douyin-archive/app/api/search/route.ts

import { NextResponse } from "next/server";
import { isAsrContentTag } from "@/lib/asr-tags";
import { json } from "@/lib/json";
import { getFileUrl } from "@/lib/minio";
import { prisma } from "@/lib/prisma";

type SearchRow = {
  id: string;
  awemeId: string;
  type: "video" | "image";
  rank: number;
  snippet: string;
  contentTags: string[];
};

function sanitizeSnippetHtml(snippet: string) {
  return snippet
    .replaceAll("&", "&")
    .replaceAll("<", "&lt;")
    .replaceAll(">", "&gt;")
    .replaceAll("&lt;mark&gt;", "<mark>")
    .replaceAll("&lt;/mark&gt;", "</mark>");
}

export async function GET(req: Request) {
  const { searchParams } = new URL(req.url);
  const q = (searchParams.get("q") || "").trim();
  const rawTag = (searchParams.get("tag") || "").trim();
  const contentTag = rawTag && isAsrContentTag(rawTag) ? rawTag : null;
  const page = Math.max(1, Number(searchParams.get("page") || 1));
  const limit = Math.min(
    50,
    Math.max(1, Number(searchParams.get("limit") || 20)),
  );
  const offset = (page - 1) * limit;

  if (rawTag && !contentTag) {
    return NextResponse.json({ error: "Invalid tag" }, { status: 400 });
  }

  if (!q && !contentTag) {
    return NextResponse.json({ results: [], total: 0 });
  }

  try {
    // 核心查询：Postgres 中文全文检索（视频转写 + 视频desc + 图文desc）+ 相关度排序 + 摘要高亮
    // 说明：
    // - 合并搜索来源：VideoTranscript.transcript_zh_tsv、Video.desc 与 ImagePost.desc
    // - 使用 UNION ALL 合并视频和图文的搜索结果
    // - rank 取两者匹配得分的最大值，snippet 优先展示转写匹配，否则展示 desc 匹配
    const rows = q
      ? await prisma.$queryRaw<SearchRow[]>`
          WITH tsq AS (
            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
          )
          -- 视频搜索结果
          SELECT
            COALESCE(vt.id, 'desc:' || v.aweme_id) AS id,
            v.aweme_id AS "awemeId",
            'video'::text AS type,
            GREATEST(
              COALESCE(ts_rank(vt.transcript_zh_tsv, tsq.query), 0),
              COALESCE(ts_rank(to_tsvector('zhcfg', v.desc), tsq.query), 0)
            ) AS rank,
            CASE
              WHEN vt.transcript_zh_tsv @@ tsq.query THEN
                ts_headline(
                  'zhcfg',
                  array_to_string(vt.transcript, ' '),
                  tsq.query,
                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
                )
              ELSE
                ts_headline(
                  'zhcfg',
                  v.desc,
                  tsq.query,
                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
                )
            END AS snippet,
            CASE
              WHEN cardinality(v.manual_content_tags) > 0 THEN v.manual_content_tags
              ELSE COALESCE(vt.content_tags, ARRAY[]::text[])
            END AS "contentTags"
          FROM "Video" v
          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id,
               tsq
          WHERE
            (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
            AND CASE
              WHEN ${contentTag}::text IS NULL THEN TRUE
              WHEN cardinality(v.manual_content_tags) > 0 THEN v.manual_content_tags @> ARRAY[${contentTag}]::text[]
              ELSE COALESCE(vt.content_tags, ARRAY[]::text[]) @> ARRAY[${contentTag}]::text[]
            END

          UNION ALL

          -- 图文搜索结果
          SELECT
            'img_desc:' || ip.aweme_id AS id,
            ip.aweme_id AS "awemeId",
            'image'::text AS type,
            ts_rank(to_tsvector('zhcfg', ip.desc), tsq.query) AS rank,
            ts_headline(
              'zhcfg',
              ip.desc,
              tsq.query,
              'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
            ) AS snippet,
            ARRAY[]::text[] AS "contentTags"
          FROM "ImagePost" ip, tsq
          WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
            AND ${contentTag}::text IS NULL

