moshing-mammut/src/lib/server/timeline.ts

import {
  HASHTAG_FILTER,
  MASTODON_ACCESS_TOKEN,
  MASTODON_INSTANCE,
  ODESLI_API_KEY,
  YOUTUBE_API_KEY
} from '$env/static/private';
import { log } from '$lib/log';
import type {
  Account,
  AccountAvatar,
  Post,
  SongThumbnailImage,
  Tag,
  TimelineEvent
} from '$lib/mastodon/response';
import { SongThumbnailImageKind } from '$lib/mastodon/response';
import type { OdesliResponse, Platform, SongInfo } from '$lib/odesliResponse';
import {
  getAvatars,
  getPosts,
  getSongThumbnails,
  removeAvatars,
  saveAvatar,
  savePost,
  saveSongThumbnail
} from '$lib/server/db';
import { createFeed, saveAtomFeed } from '$lib/server/rss';
import { sleep } from '$lib/sleep';
import crypto from 'crypto';
import fs from 'fs/promises';
import sharp from 'sharp';
import { WebSocket } from 'ws';

const URL_REGEX = new RegExp(/href="(?<postUrl>[^>]+?)" target="_blank"/gm);
const INVIDIOUS_REGEX = new RegExp(/invidious.*?watch.*?v=(?<videoId>[a-zA-Z_0-9-]+)/gm);
const YOUTUBE_REGEX = new RegExp(
  /https?:\/\/(www\.)?youtu((be.com\/.*?v=)|(\.be\/))(?<videoId>[a-zA-Z_0-9-]+)/gm
);

export class TimelineReader {
  private static _instance: TimelineReader;

  private static async isMusicVideo(videoId: string) {
    if (!YOUTUBE_API_KEY || YOUTUBE_API_KEY === 'CHANGE_ME') {
      // Assume that it *is* a music link when no YT API key is provided
      return true;
    }
    const searchParams = new URLSearchParams([
      ['part', 'snippet'],
      ['id', videoId],
      ['key', YOUTUBE_API_KEY]
    ]);
    const youtubeVideoUrl = new URL(`https://www.googleapis.com/youtube/v3/videos?${searchParams}`);
    const resp = await fetch(youtubeVideoUrl);
    const respObj = await resp.json();
    if (!respObj.items.length) {
      console.warn('Could not find video with id', videoId);
      return false;
    }

    const item = respObj.items[0];
    if (!item.snippet) {
      console.warn('Could not load snippet for video', videoId, item);
      return false;
    }
    if (item.snippet.tags?.includes('music')) {
      return true;
    }

    const categorySearchParams = new URLSearchParams([
      ['part', 'snippet'],
      ['id', item.snippet.categoryId],
      ['key', YOUTUBE_API_KEY]
    ]);
    const youtubeCategoryUrl = new URL(
      `https://www.googleapis.com/youtube/v3/videoCategories?${categorySearchParams}`
    );
    const categoryTitle: string = await fetch(youtubeCategoryUrl)
      .then((r) => r.json())
      .then((r) => r.items[0]?.snippet?.title);
    return categoryTitle === 'Music';
  }

  public static async getSongInfoInPost(post: Post): Promise<SongInfo[]> {
    const urlMatches = post.content.matchAll(URL_REGEX);
    const songs: SongInfo[] = [];
    for (const match of urlMatches) {
      if (match === undefined || match.groups === undefined) {
        log.warn('Match listed in allMatches, but either it or its groups are undefined', match);
        continue;
      }
      const urlMatch = match.groups.postUrl.toString();
      let url: URL;
      try {
        url = new URL(urlMatch);
      } catch (e) {
        log.error('URL found via Regex does not seem to be a valud url', urlMatch, e);
        continue;
      }

      // Check *all* found url and let odesli determine if it is music or not
      log.debug(`Checking ${url} if it contains song data`);
      const info = await TimelineReader.getSongInfo(url);
      log.debug(`Found song info for ${url}?`, info);
      if (info) {
        songs.push(info);
      }
    }
    return songs;
  }

  private static async getSongInfo(url: URL, remainingTries = 6): Promise<SongInfo | null> {
    if (remainingTries === 0) {
      log.error('No tries remaining. Lookup failed!');
      return null;
    }
    if (url.hostname === 'songwhip.com') {
      // song.link doesn't support songwhip links and songwhip themselves will provide metadata if you pass in a
      // Apple Music/Spotify/etc link, but won't when provided with their own link, so no way to extract song info
      // except maybe scraping their HTML
      return null;
    }

    const videoId = INVIDIOUS_REGEX.exec(url.href)?.groups?.videoId;
    const urlString =
      videoId !== undefined ? `https://youtube.com/watch?v=${videoId}` : url.toString();

