Gracefully handle dirtying of audio nodes while processing current set.

third_party/WebKit/Source/modules/webaudio/DeferredTaskHandler.cpp

Index: third_party/WebKit/Source/modules/webaudio/DeferredTaskHandler.cpp
diff --git a/third_party/WebKit/Source/modules/webaudio/DeferredTaskHandler.cpp b/third_party/WebKit/Source/modules/webaudio/DeferredTaskHandler.cpp
index d294fe8fa80585e76e0331a63cb657d395b0f72c..8cbecfc07b38c89ce60357434018d0589fa663b0 100644
--- a/third_party/WebKit/Source/modules/webaudio/DeferredTaskHandler.cpp
+++ b/third_party/WebKit/Source/modules/webaudio/DeferredTaskHandler.cpp
@@ -131,9 +131,14 @@ void DeferredTaskHandler::handleDirtyAudioNodeOutputs()
 {
     ASSERT(isGraphOwner());
 
-    for (AudioNodeOutput* output : m_dirtyAudioNodeOutputs)
+    HashSet<AudioNodeOutput*> dirtyOutputs;
+    m_dirtyAudioNodeOutputs.swap(dirtyOutputs);

[hongchan, 2016/05/23 20:43:34]

What if this swapping happens when the audio thread is accessing
m_dirtyAudionodeOutputs? Is this guaranteed to be atomic?

[sof, 2016/05/23 21:06:30]

No, the calling context will have to guarantee exclusive access. I thought
handleDeferredTasks() was only performed on the audio thread?

+
+    // Note: the updating of rendering state may cause output nodes
+    // further down the chain to be marked as dirty. These will not
+    // be processed in this render quantum.
+    for (AudioNodeOutput* output : dirtyOutputs)
         output->updateRenderingState();
-    m_dirtyAudioNodeOutputs.clear();
 }
 
 void DeferredTaskHandler::addAutomaticPullNode(AudioHandler* node)