mslxdff 0.1.236 → 0.1.238

package/src/models.js

@@ -1,294 +1,294 @@
-const KNOWN_FREE_OPENCODE_MODELS = ["big-pickle"];
-const CACHE_TTL_MS = 2 * 60 * 60 * 1000;
-const DEFAULT_REFRESH_MS = 2 * 60 * 60 * 1000;
-import { mkdirSync, writeFileSync, readFileSync } from "node:fs";
-import { dirname } from "node:path";
-import { compatFetch } from "./compat.js";
-
-export function isFreeModel(id) {
-  return (typeof id === "string" && id.endsWith("-free")) ||
-    KNOWN_FREE_OPENCODE_MODELS.includes(id);
-}
-
-export function filterFreeModels(list) {
-  const seen = new Set();
-  const out = [];
-  for (const m of list || []) {
-    if (!(m && m.id)) continue;
-    if (!isFreeModel(m.id)) continue;
-    if (seen.has(m.id)) continue;
-    seen.add(m.id);
-    out.push(m);
-  }
-  return out;
-}
-
-export function createModelsService({ baseUrl, headers, ttlMs = CACHE_TTL_MS, refreshMs = DEFAULT_REFRESH_MS, cacheFile, providers } = {}) {
-  // 多供应商聚合模式：给 `providers` 数组，则缓存并返回所有供应商的模型列表
-  if (providers?.length) {
-    let aggregate = null;
-    let aggFetchedAt = 0;
-    let aggInflight = null;
-    let aggTimer = null;
-    let lastAllowlistKey = "";
-    // 冷启动预热：读上一次 persist 的磁盘目录当首值（真机取证：全目录拉取烧 20-26s，
-    // 首个 candidatesFor 冷等全程）。aggFetchedAt 用磁盘记录的真实时间；get 对
-    // 「有旧值」的情形一律 stale-while-revalidate（立即返回 + 后台刷），主请求零等待。
-    if (cacheFile) {
-      try {
-        const raw = JSON.parse(readFileSync(cacheFile, "utf8"));
-        if (Array.isArray(raw?.data)) {
-          aggregate = { object: "list", data: raw.data };
-          aggFetchedAt = Number(raw.cachedAt) || 0;
-        }
-      } catch { /* 无缓存/坏档：按冷处理 */ }
-    }
-    async function currentAllowlistKey() {
-      try {
-        const { loadProviderAllowedModels, loadProviderAllowAnyModels } = await import("./state.js");
-        return providers.map((p) => {
-          const allowAny = loadProviderAllowAnyModels(p.id) ? "any" : "block";
-          return `${p.id}:${allowAny}:${loadProviderAllowedModels(p.id).join(",")}`;
-        }).join("|");
-      } catch {
-        return "";
-      }
-    }
-    async function aggLoad() {
-      const all = [];
-      // 单供应商拉取失败时沿用旧缓存里它的段（取证教训：此前失败=整段消失，
-      // 且 2h TTL 内无人再拉（startAutoRefresh 曾是空函数），目录会长期缺一大块）。
-      const prevByProvider = new Map();
-      for (const m of aggregate?.data || []) {
-        const pid = typeof m?.id === "string" ? m.id.split("/")[0] : "";
-        if (!pid) continue;
-        if (!prevByProvider.has(pid)) prevByProvider.set(pid, []);
-        prevByProvider.get(pid).push(m);
-      }
-      for (const p of providers) {
-        try {
-          let list = (await p.listModels?.()) ?? [];
-          // 白名单过滤：空名单且 allowAny=false → 该供应商不暴露任何模型（安全默认全拦，auto 直接跳过该供应商）
-          try {
-            const { loadProviderAllowedModels, loadProviderAllowAnyModels } = await import("./state.js");
-            const allowed = loadProviderAllowedModels(p.id);
-            const allowAny = loadProviderAllowAnyModels(p.id);
-            if (!allowed.length && !allowAny) {
-              // 该供应商被全拦，auto 直接跳过整个供应商
-              continue;
-            }
-            if (allowed.length) {
-              const allowedSet = new Set(allowed);
-              const { splitModelId } = await import("./providers/model-id.js");
-              list = list.filter((m) => {
-                if (!m || !m.id) return false;
