[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_247":248,"_249":250},"routes/_app.x.$handle.$name",{"_11":12,"_13":14,"_15":-7,"_18":19,"_17":-7,"_187":-5,"_20":188,"_22":189,"_190":191,"_192":193,"_26":194,"_24":195,"_196":-5,"_197":-5,"_198":51,"_199":200,"_201":43,"_202":-5,"_203":-5,"_204":205},"routes/_app.x.$handle.$name.code.$",{"_11":12,"_13":14,"_15":16,"_17":-7,"_18":19,"_20":21,"_22":23,"_24":25,"_26":27,"_28":-5,"_29":-5,"_30":-5,"_31":-5,"_32":-5},"handle","yuxiaoy","name","smzdm","branchName","main","version","path","README.md","project",{"_33":129,"_13":14,"_44":130,"_131":-5,"_132":51,"_133":134,"_135":134,"_136":137,"_138":-5,"_139":140,"_141":-5,"_142":51,"_143":-5,"_144":-5,"_145":146,"_147":148,"_149":150,"_151":152,"_153":154,"_155":-5,"_156":157,"_158":157,"_159":160,"_161":160,"_162":160,"_163":157,"_164":157,"_165":166,"_167":168,"_169":170,"_171":51},"tree",[54,55,56,57,58,59,60,61],"file",{"_33":34,"_13":19,"_35":36,"_37":-5,"_38":39,"_40":41,"_42":43,"_44":45,"_17":46,"_47":24,"_48":49,"_50":51,"_52":-5,"_18":53},"contents",[],"upstream","pull","ahead","behind","envVars","id","01bc8846-4959-426f-84ad-dfe33b9e6c3c","fileId","ee89951e-8f56-11f1-83cc-1607ee4eb77e","parentId","branchId","8f6ec682-59da-11f1-9bcd-ee650bb23af1","branchVersion",46,"latest",true,"createdAt",["D",1786638456173],3,"fileType","content","# SMZDM Telegram Bot\n\n值得买（SMZDM）好价推送机器人，运行在 [Val Town](https://www.val.town) 的 Deno\n运行时上。按关键词规则监控「好价」列表，命中后以富文本卡片推送到 Telegram。\n\n这是对 `yuxiaoy/smzdm` 的重构版本：存储从单个 blob JSON 换成 SQLite，抓取从固定\n5 页换成增量游标翻页，推送从纯文本 `标题: 链接`\n换成带价格/商城/热度/爆料原文/按钮的卡片。\n\n## 文件结构\n\n| 文件 | 类型 | 作用 |\n| ------------ | ------------ | ------------------------------------------ |\n| `shared.ts` | script | 环境变量、可调参数、时间与价格解析等纯函数 |\n| `db.ts` | script | SQLite 表结构、设置存取、推送历史 |\n| `smzdm.ts` | script | 好价接口客户端与增量游标翻页 |\n| `matcher.ts` | script | 规则解析与匹配 |\n| `notify.ts` | script | Telegram 卡片渲染、限流、重试 |\n| `spider.ts` | **interval** | 定时任务入口 |\n| `webhook.ts` | **http** | Telegram 命令入口（无第三方 bot 框架） |\n\n## 部署\n\n1. 在 Val Town 新建一个 val，把上面 7 个文件贴进去，注意把 `spider.ts`\n 的类型设为 **Interval**、`webhook.ts` 设为 **HTTP**，其余保持 Script。\n2. 配置环境变量（Settings → Environment Variables）：\n\n | 变量 | 必填 | 说明 |\n | ---------------- | ---- | ----------------------------------------------------------------------- |\n | `BOT_TOKEN` | 是 | @BotFather 给的 bot token |\n | `CHAT_ID` | 是 | 接收推送的 chat id |\n | `OWNER_IDS` | 建议 | 允许发命令的 Telegram 用户 id，逗号分隔。**留空时机器人会拒绝所有命令** |\n | `WEBHOOK_SECRET` | 否 | webhook 校验密钥，不填则由 token 派生 SHA-256 |\n\n3. 用浏览器打开 `webhook.ts` 的 HTTP 地址一次。GET 请求会注册 webhook\n 和命令菜单，并返回注册结果。之后每次改动 URL 都要再访问一次。\n4. 把 `spider.ts` 的 interval 设为 **10 分钟**。增量翻页在闲时只发 1 个请求，\n 忙时自动加深，所以跑得勤反而更省也更及时。\n5. 在 Telegram 里 `/add_brands 你的关键词` 开始用。\n\n`OWNER_IDS` 如果不填，任何人找到这个 bot 都会被拒绝命令——这是故意的安全默认值。\n你的用户 id 可以问 [@userinfobot](https://t.me/userinfobot)。\n\n## 规则语法\n\n一条规则是不含空格的单个 token，内部用 `-` 分隔多个条件：\n\n| 写法 | 含义 |\n| --------------------------- | ----------------------------- |\n| `apple` | 含 `apple` |\n| `apple-watch` | 同时含 `apple` 和 `watch` |\n| `sony\\|索尼-耳机` | （`sony` 或 `索尼`）且 `耳机` |\n| `apple-watch-!表带-!钢化膜` | 排除配件噪音 |\n| `nike-<300` | 价格不超过 300 |\n| `茅台->1000` | 价格不低于 1000 |\n| `/rtx\\s?40[6-9]0/` | 正则，忽略大小写 |\n| `!/翻新\\|拆机/` | 负向正则 |\n\n约束：\n\n- `-` 是分隔符，所以品牌名里带连字符的要写成正则。\n- 只有排除条件、没有正向条件的规则会被 `/add_brands`\n 拒绝，否则它会命中全站好价。\n- 价格解析不出来的（`包邮`、`1元起`）**不会**被价格条件挡掉，宁可多推不漏推。\n- 中文没有词边界，`米` 这种短词误报率很高，建议配合排除词或正则。\n\n用 `/test 商品标题 199元` 可以看每条规则命中与否以及原因。\n\n## 命令\n\n```\n/help 使用说明\n/brands 查看规则\n/add_brands a b 添加规则\n/delete_brands a 删除规则\n/delete_all_brands 清空规则\n/test 标题测试规则命中情况\n\n/malls 查看商城名单\n/block_mall 唯品会加入黑名单\n/unblock_mall 唯品会移出黑名单\n/only_mall 京东天猫超市只看这些商城（留空取消）\n/min_worthy 0 最低「值」票数\n/max_age 24 只推多久以内发布的好价\n/match_content on|off 是否同时匹配爆料原文\n\n/photo preview|photo|off 配图方式\n/quiet 23 7 静音时段（北京时间），off 关闭\n/drop 5 降价超过 5% 时复推\n/pause /resume 暂停 / 恢复\n/run 立刻抓取一次\n\n/status 运行状态\n/settings 当前配置\n/results [页码] 推送记录\n/delete_results 清空推送记录\n```\n\n推送卡片下方还有三个按钮：去购买（`go.smzdm.com`\n直达商城）、原文、以及一键把该商城加入黑名单。\n\n## 关键实现说明\n\n**为什么用 app 接口。** `www.smzdm.com` 在腾讯 Chaos\n反爬后面，对非浏览器请求返回 JS 挑战页，Val Town 的出口 IP\n拿不到数据。`api.smzdm.com/v1/youhui/list` 是开放 JSON，这个端点不需要鉴权。\n\n**为什么用 `article_date` 而不是 `article_unix_date` 排序。**\n两个字段实测对不上：同一条 `article_date` 为 `00:06:41` 的商品，其\n`article_unix_date` 换算出来差了约 12 小时；两条相隔 7 秒的商品 unix 值差了 25\n分钟。所以时间排序和游标都基于 `article_date`（东八区）。\n\n**为什么改成增量翻页。** 实测好价流 80 条只跨约 35 分钟，且存在 7\n秒一条的爆发。固定抓 100 条约等于 44\n分钟的流量，白天高峰期会静默漏掉中间的商品。现在的做法是：第 0 页先拉一次，\n若还没追上就并行抓后续页；某一页的 `article_id` 已全部见过，或没有任何比时间\n游标更新的条目，就停。上限 20 页。\n\n判断「是否继续翻页」优先看已见 id 滑动窗口（最近 600 条），时间游标只做辅助。\n`offset=0` 会出现运营置顶的老帖（实测见到 2021 年的），用「本页最旧时间」\n会在第一页就误停。翻页步进按接口返回的**原始行数**，不按过滤后的商品数——\n整页都是运营卡时继续往下走，而不是当成列表结束。\n\n后面某一页失败时不推进游标、也不写入已见 id，否则中间那一页会永远漏掉。\n\n**为什么过滤 `article_is_timeout` / `article_is_sold_out`。**\n列表里混有已失效和已售罄的条目，`offset=0`\n第一条就可能是几年前的置顶好价。不过滤就会推送早已过期的「好价」。\n\n**为什么先发送后入库。** 只有 Telegram 确认送达才写入\n`smzdm_deals_v1`；发送失败的条目\n会把游标回退到它之前，下一轮重试。旧版本是先入库再发送且忽略发送错误，一旦被限流，\n那条商品就永久丢失了。\n\n**为什么用 SQLite 而不是 blob。** `article_id` 做主键让去重精确且\nO(1)；每行带时间戳让历史按 30 天滚动过期，而不是像旧版那样攒到 150\n条就整表清空——清空后所有仍在列表里的旧商品会被当成新的重推一遍。SQLite\n也不存在两次 `setJSON` 互相覆盖的丢更新问题。\n\n## 本地测试\n\n测试放在 `tests/`，不需要 Val Town 账号，也不会碰到线上数据库或真实 Telegram：\n`fetch` 全程被桩替换，`std/sqlite` 通过 `tests/import_map.json`\n指向内存里的假实现。`.vtignore` 会让 `vt push` 跳过整个 `tests/` 目录。\n\n```bash\n# 全部用例（含真实接口的集成测试）\nBOT_TOKEN=x:y CHAT_ID=1 OWNER_IDS=42 deno test --no-prompt --allow-env \\\n --allow-net --allow-read --allow-import \\\n --import-map=tests/import_map.json tests/\n\n# 离线时跳过联网用例\nSKIP_LIVE=1 BOT_TOKEN=x:y CHAT_ID=1 OWNER_IDS=42 deno test --no-prompt \\\n --allow-env --allow-read --allow-import \\\n --import-map=tests/import_map.json tests/\n\ndeno check --allow-import *.ts\ndeno lint .\n```\n\n三个环境变量都是占位符。`shared.ts` 在导入时会校验 `BOT_TOKEN` 和 `CHAT_ID`\n存在， `OWNER_IDS=42` 是 `webhook_test.ts` 里那个「机主」的 id。\n\n| 文件 | 覆盖内容 |\n| ----------------- | ---------------------------------------------------------------- |\n| `pure_test.ts` | 价格与日期解析、静音时段、分块、规则 DSL 全部语法 |\n| `notify_test.ts` | 卡片字段、HTML 转义、实体截断、429 重试、400 降级、摘要分块 |\n| `sql_test.ts` | 从 `db.ts` 抽出 SQL 丢给真实 SQLite 校验，并验证降价 upsert 语义 |\n| `smzdm_test.ts` | 翻页步进、运营卡整页、已见 id 停页、并行抓取、后页失败 |\n| `spider_test.ts` | 完整链路：过滤、去重、溢出摘要、失败重试、降价复推、失败告警 |\n| `webhook_test.ts` | 密钥校验、机主白名单、命令与参数校验、回调按钮、GET 注册 |\n| `live_test.ts` | 真实接口：字段契约、游标翻页、时间戳异常、抓取速率 |\n\n写测试时踩到的一个坑：webhook 的 `fetch` 桩要对非 Telegram 请求透传，这样和\n`live_test.ts` 共用 isolate 时才不会把真实接口调用吃掉。\n\n`live_test.ts` 里有两个断言是「设计前提的哨兵」：如果 SMZDM 修好了\n`article_unix_date`，或者不再往列表里塞置顶老帖，它们会失败，提示对应的实现\n可以简化。\n\n## 从旧版本迁移\n\n品牌列表早已在 SQLite 的 `rules` 键里。当前代码不再读取旧的 blob key。\n\n## 已知取舍\n\n- `/min_worthy` 会挡掉刚发布的好价（票数为\n 0），而刚发布的往往是最值得抢的，所以默认 0。\n- 正向关键词始终只匹配标题。`/match_content on` 只让**排除词**同时看爆料原文，\n 避免「凑单苹果」误中 `apple`。默认关。\n- 单轮最多发 8 张卡片，多出来的合并成一条摘要，既不刷屏也不丢条目。\n- 同一商品最多推 3 次（首次 + 降价），防止价格反复波动时刷屏。\n","deleted",false,"label",[19],{"_33":34,"_13":19,"_35":36,"_37":-5,"_38":39,"_40":41,"_42":43,"_44":126,"_17":46,"_47":24,"_48":49,"_50":51,"_52":-5,"_70":127,"_18":128},{"_33":117,"_13":118,"_35":119,"_37":-5,"_38":39,"_40":120,"_42":43,"_44":121,"_17":122,"_47":89,"_48":123,"_50":51,"_52":-5,"_70":124,"_18":125},{"_33":109,"_13":110,"_35":111,"_37":-5,"_38":39,"_40":112,"_42":43,"_44":113,"_17":88,"_47":89,"_48":114,"_50":51,"_52":-5,"_70":115,"_18":116},{"_33":101,"_13":102,"_35":103,"_37":-5,"_38":39,"_40":104,"_42":43,"_44":105,"_17":46,"_47":89,"_48":106,"_50":51,"_52":-5,"_70":107,"_18":108},{"_33":93,"_13":94,"_35":95,"_37":-5,"_38":39,"_40":96,"_42":43,"_44":97,"_17":78,"_47":89,"_48":98,"_50":51,"_52":-5,"_70":99,"_18":100},{"_33":83,"_13":84,"_35":85,"_37":-5,"_38":39,"_40":86,"_42":43,"_44":87,"_17":88,"_47":89,"_48":90,"_50":51,"_52":-5,"_70":91,"_18":92},{"_33":73,"_13":74,"_35":75,"_37":-5,"_38":39,"_40":76,"_42":43,"_44":77,"_17":78,"_47":79,"_48":80,"_50":51,"_52":-5,"_70":81,"_18":82},{"_33":62,"_13":63,"_35":64,"_37":-5,"_38":39,"_40":65,"_42":43,"_44":66,"_17":67,"_47":68,"_48":69,"_50":51,"_52":-5,"_70":71,"_18":72},"d927282b-b46a-482b-92b4-f78043c5e35c","webhook.ts","8f710fe6-59da-11f1-9bcd-ee650bb23af1",51,["D",1786638456820],6,"http","// HTTP val: Telegram command interface. Owner-gated, no bot framework.\n\nimport {\n clearSentDeals,\n countSentDeals,\n getRules,\n getSettings,\n getStats,\n initDb,\n listSentDeals,\n patchSettings,\n setRules,\n} from \"./db.ts\";\nimport {\n compileRules,\n evaluateRule,\n isUsableRule,\n parseRule,\n} from \"./matcher.ts\";\nimport {\n BOT_TOKEN,\n chunkLines,\n dedupeOrdered,\n escapeHtml,\n formatLocal,\n OWNER_IDS,\n parsePrice,\n TELEGRAM_TIMEOUT_MS,\n webhookSecret,\n} from \"./shared.ts\";\n\nconst API = `https://api.telegram.org/bot${BOT_TOKEN}`;\nconst SECRET_TOKEN = await webhookSecret();\nconst RESULTS_PAGE_SIZE = 15;\n\nconst HELP = `SMZDM Bot\n\n品牌规则\n/brands - 查看规则\n/add_brands 规则1 规则2 - 添加\n/delete_brands 规则1 - 删除\n/delete_all_brands - 清空\n/test 商品标题 - 测试标题会命中哪条规则\n\n规则语法（单条规则内不能有空格）\napple 标题含 apple\napple-watch 同时含两个词\nsony|索尼-耳机 括号或关系\napple-watch-!