gerbeur/api/services/dump-service.ts

/**
 * Posts, edits and deletes dumps, and reads them as each viewer may see them.
 * `routes/dumps.ts` calls it, and so do the search, user, file and thumbnail
 * routes to read dumps. A dump's slug is made from its title when it is
 * posted, and changes only when an edit sends a new one (`slug-service.ts`). It broadcasts the changes to public dumps over the
 * WebSocket.
 *
 * The functions that change a dump don't check who asks: the routes do first.
 *
 * @module
 */

import { APIException } from "../lib/api-exception.ts";
import type {
  CreateUrlDumpRequest,
  Dump,
  DumpUrlMatch,
  UpdateDumpRequest,
} from "../../shared/dump.ts";
import { db, transaction } from "../db/connection.ts";
import {
  DUMP_SELECT_COLUMNS as SELECT_COLS,
  DUMP_SELECT_COLUMNS_ALIASED as SELECT_COLS_ALIASED,
  dumpApiToRow,
  dumpRowToApi,
  isDumpRow,
} from "../model/dump.ts";
import { type Page, pageOf } from "../lib/pagination.ts";
import {
  fetchRichContent,
  tryFetchRichContent,
} from "./rich-content-service.ts";
import { isValidHttpUrl } from "../remote/fetch.ts";
import { broadcastDumpChange } from "./ws-service.ts";
import {
  notifyMentions,
  notifyUserFollowersNewDump,
} from "./notification-service.ts";
import { changeSlug, newSlug, resolveSlug } from "./slug-service.ts";
import { canonicalizeUrl } from "../lib/canonical-url.ts";
import {
  DUMP_ALLOWED_MIME_PREFIXES,
  DUMP_ALLOWED_MIME_TYPES,
  DUMPS_DIR,
  THUMBNAILS_DIR,
} from "../config.ts";
import {
  DUMP_MAX_FILE_SIZE_BYTES,
  VALIDATION,
} from "../../shared/validation.ts";
import { linkAttachments } from "./attachment-service.ts";
import { relinkBacklinks } from "./backlink-service.ts";
import {
  assertCategoriesExist,
  attachCategoryIds,
  setDumpCategories,
} from "./category-service.ts";
import { LIKE, likePattern } from "../lib/like.ts";
import { parseRows } from "../model/rows.ts";

function isAllowedMime(mime: string): boolean {
  return DUMP_ALLOWED_MIME_PREFIXES.some((p) => mime.startsWith(p)) ||
    DUMP_ALLOWED_MIME_TYPES.has(mime);
}

/**
 * Throws a 400 when a file dump's title or comment is too long. The multipart
 * routes that post and replace a file don't check them, unlike the JSON ones.
 */
function assertFileDumpText(title?: string, comment?: string): void {
  if (title && title.trim().length > VALIDATION.DUMP_TITLE_MAX) {
    throw new APIException(
      400,
      `Title must be at most ${VALIDATION.DUMP_TITLE_MAX} characters`,
    );
  }
  if (comment && comment.length > VALIDATION.DUMP_COMMENT_MAX) {
    throw new APIException(
      400,
      `Comment must be at most ${VALIDATION.DUMP_COMMENT_MAX} characters`,
    );
  }
}

/** Returns a link's host name, the title it gets when its page has none. */
function titleFromUrl(url: string): string {
  try {
    return new URL(url).hostname.replace(/^www\./, "");
  } catch {
    return url;
  }
}

