Bug 1513973 - Enable pulling at the same time as starting audio sources. r=jib

Otherwise we risk building up a buffer in the microphone source from when
Start() is run until pulling is enabled. This manifests itself as input latency
to the user.

Differential Revision: https://phabricator.services.mozilla.com/D15194

--HG--
extra : moz-landing-system : lando

dom/media/MediaManager.cpp

@@ -4123,6 +4123,17 @@ SourceListener::InitializeAsync() {
                  MozPromiseHolder<SourceListenerPromise>& aHolder) {
                if (audioDevice) {
                  audioDevice->SetTrack(stream, kAudioTrack, principal);
+               }
+
+               if (videoDevice) {
+                 videoDevice->SetTrack(stream, kVideoTrack, principal);
+               }
+
+               // SetTrack() queued the tracks. We add them synchronously here
+               // to avoid races.
+               stream->FinishAddTracks();
+
+               if (audioDevice) {
                  nsresult rv = audioDevice->Start();
                  if (NS_FAILED(rv)) {
                    nsString log;
@@ -4143,7 +4154,6 @@ SourceListener::InitializeAsync() {
                }
 
                if (videoDevice) {
-                 videoDevice->SetTrack(stream, kVideoTrack, principal);
                  nsresult rv = videoDevice->Start();
                  if (NS_FAILED(rv)) {
                    if (audioDevice) {
@@ -4160,9 +4170,6 @@ SourceListener::InitializeAsync() {
                  }
                }
 
-               // Start() queued the tracks to be added synchronously to avoid
-               // races
-               stream->FinishAddTracks();
                LOG("started all sources");
                aHolder.Resolve(true, __func__);
              })
@@ -4186,15 +4193,8 @@ SourceListener::InitializeAsync() {
                  state->mTrackEnabled = true;
                  state->mTrackEnabledTime = TimeStamp::Now();
 
-                 if (state->mDevice->GetMediaSource() !=
-                     MediaSourceEnum::AudioCapture) {
-                   // For AudioCapture mStream is a dummy stream, so we don't
-                   // try to enable pulling - there won't be a track to enable
-                   // it for.
-                   mStream->SetPullingEnabled(state == mAudioDeviceState.get()
-                                                  ? kAudioTrack
-                                                  : kVideoTrack,
-                                              true);
+                 if (state == mVideoDeviceState.get()) {
+                   mStream->SetPullingEnabled(kVideoTrack, true);
                  }
                }
                return SourceListenerPromise::CreateAndResolve(true, __func__);

dom/media/webrtc/MediaEngineDefault.cpp

@@ -454,8 +454,16 @@ nsresult MediaEngineDefaultAudioSource::Start(
     mSineGenerator = new SineWaveGenerator(mStream->GraphRate(), mFreq);
   }
 
-  MutexAutoLock lock(mMutex);
-  mState = kStarted;
+  {
+    MutexAutoLock lock(mMutex);
+    mState = kStarted;
+  }
+
+  NS_DispatchToMainThread(
+      NS_NewRunnableFunction(__func__, [stream = mStream, track = mTrackID]() {
+        stream->SetPullingEnabled(track, true);
+      }));
+
   return NS_OK;
 }
 
@@ -469,8 +477,15 @@ nsresult MediaEngineDefaultAudioSource::Stop(
 
   MOZ_ASSERT(mState == kStarted);
 
-  MutexAutoLock lock(mMutex);
-  mState = kStopped;
+  {
+    MutexAutoLock lock(mMutex);
+    mState = kStopped;
+  }
+
+  NS_DispatchToMainThread(
+      NS_NewRunnableFunction(__func__, [stream = mStream, track = mTrackID]() {
+        stream->SetPullingEnabled(track, false);
+      }));
   return NS_OK;
 }
 

dom/media/webrtc/MediaEngineSource.h

@@ -156,6 +156,8 @@ class MediaEngineSourceInterface {
    *
    * If this is the first AllocationHandle to start, the underlying device
    * will be started.
+   *
+   * NB: Audio sources handle the enabling of pulling themselves.
    */
   virtual nsresult Start(const RefPtr<const AllocationHandle>& aHandle) = 0;
 
@@ -204,6 +206,8 @@ class MediaEngineSourceInterface {
    *
    * Double-stopping a given allocation handle is allowed and will return NS_OK.
    * This is necessary sometimes during shutdown.
+   *
+   * NB: Audio sources handle the disabling of pulling themselves.
    */
   virtual nsresult Stop(const RefPtr<const AllocationHandle>& aHandle) = 0;
 

dom/media/webrtc/MediaEngineWebRTCAudio.cpp

@@ -574,14 +574,16 @@ nsresult MediaEngineWebRTCMicrophoneSource::Start(
 
   RefPtr<MediaEngineWebRTCMicrophoneSource> that = this;
   RefPtr<MediaStreamGraphImpl> gripGraph = mStream->GraphImpl();
-  NS_DispatchToMainThread(media::NewRunnableFrom(
-      [that, graph = std::move(gripGraph), deviceID]() mutable {
+  NS_DispatchToMainThread(
+      media::NewRunnableFrom([that, graph = std::move(gripGraph), deviceID,
+                              stream = mStream, track = mTrackID]() mutable {
         if (graph) {
           graph->AppendMessage(MakeUnique<StartStopMessage>(
               that->mInputProcessing, StartStopMessage::Start));
         }
 
-        that->mStream->OpenAudioInput(deviceID, that->mInputProcessing);
+        stream->OpenAudioInput(deviceID, that->mInputProcessing);
+        stream->SetPullingEnabled(track, true);
 
         return NS_OK;
       }));
@@ -608,8 +610,9 @@ nsresult MediaEngineWebRTCMicrophoneSource::Stop(
 
   RefPtr<MediaEngineWebRTCMicrophoneSource> that = this;
   RefPtr<MediaStreamGraphImpl> gripGraph = mStream->GraphImpl();
-  NS_DispatchToMainThread(media::NewRunnableFrom(
-      [that, graph = std::move(gripGraph), stream = mStream]() mutable {
+  NS_DispatchToMainThread(
+      media::NewRunnableFrom([that, graph = std::move(gripGraph),
+                              stream = mStream, track = mTrackID]() mutable {
         if (graph) {
           graph->AppendMessage(MakeUnique<StartStopMessage>(
               that->mInputProcessing, StartStopMessage::Stop));
@@ -618,6 +621,7 @@ nsresult MediaEngineWebRTCMicrophoneSource::Stop(
         CubebUtils::AudioDeviceID deviceID = that->mDeviceInfo->DeviceID();
         Maybe<CubebUtils::AudioDeviceID> id = Some(deviceID);
         stream->CloseAudioInput(id, that->mInputProcessing);
+        stream->SetPullingEnabled(track, false);
 
         return NS_OK;
       }));