表带 排除配件\nnike-<300 价格不超过 300\n茅台->1000 价格不低于 1000\n/rtx\\\\s?40[6-9]0/ 正则\n\n过滤\n/malls - 查看商城名单\n/block_mall 商城 - 加入黑名单\n/unblock_mall 商城 - 移出黑名单\n/only_mall 京东天猫超市 - 只看这些商城（留空则取消）\n/min_worthy 数字 - 最低「值」票数（0 关闭，注意会过滤掉刚发布的好价）\n/max_age 小时 - 只推多久以内的好价\n/match_content on|off - 排除词是否同时看爆料原文（正向词始终只看标题，默认关）\n\n推送\n/photo preview|photo|off - 配图方式\n/quiet 23 7 - 该时段静音推送，off 关闭\n/drop 5 - 降价超过百分之几时复推\n/pause /resume - 暂停/恢复\n/run - 立刻抓取一次\n\n其他\n/status - 运行状态\n/settings - 当前配置\n/results [页码] - 已推送记录\n/delete_results - 清空推送记录`;\n\nconst COMMAND_MENU = [\n { command: \"help\", description: \"使用说明\" },\n { command: \"brands\", description: \"查看品牌规则\" },\n { command: \"add_brands\", description: \"添加规则，空格分隔\" },\n { command: \"delete_brands\", description: \"删除规则\" },\n { command: \"test\", description: \"测试标题是否命中规则\" },\n { command: \"malls\", description: \"查看商城名单\" },\n { command: \"status\", description: \"运行状态\" },\n { command: \"settings\", description: \"当前配置\" },\n { command: \"results\", description: \"已推送记录\" },\n { command: \"run\", description: \"立刻抓取一次\" },\n { command: \"pause\", description: \"暂停推送\" },\n { command: \"resume\", description: \"恢复推送\" },\n];\n\ninterface Ctx {\n chatId: number;\n chatType: string;\n fromId: number;\n args: string;\n}\n\nfunction words(ctx: Ctx): string[] {\n return ctx.args.trim().split(/\\s+/).filter(Boolean);\n}\n\nfunction onOff(value: string | undefined): boolean | undefined {\n if (value === \"on\" || value === \"true\" || value === \"1\") return true;\n if (value === \"off\" || value === \"false\" || value === \"0\") return false;\n return undefined;\n}\n\nasync function tg(\n method: string,\n payload: Record,\n): Promise {\n try {\n const response = await fetch(`${API}/${method}`, {\n method: \"POST\",\n headers: { \"content-type\": \"application/json\" },\n body: JSON.stringify(payload),\n signal: AbortSignal.timeout(TELEGRAM_TIMEOUT_MS),\n });\n const body = await response.json() as { ok?: boolean };\n return body.ok === true;\n } catch (error) {\n console.error(`[webhook] ${method} failed:`, error);\n return false;\n }\n}\n\nfunction reply(ctx: Ctx, text: string): Promise {\n return tg(\"sendMessage\", { chat_id: ctx.chatId, text });\n}\n\nasync function replyHtml(ctx: Ctx, text: string): Promise {\n await tg(\"sendMessage\", {\n chat_id: ctx.chatId,\n text,\n parse_mode: \"HTML\",\n link_preview_options: { is_disabled: true },\n });\n}\n\nasync function replyLong(ctx: Ctx, lines: string[]): Promise {\n for (const chunk of chunkLines(lines)) await replyHtml(ctx, chunk);\n}\n\nconst handlers: Record Promise> = {\n start: (ctx) => replyHtml(ctx, HELP),\n help: (ctx) => replyHtml(ctx, HELP),\n\n async brands(ctx) {\n const rules = await getRules();\n if (rules.length === 0) {\n await reply(ctx, \"还没有规则，用 /add_brands 添加。\");\n return;\n }\n await replyLong(ctx, [\n `当前 ${rules.length} 条规则`,\n ...rules.map((rule, i) => `${i + 1}. ${escapeHtml(rule)}`),\n ]);\n },\n\n async add_brands(ctx) {\n const incoming = words(ctx);\n if (incoming.length === 0) {\n await reply(ctx, \"用法：/add_brands 规则1 规则2 ...\");\n return;\n }\n const accepted: string[] = [];\n const rejected: string[] = [];\n for (const raw of incoming) {\n if (isUsableRule(parseRule(raw))) accepted.push(raw);\n else rejected.push(raw);\n }\n if (accepted.length > 0) {\n await setRules(dedupeOrdered([...(await getRules()), ...accepted]));\n }\n const parts: string[] = [];\n if (accepted.length > 0) parts.push(`已添加：${accepted.join(\", \")}`);\n if (rejected.length > 0) {\n parts.push(`已忽略（缺少正向关键词）：${rejected.join(\", \")}`);\n }\n await reply(ctx, parts.join(\"\\n\"));\n },\n\n async delete_brands(ctx) {\n const targets = words(ctx);\n if (targets.length === 0) {\n await reply(ctx, \"用法：/delete_brands 规则1 规则2 ...\");\n return;\n }\n const before = await getRules();\n const remaining = before.filter((rule) => !targets.includes(rule));\n await setRules(remaining);\n await reply(\n ctx,\n `删除 ${\n before.length - remaining.length\n } 条，剩余 ${remaining.length} 条。`,\n );\n },\n\n async delete_all_brands(ctx) {\n await setRules([]);\n await reply(ctx, \"已清空所有规则，用 /add_brands 重新添加。\");\n },\n\n async test(ctx) {\n const title = ctx.args.trim();\n if (!title) {\n await reply(ctx, \"用法：/test 商品标题（可带价格，如 199元）\");\n return;\n }\n const settings = await getSettings();\n const raws = await getRules();\n if (raws.length === 0) {\n await reply(ctx, \"还没有规则可测。\");\n return;\n }\n const target = {\n title,\n content: \"\",\n mall: \"\",\n price: parsePrice(title),\n };\n await replyLong(ctx, [\n `识别价格：${target.price ?? \"无\"}`,\n ...raws.map((raw) => {\n const outcome = evaluateRule(\n parseRule(raw),\n target,\n settings.matchContent,\n );\n return `${outcome.ok ? \"✅\" : \"▫️\"} ${escapeHtml(raw)} — ${\n escapeHtml(outcome.reason)\n }`;\n }),\n ]);\n },\n\n async malls(ctx) {\n const settings = await getSettings();\n await reply(\n ctx,\n `黑名单：${settings.mallBlacklist.join(\", \") || \"（空）\"}\\n` +\n `白名单：${settings.mallWhitelist.join(\", \") || \"（空，表示不限）\"}`,\n );\n },\n\n async block_mall(ctx) {\n const targets = words(ctx);\n if (targets.length === 0) {\n await reply(ctx, \"用法：/block_mall 商城名\");\n return;\n }\n const settings = await getSettings();\n const merged = dedupeOrdered([...settings.mallBlacklist, ...targets]);\n await patchSettings({ mallBlacklist: merged });\n await reply(ctx, `黑名单：${merged.join(\", \")}`);\n },\n\n async unblock_mall(ctx) {\n const targets = words(ctx);\n if (targets.length === 0) {\n await reply(ctx, \"用法：/unblock_mall 商城名\");\n return;\n }\n const settings = await getSettings();\n const remaining = settings.mallBlacklist.filter((m) =>\n !targets.includes(m)\n );\n await patchSettings({ mallBlacklist: remaining });\n await reply(ctx, `黑名单：${remaining.join(\", \") || \"（空）\"}`);\n },\n\n async only_mall(ctx) {\n const targets = words(ctx);\n await patchSettings({ mallWhitelist: targets });\n await reply(\n ctx,\n targets.length === 0\n ? \"已取消商城白名单。\"\n : `只推送这些商城：${targets.join(\", \")}`,\n );\n },\n\n async min_worthy(ctx) {\n const value = Number(words(ctx)[0]);\n if (!Number.isFinite(value) || value < 0) {\n await reply(ctx, \"用法：/min_worthy 数字（0 表示不过滤）\");\n return;\n }\n await patchSettings({ minWorthy: Math.floor(value) });\n await reply(\n ctx,\n value === 0\n ? \"已关闭热度过滤。\"\n : `只推送「值」票数 ≥ ${\n Math.floor(value)\n } 的好价。注意刚发布的好价票数为 0，会被挡下。`,\n );\n },\n\n async max_age(ctx) {\n const value = Number(words(ctx)[0]);\n if (!Number.isFinite(value) || value <= 0) {\n await reply(ctx, \"用法：/max_age 小时数\");\n return;\n }\n await patchSettings({ maxAgeHours: Math.floor(value) });\n await reply(ctx, `只推送 ${Math.floor(value)} 小时内发布的好价。`);\n },\n\n async match_content(ctx) {\n const value = onOff(words(ctx)[0]);\n if (value === undefined) {\n await reply(ctx, \"用法：/match_content on|off\");\n return;\n }\n await patchSettings({ matchContent: value });\n await reply(\n ctx,\n value\n ? \"已开启：排除词同时看爆料原文。正向关键词始终只匹配标题。\"\n : \"已关闭：规则只匹配标题。\",\n );\n },\n\n async photo(ctx) {\n const mode = words(ctx)[0];\n if (mode !== \"preview\" && mode !== \"photo\" && mode !== \"off\") {\n await reply(ctx, \"用法：/photo preview|photo|off\");\n return;\n }\n await patchSettings({ photoMode: mode });\n await reply(ctx, `配图方式：${mode}`);\n },\n\n async quiet(ctx) {\n const values = words(ctx);\n if (values[0] === \"off\") {\n await patchSettings({ quietStart: -1, quietEnd: -1 });\n await reply(ctx, \"已关闭静音时段。\");\n return;\n }\n const start = Number(values[0]);\n const end = Number(values[1]);\n const valid = (hour: number) =>\n Number.isInteger(hour) && hour >= 0 && hour <= 23;\n if (!valid(start) || !valid(end)) {\n await reply(ctx, \"用法：/quiet 23 7 或 /quiet off\");\n return;\n }\n await patchSettings({ quietStart: start, quietEnd: end });\n await reply(ctx, `${start}:00 - ${end}:00 期间静音推送（北京时间）。`);\n },\n\n async drop(ctx) {\n const percent = Number(words(ctx)[0]);\n if (!Number.isFinite(percent) || percent <= 0 || percent >= 100) {\n await reply(ctx, \"用法：/drop 5（降价超过 5% 时复推）\");\n return;\n }\n await patchSettings({ dropThreshold: percent / 100 });\n await reply(ctx, `降价超过 ${percent}% 时会再推一次。`);\n },\n\n async pause(ctx) {\n await patchSettings({ paused: true });\n await reply(ctx, \"已暂停推送，/resume 恢复。\");\n },\n\n async resume(ctx) {\n await patchSettings({ paused: false });\n await reply(ctx, \"已恢复推送。\");\n },\n\n async run(ctx) {\n await reply(ctx, \"开始抓取…\");\n const { runSpider } = await import(\"./spider.ts\");\n const summary = await runSpider({ manual: true, maxPages: 5 });\n const skipped = Object.entries(summary.skipped)\n .map(([reason, count]) => `${reason} ${count}`)\n .join(\" · \") || \"无\";\n await reply(\n ctx,\n [\n summary.note ? `提示：${summary.note}` : \"\",\n `翻页 ${summary.pages} · 扫描 ${summary.scanned} · 命中 ${summary.matched}`,\n `已推送 ${summary.sent} · 失败 ${summary.failed}`,\n `过滤：${skipped}`,\n summary.errors.length > 0 ? `错误：${summary.errors[0]}` : \"\",\n ].filter(Boolean).join(\"\\n\"),\n );\n },\n\n async status(ctx) {\n const [stats, settings, total] = await Promise.all([\n getStats(),\n getSettings(),\n countSentDeals(),\n ]);\n await reply(\n ctx,\n [\n `状态：${settings.paused ? \"已暂停\" : \"运行中\"}`,\n `上次运行：${formatLocal(stats.lastRunAt)}`,\n `上次成功：${formatLocal(stats.lastOkAt)}`,\n `上轮：翻页 ${stats.pages} · 扫描 ${stats.scanned} · 命中 ${stats.matched} · 推送 ${stats.sent}`,\n `连续失败：${stats.failStreak}`,\n stats.lastError ? `最近错误：${stats.lastError}` : \"\",\n `历史记录：${total} 条`,\n ].filter(Boolean).join(\"\\n\"),\n );\n },\n\n async settings(ctx) {\n const settings = await getSettings();\n const quiet = settings.quietStart < 0\n ? \"关闭\"\n : `${settings.quietStart}:00-${settings.quietEnd}:00`;\n await reply(\n ctx,\n [\n `规则数：${compileRules(await getRules()).length}`,\n `商城黑名单：${settings.mallBlacklist.join(\", \") || \"（空）\"}`,\n `商城白名单：${settings.mallWhitelist.join(\", \") || \"（不限）\"}`,\n `最低值票：${settings.minWorthy}`,\n `时效：${settings.maxAgeHours} 小时`,\n `匹配爆料原文：${settings.matchContent ? \"开\" : \"关\"}`,\n `配图：${settings.photoMode}`,\n `静音时段：${quiet}`,\n `降价复推阈值：${Math.round(settings.dropThreshold * 100)}%`,\n ].join(\"\\n\"),\n );\n },\n\n async results(ctx) {\n const page = Math.max(1, Number(words(ctx)[0]) || 1);\n const total = await countSentDeals();\n if (total === 0) {\n await reply(ctx, \"还没有推送记录。