UHD_Media_Tool/packages/adapters/src/sites/detail-description.ts

/**
 * 详情页简介清洗与结构化(DETAIL-1):纯函数,输入是抓取时得到的简介文本(可能残留 HTML / BBCode),
 * 输出「信息块」键值对(◎片名 / 【年代】/ 『导演』/ 片名: 等)与正文段落。
 * 清洗规则可配置:站点模板段、广告签名按正则整段或整行去掉,调用方可在默认规则上追加。
 */

export interface DetailInfoItem {
  key: string;
  value: string;
}

export interface StructuredDescription {
  infoBlock: DetailInfoItem[];
  paragraphs: string[];
}

export interface DescriptionRules {
  /** 整段去掉(对整篇文本匹配,可跨行):站点模板段、更新提示等 */
  dropBlocks: RegExp[];
  /** 整行去掉:广告、群号、转载声明、感谢签名、分隔线 */
  dropLines: RegExp[];
}

export const DEFAULT_DESCRIPTION_RULES: DescriptionRules = {
  dropBlocks: [
    // 憨憨豆瓣信息块末尾:「注意：该信息最后更新于…点击 这里 更新信息」
    /注意[:：]\s*该信息最后更新于[\s\S]{0,80}?更新信息/g,
  ],
  dropLines: [
    /^[-=_*~—─━·.]{5,}$/,
    /^(?:QQ|qq|微信|TG|Telegram|电报|Discord)\s*群?(?:号)?\s*[:：]/,
    /(?:禁止|严禁|请勿|不得).{0,12}转载|转载.{0,8}(?:请注明|须注明)/,
    /^(?:感谢|鸣谢|Thanks?\s+to)\s*.{0,20}(?:分享|发布|制作|压制|上传)/i,
    /^(?:本资源|本种子?|资源)?由\s*\S{1,24}\s*(?:小组|组)?\s*(?:制作|压制|发布|分享)\S{0,4}$/,
    /^(?:点击|更多).{0,10}(?:下载|资源).{0,20}(?:访问|请到|见)/,
  ],
};

/** 常见信息块键(去掉空白后比较);冒号式 / 【】/『』 只认这些,避免把正文里的「【第一版】」当键 */
const KNOWN_KEYS = [
  "译名",
  "片名",
  "原名",
  "又名",
  "别名",
  "中文名",
  "英文名",
  "年代",
  "年份",
  "产地",
  "国家",
  "地区",
  "国家地区",
  "制片国家/地区",
  "类别",
  "类型",
  "语言",
  "对白语言",
  "字幕",
  "上映日期",
  "首播",
  "首播日期",
  "IMDb评分",
  "IMDB评分",
  "IMDb链接",
  "IMDB链接",
  "豆瓣评分",
  "豆瓣链接",
  "TMDB",
  "TMDb链接",
  "评分",
  "票数",
  "集数",
  "季数",
  "片长",
  "单集片长",
  "视频尺寸",
  "文件格式",
  "文件大小",
  "导演",
  "编剧",
  "主演",
  "演员",
  "配音",
  "制片人",
  "标签",
  "标语",
  "简介",
  "剧情简介",
  "内容简介",
  "剧情概要",
  "剧情介绍",
  "获奖情况",
];
const KNOWN = new Set(KNOWN_KEYS.map((k) => k.toLowerCase()));
/** 值就是正文的键:内容进段落 */
const DESC_KEYS = new Set(["简介", "剧情简介", "内容简介", "剧情概要", "剧情介绍"]);
/** 人名列表:续行合并,修复「Kai Xu周也」这类元素拼接 */
const PEOPLE_KEYS = new Set(["导演", "编剧", "主演", "演员", "配音", "制片人"]);

