gerbeur/api/db/migrations/0021_page_urls.ts

/**
 * Moves the site to v4's page URLs, where dumps and playlists live under their
 * owner (`/@bob/dumps/<slug>`) and are named by an editable slug alone. A
 * slug was `slugify(title)-<8 hex of the id>`, unique site-wide; it is now the
 * bare `slugify(title)`, unique in its owner's space for its kind, and every
 * slug an item or a user leaves behind stays an alias that redirects
 * (`dump_slug_aliases`, `playlist_slug_aliases`, `user_aliases`).
 *
 * v3 URLs aren't redirected. Instead this rewrites the links to this site's
 * pages in what users wrote: dump notes, comments, chat messages, and
 * playlist and profile descriptions.
 *
 * SQLite can't make `slug` NOT NULL without rebuilding `dumps` and
 * `playlists`, which other tables reference, so the column stays nullable
 * and every writer sets it; the unique index is what the services rely on.
 *
 * @module
 */

import type { DatabaseSync } from "node:sqlite";
import { slugify } from "../../../shared/slug.ts";
import {
  dumpPath,
  feedPath,
  playlistPath,
  type ProfileTab,
  searchPath,
  userPath,
} from "../../../shared/paths.ts";
import { type SlugKind, uniqueSlug } from "../../lib/slugify.ts";
import { isReservedSlug } from "../../lib/reserved-slugs.ts";
import { isSelfHost } from "../../lib/self-url.ts";

/** An item's new path, by its v3 slug and by its id. */
type PathsByKey = Map<string, string>;

/**
 * Gives every dump or playlist of `kind` the bare slug of its title, unique
 * per owner, the oldest first so it keeps the bare one. Returns each one's new
 * path under both keys a v3 link may have used: its old slug and its id.
 */
function reslug(db: DatabaseSync, kind: SlugKind): PathsByKey {
  const table = kind === "dump" ? "dumps" : "playlists";
  const rows = db.prepare(
    `SELECT t.id, t.user_id, t.title, t.slug, u.username
       FROM ${table} t JOIN users u ON u.id = t.user_id
      ORDER BY t.created_at, t.id;`,
  ).all() as {
    id: string;
    user_id: string;
    title: string;
    slug: string | null;
    username: string;
  }[];
  // Cleared first, so that an old slug doesn't take a new one's place.
  db.exec(`UPDATE ${table} SET slug = NULL;`);
  const set = db.prepare(`UPDATE ${table} SET slug = ? WHERE id = ?;`);
  const paths: PathsByKey = new Map();
  const build = kind === "dump" ? dumpPath : playlistPath;
  for (const row of rows) {
    const slug = uniqueSlug(db, kind, row.user_id, slugify(row.title));
    set.run(slug, row.id);
    const path = build({ ownerUsername: row.username, slug });
    paths.set(row.id, path);
    if (row.slug) paths.set(row.slug, path);
  }
  return paths;
}

/**
 * Renames each category whose slug v4 reserves at the root, such as `hot`,
 * to the first of `<slug>-2`, `<slug>-3`… that is free, and logs it.
 */
function renameReservedCategories(db: DatabaseSync): void {
  const rows = db.prepare(`SELECT id, slug FROM categories;`).all() as {
    id: string;
    slug: string;
  }[];
  const taken = new Set(rows.map((r) => r.slug));
  for (const { id, slug } of rows) {
    if (!isReservedSlug(slug)) continue;
    let n = 2;
    while (taken.has(`${slug}-${n}`) || isReservedSlug(`${slug}-${n}`)) n++;
    const renamed = `${slug}-${n}`;
    taken.add(renamed);
    db.prepare(`UPDATE categories SET slug = ? WHERE id = ?;`).run(renamed, id);
    console.log(
      `[migrate] category "${slug}" is now "${renamed}": v4 reserves its slug`,
    );
  }
}

/** Where each v3 profile tab went: a tab, the overview, or a page of its own. */
const PROFILE_TAB_V4: Record<string, ProfileTab | "" | "/settings"> = {
  dumps: "",
  playlists: "playlists",
  followed: "follow",
  invitees: "invitees",
  settings: "/settings",
  manage: "manage",
};
/** The v3 full lists under a profile, which are v4 tabs. */
const PROFILE_LISTS = new Set<string>(["dumps", "upvoted", "playlists"]);

/**
 * Matches a link to a v3 page, its origin being optional: an absolute one is
 * this site's only when its host is, a relative one only as a Markdown link's
 * target. The path stops before a query or a fragment, which stay as they are.
 */
const V3_LINK_RE =
  /(https?:\/\/([^/\s)\]<>"']+))?(\/(?:dumps|playlists|users|~|search\/~)\/[^\s)\]<>"'?#]+)/g;