\");\n return;\n }\n const pages = Math.ceil(total / RESULTS_PAGE_SIZE);\n const rows = await listSentDeals(\n RESULTS_PAGE_SIZE,\n (page - 1) * RESULTS_PAGE_SIZE,\n );\n await replyLong(\n ctx,\n [\n `推送记录 ${page}/${pages}（共 ${total} 条）`,\n ...rows.map((row, index) => {\n const order = (page - 1) * RESULTS_PAGE_SIZE + index + 1;\n const price = row.priceText ? ` — ${escapeHtml(row.priceText)}` : \"\";\n return `${order}. ${\n escapeHtml(row.title)\n }${price}`;\n }),\n pages > page ? `下一页：/results ${page + 1}` : \"\",\n ].filter(Boolean),\n );\n },\n\n async delete_results(ctx) {\n await clearSentDeals();\n await reply(ctx, \"已清空推送记录。注意：已推过的好价可能会再推一次。\");\n },\n};\n\nasync function handleCallback(queryId: string, data: string) {\n if (!data.startsWith(\"bm:\")) {\n await tg(\"answerCallbackQuery\", { callback_query_id: queryId });\n return;\n }\n const mall = data.slice(3);\n const settings = await getSettings();\n if (settings.mallBlacklist.includes(mall)) {\n await tg(\"answerCallbackQuery\", {\n callback_query_id: queryId,\n text: `${mall} 已在黑名单`,\n });\n return;\n }\n await patchSettings({ mallBlacklist: [...settings.mallBlacklist, mall] });\n await tg(\"answerCallbackQuery\", {\n callback_query_id: queryId,\n text: `已屏蔽 ${mall}`,\n });\n}\n\nfunction isOwner(userId: number | undefined): boolean {\n if (OWNER_IDS.length === 0) {\n console.warn(\"[webhook] OWNER_IDS is empty; refusing all commands\");\n return false;\n }\n if (!userId || !OWNER_IDS.includes(userId)) {\n console.warn(`[webhook] ignoring update from ${userId ?? \"unknown\"}`);\n return false;\n }\n return true;\n}\n\nasync function handleUpdate(update: {\n message?: {\n text?: string;\n chat: { id: number; type: string };\n from?: { id: number };\n };\n callback_query?: { id: string; data?: string; from: { id: number } };\n}): Promise {\n const callback = update.callback_query;\n if (callback) {\n if (!isOwner(callback.from.id)) return;\n await initDb();\n await handleCallback(callback.id, callback.data ?? \"\");\n return;\n }\n\n const message = update.message;\n const text = message?.text;\n if (!message || !text) return;\n if (!isOwner(message.from?.id)) return;\n await initDb();\n\n const ctx: Ctx = {\n chatId: message.chat.id,\n chatType: message.chat.type,\n fromId: message.from?.id ?? 0,\n args: \"\",\n };\n const matched = /^\\/([a-z_]+)(?:@\\S+)?(?:\\s+([\\s\\S]*))?$/i.exec(text);\n if (!matched) {\n if (ctx.chatType === \"private\") {\n await reply(ctx, \"没有这个命令，/help 查看用法。\");\n }\n return;\n }\n ctx.args = matched[2] ?? \"\";\n const handler = handlers[matched[1].toLowerCase()];\n if (!handler) {\n if (ctx.chatType === \"private\") {\n await reply(ctx, \"没有这个命令，/help 查看用法。\");\n }\n return;\n }\n await handler(ctx);\n}\n\nasync function setup(requestUrl: string): Promise {\n const url = new URL(requestUrl);\n url.search = \"\";\n url.hash = \"\";\n const notes: string[] = [];\n if (\n await tg(\"setWebhook\", {\n url: url.toString(),\n secret_token: SECRET_TOKEN,\n allowed_updates: [\"message\", \"callback_query\"],\n })\n ) {\n notes.push(`webhook -> ${url}`);\n } else {\n notes.push(\"setWebhook failed\");\n }\n if (await tg(\"setMyCommands\", { commands: COMMAND_MENU })) {\n notes.push(\"command menu registered\");\n } else {\n notes.push(\"setMyCommands failed\");\n }\n return notes;\n}\n\nexport default async function handler(req: Request): Promise {\n if (req.method !== \"POST\") {\n const notes = await setup(req.url);\n return new Response(`SMZDM bot is up.\\n${notes.join(\"\\n\")}\\n`, {\n headers: { \"content-type\": \"text/plain; charset=utf-8\" },\n });\n }\n\n const secret = req.headers.get(\"x-telegram-bot-api-secret-token\");\n if (secret !== SECRET_TOKEN) {\n return new Response(\"unauthorized\", { status: 401 });\n }\n\n try {\n await handleUpdate(await req.json());\n } catch (error) {\n console.error(\"[webhook] update failed:\", error);\n }\n return new Response(\"ok\");\n}\n","children",[],[63],"c7b402df-ba22-4cbd-bb56-e9a691c60459","spider.ts","2ca8446c-59dd-11f1-87d5-ee650bb23af1",50,["D",1786638456777],9,"interval","// Cron val: crawl 好价, match the user's rules, push what is new.\n//\n// Set the schedule in the val.town editor. Because paging is now incremental,\n// a short interval (10 minutes) is both cheaper and fresher than the old\n// fixed-5-page run: quiet periods cost one request, busy periods dig deeper.\n\nimport {\n findSentDeals,\n initDb,\n loadSnapshot,\n mergeSeen,\n pruneSentDeals,\n type RecordInput,\n recordSentMany,\n saveKv,\n} from \"./db.ts\";\nimport { compileRules, firstMatch } from \"./matcher.ts\";\nimport { sendCard, sendDigest, sendNotice } from \"./notify.ts\";\nimport { crawl, type Deal } from \"./smzdm.ts\";\nimport {\n escapeHtml,\n isQuietHour,\n MAX_CARDS_PER_RUN,\n MAX_PAGES,\n MAX_SENDS_PER_DEAL,\n nowSec,\n RETENTION_DAYS,\n type Settings,\n} from \"./shared.ts\";\n\nexport interface RunOptions {\n maxPages?: number;\n /** Manual runs ignore the paused flag. */\n manual?: boolean;\n}\n\nexport interface RunSummary {\n pages: number;\n scanned: number;\n matched: number;\n sent: number;\n failed: number;\n skipped: Record;\n errors: string[];\n note?: string;\n}\n\ninterface Candidate {\n deal: Deal;\n rule: string;\n priceDropFrom?: number;\n}\n\nfunction toRecord(deal: Deal) {\n return {\n articleId: deal.id,\n title: deal.title,\n url: deal.url,\n mall: deal.mall,\n priceText: deal.priceText,\n priceNum: deal.price,\n worthy: deal.worthy,\n publishedAt: deal.publishedAt,\n };\n}\n\nfunction passesFilters(\n deal: Deal,\n settings: Settings,\n skipped: Record,\n): boolean {\n const bump = (reason: string) => {\n skipped[reason] = (skipped[reason] ?? 0) + 1;\n return false;\n };\n\n // The feed keeps evergreen and pinned posts around: the very first row of\n // offset 0 has been observed as a 2021 deal flagged article_is_timeout.\n if (deal.expired) return bump(\"已过期\");\n if (deal.soldOut) return bump(\"已售罄\");\n\n if (\n deal.publishedAt > 0 &&\n nowSec() - deal.publishedAt > settings.maxAgeHours * 3_600\n ) {\n return bump(\"超过时效\");\n }\n if (settings.mallBlacklist.includes(deal.mall)) return bump(\"商城黑名单\");\n if (\n settings.mallWhitelist.length > 0 &&\n !settings.mallWhitelist.includes(deal.mall)\n ) {\n return bump(\"不在商城白名单\");\n }\n if (settings.minWorthy > 0 && deal.worthy < settings.minWorthy) {\n return bump(\"热度不足\");\n }\n return true;\n}\n\nexport async function runSpider(\n options: RunOptions = {},\n): Promise {\n await initDb();\n\n const snapshot = await loadSnapshot();\n const settings = snapshot.settings;\n const skipped: Record = {};\n const summary: RunSummary = {\n pages: 0,\n scanned: 0,\n matched: 0,\n sent: 0,\n failed: 0,\n skipped,\n errors: [],\n };\n\n if (settings.paused && !options.manual) {\n summary.note = \"已暂停\";\n return summary;\n }\n\n const rules = compileRules(snapshot.rules);\n if (rules.length === 0) {\n summary.note = \"未配置品牌规则\";\n return summary;\n }\n\n const result = await crawl(\n snapshot.cursor,\n options.maxPages ?? MAX_PAGES,\n new Set(snapshot.seenIds),\n );\n summary.pages = result.pages;\n summary.scanned = result.deals.length;\n summary.errors = result.errors;\n\n const stats = snapshot.stats;\n stats.lastRunAt = nowSec();\n stats.pages = result.pages;\n stats.scanned = result.deals.length;\n\n if (result.pages === 0) {\n stats.failStreak += 1;\n stats.lastError = result.errors[0] ?? \"未知抓取失败\";\n await saveKv({ stats });\n if (stats.failStreak === 3 || stats.failStreak % 12 === 0) {\n await sendNotice(\n `⚠️ 连续 ${stats.failStreak} 次抓取失败\\n` +\n `${escapeHtml(stats.lastError)}`,\n );\n }\n return summary;\n }\n\n const recovered = stats.failStreak >= 3;\n stats.failStreak = 0;\n stats.lastError = \"\";\n stats.lastOkAt = nowSec();\n\n const eligible = result.deals.filter((deal) =>\n passesFilters(deal, settings, skipped)\n );\n\n const matched: { deal: Deal; rule: string }[] = [];\n for (const deal of eligible) {\n const rule = firstMatch(rules, {\n title: deal.title,\n content: deal.content,\n mall: deal.mall,\n price: deal.price,\n }, settings.matchContent);\n if (rule) matched.push({ deal, rule });\n }\n summary.matched = matched.length;\n\n const known = await findSentDeals(matched.map(({ deal }) => deal.id));\n const candidates: Candidate[] = [];\n for (const { deal, rule } of matched) {\n const previous = known.get(deal.id);\n if (!previous) {\n candidates.push({ deal, rule });\n continue;\n }\n if (previous.sendCount >= MAX_SENDS_PER_DEAL) continue;\n\n const base = previous.bestPrice ?? previous.priceNum;\n if (\n deal.price !