/**
 * Posts a link as `userId`, for `POST /api/dumps`, with its preview fetched
 * from the page. A public dump is broadcast as `dump_created` and notified to
 * the poster's followers. Anyone @mentioned in the note is notified even when
 * the dump is private.
 *
 * @throws {APIException} 400 when the URL isn't a public http(s) URL or a
 * category doesn't exist.
 */
export async function createUrlDump(
  request: CreateUrlDumpRequest,
  userId: string,
): Promise<Dump> {
  if (!isValidHttpUrl(request.url)) {
    throw new APIException(400, "Invalid URL");
  }
  assertCategoriesExist(request.categoryIds ?? []);

  const dumpId = crypto.randomUUID();
  const createdAt = new Date();
  const richContent = await fetchRichContent(request.url);
  // A title typed on the create form wins over the one scraped from the page:
  // the poster has seen the preview and is correcting it.
  const title = request.title?.trim() || richContent?.title ||
    titleFromUrl(request.url);
  const isPrivate = request.isPrivate ?? false;

  const categoryIds = [...new Set(request.categoryIds ?? [])];
  transaction(() => {
    const slug = newSlug("dump", userId, title);
    db.prepare(
      `INSERT INTO dumps (id, kind, title, slug, comment, user_id, created_at, url, url_canonical, rich_content, is_private)
      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?);`,
    ).run(
      dumpId,
      "url",
      title,
      slug,
      request.comment ?? null,
      userId,
      createdAt.toISOString(),
      request.url,
      canonicalizeUrl(request.url),
      richContent ? JSON.stringify(richContent) : null,
      isPrivate ? 1 : 0,
    );
    setDumpCategories(dumpId, categoryIds);
  });

  const dump: Dump = { ...fetchDump(dumpId), categoryIds };
  broadcastDumpChange(null, dump);
  if (!isPrivate) notifyUserFollowersNewDump(userId, dumpId, title);
  if (request.comment) {
    // A private dump's mentions are notified if it goes public.
    if (!isPrivate) {
      notifyMentions(userId, request.comment, "dump", dumpId, title);
    }
    linkAttachments(request.comment, "dump", dumpId);
  }
  relinkBacklinks("dump_comment", dumpId, dumpId, request.comment ?? "");
  return dump;
}

/**
 * Posts an uploaded file as `userId`, for `POST /api/dumps`, and stores it in
 * `DUMPS_DIR` under the dump's id. It is announced and notified as
 * {@link createUrlDump} does.
 *
 * @throws {APIException} 400 when the title or comment is too long, the file's
 * type isn't allowed, it exceeds 50 MB, or a category doesn't exist.
 */
export async function createFileDump(
  file: File,
  comment: string | undefined,
  userId: string,
  isPrivate = false,
  title?: string,
  categoryIds: string[] = [],
): Promise<Dump> {
  assertFileDumpText(title, comment);
  if (!isAllowedMime(file.type)) {
    throw new APIException(400, `File type '${file.type}' is not allowed`);
  }
  if (file.size > DUMP_MAX_FILE_SIZE_BYTES) {
    throw new APIException(400, "File too large (max 50 MB)");
  }
  assertCategoriesExist(categoryIds);

  const dumpId = crypto.randomUUID();
  const createdAt = new Date();
  const finalTitle = title?.trim() || file.name;
  const dedupedCategoryIds = [...new Set(categoryIds)];

  await Deno.mkdir(DUMPS_DIR, { recursive: true });

  try {
    // Streamed to disk rather than copied into one more 50 MB buffer first.
    await Deno.writeFile(`${DUMPS_DIR}/${dumpId}`, file.stream());

    transaction(() => {
      const slug = newSlug("dump", userId, finalTitle);
      db.prepare(
        `INSERT INTO dumps (id, kind, title, slug, comment, user_id, created_at, file_name, file_mime, file_size, is_private)
        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?);`,
      ).run(
        dumpId,
        "file",
        finalTitle,
        slug,
        comment ?? null,
        userId,
        createdAt.toISOString(),
        file.name,
        file.type,
        file.size,
        isPrivate ? 1 : 0,
      );
      setDumpCategories(dumpId, dedupedCategoryIds);
    });
  } catch (err) {
    // Leave no file behind when the write or the insert fails.
    await Deno.remove(`${DUMPS_DIR}/${dumpId}`).catch(() => {});
    throw err;
  }

  const dump: Dump = { ...fetchDump(dumpId), categoryIds: dedupedCategoryIds };
  broadcastDumpChange(null, dump);
  if (!isPrivate) notifyUserFollowersNewDump(userId, dumpId, finalTitle);
  if (comment) {
    // A private dump's mentions are notified if it goes public.
    if (!isPrivate) {
      notifyMentions(userId, comment, "dump", dumpId, finalTitle);
    }
    linkAttachments(comment, "dump", dumpId);
  }
  relinkBacklinks("dump_comment", dumpId, dumpId, comment ?? "");
  return dump;
}

