dsh-free-search 0.9.3 → 0.9.5

package/README.md

@@ -260,6 +260,25 @@ Error: configured web provider "ddg" is not registered
 
 作用于 **Bing**（`adlt`）、**DuckDuckGo HTML**（`adlt`）、**DuckDuckGo Lite**（`adlt`）；其它引擎没有对应参数，不受影响。设置页有下拉框，也可直接在 config 里写 `safeSearch: moderate`。
 
+#### fetch provider 接管（与其它 fetch 插件冲突）
+
+插件运行时会尝试接管 `web.fetchProvider`：
+
+- 未设置、或仍是出厂默认 `http` → 接管为 `fs-quality-fetch`（直连 → exa contents → firecrawl → crawl4ai）；
+- 已被**显式指向其它 provider**（例如 `playwright`）→ **不抢占**，只记一条日志（进程内只打一次）。
+
+如果你同时装了也会注册 fetch provider 的插件（例如 `dsh-web-fetch-playwright`、`dsh-chatgpt-subscription`），**请在 profile 的 `cordis.patch.yml` 里明确让一方退出**：
+
+```yaml
+- id: web
+  config:
+    fetchProvider: playwright   # 用别人的 fetch；本插件只接管 search
+```
+
+否则启动阶段可能出现「互相写 `web` config → web 服务重建 → 插件重复 apply」的循环（见 issue #82）。
+
+`searchProvider` 的规则相同：未设置或仍是 `deepseek-official` 时接管，显式指定其它 provider 时不抢占。
+
 #### 配置文件
 
 DSH 0.1.7-rc.1 起，配置跟随 profile 的插件条目保存：设置页与 `/free-search-engine` 都会写入当前 profile 的 `cordis.patch.yml` 中 `web-search-free`（`dsh-free-search`）条目的 `config`。
@@ -436,7 +455,7 @@ Search engine test:
 
 #### 离线自检断言（tools/）
 
-仓库自带 **17 个**零依赖断言脚本（合计 **501 项**），**不需要 API key、不联网、不需要测试框架**，Node ≥ 20 即可跑：
+仓库自带 **17 个**零依赖断言脚本（合计 **504 项**），**不需要 API key、不联网、不需要测试框架**，Node ≥ 20 即可跑：
 
 | 脚本 | 覆盖 | 运行（项数） |
 |---|---|---|
@@ -456,7 +475,7 @@ Search engine test:
 | `assert-killtree-win32.mjs` | Windows 子进程树终止（#74）：不再依赖 `taskkill` | `node tools/assert-killtree-win32.mjs`（8） |
 | `assert-killtree-portable.mjs` | 跨平台 `killTree` 便携版契约 | `node tools/assert-killtree-portable.mjs`（5） |
 | `assert-parallel-mcp.mjs` | Parallel MCP 引擎：25 s 超时/调用方取消覆盖**响应体读取**、共享 SSE 解析器（通知/多行帧）、6 类错误路径、REST 精确时间过滤 | `node tools/assert-parallel-mcp.mjs`（15） |
-| `assert-provider-takeover.mjs` | provider 接管：旧属性 API 与 rc.2 的 `config.*` ref 两套兼容、`settings.mutate(web)` 单批写入、显式指定的 provider 不被抢占 | `node tools/assert-provider-takeover.mjs`（6） |
+| `assert-provider-takeover.mjs` | provider 接管：旧属性 API 与 rc.2 的 `config.*` ref 两套兼容、`settings.mutate(web)` 单批写入、显式指定的 provider 不被抢占、apply 重入/重复注册不再中断、重复日志去重 | `node tools/assert-provider-takeover.mjs`（9） |
 
 任何一个失败会以非 0 退出码结束并列出失败项；把某个旧版本文件路径作为参数传进去，可以看到修复前必红（用来证明断言不是空转）。改动发布相关逻辑后建议先跑一遍再发版。
 
@@ -737,6 +756,25 @@ Cooldowns are per-process (reloading the plugin or restarting clears them) and a
 
 It applies to **Bing** (`adlt`), **DuckDuckGo HTML** (`adlt`), and **DuckDuckGo Lite** (`adlt`). Other engines have no such parameter and are unaffected. There is a dropdown in the settings page; you can also write `safeSearch: moderate` in the config.
 
