Support (E)AC3 passthrough

media/base/audio_splicer.cc

Index: media/base/audio_splicer.cc
diff --git a/media/base/audio_splicer.cc b/media/base/audio_splicer.cc
index 0976c89432a3baef1ce06606b295b7d840c2bb4d..88d7aeef185e91cfa422261135a5e14e2c51ab2c 100644
--- a/media/base/audio_splicer.cc
+++ b/media/base/audio_splicer.cc
@@ -183,7 +183,8 @@ bool AudioStreamSanitizer::AddInput(const scoped_refptr<AudioBuffer>& input) {
   if (!delta.is_zero())
     frames_to_fill = output_timestamp_helper_.GetFramesToTarget(timestamp);
 
-  if (frames_to_fill == 0 || std::abs(frames_to_fill) < kMinGapSize) {
+  if (frames_to_fill == 0 || std::abs(frames_to_fill) < kMinGapSize ||

[DaleCurtis, 2016/11/01 23:05:13]

Raw formats should just skip the sanitizer entirely. Notably this is going to
break things like append window trimming in MSE. So authors will have to be very
careful to ensure they align their buffers correctly during streaming.
+chcunningham.

[AndyWu, 2016/11/04 18:04:24]

Sorry, I am not sure what to do. Should I move the change to line 154 or 164?

[chcunningham, 2016/11/11 19:14:28]

This whole file is now deleted. But Dale is right about append window trimming
(see use case: http://dalecurtis.github.io/llama-demo/index.html)

Append window trimming is achieved by discarding decoded frames in
audio_discard_helper. I'm not sure how you'll support this in a passthrough
scenario. 

[AndyWu, 2016/11/11 23:49:12]

Thanks a lot for the well written article from Dale.
We can still discard bitstream in terms of "compressed audio frame", which is
576 PCM frames in MP3. That is, dropping a complete AudioBuffer data should be
acceptable in passthrough mode.

+      input->sample_format() == kSampleFormatRaw) {
     AddOutputBuffer(input);
     return true;
   }