Prevent AudioLowLatencyInputMac from changing frame sizes.

media/audio/mac/audio_low_latency_output_mac.cc

Index: media/audio/mac/audio_low_latency_output_mac.cc
diff --git a/media/audio/mac/audio_low_latency_output_mac.cc b/media/audio/mac/audio_low_latency_output_mac.cc
index 0327db0b444d0af8ba610354e236897842bb452b..9d1c673e5936f40f7cf83e2b02791ef7790ce6ff 100644
--- a/media/audio/mac/audio_low_latency_output_mac.cc
+++ b/media/audio/mac/audio_low_latency_output_mac.cc
@@ -154,6 +154,11 @@ bool AUAudioOutputStream::Configure() {
     return false;
 
   // Set the buffer frame size.
+  // WARNING: Setting this value changes the frame size for all audio units in
+  // the current process.  It's imperative that the input and output frame sizes
+  // be the same as audio_util::GetAudioHardwareBufferSize().
+  // TODO(henrika): Due to http://crrev.com/159666 this is currently not true
+  // and should be fixed, a CHECK() should be added at that time.
   UInt32 buffer_size = number_of_frames_;
   result = AudioUnitSetProperty(
       output_unit_,
@@ -234,18 +239,21 @@ OSStatus AUAudioOutputStream::Render(UInt32 number_of_frames,
   uint32 hardware_pending_bytes = static_cast<uint32>
       ((playout_latency_frames + 0.5) * format_.mBytesPerFrame);
 
-  // If we specify a buffer size which is too low, the OS will ask for more data
-  // to fulfill the hardware request, so resize the AudioBus as appropriate.
-  // This change requires AudioOutputResampler to prevent buffer size mismatches
-  // downstream, so glitch if it's not enabled.
-  if (!kDisableAudioOutputResampler &&
-      static_cast<UInt32>(audio_bus_->frames()) != number_of_frames) {
-    audio_bus_ = AudioBus::Create(audio_bus_->channels(), number_of_frames);
-  }
-
+  // Unfortunately AUAudioInputStream and AUAudioOutputStream share the frame

[Chris Rogers, 2012/10/08 21:08:26]

nit: "frame frame"

[DaleCurtis, 2012/10/08 21:17:34]

Done.

+  // frame size set by kAudioDevicePropertyBufferFrameSize above on a per
+  // process basis.  What this means is that the |number_of_frames| value may
+  // be larger or smaller than the value set during Configure().  The downstream
+  // audio pipeline does not support dynamic frame size changes, as such we must
+  // clip |frames_filled| as necessary, this will result in bad audio, but the
+  // alternative is a browser crash.
+  // TODO(henrika): This should never happen so long as we're always using the
+  // hardware sample rate and the input/output streams configure the same frame
+  // size.  This is currently not true.  See http://crbug.com/154352.  Once
+  // fixed, a CHECK() should be added and the clipping + wall of text removed.

[Chris Rogers, 2012/10/08 21:08:26]

Don't we still want to check for (audio_bus_->frames()) != number_of_frames) 
and simply memset(audio_data, 0, ...? and return early here?

[DaleCurtis, 2012/10/08 21:17:34]

Things are already broken so it doesn't really matter, but I thought keeping the
existing clipping behavior would be simpler.  One results in glitched audio the
other in silence.

   int frames_filled = std::min(source_->OnMoreData(
       audio_bus_.get(), AudioBuffersState(0, hardware_pending_bytes)),
       static_cast<int>(number_of_frames));
+
   // Note: If this ever changes to output raw float the data must be clipped and
   // sanitized since it may come from an untrusted source such as NaCl.
   audio_bus_->ToInterleaved(