+#### Fetch provider takeover (conflicts with other fetch plugins)
+
+At runtime the plugin tries to take over `web.fetchProvider`:
+
+- unset, or still the factory default `http` → taken over as `fs-quality-fetch` (direct → exa contents → firecrawl → crawl4ai);
+- **explicitly pointed at another provider** (e.g. `playwright`) → **not stolen**, one log line only (once per process).
+
+If you also run a plugin that registers a fetch provider (`dsh-web-fetch-playwright`, `dsh-chatgpt-subscription`, …), **make one of them step aside** in the profile's `cordis.patch.yml`:
+
+```yaml
+- id: web
+  config:
+    fetchProvider: playwright   # use theirs; this plugin only takes over search
+```
+
+Otherwise startup can loop on "write the `web` config → the web service rebuilds → the plugin re-applies" (see issue #82).
+
+`searchProvider` follows the same rule: taken over when unset or still `deepseek-official`, never when explicitly pointed elsewhere.
+
 #### Configuration File
 
 Since DSH 0.1.7-rc.1 the configuration is stored with the profile's plugin entry: the settings page and `/free-search-engine` both write the `config` of the `web-search-free` (`dsh-free-search`) entry in the active profile's `cordis.patch.yml`.
@@ -913,7 +951,7 @@ Restart `dsh web` after switching to apply changes.
 
 #### Offline self-check assertions (`tools/`)
 
-**Seventeen** zero-dependency assertion scripts ship with the repo (**501 assertions** total): **no API keys, no network, no test framework**, Node ≥ 20:
+**Seventeen** zero-dependency assertion scripts ship with the repo (**504 assertions** total): **no API keys, no network, no test framework**, Node ≥ 20:
 
 | Script | Covers | Run (count) |
 |---|---|---|
@@ -933,7 +971,7 @@ Restart `dsh web` after switching to apply changes.
 | `assert-killtree-win32.mjs` | Windows process-tree termination (#74): no longer relies on `taskkill` | `node tools/assert-killtree-win32.mjs` (8) |
 | `assert-killtree-portable.mjs` | portable `killTree` contract | `node tools/assert-killtree-portable.mjs` (5) |
 | `assert-parallel-mcp.mjs` | Parallel MCP engine: the 25 s timeout / caller abort covers the **body read**, shared SSE parser (notifications, multiline frames), six error paths, REST precise time filtering | `node tools/assert-parallel-mcp.mjs` (15) |
-| `assert-provider-takeover.mjs` | provider takeover: both the legacy property API and the rc.2 `config.*` refs, a single batched `settings.mutate(web)`, and explicit foreign providers left untouched | `node tools/assert-provider-takeover.mjs` (6) |
+| `assert-provider-takeover.mjs` | provider takeover: both the legacy property API and the rc.2 `config.*` refs, a single batched `settings.mutate(web)`, explicit foreign providers left untouched, apply re-entry / duplicate registrations tolerated, and repeated logs deduplicated | `node tools/assert-provider-takeover.mjs` (9) |
 
 Any failure exits non-zero with the failing items listed. Passing the path of an older file shows the pre-fix failures (proving the assertions are not vacuous). Worth running before a release.
 

package/lib/client.js

@@ -388,7 +388,7 @@ window.__ModuleLoader__.load({
     };
     const tt = (lang) => I18N[lang === "en" ? "en" : "zh"];
     // 当前插件版本（与 lib/index.js 的 PLUGIN_VERSION 保持一致；跟随上游 v0.7.5）
-    const PLUGIN_VERSION = "0.9.3";
+    const PLUGIN_VERSION = "0.9.5";
     const ENGINES = [
       { id: "auto", label: "Auto · 智能路由", badge: "AUTO", link: null },
       { id: "multi", label: "Multi Search · 多引擎", badge: "MULTI", link: null },