== null && base !== null &&\n deal.price < base * (1 - settings.dropThreshold)\n ) {\n candidates.push({ deal, rule, priceDropFrom: base });\n }\n }\n\n const silent = isQuietHour(settings, nowSec());\n const cards = candidates.slice(0, MAX_CARDS_PER_RUN);\n const overflow = candidates.slice(MAX_CARDS_PER_RUN);\n const failedPublishAt: number[] = [];\n const failedIds: string[] = [];\n const delivered: RecordInput[] = [];\n\n for (const candidate of cards) {\n const ok = await sendCard(candidate.deal, {\n rule: candidate.rule,\n priceDropFrom: candidate.priceDropFrom,\n settings,\n silent,\n });\n if (ok) {\n delivered.push(toRecord(candidate.deal));\n summary.sent++;\n } else {\n summary.failed++;\n failedPublishAt.push(candidate.deal.publishedAt);\n failedIds.push(candidate.deal.id);\n }\n }\n\n if (overflow.length > 0) {\n const ok = await sendDigest(overflow, silent);\n for (const candidate of overflow) {\n if (ok) {\n delivered.push(toRecord(candidate.deal));\n summary.sent++;\n } else {\n summary.failed++;\n failedPublishAt.push(candidate.deal.publishedAt);\n failedIds.push(candidate.deal.id);\n }\n }\n }\n\n await recordSentMany(delivered);\n\n stats.matched = summary.matched;\n stats.sent = summary.sent;\n\n const kv: Record = { stats };\n if (result.errors.length === 0) {\n let nextCursor = result.newest;\n if (failedPublishAt.length > 0) {\n nextCursor = Math.min(nextCursor, Math.min(...failedPublishAt) - 1);\n }\n nextCursor = Math.max(nextCursor, nowSec() - settings.maxAgeHours * 3_600);\n kv.cursor = nextCursor;\n const skip = new Set(failedIds);\n kv.seen_ids = mergeSeen(\n snapshot.seenIds,\n result.deals.filter((deal) => !skip.has(deal.id)).map((deal) => deal.id),\n );\n }\n\n if (nowSec() - stats.lastPruneAt >= 86_400) {\n await pruneSentDeals(RETENTION_DAYS);\n stats.lastPruneAt = nowSec();\n kv.stats = stats;\n }\n\n await saveKv(kv);\n\n if (recovered) await sendNotice(\"✅ 抓取已恢复正常\");\n\n console.log(\n `[spider] pages=${summary.pages} scanned=${summary.scanned} ` +\n `matched=${summary.matched} sent=${summary.sent} ` +\n `failed=${summary.failed}`,\n );\n return summary;\n}\n\ninterface Interval {\n lastRunAt: Date | undefined;\n}\n\nexport default async function spider(_interval?: Interval): Promise {\n await runSpider();\n}\n",[],[74],"daf75f36-b0a4-4f48-acf6-b21333d5bce1","smzdm.ts","073ebd78-8f57-11f1-9060-1607ee4eb77e",47,["D",1786638456332],4,"script","// SMZDM 好价 feed client: normalisation plus incremental cursor paging.\n//\n// Page 0 is sequential; further pages in each wave are fetched in parallel.\n// Paging stops when a page's ids are all already seen, or when nothing on it\n// is newer than the time cursor. Offset advances by the API's raw row count.\n\nimport {\n FETCH_RETRIES,\n FETCH_TIMEOUT_MS,\n INITIAL_PAGES,\n MAX_PAGES,\n PAGE_SIZE,\n PARALLEL_PAGES,\n parsePrice,\n parseSmzdmDate,\n sleep,\n SMZDM_API_URL,\n SMZDM_HEADERS,\n} from \"./shared.ts\";\n\nexport interface Deal {\n id: string;\n title: string;\n url: string;\n /** go.smzdm.com redirect straight to the mall, when present. */\n buyUrl: string;\n pic: string;\n mall: string;\n priceText: string;\n price: number | null;\n worthy: number;\n unworthy: number;\n worthyPercent: string;\n comments: number;\n collections: number;\n referrals: string;\n /** 爆料原文: how to stack coupons, often the only place the model appears. */\n content: string;\n publishedAt: number;\n soldOut: boolean;\n expired: boolean;\n}\n\ninterface RawRow {\n article_id?: string;\n article_title?: string;\n article_url?: string;\n article_link?: string;\n article_pic?: string;\n article_price?: string;\n article_date?: string;\n article_mall?: string | null;\n mall_name?: string | null;\n article_worthy?: string;\n article_unworthy?: string;\n article_worthy_percent?: string;\n article_comment?: string;\n article_collection?: string;\n article_referrals?: string;\n article_filter_content?: string;\n article_is_sold_out?: string;\n article_is_timeout?: string;\n}\n\ninterface ListResponse {\n error_code?: string;\n error_msg?: string;\n data?: { rows?: RawRow[] };\n}\n\nfunction toInt(value?: string | null): number {\n const parsed = Number(value);\n return Number.isFinite(parsed) ? parsed : 0;\n}\n\nfunction normalise(row: RawRow): Deal | null {\n const id = row.article_id?.trim();\n const title = row.article_title?.trim();\n const rawUrl = row.article_url?.trim();\n // Banner and topic cards share the feed but carry none of these, so this\n // check doubles as a filter for non-deal rows.\n if (!id || !title || !rawUrl) return null;\n\n const url = rawUrl.replace(/^http:\\/\\//i, \"https://\");\n const priceText = row.article_price?.trim() ?? \"\";\n\n return {\n id,\n title,\n url,\n buyUrl: row.article_link?.trim() || url,\n pic: row.article_pic?.trim() ?? \"\",\n mall: (row.article_mall ?? row.mall_name ?? \"\").trim(),\n priceText,\n price: parsePrice(priceText),\n worthy: toInt(row.article_worthy),\n unworthy: toInt(row.article_unworthy),\n worthyPercent: row.article_worthy_percent?.trim() ?? \"\",\n comments: toInt(row.article_comment),\n collections: toInt(row.article_collection),\n referrals: row.article_referrals?.trim() ?? \"\",\n content: row.article_filter_content?.trim() ?? \"\",\n publishedAt: parseSmzdmDate(row.article_date),\n soldOut: row.article_is_sold_out === \"1\",\n expired: row.article_is_timeout === \"1\",\n };\n}\n\nclass PageError extends Error {}\n\ninterface Page {\n deals: Deal[];\n /** Raw row count from the API, including banners we dropped. */\n rawCount: number;\n}\n\ntype PageResult = Page | { error: string };\n\nasync function fetchPage(offset: number): Promise {\n const url = `${SMZDM_API_URL}?limit=${PAGE_SIZE}&offset=${offset}`;\n let lastError = \"unknown error\";\n\n for (let attempt = 0; attempt <= FETCH_RETRIES; attempt++) {\n if (attempt > 0) await sleep(500 * attempt);\n\n let body: string;\n let status: number;\n try {\n const response = await fetch(url, {\n headers: SMZDM_HEADERS,\n signal: AbortSignal.timeout(FETCH_TIMEOUT_MS),\n });\n status = response.status;\n body = await response.text();\n } catch (error) {\n lastError = `network: ${(error as Error).message}`;\n continue;\n }\n\n // A HTML body means SMZDM rerouted us to its WAF or an error page.\n if (status !== 200 || /^\\s*<(!doctype|html|head|body)/i.test(body)) {\n lastError = `status=${status} bytes=${body.length} ` +\n `head=${JSON.stringify(body.slice(0, 60))}`;\n continue;\n }\n\n let payload: ListResponse;\n try {\n payload = JSON.parse(body) as ListResponse;\n } catch (error) {\n lastError = `invalid json: ${(error as Error).message}`;\n continue;\n }\n\n if (payload.error_code && payload.error_code !== \"0\") {\n lastError = `api ${payload.error_code}: ${payload.error_msg ?? \"\"}`;\n continue;\n }\n\n const rows = Array.isArray(payload.data?.rows) ? payload.data!.rows! : [];\n return {\n deals: rows.map(normalise).filter((deal): deal is Deal => deal !== null),\n rawCount: rows.length,\n };\n }\n\n throw new PageError(`offset=${offset} failed: ${lastError}`);\n}\n\nasync function pullPage(offset: number): Promise {\n try {\n return await fetchPage(offset);\n } catch (error) {\n return { error: (error as Error).message };\n }\n}\n\nexport interface CrawlResult {\n deals: Deal[];\n pages: number;\n /** Newest publish time seen, for the next run's cursor. */\n newest: number;\n errors: string[];\n}\n\n/**\n * Walks the feed newest-first until a page is entirely familiar (seen ids) or\n * holds nothing newer than `cursor`. Page 0 is fetched first so we know whether\n * to continue; remaining pages in each wave go out in parallel. Offset steps by\n * the API's raw row count, not the post-filter deal count.\n *\n * Returns deals deduped by article id, oldest first. `errors` is non-empty if\n * any page failed — callers must not advance the cursor or seen ring in that\n * case, or items on the failed page are skipped forever.\n */\nexport async function crawl(\n cursor: number,\n maxPages = MAX_PAGES,\n seenIds: ReadonlySet = new Set(),\n): Promise {\n const pageBudget = cursor > 0\n ? Math.min(maxPages, MAX_PAGES)\n : Math.min(maxPages, INITIAL_PAGES);\n\n const byId = new Map();\n const errors: string[] = [];\n let newest = cursor;\n let pages = 0;\n let offset = 0;\n\n const ingest = (deals: Deal[]): void => {\n for (const deal of deals) {\n if (!byId.has(deal.id)) byId.set(deal.id, deal);\n if (deal.publishedAt > newest) newest = deal.publishedAt;\n }\n };\n\n const caughtUp = (deals: Deal[], rawCount: number): boolean => {\n // Short or empty page: the feed has ended. A full page of banners is\n // PAGE_SIZE long and must not look like EOF.\n if (rawCount < PAGE_SIZE) return true;\n if (deals.length === 0) return false;\n if (seenIds.size > 0 && deals.every((deal) => seenIds.has(deal.id))) {\n return true;\n }\n // Counting fresh items rather than comparing the page's oldest timestamp\n // keeps pinned evergreen posts from aborting the walk on the first page.