/**
 * Returns a file dump's version, which the app's file and thumbnail URLs carry
 * as `?v=`. It changes whenever the file may have, so the file and thumbnail
 * routes cache a URL naming the current one for good. The frontend's
 * `dumpFileVersion` must compute the same.
 */
export function dumpFileVersion(
  dump: Pick<Dump, "fileSize" | "createdAt" | "updatedAt">,
): string {
  return `${dump.fileSize ?? 0}-${
    (dump.updatedAt ?? dump.createdAt).getTime()
  }`;
}

/**
 * Deletes the thumbnails that `routes/thumbnails.ts` generated from a dump's
 * upload, once the upload is replaced or deleted. The route makes them again
 * from a new one.
 */
async function removeGeneratedThumbnails(dumpId: string): Promise<void> {
  await Promise.all(
    [`${dumpId}.jpg`, `${dumpId}.preview.jpg`, `${dumpId}.preview.png`].map(
      (name) => Deno.remove(`${THUMBNAILS_DIR}/${name}`).catch(() => {}),
    ),
  );
}

/**
 * Returns dump `dumpId`, without its categories, or throws a 404. It doesn't
 * check privacy: the routes do, with {@link getDump}.
 */
function fetchDump(dumpId: string): Dump {
  const row = db.prepare(`SELECT ${SELECT_COLS} FROM dumps WHERE id = ?;`)
    .get(dumpId);
  if (!row || !isDumpRow(row)) {
    throw new APIException(404, "Dump not found");
  }
  return dumpRowToApi(row);
}

/**
 * Returns dump `dumpId` as `requestingUserId` may see it. The routes resolve
 * every `:dumpId` with it before acting on the dump.
 *
 * @throws {APIException} 404 when there is no such dump, or it is private and
 * not theirs.
 */
export function getDump(dumpId: string, requestingUserId?: string): Dump {
  const dump = fetchDump(dumpId);
  if (dump.isPrivate && dump.userId !== requestingUserId) {
    throw new APIException(404, "Dump not found");
  }
  attachCategoryIds([dump]);
  return dump;
}

/**
 * Returns the dump of `userId`'s whose slug is or was `slug`, as
 * `requestingUserId` may see it, for `GET /api/users/:key/dumps/:slug`, which
 * reads a dump by its page's path. Reached by a former slug, it comes with
 * its current one, which the client puts in its URL.
 *
 * @throws {APIException} 404 when there is no such dump, or it is private and
 * not theirs.
 */
export function getDumpByPath(
  userId: string,
  slug: string,
  requestingUserId?: string,
): Dump {
  const dumpId = resolveSlug("dump", userId, slug);
  if (!dumpId) throw new APIException(404, "Dump not found");
  return getDump(dumpId, requestingUserId);
}

/**
 * Returns a page of the dumps `requestingUserId` may see whose title, note, or
 * previewed title or description contains `query`, newest first, with how many
 * match in all. `GET /api/search` calls it.
 */
export function searchDumps(
  query: string,
  page: number,
  limit: number,
  requestingUserId?: string,
): { items: Dump[]; total: number } {
  if (!query.trim()) return { items: [], total: 0 };
  const pattern = likePattern(query);
  const offset = (page - 1) * limit;

  const searchClause =
    `(title ${LIKE} OR comment ${LIKE} OR json_extract(rich_content,'$.title') ${LIKE} OR json_extract(rich_content,'$.description') ${LIKE})`;
  const searchParams = [pattern, pattern, pattern, pattern] as const;

