Cap the memory usage in FFMpegDemuxer.

media/base/decoder_buffer_queue.h

Index: media/base/decoder_buffer_queue.h
diff --git a/media/base/decoder_buffer_queue.h b/media/base/decoder_buffer_queue.h
index 938db63123e5523135bfcd529907114a3799c908..ebc5ab7c7f1ef9894c1af04018f2bf6d5e3ce7de 100644
--- a/media/base/decoder_buffer_queue.h
+++ b/media/base/decoder_buffer_queue.h
@@ -51,6 +51,9 @@ class MEDIA_EXPORT DecoderBufferQueue {
   // Returns zero if the queue is empty.
   base::TimeDelta Duration();
 
+  // Returns the total size of buffers inside the queue.
+  int data_size() const { return data_size_; }

[DaleCurtis, 2014/02/19 18:59:25]

I'd just make this a size_t to minimize the CHECKs you need. The style guide now
recommends size_t for "size" types.

[damienv1, 2014/02/19 19:21:10]

I agree that size_t should have been used in a first place.
However, I used "int" for consistency with DecoderBuffer::data_size which
currently returns "int".
Does it make sense to mix both data types ?
My first feeling was to use the same type and if we want to change to the
suggested size_t type, then we would change both DecoderBuffer and
DecoderBufferQueue all together.

+
  private:
   typedef std::deque<scoped_refptr<DecoderBuffer> > Queue;
   Queue queue_;
@@ -62,6 +65,9 @@ class MEDIA_EXPORT DecoderBufferQueue {
 
   base::TimeDelta earliest_valid_timestamp_;
 
+  // Total size in bytes of buffers in the queue.
+  int data_size_;
+
   DISALLOW_COPY_AND_ASSIGN(DecoderBufferQueue);
 };