feat(search): surface PDF highlight bodies in Global Search (#864)

server/api/src/routes/search.ts

@@ -12,6 +12,30 @@
  * - `scope=own` restricts to pages under the caller's default note.
  * - `scope=shared` spans pages in notes the caller can access (owner, accepted
  *   member, or domain rule).
+ *
+ * PDF ハイライト統合 (Issue #864 / #389 follow-up):
+ * - 同じ検索 q を用いて `pdf_highlights.text` も対象に含める。
+ * - ハイライトは常に呼び出し元 (`owner_id = userId`) 所有のみを返し、scope に依らず
+ *   他ユーザーに漏れない（テーブル単体で所有者を持つ前提）。
+ * - 元ソースが `kind="pdf_local"` のもののみ対象（JOIN で防御的にフィルタ）。
+ * - レスポンスは `kind` 識別子付きの discriminated union で返す
+ *   (`kind="page"` / `kind="pdf_highlight"`)。クライアントはこれで分岐する。
+ * - 環境変数 `PDF_HIGHLIGHT_SEARCH_DISABLED=1` をセットすると、ハイライト検索だけを
+ *   無効化できる（運用上のセーフティ）。ページ検索には影響しない。
+ *
+ * PDF highlight integration (Issue #864, follow-up to #389):
+ * - The same query string also probes `pdf_highlights.text`.
+ * - Highlights are only returned to their owner — scope does NOT widen this; the
+ *   table has a denormalized `owner_id` precisely for this case.
+ * - Defensive `JOIN sources` ensures we never surface highlights whose owning
+ *   source row is somehow not `kind="pdf_local"` anymore.
+ * - The response is now a discriminated union tagged by `kind`:
+ *   `"page"` (existing rows) and `"pdf_highlight"` (new rows). Clients branch on
+ *   `kind` and the highlight rows carry `source_id` / `pdf_page` / `highlight_id`
+ *   / `derived_page_id` so the UI can deep-link back to the PDF viewer or the
+ *   derived Zedi page.
+ * - Set `PDF_HIGHLIGHT_SEARCH_DISABLED=1` to disable just the highlight part
+ *   (kill switch); page search is unaffected.
  */
 import { Hono } from "hono";
 import { sql } from "drizzle-orm";
@@ -30,6 +54,15 @@ function clampLimit(raw: string | undefined): number {
   return Math.min(Math.max(safe, 1), 100);
 }
 
+/**
+ * ハイライト検索のキルスイッチ。`PDF_HIGHLIGHT_SEARCH_DISABLED=1` or `=true` で有効。
+ * Kill switch for the highlight search branch.
+ */
+function isPdfHighlightSearchDisabled(): boolean {
+  const v = (process.env.PDF_HIGHLIGHT_SEARCH_DISABLED ?? "").trim().toLowerCase();
+  return v === "1" || v === "true";
+}
+
 const app = new Hono<AppEnv>();
 
 app.get("/", authRequired, async (c) => {
@@ -46,8 +79,14 @@ app.get("/", authRequired, async (c) => {
   const limit = clampLimit(c.req.query("limit"));
   const pattern = `%${escapeLike(query)}%`;
 
-  const searchColumns = sql`p.id, p.title, p.content_preview, p.updated_at, p.note_id,
-             pc.content_text`;
+  // 検索条件用にだけ `content_text` を WHERE に登場させるが、SELECT には含めない。
+  // SELECT に流すと API 経由でページ本文が丸ごと露出し得る（PR #873 review:
+  // CodeRabbit）。クライアントが消費するのは `content_preview` のみ。
+  //
+  // `content_text` is used in the WHERE clause for matching but is NOT in the
+  // SELECT list — otherwise the API would leak full page bodies (PR #873 review:
+  // CodeRabbit). Clients only consume `content_preview`.
+  const searchColumns = sql`p.id, p.title, p.content_preview, p.updated_at, p.note_id`;
 
   const normalizedEmail = typeof userEmailRaw === "string" ? userEmailRaw.trim().toLowerCase() : "";
   const emailDomain = extractEmailDomain(normalizedEmail);
@@ -65,10 +104,10 @@ app.get("/", authRequired, async (c) => {
         )`
       : sql``;
 