const WS = "[\\s\\u3000]";
const KEY_ALTERNATION = [...KNOWN_KEYS]
  .sort((a, b) => b.length - a.length)
  .map((k) => [...k].map((c) => c.replace(/[.*+?^${}()|[\]\\/]/g, "\\$&")).join(`${WS}*`))
  .join("|");
/** ◎ 标记:认已知键,否则取到第一个空白 / 冒号为止的短词 */
const MARK_KNOWN = new RegExp(`^◎${WS}*(${KEY_ALTERNATION})${WS}*[:：]?${WS}*(.*)$`, "i");
const MARK_ANY = /^◎[\s　]*([^\s　:：]{1,12})[\s　]*[:：]?[\s　]*(.*)$/;
const BRACKET = new RegExp(`^(?:【|『|\\[)${WS}*(${KEY_ALTERNATION})${WS}*(?:】|』|\\])${WS}*[:：]?${WS}*(.*)$`, "i");
const COLON = new RegExp(`^(${KEY_ALTERNATION})${WS}*[:：]${WS}*(.*)$`, "i");

const ENTITIES: Record<string, string> = {
  nbsp: " ",
  amp: "&",
  lt: "<",
  gt: ">",
  quot: '"',
  apos: "'",
  ensp: " ",
  emsp: " ",
};

function decodeEntities(s: string): string {
  return s.replace(/&(#x[0-9a-f]+|#\d+|[a-z]+);/gi, (all, e: string) => {
    if (e[0] === "#") {
      const code = e[1] === "x" || e[1] === "X" ? Number.parseInt(e.slice(2), 16) : Number.parseInt(e.slice(1), 10);
      return Number.isFinite(code) && code > 0 && code < 0x110000 ? String.fromCodePoint(code) : all;
    }
    return ENTITIES[e.toLowerCase()] ?? all;
  });
}

/** 去 HTML / BBCode 残留:图片整段去掉,链接留文字,块级标签换行,实体解码 */
export function stripMarkup(text: string): string {
  return decodeEntities(
    text
      .replace(/\r/g, "")
      .replace(/<(script|style)\b[\s\S]*?<\/\1>/gi, "")
      // 站点截断在标签中间:「…<br…」
      .replace(/<[a-z/][^>\n]*…?$/i, "…")
      .replace(/<img\b[^>]*>/gi, "")
      .replace(/<br\s*\/?>[ \t]*\n?/gi, "\n")
      .replace(/<\/(?:p|div|li|tr|h[1-6]|blockquote|fieldset|pre)>[ \t]*\n?/gi, "\n")
      .replace(/<[a-z/!][^>]*>/gi, "")
      .replace(/\[img(?:=[^\]]*)?\][^[]*\[\/img\]|\[img=[^\]]+\]/gi, "")
      .replace(/\[url=[^\]]*\]([\s\S]*?)\[\/url\]/gi, "$1")
      .replace(
        /\[\/?(?:b|i|u|s|color|size|font|align|center|left|right|url|quote|code|spoiler|hide|box|list|\*|pre|table|tr|td)(?:=[^\]]*)?\]/gi,
        "",
      ),
  ).replace(/ /g, " ");
}

/** 清洗:去标记、按规则删段删行、行尾空白、连续空行压成一行 */
export function cleanDescription(text: string, rules: DescriptionRules = DEFAULT_DESCRIPTION_RULES): string {
  let t = stripMarkup(text);
  for (const re of rules.dropBlocks) t = t.replace(re, "");
  return t
    .split("\n")
    .map((l) => l.replace(/[\s　]+$/, ""))
    .filter((l) => {
      const s = l.trim().replace(/^　+/, "");
      return !s || !rules.dropLines.some((re) => re.test(s));
    })
    .join("\n")
    .replace(/\n{3,}/g, "\n\n")
    .trim();
}

interface KeyHit {
  key: string;
  value: string;
}

const normKey = (k: string) => k.replace(/[\s　]+/g, "");

function matchKey(line: string): KeyHit | null {
  const m = MARK_KNOWN.exec(line) ?? MARK_ANY.exec(line) ?? BRACKET.exec(line) ?? COLON.exec(line);
  if (!m) return null;
  const key = normKey(m[1]!);
  if (!key) return null;
  // 冒号式:键后面要么没东西(值在下一行),要么是冒号分隔的值;「片名」开头的正文句子不当键
  if (!line.startsWith("◎") && !KNOWN.has(key.toLowerCase())) return null;
  return { key, value: (m[2] ?? "").trim() };
}

const isSeparator = (l: string) => /^[-=_*~—─━·.]{5,}$/.test(l);
const SENTENCE_END = /[。！？!?.…」』"”)）]$/;

