Default the generate key frame algorithm to all layers.

[jan-ivar]

From #143 (comment)

---

Preview | Diff

[fippo]

What is the purpose of blocking the return until all requested key frames have been generated? This was previously required/done to get "a" timestamp but no longer serves any purpose?

See also #145 on the overall broken assumptions

[dontcallmedom-bot]

This issue had an associated resolution in WebRTC September 2025 meeting – 16 September 2025 (PR #276: Default the generate key frame algorithm to all layers):

RESOLUTION: proceed with merging PR

[henbos]

Note that RIDs are optional, we only know that either all encodings have RIDs or none of them have RIDs, but we don't know that no RIDs equals singlecast - you could have multiple encodings without RIDs. The alternative would be encodingIndex but we can say that rids are a prerequisite

[jan-ivar]

I've now updated the algorithm to work sans rids. PTAL @henbos @youennf

[jan-ivar]

Note I also cleaned up the language a bit:

"encoder" was being used at two different conceptual levels: A) as the owner of all layers, and B) per layer. I reworked/replaced the latter with "layer" consistent with existing text in 4.10: "When transform’s [[encoder]] receives a keyframe request, ... Set rid to the RID of the appropriate layer, or undefined if the request is not for a specific layer."

[youennf]

I'll file a separate issue about the promise resolution timing, maybe we are too smart and we should simplify further the algorithm.