Repository navigation
SABR is implementable (notes + probes) — but what does a request need while StreamProtectionStatus is 2? #937
|
First off, thanks for Playlet — the codebase is genuinely a pleasure to read. Context: I'm building an unofficial Windows desktop port (Tauri + Svelte, AGPL-3.0 like upstream). Browsing, search, channels, playlists and SponsorBlock all work against InnerTube. Playback is the one thing that doesn't. This isn't a support request — I think I have something useful to contribute, plus one narrow question I can't crack. In #754 you mentioned SABR "is not supported on Playlet yet (and I'm not sure if it's even possible yet)", and #862 repeats that uncertainty. It is possible — I implemented it. Here's everything I measured, in case any of it is useful to you. SABR works, and it isn't especially hardBuilt from the protos in Three things cost me time and aren't in any doc I found:
Also: the first response is a Where I'm stuckPlayback stops at exactly 60.000 seconds, and I can't get past it on any transport:
The questionWhile I can't tell whether it's a field in From #618 I understand why the JS-execution work is offloaded to a server on Roku. A desktop app doesn't have that constraint — I have signature/nsig evaluation (via Completely understand if this is more than you want to get into, or if the honest answer is "it's policy". Either way, happy to share the SABR probe scripts — they're small, standalone Node files and they'd drop into a test harness easily if you ever want to revisit SABR support. |
Replies: 1 comment 1 reply
|
Solved — and the answer makes my question above mostly obsolete. Posting it here because I think it's directly relevant to Playlet. The 60-second wall was never a policy limit. It's an artifact of which client is being used. I was testing IOS / WEB / TV / MWEB / the embedded clients, and all of them hit it. The The recipeThis is yt-dlp's current default client (
Also send Why this might matter for Playlet specificallyThe response comes back with 32 adaptive formats, every one carrying a direct In yt-dlp's client table this client is marked Verified against a 635s video: HTTP 206 with real media at 1%, 25%, 50%, 75%, 95% and 99.9% of the file. Whole video, no wall. Same machine and IP where the IOS client was dying at exactly 60.000s minutes earlier. Corrections to my post above
Credit and caveatCredit entirely to yt-dlp — I found this by installing it, watching it download the full video on the same machine that was failing, then reading Caveat: this is policy, not protocol. VISIONOS presumably gets restricted like every client before it, so it's worth treating as a current-best default rather than a permanent fix. My SABR/UMP implementation still works as a fallback and the notes in my post above stand if you ever want it — the Happy to share any of the probe scripts. |
Solved — and the answer makes my question above mostly obsolete. Posting it here because I think it's directly relevant to Playlet.
The 60-second wall was never a policy limit. It's an artifact of which client is being used. I was testing IOS / WEB / TV / MWEB / the embedded clients, and all of them hit it. The
VISIONOSclient does not.The recipe
This is yt-dlp's current default client (
youtube/_base.py). Three parts, all required:SOCS=CAIand…