])?.at ?? 0,
+      isPreferred: id === pref,
+      isLastSuccess: id === lastSuccessId,
+      cooling: inCooldown(id, errors, now, cooldownMs, slowCooldownMs),
+      latency: normLatency(latencies[id]) ?? Number.MAX_SAFE_INTEGER,
+    }))
+    .sort(
+      (a, b) =>
+        (a.cooling ? 1 : 0) - (b.cooling ? 1 : 0) ||
+        (b.isLastSuccess ? 1 : 0) - (a.isLastSuccess ? 1 : 0) ||
+        (b.isPreferred ? 1 : 0) - (a.isPreferred ? 1 : 0) ||
+        a.latency - b.latency ||
+        a.err - b.err
+    )
+    .map((x) => x.id);
+}
+
+export function createAutoSelector({
+  loadCandidates,
+  file,
+  now = () => Date.now(),
+  cooldownMs = DEFAULT_COOLDOWN_MS,
+  slowCooldownMs = DEFAULT_SLOW_COOLDOWN_MS,
+  latencyAlpha = DEFAULT_LATENCY_ALPHA,
+  errors: seedErrors,
+  latencies: seedLatencies,
+  persist = (errors, f = file) => saveModelErrors(errors, f ? { file: f } : {}),
+  persistLatencies = (latencies, f = file) => saveModelLatencies(latencies, f ? { file: f } : {}),
+  loadPicks = () => (file ? loadModelPicks({ file }) : []),
+  persistPicks = (picks) => (file ? saveModelPicks(picks, { file }) : picks),
+} = {}) {
+  const lastErrorAt = { ...(seedErrors ?? loadModelErrors(file ? { file } : {})) };
+  const latencies = { ...(seedLatencies ?? loadModelLatencies(file ? { file } : {})) };
+
+  async function loadList() {
+    let list;
+    try {
+      const loaded = await loadCandidates?.();
+      list = Array.isArray(loaded) && loaded.length ? loaded : DEFAULT_AUTO_MODELS;
+    } catch {
+      list = DEFAULT_AUTO_MODELS;
+    }
+    return [...new Set(list)].filter(Boolean);
+  }
+
+  // 勾选集 = auto 候选池白名单：只在勾选的模型里择优；空勾选或勾选中无可用模型时回退全量
+  async function pickedPool(list) {
+    const picks = loadPicks();
+    if (!picks.length) return list;
+    const pickedSet = new Set(picks);
+    const filtered = list.filter((id) => pickedSet.has(id));
+    return filtered.length ? filtered : list;
+  }
+
+  async function candidates() {
+    const list = await loadList();
+    const pool = await pickedPool(list);
+    return rankModels(pool, lastErrorAt, { now: now(), cooldownMs, slowCooldownMs, latencies, preferred: getPreferredModel({ file: file ?? undefined }) });
+  }
+
+  async function candidatesFor(requested) {
+    if (!requested) return candidates();
+    const list = await loadList();
+    // 显式指定某模型 = 认可它，自动加入勾选集（仅当它是真实上游 free 模型时，避免垃圾 id 污染）
+    const picks = loadPicks();
+    if (list.includes(requested) && !picks.includes(requested) && persistPicks) {
+      await persistPicks([...picks, requested]);
+    }
+    const all = list.includes(requested) ? list : [requested, ...list];
+    const others = rankModels(all.filter((id) => id !== requested), lastErrorAt, {
+      now: now(),
+      cooldownMs,
+      slowCooldownMs,
+      latencies,
+      preferred: getPreferredModel({ file: file ?? undefined }),
+    });
+    // 显式指定模型：严格优先，永不因冷却被挤到最后（原设计：A deepseek 失败 → B/D deepseek 并发 → 都失败才 fallback）
+    // 冷却仅影响 auto 的择优，不影响指定模型的“很难被更改”语义
+    return [requested, ...others];
+  }
+
+  function isCooling(id) {
+    return inCooldown(id, lastErrorAt, now(), cooldownMs, slowCooldownMs);
+  }
+
+  async function recordError(id, evt = {}) {
+    if (!id) return;
+    lastErrorAt[id] = {
+      status: classifyErrorEvent(evt),
+      at: now(),
+      code: Number.isInteger(Number(evt.status)) ? Number(evt.status) : null,
+      slow: Boolean(evt.slow),
+    };
+    await persist({ ...lastErrorAt });
+  }
+
+  async function recordOk(id, evt = {}) {
+    if (!id) return;
+    lastErrorAt[id] = { status: MODEL_STATUS.NORMAL, at: now(), code: 200, slow: false };
+    await persist({ ...lastErrorAt });
+    const ms = Number(evt.latencyMs ?? evt.totalMs ?? evt.elapsedMs);
+    if (Number.isFinite(ms) && ms > 0) {
+      const prev = latencies[id]?.emaMs;
+      const ema = prev ? Math.round(prev * (1 - latencyAlpha) + ms * latencyAlpha) : Math.round(ms);
+      latencies[id] = { emaMs: ema, lastMs: Math.round(ms), at: now(), count: (latencies[id]?.count ?? 0) + 1 };
+      await persistLatencies({ ...latencies });
+    }
+  }
+
+  async function recordLatency(id, ms) {
+    if (!id || !Number.isFinite(ms) || ms <= 0) return;
+    const prev = latencies[id]?.emaMs;
