[Experimental Screenshare audio fix] Seperate logic to own files (minor change can possibly revert)

This commit is contained in:
2026-03-13 02:26:55 +01:00
parent 15c5952e29
commit 22d355a522
12 changed files with 1211 additions and 800 deletions

View File

@@ -32,6 +32,11 @@ interface SinkInputDetails extends ShortSinkInputEntry {
properties: Record<string, string>;
}
interface DescendantProcessInfo {
ids: ReadonlySet<string>;
binaryNames: ReadonlySet<string>;
}
interface PactlJsonSinkInputEntry {
index?: number | string;
properties?: Record<string, unknown>;
@@ -44,6 +49,7 @@ interface LinuxScreenShareAudioRoutingState {
screenShareLoopbackModuleId: string | null;
voiceLoopbackModuleId: string | null;
rerouteIntervalId: ReturnType<typeof setInterval> | null;
subscribeProcess: ChildProcess | null;
}
interface LinuxScreenShareMonitorCaptureState {
@@ -77,7 +83,8 @@ const routingState: LinuxScreenShareAudioRoutingState = {
restoreSinkName: null,
screenShareLoopbackModuleId: null,
voiceLoopbackModuleId: null,
rerouteIntervalId: null
rerouteIntervalId: null,
subscribeProcess: null
};
const monitorCaptureState: LinuxScreenShareMonitorCaptureState = {
captureId: null,
@@ -126,12 +133,21 @@ export async function activateLinuxScreenShareAudioRouting(): Promise<LinuxScree
routingState.screenShareLoopbackModuleId = await loadLoopbackModule(SCREEN_SHARE_MONITOR_SOURCE_NAME, restoreSinkName);
routingState.voiceLoopbackModuleId = await loadLoopbackModule(`${VOICE_SINK_NAME}.monitor`, restoreSinkName);
await setDefaultSink(SCREEN_SHARE_SINK_NAME);
await moveSinkInputs(SCREEN_SHARE_SINK_NAME, (sinkName) => !!sinkName && sinkName !== SCREEN_SHARE_SINK_NAME && sinkName !== VOICE_SINK_NAME);
// Set the default sink to the voice sink so that new app audio
// streams (received WebRTC voice) never land on the screenshare
// capture sink. This prevents the feedback loop where remote
// voice audio was picked up by parec before the reroute interval
// could move the stream away.
await setDefaultSink(VOICE_SINK_NAME);
routingState.active = true;
await rerouteAppSinkInputsToVoiceSink();
// Let the combined reroute decide placement for every existing
// stream. This avoids briefly shoving the app's own playback to the
// screenshare sink before ownership detection can move it back.
await rerouteSinkInputs();
startSinkInputRerouteLoop();
startSubscribeWatcher();
return buildRoutingInfo(true, true);
} catch (error) {
@@ -148,6 +164,7 @@ export async function activateLinuxScreenShareAudioRouting(): Promise<LinuxScree
export async function deactivateLinuxScreenShareAudioRouting(): Promise<boolean> {
const restoreSinkName = routingState.restoreSinkName;
stopSubscribeWatcher();
stopSinkInputRerouteLoop();
await stopLinuxScreenShareMonitorCapture();
@@ -166,6 +183,7 @@ export async function deactivateLinuxScreenShareAudioRouting(): Promise<boolean>
routingState.restoreSinkName = null;
routingState.screenShareLoopbackModuleId = null;
routingState.voiceLoopbackModuleId = null;
routingState.subscribeProcess = null;
return true;
}
@@ -425,34 +443,52 @@ async function setDefaultSink(sinkName: string): Promise<void> {
await runPactl('set-default-sink', sinkName);
}
async function rerouteAppSinkInputsToVoiceSink(): Promise<void> {
/**
* Combined reroute that enforces sink placement in both directions:
* - App-owned sink inputs that are NOT on the voice sink are moved there.
* - Non-app sink inputs that ARE on the voice sink are moved to the
* screenshare sink so they are captured by parec.
*
* This two-way approach, combined with the voice sink being the PulseAudio
* default, ensures that received WebRTC voice audio can never leak into the
* screenshare monitor source.
*/
async function rerouteSinkInputs(): Promise<void> {
const [
sinks,
sinkInputs,
descendantProcessIds
descendantProcessInfo
] = await Promise.all([
listSinks(),
listSinkInputDetails(),
collectDescendantProcessIds(process.pid)
collectDescendantProcessInfo(process.pid)
]);
const sinkNamesByIndex = new Map(sinks.map((sink) => [sink.index, sink.name]));
await Promise.all(
sinkInputs.map(async (sinkInput) => {
if (!isAppOwnedSinkInput(sinkInput, descendantProcessIds)) {
return;
}
const sinkName = sinkNamesByIndex.get(sinkInput.sinkIndex) ?? null;
const appOwned = isAppOwnedSinkInput(sinkInput, descendantProcessInfo);
// App-owned streams must stay on the voice sink.
if (appOwned && sinkName !== VOICE_SINK_NAME) {
try {
await runPactl('move-sink-input', sinkInput.index, VOICE_SINK_NAME);
} catch {
// Streams can disappear or be recreated while rerouting.
}
if (sinkName === VOICE_SINK_NAME) {
return;
}
try {
await runPactl('move-sink-input', sinkInput.index, VOICE_SINK_NAME);
} catch {
// Streams can disappear or be recreated while rerouting.
// Non-app streams sitting on the voice sink should be moved to the
// screenshare sink for desktop-audio capture.
if (!appOwned && sinkName === VOICE_SINK_NAME) {
try {