\n if (cursor > 0) {\n let fresh = 0;\n for (const deal of deals) if (deal.publishedAt > cursor) fresh++;\n if (fresh === 0) return true;\n }\n return false;\n };\n\n while (pages < pageBudget) {\n const batch = pages === 0\n ? 1\n : Math.min(PARALLEL_PAGES, pageBudget - pages);\n // Parallel stride uses PAGE_SIZE so a later page returning fewer rows\n // overlaps instead of leaving a gap. Sequential offset then snaps to the\n // actual rawCount so the next wave starts where this one left off.\n const offsets = pages === 0\n ? [0]\n : Array.from({ length: batch }, (_, i) => offset + i * PAGE_SIZE);\n\n const results = offsets.length === 1\n ? [await pullPage(offsets[0])]\n : await Promise.all(offsets.map(pullPage));\n\n let stop = false;\n for (let i = 0; i < results.length; i++) {\n const result = results[i];\n if (\"error\" in result) {\n errors.push(result.error);\n stop = true;\n break;\n }\n pages++;\n ingest(result.deals);\n offset = offsets[i] + result.rawCount;\n if (caughtUp(result.deals, result.rawCount)) {\n stop = true;\n break;\n }\n }\n if (stop) break;\n }\n\n const deals = [...byId.values()].sort(\n (a, b) => a.publishedAt - b.publishedAt,\n );\n return { deals, pages, newest, errors };\n}\n",[],[84],"e4b3ad12-7de9-407a-bd1b-a14a0b0d87d2","shared.ts","1b156086-59dd-11f1-9e63-ee650bb23af1",48,["D",1786638456426],"// Environment, tunables and small pure helpers shared by every file.\n//\n// Required environment variables:\n// BOT_TOKEN Telegram bot token from @BotFather\n// CHAT_ID Telegram chat that receives deal pushes\n// Optional:\n// OWNER_IDS Comma-separated Telegram user ids allowed to run commands.\n// Defaults to CHAT_ID. Anyone else is ignored.\n// WEBHOOK_SECRET Shared secret for Telegram webhook verification.\n// Defaults to a SHA-256 derived from BOT_TOKEN.\n\nfunction env(name: string): string | undefined {\n const value = Deno.env.get(name);\n return value && value.trim() ? value.trim() : undefined;\n}\n\nfunction requireEnv(name: string): string {\n const value = env(name);\n if (!value) throw new Error(`${name} env var is not set`);\n return value;\n}\n\nexport const BOT_TOKEN = requireEnv(\"BOT_TOKEN\");\nexport const CHAT_ID = requireEnv(\"CHAT_ID\");\n\nexport const OWNER_IDS: number[] = (env(\"OWNER_IDS\") ?? CHAT_ID)\n .split(/[,\\s]+/)\n .map((part) => Number(part))\n .filter((id) => Number.isFinite(id) && id !== 0);\n\n/**\n * Telegram only accepts `A-Z a-z 0-9 _ -` in secret_token. Deriving it by\n * hashing the bot token means a leaked secret does not leak half the token,\n * which is what using `BOT_TOKEN.split(\":\")[1]` directly would do.\n */\nexport async function webhookSecret(): Promise {\n const explicit = env(\"WEBHOOK_SECRET\");\n if (explicit) {\n return explicit.replace(/[^A-Za-z0-9_-]/g, \"\").slice(0, 256) || \"smzdm\";\n }\n const bytes = new TextEncoder().encode(`smzdm-webhook:${BOT_TOKEN}`);\n const digest = await crypto.subtle.digest(\"SHA-256\", bytes);\n return Array.from(new Uint8Array(digest))\n .map((b) => b.toString(16).padStart(2, \"0\"))\n .join(\"\");\n}\n\n// --- SMZDM source -----------------------------------------------------------\n//\n// api.smzdm.com is the mobile-app API. The www host sits behind Tencent Chaos\n// bot detection and returns a JS challenge stub to any non-browser caller,\n// so the app API is the only workable source. This endpoint needs no auth.\n\nexport const SMZDM_API_URL = \"https://api.smzdm.com/v1/youhui/list\";\n\nexport const SMZDM_HEADERS: Record = {\n \"user-agent\": \"smzdm_android_V11.4.6 rv:855 (Pixel 5;Android11;zh)smzdmapp\",\n \"accept\": \"application/json\",\n \"referer\": \"https://www.smzdm.com/\",\n};\n\nexport const PAGE_SIZE = 20;\n\n/** Hard ceiling on pages per run so a cursor glitch cannot loop forever. */\nexport const MAX_PAGES = 20;\n\n/** Pages to scan when no cursor exists yet (fresh install / after a reset). */\nexport const INITIAL_PAGES = 3;\n\n/** Pages fetched together after page 0. Over-fetch is cheaper than serial RTT. */\nexport const PARALLEL_PAGES = 5;\n\n/** Ring of article ids used to stop paging once a page is entirely familiar. */\nexport const SEEN_LIMIT = 600;\n\nexport const FETCH_TIMEOUT_MS = 10_000;\nexport const FETCH_RETRIES = 2;\n\n// --- Telegram delivery ------------------------------------------------------\n\nexport const TELEGRAM_TIMEOUT_MS = 15_000;\n\n/** Telegram allows ~1 message/second per chat; stay under it. */\nexport const SEND_INTERVAL_MS = 1_200;\n\nexport const TELEGRAM_RETRIES = 3;\n\n/** Full cards per run. Anything beyond this is folded into a digest message. */\nexport const MAX_CARDS_PER_RUN = 8;\n\n/** Cap re-pushes of the same article so a flapping price cannot spam. */\nexport const MAX_SENDS_PER_DEAL = 3;\n\nexport const TEXT_LIMIT = 4096;\nexport const CAPTION_LIMIT = 1024;\n\n/** Days of push history to keep for dedupe and /results. */\nexport const RETENTION_DAYS = 30;\n\nexport const TIMEZONE = \"Asia/Shanghai\";\n\n// --- User-tunable settings (persisted, editable via bot commands) -----------\n\nexport interface Settings {\n paused: boolean;\n /** Minimum 值 votes required to push. 0 disables the filter. */\n minWorthy: number;\n /** Ignore deals published longer ago than this. */\n maxAgeHours: number;\n /** Also match rules against the deal's 爆料原文 body, not just the title. */\n matchContent: boolean;\n mallBlacklist: string[];\n /** When non-empty, only these malls are pushed. */\n mallWhitelist: string[];\n /** Local hours [start, end) delivered silently. -1 disables. */\n quietStart: number;\n quietEnd: number;\n /** Re-push when the price drops by at least this fraction. */\n dropThreshold: number;\n /** How the product image is attached. */\n photoMode: \"preview\" | \"photo\" | \"off\";\n}\n\nexport const DEFAULT_SETTINGS: Settings = {\n paused: false,\n minWorthy: 0,\n maxAgeHours: 24,\n matchContent: false,\n mallBlacklist: [\"唯品会\"],\n mallWhitelist: [],\n quietStart: -1,\n quietEnd: -1,\n dropThreshold: 0.05,\n photoMode: \"preview\",\n};\n\nexport interface Stats {\n lastRunAt: number;\n lastOkAt: number;\n lastError: string;\n failStreak: number;\n pages: number;\n scanned: number;\n matched: number;\n sent: number;\n lastPruneAt: number;\n}\n\nexport const EMPTY_STATS: Stats = {\n lastRunAt: 0,\n lastOkAt: 0,\n lastError: \"\",\n failStreak: 0,\n pages: 0,\n scanned: 0,\n matched: 0,\n sent: 0,\n lastPruneAt: 0,\n};\n\n// --- Helpers ----------------------------------------------------------------\n\nexport function nowSec(): number {\n return Math.floor(Date.now() / 1000);\n}\n\nexport function sleep(ms: number): Promise {\n return new Promise((resolve) => setTimeout(resolve, ms));\n}\n\nexport function escapeHtml(text: string): string {\n return text\n .replaceAll(\"&\", \"&\")\n .replaceAll(\"<\", \"<\")\n .replaceAll(\">\", \">\");\n}\n\nexport function truncate(text: string, max: number): string {\n const trimmed = text.trim();\n return trimmed.length <= max ? trimmed : `${trimmed.slice(0, max - 1)}…`;\n}\n\n/**\n * SMZDM's `article_unix_date` does not agree with `article_date` (observed\n * skews of tens of minutes to half a day on the same row), so ordering and\n * the incremental cursor must come from `article_date`, which is Beijing time.\n */\nexport function parseSmzdmDate(value?: string | null): number {\n if (!value) return 0;\n const m = /^(\\d{4})-(\\d{2})-(\\d{2})[ T](\\d{2}):(\\d{2})(?::(\\d{2}))?/\n .exec(value.trim());\n if (!m) return 0;\n const iso = `${m[1]}-${m[2]}-${m[3]}T${m[4]}:${m[5]}:${m[6] ?? \"00\"}+08:00`;\n const parsed = Date.parse(iso);\n return Number.isFinite(parsed) ? Math.floor(parsed / 1000) : 0;\n}\n\n/** Pulls a comparable number out of strings like \"39.51元（需用券）\". */\nexport function parsePrice(text?: string | null): number | null {\n if (!text) return null;\n const cleaned = text.replaceAll(\",\", \"\");\n // Floor prices and 满减 thresholds are not the deal price.\n if (/元起|起售|满\\s*\\d/.test(cleaned)) return null;\n const yuan = /(\\d+(?:\\.\\d+)?)\\s*元/.exec(cleaned);\n if (yuan) return Number(yuan[1]);\n const symbol = /[$￥¥]\\s*(\\d+(?:\\.\\d+)?)/.exec(cleaned);\n if (symbol) return Number(symbol[1]);\n const bare = /(\\d+(?:\\.\\d+)?)/.exec(cleaned);\n return bare ? Number(bare[1]) : null;\n}\n\nconst dateFormatter = new Intl.DateTimeFormat(\"zh-CN\", {\n timeZone: TIMEZONE,\n month: \"2-digit\",\n day: \"2-digit\",\n hour: \"2-digit\",\n minute: \"2-digit\",\n hour12: false,\n});\n\nexport function formatLocal(unixSec: number): string {\n if (!unixSec) return \"未知时间\";\n return dateFormatter.format(new Date(unixSec * 1000)).replace(/\\//g, \"-\");\n}\n\nexport function localHour(unixSec: number): number {\n const formatted = new Intl.DateTimeFormat(\"en-US\", {\n timeZone: TIMEZONE,\n hour: \"2-digit\",\n hour12: false,\n }).format(new Date(unixSec * 1000));\n return Number(formatted) % 24;\n}\n\nexport function isQuietHour(settings: Settings, unixSec: number): boolean {\n const { quietStart: start, quietEnd: end } = settings;\n if (start < 0 || end < 0 || start === end) return false;\n const hour = localHour(unixSec);\n return start < end\n ? hour >= start && hour < end\n : hour >= start || hour < end;\n}\n\n/** Groups lines into messages that each stay under Telegram's length limit. */\nexport function chunkLines(lines: string[], limit = TEXT_LIMIT): string[] {\n const chunks: string[] = [];\n let current = \"\";\n for (const line of lines) {\n const candidate = current ? `${current}\\n${line}` : line;\n if (candidate.length > limit && current) {\n chunks.push(current);\n current = line;\n } else {\n current = candidate;\n }\n }\n if (current) chunks.push(current);\n return chunks;\n}\n\n/** Dedupes while preserving insertion order. */\nexport function dedupeOrdered(items: T[]): T[] {\n return [...new Set(items)];\n}\n",[],[94],"f21e3595-c10c-4619-9a15-aee0f6134a44","notify.ts","13405d34-8f57-11f1-bd72-1607ee4eb77e",44,["D",1786550432148],"// Telegram delivery: rich HTML cards, digests, throttling and retries.