Repository navigation
Sound keeps its rate, fused regions convert, feeds wait, leakies leak pictures - #5
Merged
Merged
Conversation
A sound's rate was lost at asetpts, aselect, atempo and ainterleave, as a picture's frame rate is, so a node clocked by a sound read through a setpts (the demo's takeover switch, with only `a` bound) had no window in time: its delay was uncounted and the plan was refused UNBOUNDED_LIVE_INPUT naming it. A sound's rate is its sample rate, which none of those change, so its rate is now followed through them to the filter or input that says it. The binding hint the switch is asked with carries it too, where it was null. The test now asks with the published switch's shape and binds the sound through aresample and setpts as the demo does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… takes
A region holding a leaky and the node reading its output took the
leaky's input in the default yuv420p, and nothing converted between the
leaky and the node: compose, which takes rgba, was refused at run time
("compose input 'v' accepts the pixel formats rgba, and is bound a
stream of yuv420p"). A leaky hands on the pictures it reads, as a split
does, so the edge into the region now carries the format the nodes
behind its leakies and splits take, where they agree, and the ffmpeg
writing it converts on the way in. Readers of the timing alone take
whatever arrives, as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ning The listener is nonblocking so that it can look for a second feeder between reads, and on Windows a socket accepted off a nonblocking listener is nonblocking too. The connection's read timeout then never waited: with nothing on the wire the read returned at once and the listener's loop went round again, one core for as long as a feeder was connected and pacing itself. The accepted socket is now made blocking, so a read waits its 50 ms for the feeder's next bytes. Over a 10 s feed of sound into the switch the sidecar used 0.96 of a core before and 0.01 after. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tures instead ffrwd.leaky over a coded picture drops whole groups, from a late packet to the next keyframe, which is only right in front of something that hands the packets on (a publish, a file the stream is copied into). A leaf's leaky sat on the subscribe's h264 and compose decoded after it, so a stall cost the wait for a keyframe. Now a leaky reading packets keeps them only where everything reading its output takes packets; otherwise a decode goes ahead of it, and it drops single pictures. The decode is an ffmpeg passing the pictures through `null`, which writes them in the format the leaky's readers take. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Four things the second CPU measurement of the SMART demo found, none of them the switch's fault.
asetptsand its kin do not change one, so a node clocked by its sound is no longer refused as unbounded behind them.🤖 Generated with Claude Code