fix(audio): clear audioSwitch synchronously in AudioSwitchHandler.stop()

[YashJainSC]

Summary

On Android 12+ (the CommDeviceAudioSwitch path), audio output can get stuck on the earpiece at very low volume after a disconnect() + connect() on a reused Room instance (e.g. switching between rooms without releasing/recreating the Room). The first connection routes to the speaker correctly; a subsequent reconnect intermittently does not.

Root cause

A race in AudioSwitchHandler between stop() and the next start():

• stop() nulls audioSwitch inside the posted teardown runnable (on the handler thread) while tearing down handler/thread synchronously.
• audioSwitch is a non-@Volatile field, and that write happens off the lock on the handler thread.
• A fast subsequent start() (a different thread, under the @Synchronized lock) can therefore read a stale, already-stopped switch and skip re-creation via the if (audioSwitch == null) guard.

With no new switch created, activate() never re-runs, so the route that the prior deactivate() cleared via clearCommunicationDevice() (reverting the OS to the earpiece) is never re-asserted.

Instrumented logs from a reproduction (good connect, then a bad reconnect on the same Room):

# 1st connect — OK (ends on speaker)
start() called. handlerAlive=false, threadAlive=false, audioSwitchNull=true
start(): audioSwitch == null -> creating a new switch
deviceChange: available=[Speakerphone, Earpiece], selected=Speakerphone
switch started+activated (CommDeviceAudioSwitch). selected=Speakerphone

# reconnect on the SAME Room — BAD (stuck on earpiece)
stop() called. audioSwitchNull=false
stop(): posted audioSwitch.stop() + audioSwitch=null on handler thread   # field nulled here
start() called. handlerAlive=false, threadAlive=false, audioSwitchNull=false   # 456ms later, STILL reads non-null
-> SKIPPING re-create (reusing stale, stopped switch)   # no activate(), routing not re-asserted

The audioSwitch = null write is wall-clock before the start() read, yet start() observes non-null — a memory-visibility issue, compounded by the structural asymmetry that handler/thread are nulled synchronously while audioSwitch is not.

Fix

Clear audioSwitch synchronously under the lock in stop() (capturing the instance into a local so its stop() can still be posted to the handler thread, since AbstractAudioSwitch is not threadsafe), and mark the field @Volatile. This guarantees the next start() observes the teardown and re-creates the switch.

Verification

With the fix, every reconnect on the reused Room logs start(): audioSwitch == null -> creating a new switch → activate() → selected=Speakerphone, and audio reliably routes to the speaker (no more SKIPPING re-create). Verified on-device (Android 15, Moto G45 5G) across repeated room switches.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: d0cee0d

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
client-sdk-android	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[davidliu]

Thanks for the PR!