\n//\n// Three problems with the previous implementation are fixed here. Sends are\n// serialised behind a shared interval so a burst of matches cannot trip\n// Telegram's per-chat rate limit; a 429 is honoured via `retry_after` instead\n// of being logged and dropped; and every send reports success so the caller\n// only records a deal as pushed once it actually arrived.\n\nimport type { Deal } from \"./smzdm.ts\";\nimport {\n BOT_TOKEN,\n CAPTION_LIMIT,\n CHAT_ID,\n chunkLines,\n escapeHtml,\n formatLocal,\n SEND_INTERVAL_MS,\n type Settings,\n sleep,\n TELEGRAM_RETRIES,\n TELEGRAM_TIMEOUT_MS,\n TEXT_LIMIT,\n truncate,\n} from \"./shared.ts\";\n\nconst API_BASE = `https://api.telegram.org/bot${BOT_TOKEN}`;\n\nlet nextSendAt = 0;\n\nasync function throttle(): Promise {\n const wait = nextSendAt - Date.now();\n if (wait > 0) await sleep(wait);\n nextSendAt = Date.now() + SEND_INTERVAL_MS;\n}\n\ninterface ApiResponse {\n ok: boolean;\n description?: string;\n parameters?: { retry_after?: number };\n}\n\n/** \"rejected\" means Telegram refused the payload itself, so retrying is futile. */\ntype Outcome = \"ok\" | \"failed\" | \"rejected\";\n\nasync function callApi(\n method: string,\n payload: Record,\n): Promise {\n for (let attempt = 0; attempt <= TELEGRAM_RETRIES; attempt++) {\n await throttle();\n\n let status: number;\n let body: ApiResponse;\n try {\n const response = await fetch(`${API_BASE}/${method}`, {\n method: \"POST\",\n headers: { \"content-type\": \"application/json\" },\n body: JSON.stringify(payload),\n signal: AbortSignal.timeout(TELEGRAM_TIMEOUT_MS),\n });\n status = response.status;\n body = await response.json() as ApiResponse;\n } catch (error) {\n console.warn(`[notify] ${method} network error:`, error);\n await sleep(1_000 * (attempt + 1));\n continue;\n }\n\n if (body.ok) return \"ok\";\n\n if (status === 429) {\n const retryAfter = body.parameters?.retry_after ?? 3;\n console.warn(`[notify] rate limited, waiting ${retryAfter}s`);\n await sleep((retryAfter + 1) * 1_000);\n continue;\n }\n if (status >= 500) {\n await sleep(1_000 * (attempt + 1));\n continue;\n }\n\n // 4xx other than 429 will not fix itself; stop and surface it.\n console.error(`[notify] ${method} failed ${status}: ${body.description}`);\n return \"rejected\";\n }\n\n console.error(`[notify] ${method} gave up after ${TELEGRAM_RETRIES} retries`);\n return \"failed\";\n}\n\n/**\n * Last resort when Telegram rejects a formatted payload, e.g. a title that\n * breaks HTML parsing. Losing the formatting beats losing the deal.\n */\nasync function sendPlain(text: string, silent = false): Promise {\n const outcome = await callApi(\"sendMessage\", {\n chat_id: CHAT_ID,\n text: truncate(text, TEXT_LIMIT),\n disable_notification: silent,\n link_preview_options: { is_disabled: true },\n });\n return outcome === \"ok\";\n}\n\n// --- Message building -------------------------------------------------------\n\nfunction link(url: string, label: string): string {\n return `${escapeHtml(label)}`;\n}\n\nfunction votes(deal: Deal): string {\n if (deal.worthy === 0 && deal.unworthy === 0) return \"\";\n const percent = deal.worthyPercent\n ? ` ${escapeHtml(deal.worthyPercent)}`\n : \"\";\n return `👍 ${deal.worthy} 👎 ${deal.unworthy}${percent}`;\n}\n\n/**\n * Truncates already-escaped HTML without ever cutting an entity in half, which\n * would make Telegram reject the whole message.\n */\nfunction truncateEscaped(escaped: string, max: number): string {\n if (escaped.length <= max) return escaped;\n let cut = escaped.slice(0, max - 1);\n const lastAmp = cut.lastIndexOf(\"&\");\n if (lastAmp !== -1 && !cut.slice(lastAmp).includes(\";\")) {\n cut = cut.slice(0, lastAmp);\n }\n return `${cut}…`;\n}\n\nfunction keyboard(deal: Deal): Record {\n const row: Record[] = [\n { text: \"🛒 去购买\", url: deal.buyUrl },\n { text: \"📄 原文\", url: deal.url },\n ];\n const rows = [row];\n\n const muteData = `bm:${deal.mall}`;\n // callback_data is capped at 64 bytes; skip the button rather than 400.\n if (deal.mall && new TextEncoder().encode(muteData).length <= 64) {\n rows.push([{ text: `🚫 屏蔽 ${deal.mall}`, callback_data: muteData }]);\n }\n return { inline_keyboard: rows };\n}\n\nexport interface CardOptions {\n rule: string;\n priceDropFrom?: number;\n settings: Settings;\n silent: boolean;\n}\n\nfunction buildCard(deal: Deal, options: CardOptions, limit: number): string {\n const lines: string[] = [];\n\n if (options.priceDropFrom !== undefined) {\n lines.push(\n `📉 降价 ${options.priceDropFrom} 元 → ` +\n `${deal.price ?? \"?\"} 元`,\n );\n }\n\n lines.push(`🔥 ${link(deal.url, deal.title)}`);\n if (deal.priceText) lines.push(`💰 ${escapeHtml(deal.priceText)}`);\n\n const meta = [\n deal.mall ? `🏬 ${escapeHtml(deal.mall)}` : \"\",\n votes(deal),\n deal.comments > 0 ? `💬 ${deal.comments}` : \"\",\n `🕒 ${formatLocal(deal.publishedAt)}`,\n ].filter(Boolean);\n lines.push(meta.join(\" · \"));\n\n const tags = [\n `🏷 ${escapeHtml(options.rule)}`,\n deal.referrals ? `爆料 ${escapeHtml(deal.referrals)}` : \"\",\n ].filter(Boolean);\n lines.push(tags.join(\" · \"));\n\n const head = lines.join(\"\\n\");\n if (!deal.content) return head;\n\n // The 爆料原文 explains how to stack coupons and is often the whole point of\n // the deal, so it gets whatever room is left in the message.\n const budget = limit - head.length - 8;\n if (budget < 40) return head;\n return `${head}\\n📝 ${truncateEscaped(escapeHtml(deal.content), budget)}`;\n}\n\nexport async function sendCard(\n deal: Deal,\n options: CardOptions,\n): Promise {\n const usePhoto = options.settings.photoMode === \"photo\" && deal.pic !== \"\";\n const limit = usePhoto ? CAPTION_LIMIT : TEXT_LIMIT;\n const text = buildCard(deal, options, limit);\n\n const common = {\n chat_id: CHAT_ID,\n parse_mode: \"HTML\",\n disable_notification: options.silent,\n reply_markup: keyboard(deal),\n };\n\n const outcome = usePhoto\n ? await callApi(\"sendPhoto\", { ...common, photo: deal.pic, caption: text })\n : await callApi(\"sendMessage\", {\n ...common,\n text,\n link_preview_options: options.settings.photoMode === \"preview\" && deal.pic\n ? { url: deal.pic, prefer_large_media: true, show_above_text: true }\n : { is_disabled: true },\n });\n\n if (outcome === \"ok\") return true;\n if (outcome === \"failed\") return false;\n return await sendPlain(\n `${deal.title}\\n${deal.priceText}\\n${deal.url}`,\n options.silent,\n );\n}\n\n/**\n * Overflow beyond MAX_CARDS_PER_RUN is summarised instead of dropped, so a\n * busy feed neither floods the chat nor loses matches.\n */\nexport async function sendDigest(\n entries: { deal: Deal; rule: string }[],\n silent: boolean,\n): Promise {\n if (entries.length === 0) return true;\n\n const lines = [\n `📦 另有 ${entries.length} 条命中`,\n ...entries.map(({ deal, rule }, index) => {\n const price = deal.priceText ? ` — ${escapeHtml(deal.priceText)}` : \"\";\n const mall = deal.mall ? ` @${escapeHtml(deal.mall)}` : \"\";\n return `${index + 1}. ${link(deal.url, truncate(deal.title, 60))}` +\n `${price}${mall} [${escapeHtml(rule)}]`;\n }),\n ];\n\n let allOk = true;\n for (const chunk of chunkLines(lines)) {\n const outcome = await callApi(\"sendMessage\", {\n chat_id: CHAT_ID,\n parse_mode: \"HTML\",\n text: chunk,\n disable_notification: silent,\n link_preview_options: { is_disabled: true },\n });\n const ok = outcome === \"ok\" ||\n (outcome === \"rejected\" && await sendPlain(stripTags(chunk), silent));\n allOk = allOk && ok;\n }\n return allOk;\n}\n\nfunction stripTags(html: string): string {\n return html\n .replace(/([^<]*)<\\/a>/g, \"$2 $1\")\n .replace(/<[^>]+>/g, \"\")\n .replaceAll(\"&\", \"&\")\n .replaceAll(\"<\", \"<\")\n .replaceAll(\">\", \">\");\n}\n\n/** Notice to the push chat, used for crawl-failure and recovery alerts. */\nexport async function sendNotice(text: string): Promise {\n const outcome = await callApi(\"sendMessage\", {\n chat_id: CHAT_ID,\n parse_mode: \"HTML\",\n text: truncate(text, TEXT_LIMIT),\n link_preview_options: { is_disabled: true },\n });\n if (outcome === \"ok\") return true;\n if (outcome === \"failed\") return false;\n return await sendPlain(stripTags(text));\n}\n",[],[102],"37c71d81-77bc-4f0e-8861-02b0b95695e3","matcher.ts","0d715e58-8f57-11f1-8377-1607ee4eb77e",49,["D",1786638456146],"// Rule parsing and matching.