  if (requestingUserId) {
    const rows = db.prepare(
      `SELECT ${SELECT_COLS} FROM dumps WHERE (is_private = 0 OR user_id = ?) AND ${searchClause} ORDER BY created_at DESC LIMIT ? OFFSET ?;`,
    ).all(requestingUserId, ...searchParams, limit, offset);
    const totalRow = db.prepare(
      `SELECT COUNT(*) as count FROM dumps WHERE (is_private = 0 OR user_id = ?) AND ${searchClause};`,
    ).get(requestingUserId, ...searchParams) as { count: number } | undefined;
    const items = parseRows(rows, isDumpRow, dumpRowToApi);
    attachCategoryIds(items);
    return { items, total: totalRow?.count ?? 0 };
  } else {
    const rows = db.prepare(
      `SELECT ${SELECT_COLS} FROM dumps WHERE is_private = 0 AND ${searchClause} ORDER BY created_at DESC LIMIT ? OFFSET ?;`,
    ).all(...searchParams, limit, offset);
    const totalRow = db.prepare(
      `SELECT COUNT(*) as count FROM dumps WHERE is_private = 0 AND ${searchClause};`,
    ).get(...searchParams) as { count: number } | undefined;
    const items = parseRows(rows, isDumpRow, dumpRowToApi);
    attachCategoryIds(items);
    return { items, total: totalRow?.count ?? 0 };
  }
}

/**
 * Returns a page of the dumps `requestingUserId` may see, newest first, for the
 * index feed (`GET /api/dumps`).
 *
 * @throws {APIException} 500 when a row doesn't parse as a dump.
 */
export function listDumps(
  page: number,
  limit: number,
  requestingUserId?: string,
): Page<Dump> {
  const offset = (page - 1) * limit;
  const rows = requestingUserId
    ? db.prepare(
      `SELECT ${SELECT_COLS} FROM dumps WHERE (is_private = 0 OR user_id = ?) ORDER BY created_at DESC LIMIT ? OFFSET ?;`,
    ).all(requestingUserId, limit + 1, offset)
    : db.prepare(
      `SELECT ${SELECT_COLS} FROM dumps WHERE is_private = 0 ORDER BY created_at DESC LIMIT ? OFFSET ?;`,
    ).all(limit + 1, offset);

  if (!rows || !rows.every(isDumpRow)) {
    throw new APIException(500, "Malformed dump data");
  }

  const items = rows.map(dumpRowToApi);
  attachCategoryIds(items);
  return pageOf(items, limit);
}

/**
 * Returns the dumps `requestingUserId` may see that were posted with the same
 * link as `url`, newest first, for the create form's duplicate notice
 * (`GET /api/dumps?url=`). Links are compared in their canonical form, so
 * tracking parameters, the scheme, `www.` or a YouTube share format don't hide
 * a repost.
 */
export function findDumpsByUrl(
  url: string,
  requestingUserId?: string,
  limit = 3,
): DumpUrlMatch[] {
  const canonical = canonicalizeUrl(url);
  if (!canonical) return [];

  const rows = db.prepare(
    `SELECT d.id, d.slug, d.title, d.created_at, d.vote_count, u.username,
       (SELECT COUNT(*) FROM comments WHERE dump_id = d.id AND deleted = 0) as comment_count
     FROM dumps d
     JOIN users u ON u.id = d.user_id
     WHERE d.url_canonical = ? AND (d.is_private = 0 OR d.user_id = ?)
     ORDER BY d.created_at DESC
     LIMIT ?;`,
  ).all(canonical, requestingUserId ?? null, limit) as {
    id: string;
    slug: string;
    title: string;
    created_at: string;
    vote_count: number;
    username: string;
    comment_count: number;
  }[];

  return rows.map((row) => ({
    id: row.id,
    slug: row.slug,
    title: row.title,
    ownerUsername: row.username,
    createdAt: new Date(row.created_at),
    voteCount: row.vote_count,
    commentCount: row.comment_count,
  }));
}

