fix(adapters): 观众详情页按真实页面适配(DET-2)

- 促销:详情 h1 是 font.free / font.halfdown 等 + 剩余时间 span[title],新增 umt.detail.promo 覆盖列表促销字段(以前一律判成无促销并覆盖列表的免费)
- 标题去掉 h1 里的「剩余时间：」;发布时间不再取标题区的促销到期时间
- 做种 / 下载 / 完成数、种子 Hash:umt.detail.fields.seeders / leechers / snatched / infoHash
- 副标题去掉评分徽标;标签只取最内层;上传者取 span.torrent-uploader-wrap,标签精确匹配(避开「发布者带宽」)
- 简介:去掉站点声明(page.descriptionStrip)与 MediaInfo 控件(codetop / show / hide),源码换行不再算一行
- 截图排除豆瓣海报与站点横幅
- 列表副标题改取 span.torrent-subtitle-text(以前前面粘着标签文字)
- 脱敏真实样本:详情 2 页进 differential/detail.test.ts,列表 1 页进 MP 差分

Co-Authored-By: Claude Opus 5.5 <[email protected]>

packages/adapters/src/sites/spider/detail-page.ts

@@ -43,17 +43,35 @@ const normLabel = (t: string) =>
     .replace(/[:：]+$/, "")
     .toLowerCase();
 
-/** 保留换行的文本:<br> 与块级元素换行 */
+/** 换行占位:先把 <br> / 块级换行记成它,再去掉 HTML 源码里的换行(浏览器按空白处理),最后还原 */
+const BR = "\u0000";
+
+/**
+ * 保留换行的文本:<br> 与块级元素换行。源码换行不算换行(观众的简介是 `<br />\r\n`,
+ * 两者都算会让每行之间多一空行);<pre> 里的换行保留。
+ */
 function blockText($: CheerioAPI, el: El): string {
   const c = el.clone();
   c.find("script, style").remove();
-  c.find("br").replaceWith("\n");
-  c.find("div, p, li, tr, pre, fieldset, blockquote, h1, h2, h3, h4, h5, h6").each((_, n) => {
-    $(n).append("\n");
+  c.find("pre, textarea").each((_, n) => {
+    $(n).text($(n).text().replace(/\r?\n/g, BR));
   });
-  return tidyText(c.text());
+  c.find("br").replaceWith(BR);
+  c.find("div, p, li, tr, pre, fieldset, blockquote, h1, h2, h3, h4, h5, h6").each((_, n) => {
+    $(n).append(BR);
+  });
+  const text = c
+    .text()
+    .replace(/\r?\n/g, " ")
+    // `<br />\r\n下一行` 的源码换行变成了行首一个空格
+    .replace(new RegExp(`${BR} `, "g"), BR)
+    .replaceAll(BR, "\n");
+  return tidyText(text);
 }
 
+/** NexusPHP MediaInfo 控件的外壳:标题栏 div.codetop、摘要表 div.show、原文 div.hide(观众三者相邻) */
+const MI_WIDGET = "div.codetop, div.show, div.hide";
+
 const SIZE_TAIL = /^(.*?)\s*[(（[]\s*([\d.,]+\s*[KMGTP]?i?B)\s*[)）\]]\s*$/i;
 const HAS_EXT = /\.[A-Za-z0-9]{2,5}$/;
 
@@ -113,8 +131,9 @@ export function extractDetailPage($: CheerioAPI, conf: DetailConf, base: string)
   };
   addLabels(conf.rowLabel, (e) => e.nextAll(conf.rowValue).first());
   for (const sel of p.labelSelectors ?? []) addLabels(sel, (e) => e.next());
-  const byLabel = (names: string[]): El | null =>
-    pairs.find((x) => names.some((n) => x.label.startsWith(normLabel(n))))?.value ?? null;
+  const byLabel = (names: string[], exact = false): El | null =>
+    pairs.find((x) => names.some((n) => (exact ? x.label === normLabel(n) : x.label.startsWith(normLabel(n)))))
+      ?.value ?? null;
   page.probe.labels = [...new Set(pairs.map((x) => x.label))].slice(0, 40);
 