          ORDER BY rank DESC
          OFFSET ${offset}
          LIMIT ${limit};
        `
      : await prisma.$queryRaw<SearchRow[]>`
          SELECT
            COALESCE(vt.id, 'tag:' || v.aweme_id) AS id,
            v.aweme_id AS "awemeId",
            'video'::text AS type,
            1.0::double precision AS rank,
            COALESCE(
              NULLIF(array_to_string(vt.transcript[1:2], ' '), ''),
              vt.non_speech_summary,
              v.desc,
              '标签匹配'
            ) AS snippet,
            CASE
              WHEN cardinality(v.manual_content_tags) > 0 THEN v.manual_content_tags
              ELSE COALESCE(vt.content_tags, ARRAY[]::text[])
            END AS "contentTags"
          FROM "Video" v
          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
          WHERE CASE
            WHEN cardinality(v.manual_content_tags) > 0 THEN v.manual_content_tags @> ARRAY[${contentTag}]::text[]
            ELSE COALESCE(vt.content_tags, ARRAY[]::text[]) @> ARRAY[${contentTag}]::text[]
          END
          ORDER BY v.created_at DESC
          OFFSET ${offset}
          LIMIT ${limit};
        `;

    // 查询总数（参数化，避免注入）
    const totalRows = q
      ? await prisma.$queryRaw<{ count: number }[]>`
          WITH tsq AS (
            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
          )
          SELECT (
            (SELECT COUNT(*)::int FROM "Video" v
             LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
             WHERE (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
               AND CASE
                 WHEN ${contentTag}::text IS NULL THEN TRUE
                 WHEN cardinality(v.manual_content_tags) > 0 THEN v.manual_content_tags @> ARRAY[${contentTag}]::text[]
                 ELSE COALESCE(vt.content_tags, ARRAY[]::text[]) @> ARRAY[${contentTag}]::text[]
               END)
            +
            (SELECT COUNT(*)::int FROM "ImagePost" ip
             WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
               AND ${contentTag}::text IS NULL)
          ) AS count
          FROM tsq;
        `
      : await prisma.$queryRaw<{ count: number }[]>`
          SELECT COUNT(*)::int AS count
          FROM "Video" v
          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
          WHERE CASE
            WHEN cardinality(v.manual_content_tags) > 0 THEN v.manual_content_tags @> ARRAY[${contentTag}]::text[]
            ELSE COALESCE(vt.content_tags, ARRAY[]::text[]) @> ARRAY[${contentTag}]::text[]
          END;
        `;

    // 分离视频和图文ID
    const videoIds = rows
      .filter((r) => r.type === "video")
      .map((r) => r.awemeId);
    const imagePostIds = rows
      .filter((r) => r.type === "image")
      .map((r) => r.awemeId);

    // 批量查询视频元信息
    const videos =
      videoIds.length > 0
        ? (
            await prisma.video.findMany({
              where: { aweme_id: { in: videoIds } },
              select: {
                aweme_id: true,
                desc: true,
                cover_url: true,
                video_url: true,
                duration_ms: true,
                author: true,
              },
            })
          ).map((v) => ({
            ...v,
            cover_url: getFileUrl(v.cover_url || ""),
            author: {
              ...v.author,
              avatar_url: getFileUrl(v.author.avatar_url || ""),
            },
            video_url: getFileUrl(v.video_url || ""),
          }))
        : [];

    // 批量查询图文元信息
    const imagePosts =
      imagePostIds.length > 0
        ? (
            await prisma.imagePost.findMany({
              where: { aweme_id: { in: imagePostIds } },
              select: {
                aweme_id: true,
                desc: true,
                author: true,
                images: {
                  orderBy: { order: "asc" },
                  take: 1,
                  select: {
                    url: true,
                    width: true,
                    height: true,
                  },
                },
              },
            })
          ).map((ip) => ({
            ...ip,
            author: {
              ...ip.author,
              avatar_url: getFileUrl(ip.author.avatar_url || ""),
            },
            cover_url: ip.images[0] ? getFileUrl(ip.images[0].url) : null,
          }))
        : [];

    return json({
      results: rows.map((r) => ({
        ...r,
        snippet: sanitizeSnippetHtml(r.snippet),
        video:
          r.type === "video"
            ? videos.find((v) => v.aweme_id === r.awemeId)
            : undefined,
        imagePost:
          r.type === "image"
            ? imagePosts.find((ip) => ip.aweme_id === r.awemeId)
            : undefined,
      })),
      total: totalRows?.[0]?.count ?? 0,
      page,
      limit,
    });
  } catch (err) {
    console.error("Search error:", err);
    return NextResponse.json({ error: "Search failed" }, { status: 500 });
  }
}