-                const { raw } = splitModelId(m.id, providers.map((x) => x.id));
-                return allowedSet.has(String(raw || "").trim());
-              });
-            }
-          } catch {}
-          all.push(...list);
-        } catch {
-          // 单供应商取数失败不拖垮整体：有旧数据则保留旧数据
-          const prev = prevByProvider.get(p.id);
-          if (prev?.length) all.push(...prev);
-        }
-      }
-      aggregate = { object: "list", data: all };
-      try {
-        lastAllowlistKey = await currentAllowlistKey();
-      } catch {}
-      if (cacheFile) persistModels(aggregate, cacheFile);
-      return aggregate;
-    }
-    // 对当前 aggregate（含磁盘 stale 值）做白名单二次过滤后返回；过滤失败回原值。
-    async function aggSnapshot() {
-      try {
-        const { loadProviderAllowedModels, loadProviderAllowAnyModels } = await import("./state.js");
-        const { splitModelId } = await import("./providers/model-id.js");
-        const filteredData = aggregate.data.filter((m) => {
-          if (!m || !m.id) return false;
-          const { provider, raw } = splitModelId(m.id, providers.map((x) => x.id));
-          const allowed = loadProviderAllowedModels(provider);
-          const allowAny = loadProviderAllowAnyModels(provider);
-          if (!allowed.length && !allowAny) return false;
-          if (!allowed.length) return true;
-          return allowed.includes(String(raw || "").trim());
-        });
-        if (filteredData.length !== aggregate.data.length) return { ...aggregate, data: filteredData };
-        return aggregate;
-      } catch {
-        return aggregate;
-      }
-    }
-    function aggRefreshInBackground() {
-      void aggLoadShared().then(() => { aggFetchedAt = Date.now(); }).catch(() => {});
-    }
-    async function get() {
-      const now = Date.now();
-      const fresh = aggregate && now - aggFetchedAt < ttlMs;
-      if (fresh) {
-        // 热更新白名单：allowlist 变化时，即使命中 TTL 也要重载（否则 clear 后仍返回旧过滤结果）
-        try {
-          const curKey = await currentAllowlistKey();
-          if (curKey !== lastAllowlistKey) {
-            const out = await aggLoadShared();
-            aggFetchedAt = Date.now();
-            return out;
-          }
-        } catch {}
-        return aggSnapshot();
-      }
-      // 有旧值（含冷启动磁盘 stale）：立即返回 + 后台刷新，主请求零等待。
-      // 取证教训：全目录拉取 20-26s，此前冷路径同步阻塞整个首请求。
-      if (aggregate) {
-        aggRefreshInBackground();
-        return aggSnapshot();
-      }
-      // 完全无值（首次冷跑）：只能等这一趟。
-      const out = await aggLoadShared();
-      aggFetchedAt = Date.now();
-      return out;
-    }
-    // 并发去重：aggLoad 要遍历全部供应商拉目录（真机实测 20-26s），
-    // 无去重时赛马/后台任务并发触发就各烧各的（真机日志实证同窗口重复 3 次）。
-    function aggLoadShared() {
-      if (aggInflight) return aggInflight;
-      aggInflight = (async () => {
-        try {
-          return await aggLoad();
-        } catch (e) {
-          // 失败保旧：有旧聚合就回退旧值，绝不让半次刷新把目录清空
-          if (aggregate) return aggregate;
-          throw e;
-        } finally {
-          aggInflight = null;
-        }
-      })();
-      return aggInflight;
-    }
-    function aggStartAutoRefresh(intervalMs = refreshMs) {
-      if (aggTimer) return aggStopAutoRefresh;
-      aggTimer = setInterval(() => {
-        void aggLoadShared().then(() => { aggFetchedAt = Date.now(); }).catch(() => {});
-      }, intervalMs);
-      aggTimer.unref?.();
-      return aggStopAutoRefresh;
