Two edge tests for the adaptive playout-delay controller:
- grown_target_requires_deeper_reprime: a disruption-grown target actually
gates the next re-prime (3 frames no longer enough once target is 4).
- overflow_resync_resets_clean_run: the MAX_BUFFERED overflow resync path
restarts the clean run.
Gemini-authored (junior), senior-reviewed against the real diff and
independently re-verified (cargo test --lib + clippy clean). Driven via the
headless agy --print --sandbox loop (resumed with --continue past the
orientation-tax timeout).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the fixed 3-frame (~60ms) playout delay with a feedback
controller that tunes depth to real network behavior, no wall clock
needed:
- Grow (+1 frame) on a late-arriving packet (one for a sequence already
played past) or a gap that forces Opus PLC — jitter beat the cushion.
- Shrink (-1 frame) after a long unbroken run of real frames — the link
is comfortably ahead. Fast grow, slow shrink (AIMD-style).
- Bounded to [2, 12] frames (40-240ms), well under MAX_BUFFERED_FRAMES.
- Benign silence (a talker pausing) emits none of these signals, so the
delay is untouched across quiet stretches — avoids the classic
"inflate delay because someone went quiet" bug.
- Prime-timeout safety net: since the mixer polls every ~20ms, prime
after ~500ms even under a grown target so a short utterance isn't held
forever and startup latency stays bounded.
No public API change; all logic stays in jitter.rs. Adds 8 unit tests
(grow/shrink, both bounds, silence-neutrality, prime timeout).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Covers the remaining state-transition edges: re-prime after an underrun goes
idle (must re-accumulate TARGET_DELAY_FRAMES, not resume on one packet),
duplicate-insert overwrite (no buffer growth), is_idle across fresh/buffering/
underrun, and overflow-resync when next_seq is already Some (playout head
snaps to the new front). Jitter tests 6 -> 10; test-only, no prod change.
Implemented by Gemini per next-task.md; left uncommitted per the operating-
agreement default, reviewed against the real diff and re-verified (build +
clippy --all-targets + test all green) by the senior.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add unit tests for the jitter buffer's untested pure logic: seq_before's
wrapping u32 comparison — basic ordering, the u32::MAX→0 forward wrap, and
the exact < (1<<31) half-range tipping point (0x7FFF_FFFF before, 0x8000_0000
not) — and insert's MAX_BUFFERED_FRAMES overflow path (caps depth, drops the
oldest frame, resyncs the playout head next_seq to the new front). Tests-only;
no production change.
Implemented by Gemini (junior implementer), reviewed and verified by senior
(cargo build + clippy --all-targets + cargo test all green).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replaces the fire-and-forget datagram path with sequenced packets, a
per-peer jitter buffer, and persistent per-peer send tasks. Together these
fix three intertwined weaknesses that only showed up off localhost.
Packet format: every audio frame now carries a 4-byte little-endian
sequence number header ([seq][opus payload]), the basis for reordering and
loss detection.
Jitter buffer (core/jitter.rs): incoming packets are reordered by sequence
behind a fixed ~60ms playout delay. Missing sequences with later packets
already buffered are concealed via Opus PLC (decode(None)) -- a path the
decoder supported but nothing ever invoked. Underruns go idle and re-buffer
rather than concealing indefinitely. Covered by unit tests using real
encoded frames (reorder, gap-conceal, prime, late-drop).
Transport (network/iroh_impl.rs): each peer gets one long-lived send task
fed by a shallow bounded channel (drop-oldest on backpressure), instead of
spawning a throwaway task per peer per 20ms frame. Connections are now
established reactively on peer-join and torn down on peer-leave; the
lexicographically-lower EndpointId dials so a full-mesh pair forms exactly
one shared bidirectional connection instead of two racing ones. This also
removes the previous lock-held-across-connect().await serialization.
Opus decoder: PLC output is now sized to one 20ms frame, so concealment
synthesizes 20ms instead of a 120ms burst from the oversized max buffer.
Known follow-up (Tier 2): no reconnect on transient connection loss; a
send error currently retires the peer until they rejoin.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
mollusk