package/lib/index.js

@@ -88,6 +88,17 @@ const MARKET_TO_LANG = {
 // declares), not the old `$DSH_HOME/settings.yaml` section name.
 const FREE_SEARCH_NS = "web-search-free";
 const BRIDGE_PREFIX = "/api/dsh-free-search-settings";
+// apply() 重入保护：见 apply 顶部的注释。
+const appliedContexts = new WeakSet();
+// 同一条接管/冲突日志在一个进程里只打一次：rc.2 的启动重入会把同样的 warn 刷到 65ms 一条，
+// 日志噪音会盖住真正的问题（#82）。
+const loggedOnce = new Set();
+const logOnce = (log, level, message) => {
+  if (loggedOnce.has(message)) return;
+  loggedOnce.add(message);
+  const sink = level === "warn" ? log?.warn : log?.info;
+  if (typeof sink === "function") sink.call(log, message);
+};
 // 意见 A：zhihu_global / zhihu_site 需要 ZHIHU_API_KEY，不属于免 key 引擎（不得进入 FREE_ENGINES）
 const FREE_ENGINES = ["ddg", "ddg-lite", "bing", "searxng", "anysearch"];
 const ALL_ENGINES = ["ddg", "ddg-lite", "bing", "searxng", "anysearch", "exa", "tavily", "keenable", "firecrawl", "parallel", "perplexity", "serpbase", "serply", "deepseek-official", "deepseek-tools", "you", "baidu", "kimi", "aliyun", "doubao", "tinyfish", "jina", "zhihu_global", "zhihu_site", "openai", "gemini", "claude"];
@@ -245,7 +256,7 @@ function routeEngines(query, { timeRange, disabledEngines = [], fallbackOrder =
 }
 
 // 当前插件版本（发布时与 package.json 同步；跟随上游 v0.7.5，不在 PR 里 bump）
-const PLUGIN_VERSION = "0.9.3";
+const PLUGIN_VERSION = "0.9.5";
 // 检查更新的 npm registry 元数据地址（dsh-free-search 是 npmjs 上的公开包）
 const NPM_REGISTRY_URL = "https://registry.npmjs.org/dsh-free-search/latest";
 const PLUGIN_NPM_URL = "https://www.npmjs.com/package/dsh-free-search";
@@ -3420,6 +3431,26 @@ function resolveConfig(config) {
 }
 
 function apply(ctx, config) {
+  // 重入保护：同一个 ctx 上 apply 只跑一次。rc.2 的 boot 里出现过未 dispose 的重入，
+  // 会撞上下面的重复注册（provider / route），导致 entry 反复失败把 boot 拖死。
+  if (appliedContexts.has(ctx)) {
+    ctx.logger?.warn?.("free-search: apply() already ran on this context — skipping the re-entry");
+    return;
+  }
+  appliedContexts.add(ctx);
+  // 注册类调用统一做幂等：已存在 / 重复就跳过并记日志，而不是抛错中断 apply。
+  const registerOnce = (call, provider, label) => {
+    try {
+      return call();
+    } catch (error) {
+      const message = String(error?.message ?? error);
+      if (error?.code === "WEB_DUPLICATE_PROVIDER" || /already registered|duplicate/i.test(message)) {
+        ctx.logger?.warn?.(`free-search: ${label} "${provider.id}" is already registered — reusing it`);
+        return () => {};
+      }
+      throw error;
+    }
+  };
   const current = () => resolveConfig(config);
   const logger = ctx.logger;
   // credentials 服务可能晚于本插件挂载（跨 bundle 顺序），运行期动态获取而不是 apply 时缓存
@@ -3916,6 +3947,15 @@ function apply(ctx, config) {
     });
     // Re-render the dynamic system-prompt engine list whenever this entry's config changes.
     sctx.on("settings/document-updated", (ns) => {
+      // 诊断 #82：web 的 provider 选择被写入会让 web 服务重建、依赖它的插件重新 apply。
+      // 记一条（每进程一次），方便判断启动阶段的重入是谁触发的。
+      if (String(ns) === "web") {
+        logOnce(
+          logger,
+          "info",
+          "free-search: the web config was updated — a provider write rebuilds the web service and re-applies this plugin"
+        );
+      }
       if (String(ns) === FREE_SEARCH_NS && typeof refreshPrompt === "function") refreshPrompt();
     });
   });
@@ -3932,7 +3972,7 @@ function apply(ctx, config) {
           return runEngineTest(engine, query, timeRange);
         },
         getCredentials
-      ).map((route) => sctx.webServer.register(route));
+      ).map((route) => registerOnce(() => sctx.webServer.register(route), { id: route.path }, "bridge route"));
       return () => {
         for (const dispose of disposers) dispose();
       };
@@ -3963,13 +4003,13 @@ function apply(ctx, config) {
     else web.fetchProviderId = id;
   };
 