\n//\n// A rule is one whitespace-free token made of parts joined by \"-\". The old\n// version only supported plain AND-ed substrings, which made accessory noise\n// (\"apple watch 表带\") impossible to filter and gave no way to cap price.\n//\n// apple title contains \"apple\"\n// apple-watch contains both \"apple\" and \"watch\"\n// sony|索尼-耳机 (sony OR 索尼) AND 耳机\n// apple-watch-!表带-!钢化膜 excludes accessories\n// nike-<300 price at most 300\n// 茅台->1000 price at least 1000\n// /rtx\\s?40[6-9]0/ regex, case-insensitive\n// !/翻新|拆机/ negative regex\n//\n// Known limitation, inherited from the original syntax: \"-\" is the separator,\n// so a brand containing a hyphen must be written as a regex instead.\n\nexport interface Rule {\n raw: string;\n /** Each group is a set of alternatives; every group must match. */\n includes: string[][];\n excludes: string[][];\n patterns: RegExp[];\n negPatterns: RegExp[];\n minPrice?: number;\n maxPrice?: number;\n}\n\nexport interface MatchTarget {\n title: string;\n content: string;\n mall: string;\n price: number | null;\n}\n\nconst REGEX_TOKEN = /(!?)\\/((?:[^/\\\\]|\\\\.)+)\\//g;\n\nexport function parseRule(raw: string): Rule {\n const rule: Rule = {\n raw,\n includes: [],\n excludes: [],\n patterns: [],\n negPatterns: [],\n };\n\n // Regexes are pulled out before splitting so a \"-\" inside a pattern is not\n // mistaken for the part separator.\n const remainder = raw.replace(\n REGEX_TOKEN,\n (_match, negate: string, body: string) => {\n try {\n const pattern = new RegExp(body, \"i\");\n if (negate) rule.negPatterns.push(pattern);\n else rule.patterns.push(pattern);\n } catch {\n // An invalid pattern is kept as literal text instead of exploding.\n rule.includes.push([body.toLowerCase()]);\n }\n return \"-\";\n },\n );\n\n for (const part of remainder.split(\"-\")) {\n const token = part.trim();\n if (!token) continue;\n\n const upper = /^<=?\\s*(\\d+(?:\\.\\d+)?)$/.exec(token);\n if (upper) {\n rule.maxPrice = Number(upper[1]);\n continue;\n }\n const lower = /^>=?\\s*(\\d+(?:\\.\\d+)?)$/.exec(token);\n if (lower) {\n rule.minPrice = Number(lower[1]);\n continue;\n }\n\n const negated = token.startsWith(\"!\");\n const body = negated ? token.slice(1) : token;\n const alternatives = body\n .split(\"|\")\n .map((alt) => alt.trim().toLowerCase())\n .filter(Boolean);\n if (alternatives.length === 0) continue;\n\n if (negated) rule.excludes.push(alternatives);\n else rule.includes.push(alternatives);\n }\n\n return rule;\n}\n\n/** A rule with nothing positive would match every deal in the feed. */\nexport function isUsableRule(rule: Rule): boolean {\n return rule.includes.length > 0 || rule.patterns.length > 0;\n}\n\nexport interface MatchOutcome {\n ok: boolean;\n reason: string;\n}\n\nexport function evaluateRule(\n rule: Rule,\n target: MatchTarget,\n matchContent: boolean,\n): MatchOutcome {\n const titleHay = target.title.toLowerCase();\n // Positive keywords stay on the title so 爆料原文里的「凑单苹果」不能误中.\n // Exclusions may also search the body when matchContent is on.\n const excludeHay = matchContent\n ? `${titleHay}\\n${target.content.toLowerCase()}`\n : titleHay;\n\n for (const group of rule.includes) {\n if (!group.some((alt) => titleHay.includes(alt))) {\n return { ok: false, reason: `缺少关键词 ${group.join(\"|\")}` };\n }\n }\n for (const pattern of rule.patterns) {\n if (!pattern.test(titleHay)) {\n return { ok: false, reason: `不匹配 ${pattern.source}` };\n }\n }\n for (const group of rule.excludes) {\n const hit = group.find((alt) => excludeHay.includes(alt));\n if (hit) return { ok: false, reason: `命中排除词 ${hit}` };\n }\n for (const pattern of rule.negPatterns) {\n if (pattern.test(excludeHay)) {\n return { ok: false, reason: `命中排除正则 ${pattern.source}` };\n }\n }\n\n // An unparseable price (\"包邮\", \"1元起\") passes the bounds rather than being\n // dropped, so a price constraint never hides a deal we cannot price.\n if (target.price !== null) {\n if (rule.maxPrice !== undefined && target.price > rule.maxPrice) {\n return { ok: false, reason: `价格 ${target.price} > ${rule.maxPrice}` };\n }\n if (rule.minPrice !== undefined && target.price < rule.minPrice) {\n return { ok: false, reason: `价格 ${target.price} < ${rule.minPrice}` };\n }\n }\n\n return { ok: true, reason: \"命中\" };\n}\n\n/** Returns the raw text of the first matching rule, or null. */\nexport function firstMatch(\n rules: Rule[],\n target: MatchTarget,\n matchContent: boolean,\n): string | null {\n for (const rule of rules) {\n if (evaluateRule(rule, target, matchContent).ok) return rule.raw;\n }\n return null;\n}\n\nexport function compileRules(raws: string[]): Rule[] {\n return raws.map(parseRule).filter(isUsableRule);\n}\n",[],[110],"1febd27d-9bee-4863-b9af-fd6494353665","db.ts","1b89192c-8f57-11f1-a00b-1607ee4eb77e",45,["D",1786638456082],5,"// Persistence layer, backed by val.town's SQLite (std/sqlite).\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite\";\nimport {\n DEFAULT_SETTINGS,\n EMPTY_STATS,\n nowSec,\n SEEN_LIMIT,\n type Settings,\n type Stats,\n} from \"./shared.ts\";\n\nconst DEALS_TABLE = \"smzdm_deals_v1\";\nconst KV_TABLE = \"smzdm_kv_v1\";\n\nexport interface SentDeal {\n articleId: string;\n title: string;\n url: string;\n mall: string;\n priceText: string;\n priceNum: number | null;\n bestPrice: number | null;\n worthy: number;\n publishedAt: number;\n firstSentAt: number;\n lastSentAt: number;\n sendCount: number;\n}\n\ninterface ExecuteResult {\n columns?: string[];\n rows?: unknown[];\n rowsAffected?: number;\n}\n\n/** Everything we bind is a scalar, which keeps us inside libSQL's InValue. */\ntype SqlArg = string | number | null;\n\n/**\n * std/sqlite returns rows as positional arrays; some versions also expose\n * named access. Normalise to plain objects so callers never depend on it.\n */\nfunction toObjects(result: ExecuteResult): Record[] {\n const columns = result.columns ?? [];\n const rows = result.rows ?? [];\n return rows.map((row) => {\n const record: Record = {};\n columns.forEach((column, index) => {\n record[column] = Array.isArray(row)\n ? row[index]\n : (row as Record)[column];\n });\n return record;\n });\n}\n\nasync function query(\n sql: string,\n args: SqlArg[] = [],\n): Promise[]> {\n const result = await sqlite.execute({ sql, args });\n return toObjects(result as unknown as ExecuteResult);\n}\n\nasync function run(sql: string, args: SqlArg[] = []): Promise {\n await sqlite.execute({ sql, args });\n}\n\nasync function batch(\n statements: { sql: string; args: SqlArg[] }[],\n): Promise {\n if (statements.length === 0) return;\n const client = sqlite as {\n execute: (statement: { sql: string; args: SqlArg[] }) => Promise;\n batch?: (statements: { sql: string; args: SqlArg[] }[]) => Promise;\n };\n if (statements.length > 1 && client.batch) {\n await client.batch(statements);\n return;\n }\n for (const statement of statements) await client.execute(statement);\n}\n\nlet ready: Promise | undefined;\n\nexport function initDb(): Promise {\n // A failed init must not be cached, or one transient error would break the\n // isolate for its whole lifetime.\n ready ??= createSchema().catch((error) => {\n ready = undefined;\n throw error;\n });\n return ready;\n}\n\nasync function createSchema(): Promise {\n await run(`\n CREATE TABLE IF NOT EXISTS ${DEALS_TABLE} (\n article_id TEXT PRIMARY KEY,\n title TEXT NOT NULL,\n url TEXT NOT NULL,\n mall TEXT NOT NULL DEFAULT '',\n price_text TEXT NOT NULL DEFAULT '',\n price_num REAL,\n best_price REAL,\n worthy INTEGER NOT NULL DEFAULT 0,\n published_at INTEGER NOT NULL DEFAULT 0,\n first_sent_at INTEGER NOT NULL,\n last_sent_at INTEGER NOT NULL,\n send_count INTEGER NOT NULL DEFAULT 1\n )\n `);\n await run(\n `CREATE INDEX IF NOT EXISTS idx_${DEALS_TABLE}_first_sent\n ON ${DEALS_TABLE}(first_sent_at)`,\n );\n await run(`\n CREATE TABLE IF NOT EXISTS ${KV_TABLE} (\n key TEXT PRIMARY KEY,\n value TEXT NOT NULL\n )\n `);\n}\n\n// --- Key/value settings -----------------------------------------------------\n\nasync function readKv(key: string): Promise {\n const rows = await query(\n `SELECT value FROM ${KV_TABLE} WHERE key = ?`,\n [key],\n );\n if (rows.length === 0) return undefined;\n try {\n return JSON.parse(String(rows[0].value)) as T;\n } catch {\n return undefined;\n }\n}\n\nasync function writeKv(key: string, value: unknown): Promise {\n await saveKv({ [key]: value });\n}\n\nexport async function saveKv(entries: Record): Promise {\n const statements = Object.entries(entries).map(([key, value]) => ({\n sql: `INSERT INTO ${KV_TABLE} (key, value) VALUES (?, ?)\n ON CONFLICT(key) DO UPDATE SET value = excluded.value`,\n args: [key, JSON.stringify(value)] as SqlArg[],\n }));\n await batch(statements);\n}\n\nexport interface Snapshot {\n settings: Settings;\n rules: string[];\n cursor: number;\n seenIds: string[];\n stats: Stats;\n}\n\n/** One round-trip for every KV the spider needs. */\nexport async function loadSnapshot(): Promise {\n const rows = await query(`SELECT key, value FROM ${KV_TABLE}`);\n const map = new Map();\n for (const row of rows) {\n try {\n map.set(String(row.key), JSON.parse(String(row.value)));\n } catch {\n // Ignore a corrupt row rather than failing the whole run.