fix(desktop): recover dead mic on buffered PTT silent turns (#9081)#9390
Open
kodjima33 wants to merge 1 commit into
Open
fix(desktop): recover dead mic on buffered PTT silent turns (#9081)#9390kodjima33 wants to merge 1 commit into
kodjima33 wants to merge 1 commit into
Conversation
The two buffered push-to-talk silent-turn exits — commitBufferedRealtimeHubTurn
(source "buffered_hub") and transcribeBufferedWarmWaitAudio ("warm_wait_fallback"),
both hit when the realtime hub is still warming at release — discarded near-silent
turns without ever feeding silentMicRecoveryPolicy.recordDiscardedTurn or requesting
a CoreAudio capture rebuild. The primary hub/omni/batch gates already do this. A
user whose hold-mode turns landed on the warm-wait path could accumulate silent
turns forever with recovery_action=none / recovery_result=not_attempted — exactly
the telemetry in #9081 (169 users, 0.12.0, hold mode).
Fix: wire both buffered paths into the same dead-mic recovery the primary paths use
— count each discard toward the threshold, attempt a capture rebuild once tripped,
and emit the real recovery_action/recovery_result. Also reset the counter on the
buffered success paths so a recovered mic clears accumulated dead turns.
Regression test (source-scrape, matching PTTAudioCaptureRaceTests since the manager
is a @mainactor singleton): every recordPTTSilentTurn site must wire recovery.
Verification: xcrun swift build -c debug succeeded (Build complete!, 177s).
UNVERIFIED at runtime: could not reproduce a physical dead-mic + hub-warm-wait turn.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bug (#9081)
Desktop telemetry showed
ptt_audio_capture_silent_turnwithrecovery_action=none/recovery_result=not_attemptedfor 169 users on 0.12.0 in hold mode — push-to-talk turns recorded silence and nothing was done to recover the mic.Root cause
There are four silent-turn discard sites in
PushToTalkManager. The two primary gates (finalize()hub path and the omni/batch path) already runsilentMicRecoveryPolicy.recordDiscardedTurn(...)and, once consecutive dead-mic turns cross the threshold, callrequestCoreAudioCaptureRecovery(...)to rebuild CoreAudio capture.The two buffered exits did not:
commitBufferedRealtimeHubTurn()—source: "buffered_hub"transcribeBufferedWarmWaitAudio()—source: "warm_wait_fallback"Both are taken when the user releases before the realtime hub has finished warming — the common hold-mode path. They discarded near-silent turns and passed no recovery args, so
recordPTTSilentTurnfell back to itsrecoveryAction: "none"default and the dead-mic counter never advanced. A user stuck on this path could never trip recovery.Fix
Wire both buffered paths into the exact same dead-mic recovery the primary paths use:
recordDiscardedTurn,recovery_action/recovery_result,Surgical, mirrors the existing sibling pattern.
additions+deletions = 80across 1 source file + 1 test + 1 changelog.Test
PTTSilentTurnRecoveryWiringTests— sincePushToTalkManageris a@MainActorsingleton not constructible in a unit test, this follows the repo's existing source-scrape pattern (PTTAudioCaptureRaceTests): asserts everyrecordPTTSilentTurnsite wires recovery, so a future unwired site fails CI.Verification
xcrun swift build -c debug --package-path Desktop→ Build complete! (177s).🤖 automated by hourly watchdog; opened for review, not merged.