-    }
-    function aggStopAutoRefresh() {
-      if (aggTimer) { clearInterval(aggTimer); aggTimer = null; }
-    }
-    return { get, startAutoRefresh: aggStartAutoRefresh, stopAutoRefresh: aggStopAutoRefresh };
-  }
-  let cache = null;
-  let fetchedAt = 0;
-  let inflight = null;
-  let timer = null;
-
-  async function load() {
-    let data = await fetchUpstreamModels({ baseUrl, headers });
-    // 白名单过滤：opencode 亦支持 allowlist
-    try {
-      const { loadProviderAllowedModels } = await import("./state.js");
-      const allowed = loadProviderAllowedModels("opencode");
-      if (allowed.length) {
-        const allowedSet = new Set(allowed);
-        data = { ...data, data: (data.data || []).filter((m) => m && m.id && allowedSet.has(String(m.id).trim())) };
-      }
-    } catch {}
-    cache = data;
-    fetchedAt = Date.now();
-    if (cacheFile) persistModels(data, cacheFile);
-    return data;
-  }
-
-  async function get() {
-    const now = Date.now();
-    if (cache && now - fetchedAt < ttlMs) return cache;
-    if (inflight) return inflight;
-
-    inflight = (async () => {
-      try {
-        return await load();
-      } catch (err) {
-        // serve stale on failure if we have it, else rethrow
-        if (cache) return cache;
-        throw err;
-      } finally {
-        inflight = null;
-      }
-    })();
-    return inflight;
-  }
-
-  function startAutoRefresh(intervalMs = refreshMs) {
-    if (timer) return stopAutoRefresh;
-    timer = setInterval(() => {
-      void load().catch(() => {
-        // keep serving stale cache on background refresh failure
-      });
-    }, intervalMs);
-    timer.unref?.();
-    return stopAutoRefresh;
-  }
-
-  function stopAutoRefresh() {
-    if (timer) {
-      clearInterval(timer);
-      timer = null;
-    }
-  }
-
-  return { get, startAutoRefresh, stopAutoRefresh };
-}
-
-async function fetchUpstreamModels({ baseUrl, headers, connectTimeoutMs = 30_000 }) {
-  const url = `${baseUrl}/zen/v1/models`;
-  for (let attempt = 0; ; attempt++) {
-    const res = await attemptFetch(url, headers, connectTimeoutMs);
-    if (res instanceof Error) {
-      if (attempt < NETWORK_RETRIES) continue;
-      throw res;
-    }
-    if (isRetryable(res.status) && attempt < STATUS_RETRIES) {
-      await sleep(2000);
-      continue;
-    }
-    if (!res.ok) throw new Error(`models fetch failed: HTTP ${res.status}`);
-    const json = await res.json().catch(() => ({}));
-    const raw = Array.isArray(json) ? json : json.data ?? json.models ?? [];
-    return { object: "list", data: filterFreeModels(raw) };
-  }
-}
-
-async function attemptFetch(url, headers, connectTimeoutMs) {
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(), connectTimeoutMs);
-  try {
-    return await compatFetch(url, { headers, signal: controller.signal });
-  } catch (err) {
-    return err;
-  } finally {
-    clearTimeout(timer);
-  }
-}
-
-function isRetryable(status) {
-  return status === 429 || status === 502 || status === 503 || status === 504;
-}
-
-function persistModels(data, cacheFile) {
-  try {
-    mkdirSync(dirname(cacheFile), { recursive: true });
-    writeFileSync(cacheFile, JSON.stringify({ cachedAt: Date.now(), ...data }));
-  } catch {
-    // persistence is best-effort
-  }
-}
-
-function sleep(ms) {
-  return new Promise((r) => setTimeout(r, ms));
-}
-