await runPactl('move-sink-input', sinkInput.index, SCREEN_SHARE_SINK_NAME);
} catch {
// Streams can disappear or be recreated while rerouting.
}
}
})
);
@@ -515,7 +551,7 @@ function startSinkInputRerouteLoop(): void {
}
routingState.rerouteIntervalId = setInterval(() => {
void rerouteAppSinkInputsToVoiceSink();
void rerouteSinkInputs();
}, REROUTE_INTERVAL_MS);
}
@@ -528,13 +564,108 @@ function stopSinkInputRerouteLoop(): void {
routingState.rerouteIntervalId = null;
}
/**
* Spawns `pactl subscribe` to receive PulseAudio events in real time.
* When a new or changed sink-input is detected, a reroute is triggered
* immediately instead of waiting for the next interval tick. This
* drastically reduces the time non-app desktop audio spends on the
* voice sink before being moved to the screenshare sink.
*/
function startSubscribeWatcher(): void {
if (routingState.subscribeProcess) {
return;
}
let proc: ChildProcess;
try {
proc = spawn('pactl', ['subscribe'], {
env: process.env,
stdio: [
'ignore',
'pipe',
'ignore'
]
});
} catch {
// If pactl subscribe fails to spawn, the interval loop still covers us.
return;
}
routingState.subscribeProcess = proc;
let pending = false;
proc.stdout?.on('data', (chunk: Buffer) => {
if (!routingState.active) {
return;
}
const text = chunk.toString();
if (/Event '(?:new|change)' on sink-input/.test(text)) {
if (!pending) {
pending = true;
// Batch rapid-fire events with a short delay.
setTimeout(() => {
pending = false;
void rerouteSinkInputs();
}, 50);
}
}
});
proc.on('close', () => {
if (routingState.subscribeProcess === proc) {
routingState.subscribeProcess = null;
}
});
proc.on('error', () => {
if (routingState.subscribeProcess === proc) {
routingState.subscribeProcess = null;
}
});
}
function stopSubscribeWatcher(): void {
const proc = routingState.subscribeProcess;
if (!proc) {
return;
}
routingState.subscribeProcess = null;
if (!proc.killed) {
proc.kill('SIGTERM');
}
}
function isAppOwnedSinkInput(
sinkInput: SinkInputDetails,
descendantProcessIds: ReadonlySet<string>
descendantProcessInfo: DescendantProcessInfo
): boolean {
const processId = sinkInput.properties['application.process.id'];
return typeof processId === 'string' && descendantProcessIds.has(processId);
if (typeof processId === 'string' && descendantProcessInfo.ids.has(processId)) {
return true;
}
const processBinary = normalizeProcessBinary(sinkInput.properties['application.process.binary']);
if (processBinary && descendantProcessInfo.binaryNames.has(processBinary)) {
return true;
}
const applicationName = normalizeProcessBinary(sinkInput.properties['application.name']);
if (applicationName && descendantProcessInfo.binaryNames.has(applicationName)) {
return true;
}
return false;
}
async function moveSinkInputs(
@@ -697,31 +828,45 @@ async function listSinkInputDetails(): Promise<SinkInputDetails[]> {
return entries.filter((entry) => !!entry.sinkIndex);
}
async function collectDescendantProcessIds(rootProcessId: number): Promise<Set<string>> {
const { stdout } = await execFileAsync('ps', ['-eo', 'pid=,ppid='], {
async function collectDescendantProcessInfo(rootProcessId: number): Promise<DescendantProcessInfo> {
const { stdout } = await execFileAsync('ps', ['-eo', 'pid=,ppid=,comm='], {
env: process.env
});
const childrenByParentId = new Map<string, string[]>();
const binaryNameByProcessId = new Map<string, string>();
stdout
.split(/\r?\n/)
.map((line) => line.trim())
.filter(Boolean)
.forEach((line) => {
const [pid, ppid] = line.split(/\s+/);
const match = line.match(/^(\d+)\s+(\d+)\s+(.+)$/);
if (!pid || !ppid) {
if (!match) {
return;
}
const [
,
pid,
ppid,
command
] = match;
const siblings = childrenByParentId.get(ppid) ?? [];
siblings.push(pid);
childrenByParentId.set(ppid, siblings);
const normalizedBinaryName = normalizeProcessBinary(command);
if (normalizedBinaryName) {
binaryNameByProcessId.set(pid, normalizedBinaryName);
}
});
const rootId = `${rootProcessId}`;
const descendantIds = new Set<string>([rootId]);
const descendantBinaryNames = new Set<string>();
const queue = [rootId];
while (queue.length > 0) {
@@ -731,6 +876,12 @@ async function collectDescendantProcessIds(rootProcessId: number): Promise<Set<s
continue;
}
const binaryName = binaryNameByProcessId.get(currentId);
if (binaryName) {
descendantBinaryNames.add(binaryName);
}
for (const childId of childrenByParentId.get(currentId) ?? []) {
if (descendantIds.has(childId)) {
continue;
@@ -741,7 +892,30 @@ async function collectDescendantProcessIds(rootProcessId: number): Promise<Set<s
}
}
return descendantIds;
return {
ids: descendantIds,
binaryNames: descendantBinaryNames
};
}
function normalizeProcessBinary(value: string | undefined): string | null {
if (!value) {
return null;
}
const trimmed = value.trim();
if (!trimmed) {
return null;
}
const basename = trimmed
.split(/[\\/]/)
.pop()
?.trim()
.toLowerCase() ?? '';
return basename || null;
}
function stripSurroundingQuotes(value: string): string {