When auto-advancing through a queue or playlist — or stepping with
next/prev — the next video previously jumped to its saved resume
timestamp. A freshly selected track now starts at the beginning, or at
the A marker when an A-B loop is set for it. Resuming still applies when
you reopen a single video directly.
A-B markers set while playing from a playlist are now stored on that
playlist's own copy of the video (entry.ab), so each playlist keeps its
own loop and the markers sync to the database alongside the playlist.
Non-playlist playback keeps using the global per-video marker map.
Bumps service worker to v1.0.3 to bust the client cache.
Adds an About section at the bottom of Settings displaying the app
version. In WEB mode it fetches /api/version and appends the live
server build tag so you can confirm which deploy is running.
The sidebar contained two .playlist-list blocks each with flex:1 and its own
overflow-y:auto, nested inside the sidebar's own scroll. On screens too short
to fit everything, those inner flex regions collapsed toward zero height and
the playlists became unreachable. Drop the inner scroll/flex-grow and let the
whole sidebar scroll as a single unit at any screen size.
Server stamps a BUILD_TAG (stable per process, changes on restart/deploy)
into GET /api/version alongside the existing version string. Cache-Control
is set to no-store so the response is never cached by the SW or browser.
Client (WEB mode only) baselines the tag on first fetch after boot, then
rechecks every 5 minutes. On mismatch it calls the existing showUpdateBanner()
so the user sees the 'Update ready — Reload now' toast and can reload at will.
This mirrors the build-id polling pattern used across the BukidBountyApp
derivatives to avoid stale UI after deploys.
Existing PWA installs cache the app shell (app.js et al.) cache-first under
ytplayer-v1.0.0. Without a version bump they'd keep serving the old app.js
that lacked WEB-mode OPFS support, so the offline-cache fix would never reach
already-installed clients. Bumping forces activate → old cache delete →
re-fetch, and triggers the in-app 'Update ready' banner.
The PWA conversion shipped sw.js, opfs.js, manifest and icons, but app.js
was never updated to use them — it still routed cache_* calls through the
native bridge (call('cache.list', ...)), which throws 'No native bridge
available' in a browser. That surfaced as 'Offline cache unavailable /
not available in this build' on the Saved videos and Settings pages.
- Add WEB mode detection (no Tauri/zero bridge present)
- Route cache_* through OPFS (opfsDownload/Status/List/Delete/Clear)
- Proxy downloads via /api/download so the browser never hits YouTube CDN
- Load fingerprint.js + opfs.js; add manifest link and PWA meta tags
- Relax CSP to allow blob: media/img and worker-src 'self' for the SW
- Join dokploy-network so homelab Traefik can discover the container
- Remove host port binding (Traefik routes directly via service discovery)
- Add HTTP→HTTPS redirect and HTTPS router with letsencrypt cert
- VPS Traefik passthrough for worship.hesed.sbs already added on server