-  let results;
+  let pageRows: unknown[] = [];
 
   if (scope === "shared") {
-    results = await db.execute(sql`
+    const sharedResults = await db.execute(sql`
       SELECT ${searchColumns}
       FROM pages p
       LEFT JOIN page_contents pc ON pc.page_id = p.id
@@ -98,12 +137,17 @@ app.get("/", authRequired, async (c) => {
       ORDER BY p.updated_at DESC
       LIMIT ${limit}
     `);
+    pageRows = sharedResults.rows;
   } else {
     const defaultNote = await getDefaultNoteOrNull(db, userId);
     if (!defaultNote) {
-      return c.json({ results: [] });
+      // デフォルトノートが無い場合でもハイライト検索は走り得るので、ページ部だけ空配列に。
+      // Even without a default note, highlight search can still run, so only the
+      // page branch short-circuits here.
+      const highlightRows = await runPdfHighlightSearch(db, userId, pattern, limit);
+      return c.json({ results: highlightRows });
     }
-    results = await db.execute(sql`
+    const ownResults = await db.execute(sql`
       SELECT ${searchColumns}
       FROM pages p
       LEFT JOIN page_contents pc ON pc.page_id = p.id
@@ -116,9 +160,103 @@ app.get("/", authRequired, async (c) => {
       ORDER BY p.updated_at DESC
       LIMIT ${limit}
     `);
+    pageRows = ownResults.rows;
   }
 
-  return c.json({ results: results.rows });
+  // 契約フィールドのみを明示マップして response に流す。SQL の SELECT に直接含まれない
+  // カラム (owner_id / thumbnail_url / source_url) は明示的に null/undefined で埋めて
+  // 型 (`SearchPageResultRow`) との整合を取る。raw row を spread で流すと将来 SELECT
+  // を増やしたとき静かに API が漏れるので、PR #873 review (CodeRabbit) で明示化した。
+  //
+  // Map only the contracted fields explicitly. We do not spread raw SQL rows
+  // because any future SELECT addition would silently widen the API payload
+  // (PR #873 review: CodeRabbit). Columns not in the current SELECT are
+  // emitted as `null`/`undefined` to stay aligned with `SearchPageResultRow`.
+  const taggedPageRows = pageRows.map((row) => {
+    const r = row as {
+      id: string;
+      note_id: string;
+      title: string | null;
+      content_preview: string | null;
+      updated_at: string;
+    };
+    return {
+      kind: "page" as const,
+      id: r.id,
+      note_id: r.note_id,
+      // `owner_id` / `thumbnail_url` / `source_url` は将来 SELECT に追加する想定の
+      // プレースホルダ。現状の SQL では返らないので明示的に null/undefined を入れる。
+      // Placeholders for columns not yet in SELECT; emitted explicitly so the
+      // payload shape stays stable for the discriminated union.
+      owner_id: null,
+      title: r.title,
+      content_preview: r.content_preview,
+      thumbnail_url: null,
+      source_url: null,
+      updated_at: r.updated_at,
+    };
+  });
+
+  const highlightRows = await runPdfHighlightSearch(db, userId, pattern, limit);
+
+  // ページ検索とハイライト検索のそれぞれが `LIMIT ${limit}` を持つので、無加工に連結
+  // すると endpoint は最大 2*limit 行返してしまう (PR #873 review: codex)。クライアント
+  // 契約上 `limit` はハード上限なので、結合後に再度クリップする。元の DESC 順は
+  // 種別ごとに維持されるが、種別を跨いだ並び替えはクライアント側のスコアリングに任せる。
+  //
+  // The page query and the highlight query each apply `LIMIT ${limit}`, so a
+  // naïve concat would return up to 2*limit rows (PR #873 review: codex).
+  // We re-clip to `limit` after the merge to keep the contract a hard cap;
+  // cross-kind ordering is left to the client-side scoring layer.
+  const merged = [...taggedPageRows, ...highlightRows].slice(0, limit);
+
+  return c.json({ results: merged });
 });
 
+/**
+ * `pdf_highlights` を所有検証付きで検索し、`kind="pdf_highlight"` 行を返す。
+ * 戻り値は `c.json` にそのまま流せる形に整える（snake_case のキー）。
+ *
+ * Searches `pdf_highlights` for the caller's own highlights only and returns
+ * a list of `kind: "pdf_highlight"` rows shaped for `c.json` (snake_case).
+ *
+ * @param db    リクエストの drizzle DB ハンドル。Drizzle DB handle from the request.
+ * @param userId 呼び出し元ユーザー ID。Owner filter — only the caller's rows are returned.
+ * @param pattern ILIKE 用にエスケープ済みのパターン。Pre-escaped ILIKE pattern.
+ * @param limit 結果上限。Maximum number of rows to return.
+ */
+async function runPdfHighlightSearch(
+  db: AppEnv["Variables"]["db"],
+  userId: string,
+  pattern: string,
+  limit: number,
+): Promise<Array<Record<string, unknown>>> {
+  if (isPdfHighlightSearchDisabled()) return [];
+
+  const result = await db.execute(sql`
+    SELECT
+      h.id          AS highlight_id,
+      h.source_id   AS source_id,
+      h.owner_id    AS owner_id,
+      h.pdf_page    AS pdf_page,
+      h.text        AS text,
+      h.derived_page_id AS derived_page_id,
+      h.updated_at  AS updated_at,
+      s.display_name AS source_display_name,
+      s.title       AS source_title
+    FROM pdf_highlights h
+    INNER JOIN sources s ON s.id = h.source_id
+    WHERE h.owner_id = ${userId}
+      AND s.kind = 'pdf_local'
+      AND h.text ILIKE ${pattern}
+    ORDER BY h.updated_at DESC
+    LIMIT ${limit}
+  `);
+
+  return result.rows.map((row) => ({
+    ...(row as Record<string, unknown>),
+    kind: "pdf_highlight" as const,
+  }));
+}
+
 export default app;