    const odesliParams = new URLSearchParams();
    odesliParams.append('url', urlString);
    odesliParams.append('userCountry', 'DE');
    odesliParams.append('songIfSingle', 'true');
    if (ODESLI_API_KEY && ODESLI_API_KEY !== 'CHANGE_ME') {
      odesliParams.append('key', ODESLI_API_KEY);
    }
    const odesliApiUrl = `https://api.song.link/v1-alpha.1/links?${odesliParams}`;
    try {
      const response = await fetch(odesliApiUrl);
      if (response.status === 429) {
        throw new Error('Rate limit reached', { cause: 429 });
      }
      const odesliInfo: OdesliResponse = await response.json();
      if (!odesliInfo || !odesliInfo.entitiesByUniqueId || !odesliInfo.entityUniqueId) {
        return null;
      }
      const info = odesliInfo.entitiesByUniqueId[odesliInfo.entityUniqueId];
      const platform: Platform = 'youtube';
      if (info.platforms.includes(platform)) {
        const youtubeId =
          videoId ??
          YOUTUBE_REGEX.exec(url.href)?.groups?.videoId ??
          new URL(odesliInfo.pageUrl).pathname.split('/y/').pop();
        if (youtubeId === undefined) {
          log.warn('Looks like a youtube video, but could not extract a video id', url, odesliInfo);
          return null;
        }
        const isMusic = await TimelineReader.isMusicVideo(youtubeId);
        if (!isMusic) {
          log.debug('Probably not a music video', url);
          return null;
        }
      }
      return {
        ...info,
        pageUrl: odesliInfo.pageUrl,
        youtubeUrl: odesliInfo.linksByPlatform[platform]?.url,
        postedUrl: url.toString()
      } as SongInfo;
    } catch (e) {
      if (e instanceof Error && e.cause === 429) {
        log.warn('song.link rate limit reached. Trying again in 10 seconds');
        await sleep(10_000);
        return await this.getSongInfo(url, remainingTries - 1);
      }
      log.error(`Failed to load ${url} info from song.link`, e);
      return null;
    }
  }

  private static async resizeAvatar(
    baseName: string,
    size: number,
    suffix: string,
    folder: string,
    sharpAvatar: sharp.Sharp
  ): Promise<string | null> {
    const fileName = `${folder}/${baseName}_${suffix}`;
    const exists = await fs
      .access(fileName, fs.constants.F_OK)
      .then(() => true)
      .catch(() => false);
    if (exists) {
      log.debug('File already exists', fileName);
      return null;
    }
    log.debug('Saving avatar', fileName);
    await sharpAvatar.resize(size).toFile(fileName);
    return fileName;
  }

  private static resizeAvatarPromiseMaker(
    avatarFilenameBase: string,
    baseSize: number,
    maxPixelDensity: number,
    accountUrl: string,
    formats: string[],
    avatar: ArrayBuffer
  ): Promise<void>[] {
    const sharpAvatar = sharp(avatar);
    const promises: Promise<void>[] = [];
    for (let i = 1; i <= maxPixelDensity; i++) {
      promises.push(
        ...formats.map((f) =>
          TimelineReader.resizeAvatar(
            avatarFilenameBase,
            baseSize * i,
            `${i}x.${f}`,
            'avatars',
            sharpAvatar
          )
            .then(
              (fn) =>
                ({
                  accountUrl: accountUrl,
                  file: fn,
                  sizeDescriptor: `${i}x`
                }) as AccountAvatar
            )
            .then(saveAvatar)
        )
      );
    }
    return promises;
  }

  private static resizeThumbnailPromiseMaker(
    filenameBase: string,
    baseSize: number,
    maxPixelDensity: number,
    songThumbnailUrl: string,
    formats: string[],
    image: ArrayBuffer,
    kind: SongThumbnailImageKind
  ): Promise<void>[] {
    const sharpAvatar = sharp(image);
    const promises: Promise<void>[] = [];
    for (let i = 1; i <= maxPixelDensity; i++) {
      promises.push(
        ...formats.map((f) =>
          TimelineReader.resizeAvatar(
            filenameBase,
            baseSize * i,
            `${i}x.${f}`,
            'thumbnails',
            sharpAvatar
          )
            .then(
              (fn) =>
                ({
                  songThumbnailUrl: songThumbnailUrl,
                  file: fn,
                  sizeDescriptor: `${i}x`,
                  kind: kind
                }) as SongThumbnailImage
            )
            .then(saveSongThumbnail)
        )
      );
    }
    return promises;
  }

  private static async saveAvatar(account: Account) {
    try {
      const existingAvatars = await getAvatars(account.url, 1);
      const existingAvatarBase = existingAvatars.shift()?.file.split('/').pop()?.split('_').shift();
      const avatarFilenameBase =
        new URL(account.avatar).pathname.split('/').pop()?.split('.').shift() ?? account.acct;