-const NETWORK_RETRIES = 2;
+const KNOWN_FREE_OPENCODE_MODELS = ["big-pickle"];
+const CACHE_TTL_MS = 2 * 60 * 60 * 1000;
+const DEFAULT_REFRESH_MS = 2 * 60 * 60 * 1000;
+import { mkdirSync, writeFileSync, readFileSync } from "node:fs";
+import { dirname } from "node:path";
+import { compatFetch } from "./compat.js";
+
+export function isFreeModel(id) {
+  return (typeof id === "string" && id.endsWith("-free")) ||
+    KNOWN_FREE_OPENCODE_MODELS.includes(id);
+}
+
+export function filterFreeModels(list) {
+  const seen = new Set();
+  const out = [];
+  for (const m of list || []) {
+    if (!(m && m.id)) continue;
+    if (!isFreeModel(m.id)) continue;
+    if (seen.has(m.id)) continue;
+    seen.add(m.id);
+    out.push(m);
+  }
+  return out;
+}
+
+export function createModelsService({ baseUrl, headers, ttlMs = CACHE_TTL_MS, refreshMs = DEFAULT_REFRESH_MS, cacheFile, providers } = {}) {
+  // 多供应商聚合模式：给 `providers` 数组，则缓存并返回所有供应商的模型列表
+  if (providers?.length) {
+    let aggregate = null;
+    let aggFetchedAt = 0;
+    let aggInflight = null;
+    let aggTimer = null;
+    let lastAllowlistKey = "";
+    // 冷启动预热：读上一次 persist 的磁盘目录当首值（真机取证：全目录拉取烧 20-26s，
+    // 首个 candidatesFor 冷等全程）。aggFetchedAt 用磁盘记录的真实时间；get 对
+    // 「有旧值」的情形一律 stale-while-revalidate（立即返回 + 后台刷），主请求零等待。
+    if (cacheFile) {
+      try {
+        const raw = JSON.parse(readFileSync(cacheFile, "utf8"));
+        if (Array.isArray(raw?.data)) {
+          aggregate = { object: "list", data: raw.data };
+          aggFetchedAt = Number(raw.cachedAt) || 0;
+        }
+      } catch { /* 无缓存/坏档：按冷处理 */ }
+    }
+    async function currentAllowlistKey() {
+      try {
+        const { loadProviderAllowedModels, loadProviderAllowAnyModels } = await import("./state.js");
+        return providers.map((p) => {
+          const allowAny = loadProviderAllowAnyModels(p.id) ? "any" : "block";
+          return `${p.id}:${allowAny}:${loadProviderAllowedModels(p.id).join(",")}`;
+        }).join("|");
+      } catch {
+        return "";
+      }
+    }
+    async function aggLoad() {
+      const all = [];
+      // 单供应商拉取失败时沿用旧缓存里它的段（取证教训：此前失败=整段消失，
+      // 且 2h TTL 内无人再拉（startAutoRefresh 曾是空函数），目录会长期缺一大块）。
+      const prevByProvider = new Map();
+      for (const m of aggregate?.data || []) {
+        const pid = typeof m?.id === "string" ? m.id.split("/")[0] : "";
+        if (!pid) continue;
+        if (!prevByProvider.has(pid)) prevByProvider.set(pid, []);
+        prevByProvider.get(pid).push(m);
+      }
+      for (const p of providers) {
+        try {
+          let list = (await p.listModels?.()) ?? [];
+          // 白名单过滤：空名单且 allowAny=false → 该供应商不暴露任何模型（安全默认全拦，auto 直接跳过该供应商）
+          try {
+            const { loadProviderAllowedModels, loadProviderAllowAnyModels } = await import("./state.js");
+            const allowed = loadProviderAllowedModels(p.id);
+            const allowAny = loadProviderAllowAnyModels(p.id);
+            if (!allowed.length && !allowAny) {
+              // 该供应商被全拦，auto 直接跳过整个供应商
+              continue;
+            }
+            if (allowed.length) {
+              const allowedSet = new Set(allowed);
+              const { splitModelId } = await import("./providers/model-id.js");
+              list = list.filter((m) => {
+                if (!m || !m.id) return false;
+                const { raw } = splitModelId(m.id, providers.map((x) => x.id));