   // 简介容器
@@ -170,10 +189,30 @@ export function extractDetailPage($: CheerioAPI, conf: DetailConf, base: string)
   page.probe.imageCount = imgs.length;
   page.screenshots = imgs.slice(0, MAX_SCREENSHOTS);
 
-  // 简介纯文本:去掉 MediaInfo 块与图片
+  // 简介纯文本:去掉站点固定内容、MediaInfo 块(连同控件标题栏 / 摘要表)与图片
   if (desc?.length) {
     const c = desc.clone();
-    if (miBlockIndex >= 0) c.find(MI_BLOCKS).eq(miBlockIndex).remove();
+    if (miBlockIndex >= 0) {
+      const isMi = (n: Parameters<typeof $>[0]) => looksLikeMediaInfo(blockText($, $(n)));
+      // 同一份 MediaInfo 常出现两次(摘要 div.show 与原文 div.hide):另外去掉最内层的 MediaInfo 块,
+      // 在控件外壳(div.codetop / div.show / div.hide)里时连同相邻外壳一起去掉
+      const extra = c
+        .find(MI_BLOCKS)
+        .filter((_, n) => isMi(n) && !$(n).find(MI_BLOCKS).toArray().some(isMi))
+        .map((_, n) => {
+          const e = $(n);
+          const shell = e.parent().is(MI_WIDGET) ? e.parent() : e;
+          const out = [shell.get(0)!];
+          if (!shell.is(MI_WIDGET)) return out;
+          for (let x = shell.prev(); x.is(MI_WIDGET) && out.length < 3; x = x.prev()) out.push(x.get(0)!);
+          for (let x = shell.next(); x.is(MI_WIDGET) && out.length < 4; x = x.next()) out.push(x.get(0)!);
+          return out;
+        })
+        .get();
+      c.find(MI_BLOCKS).eq(miBlockIndex).remove();
+      for (const w of extra) $(w).remove();
+    }
+    if (p.descriptionStrip) c.find(p.descriptionStrip).remove();
     c.find("img").remove();
     const t = blockText($, c);
     if (t) page.descriptionText = t.slice(0, 20_000);
@@ -221,17 +260,16 @@ export function extractDetailPage($: CheerioAPI, conf: DetailConf, base: string)
   // 标签
   const tagEl = p.tags ? $(p.tags) : byLabel(LABELS.tags);
   if (tagEl?.length) {
-    const parts = tagEl.find("span, a").length
-      ? tagEl
-          .find("span, a")
-          .map((_, n) => $(n).text().trim())
-          .get()
-      : tagEl.text().split(/[\s,，|/]+/);
+    // 只取最内层的 span / a(观众外面还包了一层 span.detail-tag-list,取外层会把全部标签连成一个)
+    const leaves = tagEl.find("span, a").filter((_, n) => $(n).find("span, a").length === 0);
+    const parts = leaves.length ? leaves.map((_, n) => $(n).text().trim()).get() : tagEl.text().split(/[\s,，|/]+/);
     page.tags = [...new Set(parts.map((t) => t.trim()).filter((t) => t && t.length <= 20))].slice(0, 20);
   }
 
   // 上传者
-  const upEl = p.uploader ? $(p.uploader).first() : byLabel(LABELS.uploader);
+  // 上传者标签精确匹配:观众另有「发布者带宽」一行
+  const upSel = p.uploader ? $(p.uploader).first() : null;
+  const upEl = upSel?.length ? upSel : byLabel(LABELS.uploader, true);
   if (upEl?.length) {
     const a = upEl.is("a") ? upEl : upEl.find("a[href*='userdetails.php']").first();
     const name = (a.length ? a.text() : upEl.text()).replace(/\s+/g, " ").trim();