/** 相邻行合并成段:英文硬换行(下一行小写开头且上一行没有句末标点)接成一行,其余每行一段 */
function toParagraphs(lines: string[]): string[] {
  const out: string[] = [];
  let prevJoinable = false;
  for (const raw of lines) {
    const l = raw.replace(/^[\s　]+/, "").trim();
    if (!l) {
      prevJoinable = false;
      continue;
    }
    if (prevJoinable && /^[a-z]/.test(l) && out.length) out[out.length - 1] += ` ${l}`;
    else out.push(l);
    prevJoinable = /[A-Za-z,]$/.test(l) && !SENTENCE_END.test(l);
  }
  return out;
}

function tidyValue(key: string, lines: string[]): string {
  const parts = lines.map((l) => l.replace(/[\s　]+/g, " ").trim()).filter(Boolean);
  if (PEOPLE_KEYS.has(key)) {
    return parts
      .flatMap((p) => p.replace(/([a-z])(?=[一-鿿])/g, "$1\u0000").split("\u0000"))
      .map((p) => p.trim())
      .filter(Boolean)
      .join(" / ");
  }
  return parts.every((p) => p.length <= 40) ? parts.join(" / ") : parts.join("\n");
}

const URL_ONLY = /^(?:https?:\/\/\S+|www\.\S+)$/i;
const isLinkKey = (k: string) => /(?:链接|网址|link)$/i.test(k) || /^tmdb$/i.test(k);

/**
 * 信息块 + 段落。信息块的键:◎ 开头任意短键;【】『』[] 包裹或「键:」开头只认常见键。
 * 键后的值可在同一行,也可在下一个非空行(憨憨豆瓣块);续行(主演多行)并入上一键,空行结束。
 * 简介类键的内容进段落;信息块之前的零碎短行(海报卡片的片名 / 年份 / 评分)在有信息块时丢弃。
 */
export function structureDescription(
  text: string | null | undefined,
  rules: DescriptionRules = DEFAULT_DESCRIPTION_RULES,
): StructuredDescription {
  if (!text?.trim()) return { infoBlock: [], paragraphs: [] };
  const lines = cleanDescription(text, rules).split("\n");
  const info: { key: string; lines: string[] }[] = [];
  const lead: string[] = [];
  const body: string[] = [];
  let cur: { key: string; lines: string[] } | "desc" | null = null;
  const pushBody = (l: string) => (info.length ? body : lead).push(l);

  for (const line of lines) {
    const trimmed = line.trim();
    if (isSeparator(trimmed)) {
      if (cur && cur !== "desc" && cur.lines.length) cur = null;
      if (cur === "desc") body.push("");
      continue;
    }
    const hit = trimmed ? matchKey(trimmed) : null;
    if (hit) {
      if (DESC_KEYS.has(hit.key)) {
        cur = "desc";
        body.push("");
        if (hit.value) body.push(hit.value);
      } else {
        const entry = { key: hit.key, lines: hit.value ? [hit.value] : [] };
        info.push(entry);
        cur = entry;
      }
      continue;
    }
    if (!trimmed) {
      if (cur === "desc") body.push("");
      else if (cur?.lines.length) cur = null;
      else if (!cur) pushBody("");
      continue;
    }
    if (cur && cur !== "desc") cur.lines.push(line);
    else pushBody(line);
  }

  const seen = new Set<string>();
  const infoBlock: DetailInfoItem[] = [];
  for (const e of info) {
    const value = tidyValue(e.key, e.lines);
    if (!value || seen.has(e.key)) continue;
    if (isLinkKey(e.key) && value.split(/\s+/).every((v) => URL_ONLY.test(v) || !v)) continue;
    seen.add(e.key);
    infoBlock.push({ key: e.key, value });
  }
  // 信息块前面只有零碎短行(卡片摘要)就丢掉;有像样的句子则保留为正文
  const leadParas = toParagraphs(lead);
  const keepLead = !infoBlock.length || leadParas.some((p) => p.length > 24 || /[。！？!?]$/.test(p));
  const paragraphs = [...(keepLead ? leadParas : []), ...toParagraphs(body)];
  return { infoBlock, paragraphs };
}