Prevent SourceBufferRange.waiting_for_keyframe_ from being lost on range merges.

media/filters/source_buffer_stream.cc

Index: media/filters/source_buffer_stream.cc
diff --git a/media/filters/source_buffer_stream.cc b/media/filters/source_buffer_stream.cc
index 3f1b42d82662bc262607896ee3c1daa15e58e157..f50a397aa0be608704a910b536414967af589ab6 100644
--- a/media/filters/source_buffer_stream.cc
+++ b/media/filters/source_buffer_stream.cc
@@ -1498,8 +1498,13 @@ void SourceBufferRange::AppendRangeToEnd(const SourceBufferRange& range,
   DCHECK(CanAppendRangeToEnd(range));
   DCHECK(!buffers_.empty());
 
-  if (transfer_current_position)
-    next_buffer_index_ = range.next_buffer_index_ + buffers_.size();
+  if (transfer_current_position) {
+    if (range.next_buffer_index_ >= 0)

[scherkus (not reviewing), 2013/01/07 23:34:06]

aside from not updating w_f_k_/n_k_t_, was there also a bug with
next_buffer_index_ potentially being -1? do we need a test for that too or
should it be a CHECK()?

[acolwell GONE FROM CHROMIUM, 2013/01/08 00:46:39]

The test exposes both problems. When waiting_for_keyframe_ is true the index was
negative. That is part of the reason why buffers were being returned out of
order when ranges were merged.

+      next_buffer_index_ = range.next_buffer_index_ + buffers_.size();
+
+    waiting_for_keyframe_ = range.waiting_for_keyframe_;
+    next_keyframe_timestamp_ = range.next_keyframe_timestamp_;
+  }
 
   AppendBuffersToEnd(range.buffers_);
 }