+                return allowedSet.has(String(raw || "").trim());
+              });
+            }
+          } catch {}
+          all.push(...list);
+        } catch {
+          // 单供应商取数失败不拖垮整体：有旧数据则保留旧数据
+          const prev = prevByProvider.get(p.id);
+          if (prev?.length) all.push(...prev);
+        }
+      }
+      aggregate = { object: "list", data: all };
+      try {
+        lastAllowlistKey = await currentAllowlistKey();
+      } catch {}
+      if (cacheFile) persistModels(aggregate, cacheFile);
+      return aggregate;
+    }
+    // 对当前 aggregate（含磁盘 stale 值）做白名单二次过滤后返回；过滤失败回原值。
+    async function aggSnapshot() {
+      try {
+        const { loadProviderAllowedModels, loadProviderAllowAnyModels } = await import("./state.js");
+        const { splitModelId } = await import("./providers/model-id.js");
+        const filteredData = aggregate.data.filter((m) => {
+          if (!m || !m.id) return false;
+          const { provider, raw } = splitModelId(m.id, providers.map((x) => x.id));
+          const allowed = loadProviderAllowedModels(provider);
+          const allowAny = loadProviderAllowAnyModels(provider);
+          if (!allowed.length && !allowAny) return false;
+          if (!allowed.length) return true;
+          return allowed.includes(String(raw || "").trim());
+        });
+        if (filteredData.length !== aggregate.data.length) return { ...aggregate, data: filteredData };
+        return aggregate;
+      } catch {
+        return aggregate;
+      }
+    }
+    function aggRefreshInBackground() {
+      void aggLoadShared().then(() => { aggFetchedAt = Date.now(); }).catch(() => {});
+    }
+    async function get() {
+      const now = Date.now();
+      const fresh = aggregate && now - aggFetchedAt < ttlMs;
+      if (fresh) {
+        // 热更新白名单：allowlist 变化时，即使命中 TTL 也要重载（否则 clear 后仍返回旧过滤结果）
+        try {
+          const curKey = await currentAllowlistKey();
+          if (curKey !== lastAllowlistKey) {
+            const out = await aggLoadShared();
+            aggFetchedAt = Date.now();
+            return out;
+          }
+        } catch {}
+        return aggSnapshot();
+      }
+      // 有旧值（含冷启动磁盘 stale）：立即返回 + 后台刷新，主请求零等待。
+      // 取证教训：全目录拉取 20-26s，此前冷路径同步阻塞整个首请求。
+      if (aggregate) {
+        aggRefreshInBackground();
+        return aggSnapshot();
+      }
+      // 完全无值（首次冷跑）：只能等这一趟。
+      const out = await aggLoadShared();
+      aggFetchedAt = Date.now();
+      return out;
+    }
+    // 并发去重：aggLoad 要遍历全部供应商拉目录（真机实测 20-26s），
+    // 无去重时赛马/后台任务并发触发就各烧各的（真机日志实证同窗口重复 3 次）。
+    function aggLoadShared() {
+      if (aggInflight) return aggInflight;
+      aggInflight = (async () => {
+        try {
+          return await aggLoad();
+        } catch (e) {
+          // 失败保旧：有旧聚合就回退旧值，绝不让半次刷新把目录清空
+          if (aggregate) return aggregate;
+          throw e;
+        } finally {
+          aggInflight = null;
+        }
+      })();
+      return aggInflight;
+    }
+    function aggStartAutoRefresh(intervalMs = refreshMs) {
+      if (aggTimer) return aggStopAutoRefresh;
+      aggTimer = setInterval(() => {
+        void aggLoadShared().then(() => { aggFetchedAt = Date.now(); }).catch(() => {});
+      }, intervalMs);
+      aggTimer.unref?.();
+      return aggStopAutoRefresh;
+    }
+    function aggStopAutoRefresh() {
