Sulkta-Coop/torttube
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
torttube/docs/upstream.md
Kayos 9ed0aae2d0 M7 DONE via delegation — pv.youtube plays HD with audio 
After hitting the segment-timing wall on our hand-rolled DASH MPD
(audio drifted -25s -> -44s behind video on long content), pivoted to
delegating playback to plugin.video.youtube v7.4.3 which already has
years of sidx-parsed SegmentTimeline + multi-client fallback work.

torttube._play() now:
  1. Tries _delegate_to_pv_youtube(yt_id) — sets a resolved URL of
     'plugin://plugin.video.youtube/play/?video_id=<id>'. Kodi
     chain-resolves to pv.youtube which builds the proper MPD and
     hands inputstream.adaptive a correctly-aligned manifest. Default.
  2. Falls back to our DASH builder (still in code, gated by
     'dash_enabled' setting + dash.on marker) if pv.youtube is absent.
  3. Falls through to yt-dlp progressive 360p as the final safety net.

When delegating, we skip our SponsorBlock monitor — pv.youtube has its
own and would double-skip otherwise.

Cobb-verified live on Livingroom Pi: LTT 'Trump Phone' (which crashed
our DASH with audio sync errors growing to -44s) now plays HD with
audio synced. 'Please sign in' message in log is from the tv_unplugged
Innertube client; pv.youtube falls back to a working client
automatically — no user account required.

Settings: prefer_pv_youtube boolean (default true). Addon v0.0.11.
Reference: https://kodi.wiki/view/Add-on:YouTube
2026-05-23 11:58:12 -07:00

5.3 KiB
Raw Permalink Blame History

Upstream contributions log

One row per PR/issue we file. Outcomes recorded honestly — merged, closed, abandoned, whatever happened.

Active

(none yet — opens with M1 development)

Filed

Date Project PR/Issue Title Status Outcome
none yet

Investigation notes (not yet a PR — to be evaluated)

  • inputstream.adaptive — file:// not supported by libcurl downloader (kodi addon, xbmc/inputstream.adaptive on github) Hit during torttube M7 DASH work 2026-05-23. Loading an MPD from a local filesystem path via file:///storage/.kodi/temp/torttube/<id>.mpd fails with CURLOpen returned an error, download failed. Workaround confirmed: bind a localhost ThreadingHTTPServer on the LAN IP (not 127.0.0.1 — that also fails in some configs) and pass http://<lan-ip>:<port> to setResolvedUrl. plugin.video.youtube uses this pattern via a long-lived service addon. Worth filing an enhancement to either accept file:// or document the LAN-IP HTTP-server pattern in the inputstream.adaptive docs.

  • Architecture resolution: torttube delegates playback to plugin.video.youtube 2026-05-23. After spending an iteration on a native DASH MPD builder + HTTP manifest server, conceded that solving segment-timing alignment correctly (sidx-parse, SegmentTimeline emission, audio-rate detection, init-segment presentationTimeOffset) is multiple-week work. plugin.video.youtube has years of that work already. torttube now does what it's faster at (Rust rustypipe search/browse, SponsorBlock skip with fewer false positives) and hands the video_id to pv.youtube for actual playback. This is the "stand on giants' shoulders" call rather than reinvent. Native DASH code path stays in the addon as a fallback when pv.youtube is absent (or as a testbed for someday revisiting). Reference: https://kodi.wiki/view/Add-on:YouTube

  • DASH segment timing for googlevideo SegmentBase URLs — Hit 2026-05-23. My MPD with one Representation per video/audio (using SegmentBase with indexRange to the sidx box of the static MP4) parses cleanly and segments fetch correctly once the User-Agent=Mozilla/...&Origin=https://www.youtube.com&Referer=https://www.youtube.com/ headers are set via inputstream.adaptive.stream_headers. BUT audio drifts badly behind video (-25s growing to -44s within seconds of playback start). Hypothesis: inputstream.adaptive needs explicit per-segment timing (via <SegmentTimeline><S t= d= /> entries) or presentationTimeOffset to align separated audio + video streams correctly. plugin.video.youtube derives these by parsing the sidx box of each representation. Possible upstream PRs: (a) inputstream.adaptive should auto-derive segment timing from sidx when SegmentBase + indexRange is present, OR (b) document the requirement for SegmentTimeline on separated A/V. For torttube we'll need to either parse sidx ourselves (extra HTTP HEAD + binary parse) or fork plugin.video.youtube's MPD-builder.

  • Kodi — "Logic error due to two concurrent busydialogs" fatal Reproduced during rapid back-to-back Player.Open calls while a previous play's BusyDialog was still dismissing. Log message itself says "this is a known issue", but the app HARD-exits rather than silently dropping one of the dialogs. Look up the upstream tracker before filing.

Watching (not ours, but relevant to torttube)

Project PR/Issue Title Why we care
rustypipe PR #77 "Some fixes" (Schmiddiii) Targeted fixes for YouTube changes ~2026-05-21: video metadata parsing (channel-tag removal) + duration parsing (thumbnail overlay field renames). +15/-11 across 2 files. Stalled in CI-needs-approval. We've independently verified rustypipe-0.11.4 (which predates this PR) still works for player(), search(), and channel_videos() against current YouTube as of 2026-05-23 — Rick Astley video, LTT search returning 19 results, LTT channel browse returning 30 videos. Suggests the breaks PR #77 targets are in code paths we don't exercise. TODO: comment on the PR with this test data + offer to mirror to GitHub if codeberg CI stays blocked. Gated on creating a Sulkta-Coop codeberg account.
NPE #1339 n-parameter deobfuscation broken Core to playback. Fix here = fix in our Tier-1.
NPE #1444 Distinguish unavailable vs unextractable Clean typed-error PR target.
NPE #1360 Refactor link handlers "help wanted" label.
NPE #1357 JDoc checks in PR pipeline "good first issue" — smallest credible first PR.

Posture

  • Every sidecar bug we trace ends with one question: is the root cause in rustypipe / NPE / yt-dlp? If yes, fix it there first; pull the fix into our sidecar via dep bump or backport.
  • We are not pretending to be on the rustypipe / NPE / yt-dlp maintainer level. We're a downstream consumer that does its share.
  • A merged PR > a forked patch. A fork is the last resort, not the first move.