/**
 * Applies `request` to a dump, for `PUT /api/dumps/:dumpId`. A new link has its
 * preview fetched and gives the dump its page's title, unless `request` sets
 * one. A file dump ignores `url`. A new `slug` goes through `changeSlug`,
 * which `exempt` (a moderator) frees from its rate limit; a new title leaves
 * the slug as it is. Clients get the dump as `dump_updated`, or a
 * `dump_deleted` when it turns private and a `dump_created` when it turns
 * public. Anyone @mentioned in the note is notified.
 *
 * @throws {APIException} 404 when there is no such dump, 400 when a new URL
 * isn't a public http(s) URL, a category doesn't exist or the slug isn't
 * valid, 409 when the slug is taken, 429 when it changed too often lately.
 */
export async function updateDump(
  dumpId: string,
  request: UpdateDumpRequest,
  exempt = false,
): Promise<Dump> {
  const dump = fetchDump(dumpId);
  if (request.categoryIds !== undefined) {
    assertCategoriesExist(request.categoryIds);
  }
  if (dump.kind === "url" && !isValidHttpUrl(request.url ?? dump.url!)) {
    throw new APIException(400, "Invalid URL");
  }
  // First, so that a refused slug leaves the rest of the dump unchanged.
  if (request.slug !== undefined) {
    changeSlug("dump", dump, request.slug, exempt);
  }
  const slug = request.slug ?? dump.slug;

  const now = new Date();

  if (dump.kind === "file") {
    const title = request.title?.trim() || dump.title;
    const updatedDump: Dump = {
      ...dump,
      title,
      slug,
      comment: "comment" in request
        ? (request.comment ?? undefined)
        : dump.comment,
      isPrivate: "isPrivate" in request
        ? (request.isPrivate ?? false)
        : dump.isPrivate,
      updatedAt: now,
    };
    db.prepare(
      `UPDATE dumps SET title = ?, comment = ?, is_private = ?, updated_at = ? WHERE id = ?;`,
    ).run(
      updatedDump.title,
      updatedDump.comment ?? null,
      updatedDump.isPrivate ? 1 : 0,
      now.toISOString(),
      dumpId,
    );
    if (request.categoryIds !== undefined) {
      const ids = [...new Set(request.categoryIds)];
      setDumpCategories(dumpId, ids);
      updatedDump.categoryIds = ids;
    } else {
      attachCategoryIds([updatedDump]);
    }
    broadcastDumpChange(dump, updatedDump);
    if (updatedDump.comment) {
      if (!updatedDump.isPrivate) {
        notifyMentions(
          dump.userId,
          updatedDump.comment,
          "dump",
          dumpId,
          updatedDump.title,
        );
      }
      linkAttachments(updatedDump.comment, "dump", dumpId);
    }
    relinkBacklinks("dump_comment", dumpId, dumpId, updatedDump.comment ?? "");
    return updatedDump;
  }

  const newUrl = request.url ?? dump.url!;

  let { richContent, title } = dump;

  if (newUrl !== dump.url) {
    richContent = await fetchRichContent(newUrl);
    title = richContent?.title ?? titleFromUrl(newUrl);
  }

  if (request.title?.trim()) {
    title = request.title.trim();
  }

  const updatedDump: Dump = {
    ...dump,
    title,
    slug,
    comment: "comment" in request
      ? (request.comment ?? undefined)
      : dump.comment,
    url: newUrl,
    richContent,
    isPrivate: "isPrivate" in request
      ? (request.isPrivate ?? false)
      : dump.isPrivate,
    updatedAt: now,
  };

  const row = dumpApiToRow(updatedDump);
  const result = db.prepare(
    `UPDATE dumps SET title = ?, comment = ?, url = ?, url_canonical = ?, rich_content = ?, is_private = ?, updated_at = ? WHERE id = ?;`,
  ).run(
    row.title,
    row.comment,
    row.url,
    row.url ? canonicalizeUrl(row.url) : null,
    row.rich_content,
    row.is_private,
    now.toISOString(),
    row.id,
  );