/**
 * Returns the v4 path of the v3 page at `path`, or null when it names nothing
 * known, which then stays as written: a deleted dump's link is dead either way.
 */
function v4PathOf(
  path: string,
  dumps: PathsByKey,
  playlists: PathsByKey,
  username: (name: string) => string | undefined,
): string | null {
  let segments: string[];
  try {
    segments = path.split("/").slice(1).filter(Boolean).map(
      decodeURIComponent,
    );
  } catch {
    return null;
  }
  const [root, key, ...rest] = segments;
  switch (root) {
    case "dumps": {
      const target = dumps.get(key);
      if (!target) return null;
      if (rest.length === 0) return target;
      return rest.length === 1 && rest[0] === "edit" ? `${target}/edit` : null;
    }
    case "playlists":
      return rest.length === 0 ? playlists.get(key) ?? null : null;
    case "users": {
      const name = username(key) ?? key;
      if (rest.length === 0) return userPath(name);
      if (rest.length === 1 && PROFILE_LISTS.has(rest[0])) {
        return userPath(name, rest[0] as ProfileTab);
      }
      if (rest.length === 2 && rest[0] === "~" && rest[1] in PROFILE_TAB_V4) {
        const tab = PROFILE_TAB_V4[rest[1]];
        if (tab === "/settings") return tab;
        return tab ? userPath(name, tab) : userPath(name);
      }
      return null;
    }
    case "~":
      if (rest.length > 0) return null;
      return key === "followed"
        ? feedPath(undefined, "follow")
        : ["hot", "new", "journal"].includes(key)
        ? feedPath(undefined, key as "hot")
        : null;
    case "search":
      // `/search/~/users`, split as "search", "~", "users".
      return key === "~" && (rest[0] === "users" || rest[0] === "playlists") &&
          rest.length === 1
        ? searchPath(rest[0])
        : null;
  }
  return null;
}

/**
 * Rewrites the links to v3 pages of this site in the text column `column` of
 * `table`, in every row that has one.
 */
function rewriteLinks(
  db: DatabaseSync,
  table: string,
  column: string,
  toV4: (path: string) => string | null,
): void {
  const rows = db.prepare(
    `SELECT id, ${column} AS text FROM ${table} WHERE ${column} IS NOT NULL;`,
  ).all() as { id: string; text: string }[];
  const update = db.prepare(`UPDATE ${table} SET ${column} = ? WHERE id = ?;`);
  for (const { id, text } of rows) {
    const rewritten = text.replace(
      V3_LINK_RE,
      (match, origin: string | undefined, host: string, path: string, at) => {
        if (origin ? !isSelfHost(host) : text[at - 1] !== "(") return match;
        const v4 = toV4(path);
        return v4 ? `${origin ?? ""}${v4}` : match;
      },
    );
    if (rewritten !== text) update.run(rewritten, id);
  }
}

/**
 * Run once by `runMigrations`. On a database made from schema.sql the tables
 * and indexes exist and there is nothing to re-slug or rewrite.
 */
export function up(db: DatabaseSync): void {
  db.exec(`
    CREATE TABLE IF NOT EXISTS dump_slug_aliases (
      user_id    TEXT NOT NULL,
      slug       TEXT NOT NULL,
      dump_id    TEXT NOT NULL REFERENCES dumps(id) ON DELETE CASCADE,
      created_at TEXT NOT NULL,
      PRIMARY KEY (user_id, slug)
    );
    CREATE INDEX IF NOT EXISTS idx_dump_slug_aliases_dump
      ON dump_slug_aliases(dump_id, created_at);
    CREATE TABLE IF NOT EXISTS playlist_slug_aliases (
      user_id     TEXT NOT NULL,
      slug        TEXT NOT NULL,
      playlist_id TEXT NOT NULL REFERENCES playlists(id) ON DELETE CASCADE,
      created_at  TEXT NOT NULL,
      PRIMARY KEY (user_id, slug)
    );
    CREATE INDEX IF NOT EXISTS idx_playlist_slug_aliases_playlist
      ON playlist_slug_aliases(playlist_id, created_at);
    CREATE TABLE IF NOT EXISTS user_aliases (
      username   TEXT NOT NULL,
      user_id    TEXT NOT NULL REFERENCES users(id) ON DELETE CASCADE,
      created_at TEXT NOT NULL,
      PRIMARY KEY (username)
    );
    CREATE UNIQUE INDEX IF NOT EXISTS idx_user_aliases_lower
      ON user_aliases(lower(username));
    CREATE INDEX IF NOT EXISTS idx_user_aliases_user
      ON user_aliases(user_id, created_at);
    DROP INDEX IF EXISTS idx_dumps_slug;
    DROP INDEX IF EXISTS idx_playlists_slug;
  `);

  const dumps = reslug(db, "dump");
  const playlists = reslug(db, "playlist");
  db.exec(`
    CREATE UNIQUE INDEX IF NOT EXISTS idx_dumps_user_slug
      ON dumps(user_id, slug);
    CREATE UNIQUE INDEX IF NOT EXISTS idx_playlists_user_slug
      ON playlists(user_id, slug);
  `);

  renameReservedCategories(db);

  const findUser = db.prepare(
    `SELECT username FROM users WHERE lower(username) = lower(?);`,
  );
  const username = (name: string) =>
    (findUser.get(name) as { username: string } | undefined)?.username;
  const toV4 = (path: string) => v4PathOf(path, dumps, playlists, username);
  for (
    const [table, column] of [
      ["dumps", "comment"],
      ["comments", "body"],
      ["chat_messages", "body"],
      ["playlists", "description"],
      ["users", "description"],
    ]
  ) rewriteLinks(db, table, column, toV4);
}