+    const ema = prev ? Math.round(prev * (1 - latencyAlpha) + ms * latencyAlpha) : Math.round(ms);
+    latencies[id] = { emaMs: ema, lastMs: Math.round(ms), at: now(), count: (latencies[id]?.count ?? 0) + 1 };
+    await persistLatencies({ ...latencies });
+  }
+
+  function statuses() {
+    return { ...lastErrorAt };
+  }
+
+  function latencyStatuses() {
+    return { ...latencies };
+  }
+
+  return {
+    candidates,
+    candidatesFor,
+    recordError,
+    recordOk,
+    recordLatency,
+    statuses,
+    latencyStatuses,
+    isCooling,
+    errors: () => ({ ...lastErrorAt }),
+    latencies: () => ({ ...latencies }),
+  };
+}

package/src/bench/cline-bench.js

@@ -1,42 +1,42 @@
-import { clineHeaders } from "../providers/cline/headers.js";
-import { computeMetrics } from "../metrics.js";
-import { createTransport } from "../transport/index.js";
-
-function sseContent(obj) {
-  const c = obj?.choices?.[0]?.delta?.content || obj?.choices?.[0]?.message?.content || "";
-  return typeof c === "string" ? c : "";
-}
-
-export async function clineBenchOne({ baseUrl, model, accessToken, prompt, maxTokens, timeoutMs, fetchImpl }) {
-  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
-  let ttfbMs = null;
-  let totalMs = null;
-  let content = "";
-  try {
-    const sid = `sess_bench_${Date.now()}`;
-    const res = await tr.request({
-      url: `${baseUrl}/api/v1/chat/completions`,
-      method: "POST",
-      headers: { ...clineHeaders(sid, accessToken), Accept: "text/event-stream" },
-      body: { model, messages: [{ role: "user", content: prompt }], stream: true, max_tokens: maxTokens, session_id: sid, reasoning_effort: "high" },
-      stream: true,
-    });
-    if (!res.ok) {
-      let txt = "";
-      try { txt = await res.text(); } catch {}
-      const label = res.status === 401 ? "鉴权失败" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
-      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: res.totalMs, tps: null, charsPerSec: null, tokens: null };
-    }
-    for await (const ev of res.stream()) {
-      try { content += sseContent(JSON.parse(ev)); } catch {}
-    }
-    ttfbMs = res.ttfbMs;
-    totalMs = res.totalMs;
-    const chars = content.length;
-    const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
-    return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
-  } catch (e) {
-    const msg = e?.message || String(e);
-    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
-  }
-}
+import { clineHeaders } from "../providers/cline/headers.js";
+import { computeMetrics } from "../metrics.js";
+import { createTransport } from "../transport/index.js";
+
+function sseContent(obj) {
+  const c = obj?.choices?.[0]?.delta?.content || obj?.choices?.[0]?.message?.content || "";
+  return typeof c === "string" ? c : "";
+}
+
+export async function clineBenchOne({ baseUrl, model, accessToken, prompt, maxTokens, timeoutMs, fetchImpl }) {
+  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
+  let ttfbMs = null;
+  let totalMs = null;
+  let content = "";
+  try {
+    const sid = `sess_bench_${Date.now()}`;
+    const res = await tr.request({
+      url: `${baseUrl}/api/v1/chat/completions`,
+      method: "POST",
+      headers: { ...clineHeaders(sid, accessToken), Accept: "text/event-stream" },
+      body: { model, messages: [{ role: "user", content: prompt }], stream: true, max_tokens: maxTokens, session_id: sid, reasoning_effort: "high" },
+      stream: true,
+    });
+    if (!res.ok) {
+      let txt = "";
+      try { txt = await res.text(); } catch {}
+      const label = res.status === 401 ? "鉴权失败" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
+      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: res.totalMs, tps: null, charsPerSec: null, tokens: null };
+    }
+    for await (const ev of res.stream()) {
+      try { content += sseContent(JSON.parse(ev)); } catch {}
+    }
+    ttfbMs = res.ttfbMs;
+    totalMs = res.totalMs;
+    const chars = content.length;
+    const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
+    return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
+  } catch (e) {
+    const msg = e?.message || String(e);
+    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+  }
+}