Remove scoped_refptr::release which is confusing and causes leaks.

cc/occlusion_tracker_unittest.cc

Index: cc/occlusion_tracker_unittest.cc
diff --git a/cc/occlusion_tracker_unittest.cc b/cc/occlusion_tracker_unittest.cc
index d87b82c1e164650840015c945762f834e97c0567..a73772c16be3955c6ac9aedc899ca462d6e44b3b 100644
--- a/cc/occlusion_tracker_unittest.cc
+++ b/cc/occlusion_tracker_unittest.cc
@@ -134,12 +134,16 @@ struct OcclusionTrackerTestMainThreadTypes {
 
     static LayerPtrType passLayerPtr(ContentLayerPtrType& layer)
     {
-        return layer.release();
+        LayerPtrType ref(layer);
+        layer = NULL;

[danakj, 2012/12/06 05:00:09]

The NULL assignment seems unneeded. The callsite will hold the ref and release
it without this. Can't this just "return LayerPtrType(layer);"?

[piman, 2012/12/06 05:16:00]

I wanted to keep the semantics that it cleared the passed parameter. This is
what happens on the impl thread versions as well. But you're right that the
current call sites don't rely on it. I don't feel strongly either way.

+        return ref;
     }
 
     static LayerPtrType passLayerPtr(LayerPtrType& layer)
     {
-        return layer.release();
+        LayerPtrType ref(layer);
+        layer = NULL;
+        return ref;
     }
 
     static void destroyLayer(LayerPtrType& layer)