  if (result.changes === 0) {
    throw new APIException(404, "Dump not found");
  }

  if (request.categoryIds !== undefined) {
    const ids = [...new Set(request.categoryIds)];
    setDumpCategories(dumpId, ids);
    updatedDump.categoryIds = ids;
  } else {
    attachCategoryIds([updatedDump]);
  }

  broadcastDumpChange(dump, updatedDump);
  if (updatedDump.comment) {
    if (!updatedDump.isPrivate) {
      notifyMentions(
        dump.userId,
        updatedDump.comment,
        "dump",
        dumpId,
        updatedDump.title,
      );
    }
    linkAttachments(updatedDump.comment, "dump", dumpId);
  }
  relinkBacklinks("dump_comment", dumpId, dumpId, updatedDump.comment ?? "");
  return updatedDump;
}

/**
 * Replaces a file dump's upload and its comment, for
 * `PUT /api/dumps/:dumpId/file`. A `title`, `slug`, `categoryIds` or
 * `isPrivate` left out stays as it is. Visibility is taken here so that a new
 * file and a change of visibility land together. The slug changes as in
 * {@link updateDump}, before the file is written. The old file is put back if
 * the database update fails, and clients are told as {@link updateDump} tells
 * them.
 *
 * @throws {APIException} 404 when there is no such dump, 400 when it isn't a
 * file dump, the title or comment is too long, the file's type isn't allowed,
 * it exceeds 50 MB, a category doesn't exist or the slug isn't valid, 409 when
 * the slug is taken, 429 when it changed too often lately.
 */
export async function replaceFileDump(
  dumpId: string,
  file: File,
  comment: string | undefined,
  title?: string,
  categoryIds?: string[],
  isPrivate?: boolean,
  slug?: string,
  exempt = false,
): Promise<Dump> {
  assertFileDumpText(title, comment);
  if (!isAllowedMime(file.type)) {
    throw new APIException(400, `File type '${file.type}' is not allowed`);
  }
  if (file.size > DUMP_MAX_FILE_SIZE_BYTES) {
    throw new APIException(400, "File too large (max 50 MB)");
  }

  const dump = fetchDump(dumpId);
  if (dump.kind !== "file") {
    throw new APIException(400, "Not a file dump");
  }
  if (categoryIds !== undefined) assertCategoriesExist(categoryIds);
  if (slug !== undefined) changeSlug("dump", dump, slug, exempt);

  // The new file is written beside the old one and renamed over it (atomic: a
  // reader sees one or the other). A hard link keeps the old one reachable so
  // a failed update can put it back, without holding either in memory.
  const filePath = `${DUMPS_DIR}/${dumpId}`;
  const tmpPath = `${filePath}.${crypto.randomUUID()}.tmp`;
  const backupPath = `${tmpPath}.old`;
  await Deno.mkdir(DUMPS_DIR, { recursive: true });
  try {
    await Deno.writeFile(tmpPath, file.stream());
  } catch (err) {
    await Deno.remove(tmpPath).catch(() => {});
    throw err;
  }
  const hadOld = await Deno.link(filePath, backupPath).then(
    () => true,
    () => false,
  );
  await Deno.rename(tmpPath, filePath);

  const now = new Date();
  const finalTitle = title?.trim() || dump.title;
  const finalPrivate = isPrivate ?? dump.isPrivate;
  try {
    transaction(() => {
      db.prepare(
        `UPDATE dumps SET title = ?, file_name = ?, file_mime = ?, file_size = ?, comment = ?, is_private = ?, updated_at = ? WHERE id = ?;`,
      ).run(
        finalTitle,
        file.name,
        file.type,
        file.size,
        comment ?? null,
        finalPrivate ? 1 : 0,
        now.toISOString(),
        dumpId,
      );
      if (categoryIds !== undefined) {
        setDumpCategories(dumpId, [...new Set(categoryIds)]);
      }
    });
  } catch (err) {
    if (hadOld) await Deno.rename(backupPath, filePath).catch(() => {});
    else await Deno.remove(filePath).catch(() => {});
    throw err;
  }
  if (hadOld) await Deno.remove(backupPath).catch(() => {});
  await removeGeneratedThumbnails(dumpId);

