README.md

   1 # Unnamed Youtube Frontend
   2
   3 Note: this is a work in progress, still unreleased software. Feel free to look around, but keep in mind that this repository will move servers for its public release.
   4
   5 Contributors wanted! Please send me an email (see commit log) or contact `girst` on irc.freenode.net (hanging out in `#invidious` for now)
   6
   7 Test Instance: http://delta.gir.st:8000/ (will move in the future)
   8
   9 TODO:
  10  - task queue for refesh-cipher, pull-subs, update-websub
  11     - should handle delays itself
  12     - would allow us to just put a websub/pullsub request into the queue
  13       whenever a user subscribes to a channel, without worrying about
  14       ratelimiting ourselves.
  15  - webhook: as blueprint; main app looks in envvar/config what to load?
  16  - reddit blueprint:
  17    - pin/hide videos (linking their reddit comments)
  18  - abstract database access
  19    we want to be able to choose between at least sqlite and postgres
  20  - implement 'dangerous' functions as disablable blueprint
  21    - search through invidious-api
  22    - proxy for subtitles, hls/dash manifests, ...
  23      - subtitles on /watch
  24      - hls.js for livestreams
  25    - proxy for geolocked videos?
  26  - modern css frontend (flexbox)
  27  - document all the things
  28  - quality of life improvements:
  29    - asynchronically call update-subs and pull-subs on subscribing, iff necessary
  30    - get_video_info on incoming websub-webhook to get exact published date
  31  - development-friendly error handling:
  32    - log unkown info/endcards to file
  33    - save get_video_info/reddit.json/etc with traceback
  34  - proper support for sponsorblock (cached db?)
  35  - allow pinning non-subscribed videos
  36    would make reddit easier, less-surprising when unsubscribing from channels with pinned videos
  37  - cleanups:
  38    - move subsystem-specific stuff out of common/common.py (e.g. reddit)
  39    - remove jsonify()
  40    - purge magic strings (e.g guest token)
  41    - fix all the TODOs and XXXs
  42    - clean up login stuff
  43      https://flask-login.readthedocs.io/en/latest/
  44      https://blog.miguelgrinberg.com/post/the-flask-mega-tutorial-part-v-user-logins
  45    - jinja2: for/else
  46    - welcome message: maybe make dismissable, show on all pages?
  47  - find a name!
  48  - /channel/UE.../videos 404s
  49  - invidious search: api returns 'nice' author strings (e.g. numberphile instead of UC...)
  50    if not channel_id.match(...): redirect('/user/...') (in invidious blueprint)
  51  - reddit.j2: titles are double-html-escaped
  52
  53 stuff to look at:
  54  - app.config.from_pyfile('the-config.cfg')
  55    NOTE: this will not work for the non-python/non-flask parts
  56
  57 wishlist:
  58  - proxy googlevideo (and probably thumnails) responses
  59
  60 # Installation
  61
  62 see INSTALL file
  63
  64 # Usage
  65
  66 Since user authentication is not yet implemented, users are for now identified by a randomized token. To "log in", just visit `/feed/subscriptions?token=....`. at the bottom, click 'Manage Subscriptions' and paste the UC...-URLs of channels you want to subscribe to.
  67
  68 If no token is specified, it defaults to `guest`. Since this is a readonly account, you can populate it by directly writing into the database:
  69
  70 first, prepare a file (guest.csv) of channel ids like so:
  71     UCxxxxxxxxxxxxxxxxxxxxxxxx,guest,channel
  72 then from the sqlite3 console, issue:
  73     .mode csv
  74     .import guest.csv subscriptions
  75
  76
  77 # Notes
  78  - caching external api requests
  79    currently using in-memory-backend, which is purged every 10minutes. a real installation should use redis, which should scale better and handle purging for us.
  80    the in-memory cache conflicts with gunicorn's worker model: which is a sepreate process for each request. switch to mulitple gthreads on a single worker to avoid getting the cache torn down after every request (note that threads may leak memory, if my observations from $dayjob are still valid).
  81  - neater sqlite output:
  82     .mode column
  83     .headers on