[Device Service] Eliminate content::ScreenOrientationObserver.

third_party/WebKit/Source/modules/screen_orientation/ScreenOrientationDispatcher.h

Index: third_party/WebKit/Source/modules/screen_orientation/ScreenOrientationDispatcher.h
diff --git a/third_party/WebKit/Source/modules/screen_orientation/ScreenOrientationDispatcher.h b/third_party/WebKit/Source/modules/screen_orientation/ScreenOrientationDispatcher.h
index 4fcd8704a333184ec9b5ff72b9d302fadae37023..d2cbf39e288b8589b083a734234f5d163d17ec4a 100644
--- a/third_party/WebKit/Source/modules/screen_orientation/ScreenOrientationDispatcher.h
+++ b/third_party/WebKit/Source/modules/screen_orientation/ScreenOrientationDispatcher.h
@@ -6,6 +6,7 @@
 #define ScreenOrientationDispatcher_h
 
 #include "core/frame/PlatformEventDispatcher.h"
+#include "device/screen_orientation/public/interfaces/screen_orientation.mojom-blink.h"
 #include "platform/heap/Handle.h"
 
 namespace blink {
@@ -19,13 +20,16 @@ namespace blink {
 // ScreenOrientationDispatcher is listening, that means that the platform should
 // be polling if required.
 class ScreenOrientationDispatcher final
-    : public GarbageCollected<ScreenOrientationDispatcher>,
+    : public GarbageCollectedFinalized<ScreenOrientationDispatcher>,

[mlamouri (slow - plz ping), 2017/03/16 12:11:47]

Is it worth making the object Finalized just for a DCHECK()? Finalized comes
with a cost.

[leonhsl(Using Gerrit), 2017/03/17 03:01:12]

Seems similar codes are already existing inside Blink codebase,, so for our
case, for now I'd like to keep this DCHECK() for safety, we could remove this
later at any time when we're sure it's unnecessary :)

[kinuko, 2017/03/17 04:38:10]

I also wonder if we want to do this.  Actually it looks this dispatcher is
DEFINE_STATIC_LOCAL'ed instance, don't we just leak it then?

[leonhsl(Using Gerrit), 2017/03/17 06:21:38]

Oh I checked around DEFINE_STATIC_LOCAL, and got something new to me:
"
// Use |DEFINE_STATIC_LOCAL()| to declare and define a static local variable
// (|static T;|) so that it is leaked and its destructors are not called at
// exit. 
"
So it means that our destructor will never get called, means that DCHECK will
never be reached. 
And, to think further, this no destruction behavior should have no bad effect to
us, even if m_listener is alive, when renderer process exits, the mojo
connection peer side can still get notified that underlying message pipe broke.

I was taking BatteryDispatcher at [1] as an example, so we could also avoid
GarbageCollectedFinalized there and maybe many other places..
[1]
https://cs.chromium.org/chromium/src/third_party/WebKit/Source/modules/batter...

       public PlatformEventDispatcher {
   USING_GARBAGE_COLLECTED_MIXIN(ScreenOrientationDispatcher);
+  WTF_MAKE_NONCOPYABLE(ScreenOrientationDispatcher);
 
  public:
   static ScreenOrientationDispatcher& instance();
 
+  ~ScreenOrientationDispatcher();
+
   DECLARE_VIRTUAL_TRACE();
 
  private:
@@ -34,6 +38,8 @@ class ScreenOrientationDispatcher final
   // Inherited from PlatformEventDispatcher.
   void startListening() override;
   void stopListening() override;
+
+  device::mojom::blink::ScreenOrientationListenerPtr m_listener;
 };
 
 }  // namespace blink