\n }\n }\n return {\n settings: {\n ...DEFAULT_SETTINGS,\n ...(map.get(\"settings\") as Partial | undefined),\n },\n rules: Array.isArray(map.get(\"rules\")) ? map.get(\"rules\") as string[] : [],\n cursor: Number(map.get(\"cursor\") ?? 0) || 0,\n seenIds: Array.isArray(map.get(\"seen_ids\"))\n ? map.get(\"seen_ids\") as string[]\n : [],\n stats: {\n ...EMPTY_STATS,\n ...(map.get(\"stats\") as Partial | undefined),\n },\n };\n}\n\nexport async function getSettings(): Promise {\n const stored = await readKv>(\"settings\");\n return { ...DEFAULT_SETTINGS, ...(stored ?? {}) };\n}\n\nexport async function patchSettings(\n patch: Partial,\n): Promise {\n const merged = { ...(await getSettings()), ...patch };\n await writeKv(\"settings\", merged);\n return merged;\n}\n\nexport async function getRules(): Promise {\n return (await readKv(\"rules\")) ?? [];\n}\n\nexport async function setRules(rules: string[]): Promise {\n await writeKv(\"rules\", rules);\n}\n\nexport async function getCursor(): Promise {\n return (await readKv(\"cursor\")) ?? 0;\n}\n\nexport async function setCursor(publishedAt: number): Promise {\n await writeKv(\"cursor\", publishedAt);\n}\n\nexport async function getSeenIds(): Promise {\n return (await readKv(\"seen_ids\")) ?? [];\n}\n\nexport function mergeSeen(previous: string[], ids: string[]): string[] {\n const have = new Set(previous);\n const next = [...previous];\n for (const id of ids) {\n if (have.has(id)) continue;\n have.add(id);\n next.push(id);\n }\n return next.length > SEEN_LIMIT ? next.slice(next.length - SEEN_LIMIT) : next;\n}\n\nexport async function getStats(): Promise {\n const stored = await readKv>(\"stats\");\n return { ...EMPTY_STATS, ...(stored ?? {}) };\n}\n\nexport async function setStats(stats: Stats): Promise {\n await writeKv(\"stats\", stats);\n}\n\n// --- Push history -----------------------------------------------------------\n\nfunction toSentDeal(row: Record): SentDeal {\n const number = (value: unknown): number | null =>\n value === null || value === undefined ? null : Number(value);\n return {\n articleId: String(row.article_id),\n title: String(row.title ?? \"\"),\n url: String(row.url ?? \"\"),\n mall: String(row.mall ?? \"\"),\n priceText: String(row.price_text ?? \"\"),\n priceNum: number(row.price_num),\n bestPrice: number(row.best_price),\n worthy: Number(row.worthy ?? 0),\n publishedAt: Number(row.published_at ?? 0),\n firstSentAt: Number(row.first_sent_at ?? 0),\n lastSentAt: Number(row.last_sent_at ?? 0),\n sendCount: Number(row.send_count ?? 0),\n };\n}\n\n/** One round-trip lookup for a whole batch of candidates. */\nexport async function findSentDeals(\n articleIds: string[],\n): Promise> {\n const found = new Map();\n if (articleIds.length === 0) return found;\n\n const batchSize = 200;\n for (let start = 0; start < articleIds.length; start += batchSize) {\n const batch = articleIds.slice(start, start + batchSize);\n const placeholders = batch.map(() => \"?\").join(\",\");\n const rows = await query(\n `SELECT * FROM ${DEALS_TABLE} WHERE article_id IN (${placeholders})`,\n batch,\n );\n for (const row of rows) {\n const deal = toSentDeal(row);\n found.set(deal.articleId, deal);\n }\n }\n return found;\n}\n\nexport interface RecordInput {\n articleId: string;\n title: string;\n url: string;\n mall: string;\n priceText: string;\n priceNum: number | null;\n worthy: number;\n publishedAt: number;\n}\n\nconst UPSERT_DEAL = `\nINSERT INTO ${DEALS_TABLE} (\n article_id, title, url, mall, price_text, price_num, best_price,\n worthy, published_at, first_sent_at, last_sent_at, send_count\n) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 1)\nON CONFLICT(article_id) DO UPDATE SET\n title = excluded.title,\n url = excluded.url,\n mall = excluded.mall,\n price_text = excluded.price_text,\n price_num = excluded.price_num,\n best_price = CASE\n WHEN excluded.price_num IS NOT NULL\n AND (best_price IS NULL OR excluded.price_num < best_price)\n THEN excluded.price_num ELSE best_price END,\n worthy = excluded.worthy,\n last_sent_at = excluded.last_sent_at,\n send_count = send_count + 1`;\n\nfunction upsertArgs(input: RecordInput, timestamp: number): SqlArg[] {\n return [\n input.articleId,\n input.title,\n input.url,\n input.mall,\n input.priceText,\n input.priceNum,\n input.priceNum,\n input.worthy,\n input.publishedAt,\n timestamp,\n timestamp,\n ];\n}\n\nexport async function recordSent(input: RecordInput): Promise {\n await recordSentMany([input]);\n}\n\nexport async function recordSentMany(inputs: RecordInput[]): Promise {\n if (inputs.length === 0) return;\n const timestamp = nowSec();\n await batch(\n inputs.map((input) => ({\n sql: UPSERT_DEAL,\n args: upsertArgs(input, timestamp),\n })),\n );\n}\n\nexport async function listSentDeals(\n limit: number,\n offset: number,\n): Promise {\n const rows = await query(\n `SELECT * FROM ${DEALS_TABLE}\n ORDER BY first_sent_at DESC LIMIT ? OFFSET ?`,\n [limit, offset],\n );\n return rows.map(toSentDeal);\n}\n\nexport async function countSentDeals(): Promise {\n const rows = await query(`SELECT COUNT(*) AS total FROM ${DEALS_TABLE}`);\n return Number(rows[0]?.total ?? 0);\n}\n\nexport async function clearSentDeals(): Promise {\n await run(`DELETE FROM ${DEALS_TABLE}`);\n}\n\n/** Ages out old rows so the table cannot grow without bound. */\nexport async function pruneSentDeals(retentionDays: number): Promise {\n await run(\n `DELETE FROM ${DEALS_TABLE} WHERE first_sent_at < ?`,\n [nowSec() - retentionDays * 86_400],\n );\n}\n",[],[118],["D",1786638456173],[],[19],"8f6e1d7c-59da-11f1-9bcd-ee650bb23af1",["D",1779893066929],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","","imageUrl","userId","96c2793d-57c8-41d2-a49a-d1ee82f9dd24","pinnedFileId","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","93515af8-c7b7-459c-9bf2-3bb3df310a75","tursoName","8f6e1d7c59da11f19bcdee650bb23af1","tursoHostname","8f6e1d7c59da11f19bcdee650bb23af1-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789494036282],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_11":12,"_186":182},"org",{"_11":12,"_180":-5,"_181":182,"_139":140,"_183":184,"_185":51},"branch",{"_17":65,"_13":16,"_33":39,"_172":129,"_173":-5,"_174":-5,"_175":-5,"_176":-5,"_177":-5,"_178":14,"_179":12},"isOwned","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","avatar","https://img.clerk.com/eyJ0eXBlIjoiZGVmYXVsdCIsImlpZCI6Imluc18yTm10cGhQUklwU0tqZHNiOHNnNGd2ZjJURFUiLCJyaWQiOiJ1c2VyXzJObXd1QWdCZmVtaUh0bGU5TkswOG1sM0xheiIsImluaXRpYWxzIjoiRlkifQ","tier","free","canRestrictVals","profileImageUrl","adoptOrgId",{"_33":129,"_13":14,"_44":242,"_131":-5,"_132":51,"_133":134,"_135":134,"_136":137,"_138":-5,"_139":140,"_141":-5,"_142":51,"_143":-5,"_144":-5,"_145":146,"_147":148,"_149":150,"_151":152,"_153":243,"_155":-5,"_156":157,"_158":157,"_159":160,"_161":160,"_162":160,"_163":157,"_164":157,"_165":244,"_167":245,"_169":246,"_171":51},[218,219,220,221,222,223,224,225],"files",[218,219,220,221,222,223,224,225],"counts",{"_208":146,"_209":210,"_211":146,"_32":212,"_213":214},[],{"_33":34,"_38":39,"_35":36,"_13":19,"_48":49,"_47":24,"_17":46,"_37":-5,"_44":206,"_40":41,"_52":-5,"_18":207},"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},["D",1786638456173],[19],"pulls","branches",1,"forks",{"_215":46,"_217":146},"connections",{"_215":146,"_216":146},"count","unfilledCount","emptyCount",{"_33":34,"_13":19,"_35":36,"_37":-5,"_38":39,"_40":41,"_42":43,"_44":126,"_17":46,"_47":24,"_48":49,"_50":51,"_52":-5,"_70":240,"_18":241},{"_33":117,"_13":118,"_35":119,"_37":-5,"_38":39,"_40":120,"_42":43,"_44":121,"_17":122,"_47":89,"_48":123,"_50":51,"_52":-5,"_70":238,"_18":239},{"_33":109,"_13":110,"_35":111,"_37":-5,"_38":39,"_40":112,"_42":43,"_44":113,"_17":88,"_47":89,"_48":114,"_50":51,"_52":-5,"_70":236,"_18":237},{"_33":101,"_13":102,"_35":103,"_37":-5,"_38":39,"_40":104,"_42":43,"_44":105,"_17":46,"_47":89,"_48":106,"_50":51,"_52":-5,"_70":234,"_18":235},{"_33":93,"_13":94,"_35":95,"_37":-5,"_38":39,"_40":96,"_42":43,"_44":97,"_17":78,"_47":89,"_48":98,"_50":51,"_52":-5,"_70":232,"_18":233},{"_33":83,"_13":84,"_35":85,"_37":-5,"_38":39,"_40":86,"_42":43,"_44":87,"_17":88,"_47":89,"_48":90,"_50":51,"_52":-5,"_70":230,"_18":231},{"_33":73,"_13":74,"_35":75,"_37":-5,"_38":39,"_40":76,"_42":43,"_44":77,"_17":78,"_47":79,"_48":80,"_50":51,"_52":-5,"_70":228,"_18":229},{"_33":62,"_13":63,"_35":64,"_37":-5,"_38":39,"_40":65,"_42":43,"_44":66,"_17":67,"_47":68,"_48":69,"_50":51,"_52":-5,"_70":226,"_18":227},[],[63],[],[74],[],[84],[],[94],[],[102],[],[110],[],[118],[],[19],["D",1779893066929],["D",1789494036282],{"_11":12,"_186":182},{"_11":12,"_180":-5,"_181":182,"_139":140,"_183":184,"_185":51},{"_17":65,"_13":16,"_33":39,"_172":129,"_173":-5,"_174":-5,"_175":-5,"_176":-5,"_177":-5,"_178":14,"_179":12},"ctx",{"_165":-5,"_203":280,"_281":282,"_283":284,"_285":51,"_286":51,"_287":51,"_288":43,"_289":-5,"_290":-7,"_291":-7,"_292":146,"_293":51,"_294":51,"_295":-5,"_296":51,"_297":51,"_298":51,"_299":-5},"ENV",{"_251":252,"_253":254,"_255":16,"_256":257,"_258":259,"_260":261,"_262":263,"_264":265,"_266":267,"_268":269,"_270":271,"_272":273,"_274":275,"_276":277,"_278":279},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]