+      if (aggTimer) { clearInterval(aggTimer); aggTimer = null; }
+    }
+    return { get, startAutoRefresh: aggStartAutoRefresh, stopAutoRefresh: aggStopAutoRefresh };
+  }
+  let cache = null;
+  let fetchedAt = 0;
+  let inflight = null;
+  let timer = null;
+
+  async function load() {
+    let data = await fetchUpstreamModels({ baseUrl, headers });
+    // 白名单过滤：opencode 亦支持 allowlist
+    try {
+      const { loadProviderAllowedModels } = await import("./state.js");
+      const allowed = loadProviderAllowedModels("opencode");
+      if (allowed.length) {
+        const allowedSet = new Set(allowed);
+        data = { ...data, data: (data.data || []).filter((m) => m && m.id && allowedSet.has(String(m.id).trim())) };
+      }
+    } catch {}
+    cache = data;
+    fetchedAt = Date.now();
+    if (cacheFile) persistModels(data, cacheFile);
+    return data;
+  }
+
+  async function get() {
+    const now = Date.now();
+    if (cache && now - fetchedAt < ttlMs) return cache;
+    if (inflight) return inflight;
+
+    inflight = (async () => {
+      try {
+        return await load();
+      } catch (err) {
+        // serve stale on failure if we have it, else rethrow
+        if (cache) return cache;
+        throw err;
+      } finally {
+        inflight = null;
+      }
+    })();
+    return inflight;
+  }
+
+  function startAutoRefresh(intervalMs = refreshMs) {
+    if (timer) return stopAutoRefresh;
+    timer = setInterval(() => {
+      void load().catch(() => {
+        // keep serving stale cache on background refresh failure
+      });
+    }, intervalMs);
+    timer.unref?.();
+    return stopAutoRefresh;
+  }
+
+  function stopAutoRefresh() {
+    if (timer) {
+      clearInterval(timer);
+      timer = null;
+    }
+  }
+
+  return { get, startAutoRefresh, stopAutoRefresh };
+}
+
+async function fetchUpstreamModels({ baseUrl, headers, connectTimeoutMs = 30_000 }) {
+  const url = `${baseUrl}/zen/v1/models`;
+  for (let attempt = 0; ; attempt++) {
+    const res = await attemptFetch(url, headers, connectTimeoutMs);
+    if (res instanceof Error) {
+      if (attempt < NETWORK_RETRIES) continue;
+      throw res;
+    }
+    if (isRetryable(res.status) && attempt < STATUS_RETRIES) {
+      await sleep(2000);
+      continue;
+    }
+    if (!res.ok) throw new Error(`models fetch failed: HTTP ${res.status}`);
+    const json = await res.json().catch(() => ({}));
+    const raw = Array.isArray(json) ? json : json.data ?? json.models ?? [];
+    return { object: "list", data: filterFreeModels(raw) };
+  }
+}
+
+async function attemptFetch(url, headers, connectTimeoutMs) {
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(), connectTimeoutMs);
+  try {
+    return await compatFetch(url, { headers, signal: controller.signal });
+  } catch (err) {
+    return err;
+  } finally {
+    clearTimeout(timer);
+  }
+}
+
+function isRetryable(status) {
+  return status === 429 || status === 502 || status === 503 || status === 504;
+}
+
+function persistModels(data, cacheFile) {
+  try {
+    mkdirSync(dirname(cacheFile), { recursive: true });
+    writeFileSync(cacheFile, JSON.stringify({ cachedAt: Date.now(), ...data }));
+  } catch {
+    // persistence is best-effort
+  }
+}
+
+function sleep(ms) {
+  return new Promise((r) => setTimeout(r, ms));
+}
+
+const NETWORK_RETRIES = 2;
 const STATUS_RETRIES = 2;