  if (categoryIds !== undefined) {
    dump.categoryIds = [...new Set(categoryIds)];
  } else {
    attachCategoryIds([dump]);
  }

  if (comment) {
    if (!finalPrivate) {
      notifyMentions(dump.userId, comment, "dump", dumpId, finalTitle);
    }
    linkAttachments(comment, "dump", dumpId);
  }
  relinkBacklinks("dump_comment", dumpId, dumpId, comment ?? "");
  const updatedDump: Dump = {
    ...dump,
    title: finalTitle,
    slug: slug ?? dump.slug,
    fileName: file.name,
    fileMime: file.type,
    fileSize: file.size,
    comment,
    isPrivate: finalPrivate,
    updatedAt: now,
  };
  broadcastDumpChange(dump, updatedDump);
  return updatedDump;
}

/**
 * Returns a page of `userId`'s dumps, newest first, for
 * `GET /api/users/:key/dumps`. The route sets `includePrivate` when users
 * ask for their own.
 *
 * @throws {APIException} 500 when a row doesn't parse as a dump.
 */
export function getDumpsByUser(
  userId: string,
  page: number,
  limit: number,
  includePrivate: boolean,
): Page<Dump> {
  const offset = (page - 1) * limit;
  const privacyFilter = includePrivate ? "" : " AND is_private = 0";
  const rows = db.prepare(
    `SELECT ${SELECT_COLS} FROM dumps WHERE user_id = ?${privacyFilter} ORDER BY created_at DESC LIMIT ? OFFSET ?;`,
  ).all(userId, limit + 1, offset);
  if (!rows.every(isDumpRow)) {
    throw new APIException(500, "Malformed dump data");
  }
  const items = rows.map(dumpRowToApi);
  attachCategoryIds(items);
  return pageOf(items, limit);
}

/**
 * Returns a page of the dumps `userId` upvoted, latest vote first, for
 * `GET /api/users/:key/upvoted`. Their own private dumps are included only
 * when they are the one asking.
 *
 * @throws {APIException} 500 when a row doesn't parse as a dump.
 */
export function getVotedDumpsByUser(
  userId: string,
  page: number,
  limit: number,
  requestingUserId: string | null,
): Page<Dump> {
  const offset = (page - 1) * limit;
  const dumpCols = SELECT_COLS_ALIASED;
  let rows: unknown[];

  if (requestingUserId === userId) {
    rows = db.prepare(
      `SELECT ${dumpCols}
       FROM dumps d
       INNER JOIN votes v ON d.id = v.dump_id
       WHERE v.user_id = ? AND (d.is_private = 0 OR d.user_id = ?)
       ORDER BY v.created_at DESC LIMIT ? OFFSET ?;`,
    ).all(userId, userId, limit + 1, offset);
  } else {
    rows = db.prepare(
      `SELECT ${dumpCols}
       FROM dumps d
       INNER JOIN votes v ON d.id = v.dump_id
       WHERE v.user_id = ? AND d.is_private = 0
       ORDER BY v.created_at DESC LIMIT ? OFFSET ?;`,
    ).all(userId, limit + 1, offset);
  }

  if (!rows.every(isDumpRow)) {
    throw new APIException(500, "Malformed dump data");
  }
  const items = rows.map(dumpRowToApi);
  attachCategoryIds(items);
  return pageOf(items, limit);
}

