Android media: VideoFrame should not store so many sync points.

content/renderer/media/android/webmediaplayer_android.cc

Index: content/renderer/media/android/webmediaplayer_android.cc
diff --git a/content/renderer/media/android/webmediaplayer_android.cc b/content/renderer/media/android/webmediaplayer_android.cc
index cddec41d3d2c34502f88734081ea88c859cc64a2..4e865bcd5d7d0055fff5be37973e162434d2ede5 100644
--- a/content/renderer/media/android/webmediaplayer_android.cc
+++ b/content/renderer/media/android/webmediaplayer_android.cc
@@ -96,6 +96,7 @@ WebMediaPlayerAndroid::WebMediaPlayerAndroid(
       client_(client),
       delegate_(delegate),
       buffered_(static_cast<size_t>(1)),
+      main_loop_(base::MessageLoopProxy::current()),
       media_loop_(media_loop),
       ignore_metadata_duration_change_(false),
       pending_seek_(false),
@@ -1031,6 +1032,7 @@ void WebMediaPlayerAndroid::DrawRemotePlaybackText(
 }
 
 void WebMediaPlayerAndroid::ReallocateVideoFrame() {
+  DCHECK(main_thread_checker_.CalledOnValidThread());
   if (needs_external_surface_) {
     // VideoFrame::CreateHoleFrame is only defined under VIDEO_HOLE.
 #if defined(VIDEO_HOLE)
@@ -1047,6 +1049,7 @@ void WebMediaPlayerAndroid::ReallocateVideoFrame() {
   } else if (!is_remote_ && texture_id_) {
     GLES2Interface* gl = stream_texture_factory_->ContextGL();
     GLuint texture_id_ref = 0;
+    // FIXME: reuse texture pool like gpu_video_decoder.

[dshwang, 2014/06/03 15:39:40]

Currently every frame create new texture. We can avoid via texture pool like
gpu_video_decoder.
If you agree on this approach, I'll improve about it.

     gl->GenTextures(1, &texture_id_ref);
     GLuint texture_target = kGLTextureExternalOES;
     gl->BindTexture(texture_target, texture_id_ref);
@@ -1109,6 +1112,13 @@ scoped_refptr<media::VideoFrame> WebMediaPlayerAndroid::GetCurrentFrame() {
 
 void WebMediaPlayerAndroid::PutCurrentFrame(
     const scoped_refptr<media::VideoFrame>& frame) {
+  // Create new VideoFrame when a client return the previous frame.
+  DCHECK(!main_thread_checker_.CalledOnValidThread());
+  // FIXME: instead of here, VideoFrameProvider::Client::DidReceiveFrame() can
+  // handle it.

[dshwang, 2014/06/03 15:39:40]

I use PutCurrentFrame() as the end of each frame, but it feels alien because
other decoders do same thing when the frame is ready.
We can match this impl to other decoder via
VideoFrameProviderClientImpl::DidReceiveFrame() callback.

TL;DR
However, if VideoFrameProviderClientImpl::DidReceiveFrame() calls
WebMediaPlayerAndroid::DidReceiveFrame(), it looks weird too because
VideoFrameProviderClientImpl is subclass of VideoFrameProvider::Client and
WebMediaPlayerAndroid is subclass of VideoFrameProvider. Is it weird for
VideoFrameProvider::Client::DidReceiveFrame() to call
VideoFrameProvider::DidReceiveFrame()?

+  main_loop_->PostTask(FROM_HERE,
+                       base::Bind(&WebMediaPlayerAndroid::ReallocateVideoFrame,
+                                  weak_factory_.GetWeakPtr()));
 }
 
 void WebMediaPlayerAndroid::TryCreateStreamTextureProxyIfNeeded() {