PublishTrack: populate SimulcastCodecs with primary codec for video tracks (fixes H.265 single-stream forwarding)

[aphexcx]

Summary

PublishTrack for video tracks without an explicit backup codec was leaving AddTrackRequest.SimulcastCodecs empty. The server's TrackPublishedResponse therefore echoed back an empty TrackInfo.MimeType (the comment at publication.go:78-105 describes this dependency), so trackPublicationBase.MimeType() returned "" for every single-stream video publish.

setPublishingCodecsQuality (publication.go:485-528) then misclassified every subscribed codec as backup — strings.HasSuffix("", "h265") is always false — and emitted "subscriber requested backup codec but no track found" instead of running the primary-codec branch.

For H.264 the SFU still forwarded RTP via defaults, so the bug was invisible; for H.265 (and presumably other non-default codecs) the SFU never started forwarding, the subscriber's track.muted stayed true, and zero bytes were received even though SDP negotiation completed cleanly.

Fix

Add a single else if branch in PublishTrack that populates SimulcastCodecs[0] with {Codec: primaryCodec.MimeType, Cid: track.ID()} when:

• there's no explicit backup codec (existing branch already handles that case at localparticipant.go:159-176)
• kind == TrackKindVideo
• primaryCodec.MimeType != ""

This mirrors what PublishSimulcastTrack already does at localparticipant.go:311-328. After the fix:

• MimeType() returns the actual codec MIME (video/H265, video/VP8, etc.)
• setPublishingCodecsQuality correctly takes the primary-codec branch
• The spurious subscriber requested backup codec but no track found warnings stop firing for normal single-stream publishes
• H.265 subscribers actually receive RTP

Audio is correctly excluded — audio codec updates dispatch through handleSubscribedAudioCodecUpdate / setAudioCodecSubscribed, not setPublishingCodecsQuality.

Test

Adds TestPublishTrackSingleStreamCodecMime covering H.264, H.265, and VP8 single-stream PublishTrack. For each:

1. Asserts trackPub.MimeType() == codec.MimeType after publish
2. Subscribes from a second participant and asserts the track is received within 10s (regression: H.265 was previously stuck on track.muted=true and never delivered)

Test follows the existing TestSimulcastCodec pattern (same createAgent / pubNullTrack helpers, same Docker-server mage test runner). Compiles cleanly; functional verification requires the LiveKit server harness.

Context / how this was found

Discovered while spiking LiveKit as a candidate SFU for low-latency robot teleop. The camera produces hardware H.265 (Rockchip mpph265enc); re-encoding is too slow for our latency budget. lk room join --publish h265://host:port (livekit-cli v2.16.2) negotiated SDP, fired TrackSubscribed in Chrome 147, but the receiver's track stayed muted and zero RTP arrived. Same pipeline with H.264 worked.

Initial issue reported against livekit-cli: livekit/livekit-cli#837. After tracing through the SDK with help from a code-review agent, the actual bug is here in server-sdk-go's PublishTrack AddTrackRequest assembly — livekit-cli is just a caller that happens to hit it. Fixing it here also helps publishFile and any other PublishTrack caller.

I verified the build with go build ./..., go vet ./..., and go test -c . (test binary builds clean). I haven't run the integration test against a live LiveKit server in CI — happy to iterate on the test if maintainers want a different shape.

Notes for reviewers

• The new branch only runs when pubOptions.backupCodecTrack == nil (existing branch handles the backup case) and primaryCodec.MimeType != "" (skips the case where the SDK can't infer a codec).
• No encryption guard, mirroring how PublishSimulcastTrack already populates SimulcastCodecs unconditionally — the existing E2EE guard in the backup-codec branch is specifically about setBackupCodecTrack, not about SimulcastCodecs[0] metadata.
• The change is metadata-only; no behavior change to RTP packetization, RID extension, or transceiver setup.

[CLAassistant]

All committers have signed the CLA.