/**
 * Fetches a URL dump's preview again, bypassing the cache, and replaces its
 * title and preview, for `POST /api/dumps/:dumpId/metadata`. The slug stays.
 *
 * @throws {APIException} 404 when there is no such dump, 400 when it isn't a
 * URL dump, 502 when the page can't be reached.
 */
export async function refreshDumpMetadata(dumpId: string): Promise<Dump> {
  const dump = fetchDump(dumpId);

  if (dump.kind !== "url" || !dump.url) {
    throw new APIException(400, "Only URL dumps support metadata refresh");
  }

  // A failed fetch yields a stub with no title or thumbnail. Written over the
  // dump's metadata, it would rename the dump to its host name.
  const { ok, content: richContent } = await tryFetchRichContent(dump.url, {
    fresh: true,
  });
  if (!ok) {
    throw new APIException(
      502,
      "Could not reach the page to refresh its metadata",
    );
  }

  const title = richContent?.title ?? titleFromUrl(dump.url);

  const updatedDump: Dump = { ...dump, title, richContent };
  const row = dumpApiToRow(updatedDump);
  db.prepare(
    `UPDATE dumps SET title = ?, rich_content = ? WHERE id = ?;`,
  ).run(row.title, row.rich_content, row.id);

  broadcastDumpChange(dump, updatedDump);

  return updatedDump;
}

/**
 * Returns a dump's id and custom thumbnail type, or undefined without one, for
 * the thumbnail route. It doesn't check privacy: the route calls
 * {@link getDump} first.
 *
 * @throws {APIException} 404 when there is no such dump.
 */
export function getDumpThumbnailInfo(
  dumpId: string,
): { id: string; mime: string } | undefined {
  const dump = fetchDump(dumpId);
  return dump.thumbnailMime
    ? { id: dump.id, mime: dump.thumbnailMime }
    : undefined;
}

/**
 * Records that a dump has a custom thumbnail of type `mime`, for
 * `PUT /api/dumps/:dumpId/thumbnail`, once the route has written the file.
 *
 * @throws {APIException} 404 when there is no such dump.
 */
export function setDumpThumbnail(dumpId: string, mime: string): Dump {
  const dump = fetchDump(dumpId);
  db.prepare(`UPDATE dumps SET custom_thumbnail_mime = ? WHERE id = ?;`).run(
    mime,
    dump.id,
  );
  const updatedDump: Dump = { ...dump, thumbnailMime: mime };
  broadcastDumpChange(dump, updatedDump);
  return updatedDump;
}

/**
 * Removes a dump's custom thumbnail and its file, for
 * `DELETE /api/dumps/:dumpId/thumbnail`.
 *
 * @throws {APIException} 404 when there is no such dump.
 */
export async function removeDumpThumbnail(dumpId: string): Promise<Dump> {
  const dump = fetchDump(dumpId);
  db.prepare(`UPDATE dumps SET custom_thumbnail_mime = NULL WHERE id = ?;`)
    .run(dump.id);
  await Deno.remove(`${THUMBNAILS_DIR}/${dump.id}-custom`).catch(() => {});
  const updatedDump: Dump = { ...dump, thumbnailMime: undefined };
  broadcastDumpChange(dump, updatedDump);
  return updatedDump;
}

/**
 * Deletes a dump with its files and thumbnails, for
 * `DELETE /api/dumps/:dumpId`, and broadcasts `dump_deleted`.
 *
 * @throws {APIException} 404 when there is no such dump.
 */
export async function deleteDump(dumpId: string): Promise<void> {
  const dump = fetchDump(dumpId);

  const result = db.prepare(`DELETE FROM dumps WHERE id = ?;`).run(dumpId);

  if (result.changes === 0) {
    throw new APIException(404, "Dump not found");
  }

  if (dump.kind === "file") {
    await Deno.remove(`${DUMPS_DIR}/${dump.id}`).catch(() => {});
    await removeGeneratedThumbnails(dump.id);
  }
  if (dump.thumbnailMime) {
    await Deno.remove(`${THUMBNAILS_DIR}/${dump.id}-custom`).catch(() => {});
  }

  broadcastDumpChange(dump, null);
}