-  ctx.web.registerSearchProvider(provider);
+  registerOnce(() => ctx.web.registerSearchProvider(provider), provider, "search provider");
 
   // web_fetch 质量链接管：注册 fs-quality-fetch（直连 → exa contents → firecrawl → crawl4ai），
   // 未设置 fetchProviderId、或仍是出厂默认 http 时接管；已被显式指向其他 provider 时不抢占。
   // 与 web_search 的接管规则同构：运行时兜底，不依赖 bundle patch 是否落地。
   const fetchProvider = createFetchProvider({ resolveApiKey, logger });
-  ctx.web.registerFetchProvider(fetchProvider);
+  registerOnce(() => ctx.web.registerFetchProvider(fetchProvider), fetchProvider, "fetch provider");
 
   // 接管必须先等 settings 服务就绪：rc.2 起 provider 选择是 config 的 volatile ref,
   // 只能通过 settings.mutate 写；apply 期间 settings 尚未挂载，直接取会永远拿不到。
@@ -3980,13 +4020,17 @@ function apply(ctx, config) {
     const fetchSelection = providerSelection(ctx.web, "fetchProvider");
     if (!fetchSelection.current || fetchSelection.current === "http") {
       takeOverOps(ctx.web, "fetchProvider", fetchProvider.id, ops);
-      logger.info(
+      logOnce(
+        logger,
+        "info",
         fetchSelection.current === undefined || fetchSelection.current === null
           ? `free-search: web.fetchProvider was unset, taking over as "${fetchProvider.id}" (direct → exa → firecrawl → crawl4ai)`
           : `free-search: taking over from the default fetch provider "${fetchSelection.current}" as "${fetchProvider.id}" (direct → exa → firecrawl → crawl4ai)`
       );
     } else if (fetchSelection.current !== fetchProvider.id) {
-      logger.warn(
+      logOnce(
+        logger,
+        "warn",
         `free-search: fetch is served by provider "${fetchSelection.current}", not "${fetchProvider.id}" — web_fetch keeps its current provider. ` +
           `To use this plugin's fetch chain, set this in the profile's cordis.patch.yml:\n` +
           `- id: web\n  config:\n    fetchProvider: ${fetchProvider.id}`
@@ -4002,13 +4046,17 @@ function apply(ctx, config) {
     const searchSelection = providerSelection(ctx.web, "searchProvider");
     if (!searchSelection.current || searchSelection.current === OFFICIAL_SEARCH_PROVIDER_ID) {
       takeOverOps(ctx.web, "searchProvider", provider.id, ops);
-      logger.info(
+      logOnce(
+        logger,
+        "info",
         searchSelection.current === undefined || searchSelection.current === null
           ? `free-search: web.searchProvider was unset, taking over as "${provider.id}"`
           : `free-search: taking over from the default provider "${searchSelection.current}" as "${provider.id}"`
       );
     } else if (searchSelection.current !== provider.id) {
-      logger.warn(
+      logOnce(
+        logger,
+        "warn",
         `free-search: search is served by provider "${searchSelection.current}", not "${provider.id}" — web_search will NOT use this plugin. ` +
           `Add this to the profile's cordis.patch.yml (patch replaces the whole entry config, so keep any other web.* fields):\n` +
           `- id: web\n  config:\n    searchProvider: ${provider.id}\n    fetchProvider: http`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-free-search",
-  "version": "0.9.3",
+  "version": "0.9.5",
   "description": "Free web search for DeepSeek Harness: 27 engines (Bing/DuckDuckGo/AnySearch/SearXNG/Exa/Tavily/Keenable/Firecrawl/Parallel keyless; Zhihu global/site via MCP and Perplexity/SerpBase/DeepSeek (official + native tools/search)/You.com/Baidu/Kimi/Aliyun/Doubao with key; OpenAI/Gemini/Claude model built-in search, explicit-only) + auto smart routing + global engine enable/disable & fallback priority + editable zh/en language pools + multi mode (parallel full language-pool merge) + multi_search cross-source tool + failure-aware fallback (quota/auth cooldown, transient retry, fallbackOn) + time filtering + platform search + domain blocklist (blockedDomains) + web_fetch, with web settings UI.",
   "type": "module",
   "main": "lib/index.js",