      // User's avatar changed. Remove the old one!
      if (existingAvatarBase && existingAvatarBase !== avatarFilenameBase) {
        await removeAvatars(account.url);
        const avatarsToDelete = (await fs.readdir('avatars'))
          .filter((x) => x.startsWith(existingAvatarBase + '_'))
          .map((x) => {
            log.debug('Removing existing avatar file', x);
            return x;
          })
          .map((x) => fs.unlink('avatars/' + x));
        await Promise.allSettled(avatarsToDelete);
      }
      const avatarResponse = await fetch(account.avatar);
      const avatar = await avatarResponse.arrayBuffer();
      await Promise.all(
        TimelineReader.resizeAvatarPromiseMaker(
          avatarFilenameBase,
          50,
          3,
          account.url,
          ['webp', 'avif', 'jpeg'],
          avatar
        )
      );
    } catch (e) {
      console.error('Could not resize and save avatar for', account.acct, account.avatar, e);
    }
  }

  private static async saveSongThumbnails(songs: SongInfo[]) {
    for (const song of songs) {
      if (!song.thumbnailUrl) {
        continue;
      }
      try {
        const existingThumbs = await getSongThumbnails(song);
        if (existingThumbs.length) {
          continue;
        }
        const fileBaseName = crypto.createHash('sha256').update(song.thumbnailUrl).digest('hex');
        const imageResponse = await fetch(song.thumbnailUrl);
        const avatar = await imageResponse.arrayBuffer();
        await Promise.all(
          TimelineReader.resizeThumbnailPromiseMaker(
            fileBaseName + '_large',
            200,
            3,
            song.thumbnailUrl,
            ['webp', 'avif', 'jpeg'],
            avatar,
            SongThumbnailImageKind.Big
          )
        );
        await Promise.all(
          TimelineReader.resizeThumbnailPromiseMaker(
            fileBaseName + '_small',
            60,
            3,
            song.thumbnailUrl,
            ['webp', 'avif', 'jpeg'],
            avatar,
            SongThumbnailImageKind.Small
          )
        );
      } catch (e) {
        console.error(
          'Could not resize and save song thumbnail for',
          song.pageUrl,
          song.thumbnailUrl,
          e
        );
      }
    }
  }

  private startWebsocket() {
    const socket = new WebSocket(
      `wss://${MASTODON_INSTANCE}/api/v1/streaming?type=subscribe&stream=public:local&access_token=${MASTODON_ACCESS_TOKEN}`
    );
    socket.onopen = () => {
      log.log('Connected to WS');
    };
    socket.onmessage = async (event) => {
      try {
        const data: TimelineEvent = JSON.parse(event.data.toString());
        if (data.event !== 'update') {
          log.log('Ignoring ES event', data.event);
          return;
        }
        const post: Post = JSON.parse(data.payload);

        const hashttags: string[] = HASHTAG_FILTER.split(',');
        const found_tags: Tag[] = post.tags.filter((t: Tag) => hashttags.includes(t.name));

        const songs = await TimelineReader.getSongInfoInPost(post);

        // If we don't have any tags or non-youtube urls, check youtube
        // YT is handled separately, because it requires an API call and therefore is slower
        if (songs.length === 0 && found_tags.length === 0) {
          log.log('Ignoring post', post.url);
          return;
        }

        await savePost(post, songs);

        await TimelineReader.saveAvatar(post.account);
        await TimelineReader.saveSongThumbnails(songs);

        log.debug('Saved post', post.url);

        const posts = await getPosts(null, null, 100);
        await saveAtomFeed(createFeed(posts));
      } catch (e) {
        log.error('error message', event, event.data, e);
      }
    };
    socket.onclose = (event) => {
      log.warn(
        `Websocket connection to ${MASTODON_INSTANCE} closed. Code: ${event.code}, reason: '${event.reason}'`,
        event
      );
      setTimeout(() => {
        log.info(`Attempting to reconenct to WS`);
        this.startWebsocket();
      }, 10000);
    };
    socket.onerror = (event) => {
      log.error(
        `Websocket connection to ${MASTODON_INSTANCE} failed. ${event.type}: ${event.error}, message: '${event.message}'`
      );
    };
  }

  private constructor() {
    log.log('Constructing timeline object');
    this.startWebsocket();
  }

  public static init() {
    log.log('Timeline object init');
    if (this._instance === undefined) {
      this._instance = new TimelineReader();
    }
  }

  public static get instance(): TimelineReader {
    TimelineReader.init();
    return this._instance;
  }
}