Fixed a lot threading issues during tear down of <video>

Fixed a lot of dead locks during tead down of <video> due
to DataSourceImpl.
Most of the issues come from that during a tab close
RenderThread is destroyed and new tasks posted on it
will not executed, but DataSourceImpl is waiting for
those tasks to finish to complete stopping. Another
dead lock comes from that when RenderThread is
destroyed the owner loop of it (a IO Message Loop) is
being destroyed too and DataSourceImpl shouldn't post
tasks to that message loop when stopping.

Review URL: http://codereview.chromium.org/42675

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@12720 0039d316-1c4b-4281-b951-d872f2087c98

2 files changed, 4 insertions, 5 deletions

diff --git a/media/base/filter_host_impl.cc b/media/base/filter_host_impl.cc
index 74eaa16..cba2eb1 100644
--- a/media/base/filter_host_impl.cc
+++ b/media/base/filter_host_impl.cc
@@ -61,7 +61,6 @@ void FilterHostImpl::InitializationComplete() {
 }
 
 void FilterHostImpl::PostTask(Task* task) {
-  DCHECK(!stopped_);
   if (stopped_) {
     delete task;
   } else {
diff --git a/media/base/pipeline_impl.cc b/media/base/pipeline_impl.cc
index a36da6c..5bc3faa 100644
--- a/media/base/pipeline_impl.cc
+++ b/media/base/pipeline_impl.cc
@@ -401,10 +401,10 @@ void PipelineThread::StopTask() {
   if (PipelineOk()) {
     pipeline_->error_ = PIPELINE_STOPPING;
   }
-  FilterHostVector::reverse_iterator riter = filter_hosts_.rbegin();
-  while (riter != filter_hosts_.rend()) {
-    (*riter)->Stop();
-    ++riter;
+  FilterHostVector::iterator iter = filter_hosts_.begin();
+  while (iter != filter_hosts_.end()) {
+    (*iter)->Stop();
+    ++iter;
   }
   if (host_initializing_) {
     host_initializing_ = NULL;