src/components/layout/Header/HeaderSearchBar.tsx

@@ -3,6 +3,7 @@ import { Search } from "lucide-react";
 import { Popover, PopoverAnchor } from "@zedi/ui";
 import { Input } from "@zedi/ui";
 import { useGlobalSearchContext } from "@/contexts/GlobalSearchContext";
+import type { GlobalSearchResultItem } from "@/hooks/useGlobalSearch";
 import { useGlobalSearchShortcut } from "@/hooks/useGlobalSearchShortcut";
 import { HeaderSearchDropdownContent } from "./HeaderSearchDropdownContent";
 import { cn } from "@zedi/ui";
@@ -15,10 +16,10 @@ interface SearchKeyDownParams {
   totalItems: number;
   activeIndex: number;
   itemCount: number;
-  searchResults: Array<{ pageId: string; noteId?: string }>;
+  searchResults: GlobalSearchResultItem[];
   setActiveIndex: (value: number | ((prev: number) => number)) => void;
   closeDropdown: () => void;
-  onSelectItem: (pageId: string, noteId?: string) => void;
+  onSelectItem: (item: GlobalSearchResultItem) => void;
   handleSearchSubmit: () => void;
   inputRef: React.RefObject<HTMLInputElement | null>;
 }
@@ -65,7 +66,7 @@ function handleSearchKeyDown(
         handleSearchSubmit();
       } else if (activeIndex >= 0 && activeIndex < itemCount) {
         const item = searchResults[activeIndex];
-        if (item) onSelectItem(item.pageId, item.noteId);
+        if (item) onSelectItem(item);
       }
       break;
     }
@@ -81,61 +82,25 @@ function handleSearchKeyDown(
  *
  */
 export function HeaderSearchBar() {
-  /**
-   *
-   */
   const { query, setQuery, searchResults, hasQuery, handleSelect, handleSearchSubmit } =
     useGlobalSearchContext();
 
-  /**
-   *
-   */
   const [dropdownOpen, setDropdownOpen] = useState(false);
-  /**
-   *
-   */
   const [activeIndex, setActiveIndex] = useState(-1);
-  /**
-   *
-   */
   const inputRef = useRef<HTMLInputElement>(null);
-  /**
-   *
-   */
   const listRef = useRef<HTMLUListElement>(null);
-  /**
-   *
-   */
   const footerRef = useRef<HTMLButtonElement>(null);
 
-  /**
-   *
-   */
   const handleShortcutFocus = useCallback(() => {
     inputRef.current?.focus();
     inputRef.current?.select();
   }, []);
   useGlobalSearchShortcut(handleShortcutFocus);
 
-  /**
-   *
-   */
   const showResults = hasQuery && searchResults.length > 0;
-  /**
-   *
-   */
   const showEmpty = hasQuery && searchResults.length === 0;
-  /**
-   *
-   */
   const hasContent = showResults || showEmpty;
-  /**
-   *
-   */
   const itemCount = showResults ? searchResults.length : 0;
-  /**
-   *
-   */
   const totalItems = hasQuery ? itemCount + 1 : itemCount;
 
   useEffect(() => {
@@ -151,45 +116,30 @@ export function HeaderSearchBar() {
     if (activeIndex === itemCount) {
       footerRef.current?.scrollIntoView({ block: "nearest" });
     } else {
-      /**
-       *
-       */
       const items = listRef.current?.querySelectorAll("[role='option']");
       items?.[activeIndex]?.scrollIntoView({ block: "nearest" });
     }
   }, [activeIndex, itemCount]);
 
-  /**
-   *
-   */
   const closeDropdown = useCallback(() => {
     setDropdownOpen(false);
     setActiveIndex(-1);
   }, []);
 
-  /**
-   *
-   */
   const onSelectItem = useCallback(
-    (pageId: string, noteId?: string) => {
-      handleSelect(pageId, noteId);
+    (item: GlobalSearchResultItem) => {
+      handleSelect(item);
       closeDropdown();
     },
     [handleSelect, closeDropdown],
   );
 
-  /**
-   *
-   */
   const getOptionId = useCallback(
     (index: number) =>
       index === itemCount ? "header-search-footer" : `header-search-option-${index}`,
     [itemCount],
   );
 
-  /**
-   *
-   */
   const activeDescendant = activeIndex >= 0 ? getOptionId(activeIndex) : undefined;
 
   return (