Remove the IPC to request DemuxerConfigs when config changes

media/base/android/media_decoder_job.cc

Index: media/base/android/media_decoder_job.cc
diff --git a/media/base/android/media_decoder_job.cc b/media/base/android/media_decoder_job.cc
index c57da3af1729fe0df9ba72e8613742dd38d97f68..1285440c4732c6803e7b60763d2f91f5fe70ae5a 100644
--- a/media/base/android/media_decoder_job.cc
+++ b/media/base/android/media_decoder_job.cc
@@ -102,7 +102,7 @@ void MediaDecoderJob::Prefetch(const base::Closure& prefetch_cb) {
   RequestData(prefetch_cb);
 }
 
-bool MediaDecoderJob::Decode(
+scoped_ptr<DemuxerConfigs> MediaDecoderJob::Decode(
     base::TimeTicks start_time_ticks,
     base::TimeDelta start_presentation_timestamp,
     const DecoderCallback& callback) {
@@ -117,18 +117,28 @@ bool MediaDecoderJob::Decode(
                            base::Unretained(this),
                            start_time_ticks,
                            start_presentation_timestamp));
-    return true;
+    return scoped_ptr<DemuxerConfigs>();
   }
 
   if (DemuxerStream::kConfigChanged == CurrentAccessUnit().status) {
     // Clear received data because we need to handle a config change.

[wolenetz, 2014/05/08 17:41:47]

nit: we no longer clear the data. Sorry I didn't catch this nit earlier...

[qinmin, 2014/05/08 17:48:37]

Done.

     decode_cb_.Reset();
-    ClearData();
-    return false;
+    size_t index = CurrentReceivedDataChunkIndex();
+    if (received_data_[index].demuxer_configs.size() > 0) {
+      return scoped_ptr<DemuxerConfigs>(new DemuxerConfigs(
+          received_data_[index].demuxer_configs[0]));
+    } else {
+      ui_task_runner_->PostTask(FROM_HERE,
+                                base::Bind(&MediaDecoderJob::OnDecodeCompleted,
+                                           base::Unretained(this),
+                                           MEDIA_CODEC_OK,

[wolenetz, 2014/05/08 17:41:47]

nit: I think this is ok, but we would miss learning about broken demuxer that is
incorrectly signalling config change without providing config. Would
MEDIA_CODEC_ERROR be better? I liked the previous CHECK because it would
explicitly crash on this incorrect demuxer behavior. wdyt?

[qinmin, 2014/05/08 17:48:37]

The crash will crash the browser process, probably a MEDIA_CODEC_ERROR will be
better.

On 2014/05/08 17:41:47, wolenetz wrote:

[wolenetz, 2014/05/08 20:05:25]

nit: Can we do a CHECK, then, in media_source_delegate, to get crash for this
situation emitted there? This crash should not occur, so if it does, it may be
low-frequency and MEDIA_CODEC_ERROR to a few users might not get attention
directed to getting a fix. wdyt?

[qinmin, 2014/05/08 20:50:07]

Sounds ok to me. modified the media_source_delegate.cc to expect
GetDemuxerConfigFromStream() always return true in when config changes.

On 2014/05/08 20:05:25, wolenetz wrote:

+                                           kNoTimestamp(), kNoTimestamp()));
+      return scoped_ptr<DemuxerConfigs>();
+    }
   }
 
   DecodeCurrentAccessUnit(start_time_ticks, start_presentation_timestamp);
-  return true;
+  return scoped_ptr<DemuxerConfigs>();
 }
 
 void MediaDecoderJob::StopDecode() {
@@ -462,11 +472,15 @@ void MediaDecoderJob::OnDecodeCompleted(
 const AccessUnit& MediaDecoderJob::CurrentAccessUnit() const {
   DCHECK(ui_task_runner_->BelongsToCurrentThread());
   DCHECK(HasData());
-  int index = NoAccessUnitsRemainingInChunk(true) ?
-      inactive_demuxer_data_index() : current_demuxer_data_index_;
+  size_t index = CurrentReceivedDataChunkIndex();
   return received_data_[index].access_units[access_unit_index_[index]];
 }
 
+size_t MediaDecoderJob::CurrentReceivedDataChunkIndex() const {
+  return NoAccessUnitsRemainingInChunk(true) ?
+      inactive_demuxer_data_index() : current_demuxer_data_index_;
+}
+
 bool MediaDecoderJob::NoAccessUnitsRemainingInChunk(
     bool is_active_chunk) const {
   DCHECK(ui_task_runner_->BelongsToCurrentThread());