mrwnslz
08baf970ca
Neue App feeds: Quellen sind entweder Nachrichten oder Blog. - Abruf wie bei Podcasts (ETag, Body-Hash, Backoff), refresh_newsfeeds für den Cron - Anrisstexte: der Artikel wird von der Seite geholt (trafilatura), HTML bereinigt, relative Links absolut - Nachrichten: gleiche Meldung aus mehreren Quellen wird eine Story (tf-idf, je Artikel verglichen), Briefing „Seit letztem Mal“ (Top-Themen, dann nach Ressort), verfallen nach news_keep_days (Standard 2) - Blogs: Ungelesen-Liste je Blog, bleiben bis gelesen - API unter /feeds/: Quellen suchen/abonnieren/OPML, News, Stories, Blogs, Artikel, gelesen, merken Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
23 lines
928 B
Text
23 lines
928 B
Text
# Pinned to exact versions. These are what production runs and what the test
|
|
# suite is green against — with `>=` the image you get depends on the day you
|
|
# build it, which is how a rebuild once swapped in a gunicorn that could not
|
|
# boot the gevent worker at all.
|
|
#
|
|
# Nothing here updates on its own any more, so bump deliberately: change a
|
|
# version, let CI run, then deploy.
|
|
Django==6.1
|
|
pylast==7.1.0
|
|
requests==2.34.2
|
|
python-dotenv==1.2.3
|
|
whitenoise==6.12.0
|
|
feedparser==6.0.14
|
|
gevent==26.8.0
|
|
# gunicorn 26.2.0 imports packaging.version in ggevent.py while declaring no
|
|
# dependencies of its own, so packaging has to be requested explicitly.
|
|
gunicorn==26.2.0
|
|
packaging==26.3
|
|
piper-tts==1.7.0
|
|
# ffmpeg (a static binary in the wheel) finds the pauses for skipping silence in the browser.
|
|
imageio-ffmpeg==0.6.0
|
|
# The feed reader takes the article out of a news page (main text without menus, ads, comments).
|
|
trafilatura==2.3.0
|