Screen Orientation: fire event on Window instead of Screen.

Source/modules/screen_orientation/ScreenOrientationController.cpp

Index: Source/modules/screen_orientation/ScreenOrientationController.cpp
diff --git a/Source/modules/screen_orientation/ScreenOrientationController.cpp b/Source/modules/screen_orientation/ScreenOrientationController.cpp
index 9f1686db67f09080987a4209641514ece6d509f6..0de3398ec4bf0517ff840d695ff954157ac11cd0 100644
--- a/Source/modules/screen_orientation/ScreenOrientationController.cpp
+++ b/Source/modules/screen_orientation/ScreenOrientationController.cpp
@@ -41,7 +41,7 @@ void ScreenOrientationController::dispatchOrientationChangeEvent()
     if (m_document.domWindow()
         && !m_document.activeDOMObjectsAreSuspended()
         && !m_document.activeDOMObjectsAreStopped())
-        m_document.domWindow()->screen().dispatchEvent(Event::create(EventTypeNames::orientationchange));
+        m_document.domWindow()->dispatchEvent(Event::create(EventTypeNames::orientationchange));

[Inactive, 2014/04/17 20:29:48]

Doesn't this event get fired twice now? (once here and once in
core/frame/LocalFrame.cpp).

It is sad this is not covered by layout tests yet.

[mlamouri (slow - plz ping), 2014/04/17 20:34:33]

That is correct. I will make window.orientation implementation entirely based on
screen orientation, at which point it will be fixed. We currently have two ways
to tell the renderer process that the screen orientation has changed. I will
merge those two ways to only one, passing a WebScreenInfo with the updated
values. I have changed WebScreenInfo accordingly and I have patches to populate
WebScreenInfo.

I don't think it's a big deal to have the event sent twice in ToT while this is
being merged.

[Inactive, 2014/04/17 20:41:18]

I'd like to hear Adam's opinion on this. I don't really like that we fire the
event twice during the transition period. Also, having window.orientation depend
on ScreenOrientation might be a bit tricky as ScreenOrientation is behind a
RuntimeEnabled flag (i.e. we'll need to refactoring to make sure the event gets
fired on Window even if the RuntimeEnabled is disabled.

Finally, I was really hoping we would have layout test coverage for screen
orientation events before we change this part of the code.

[abarth-chromium, 2014/04/17 23:36:47]

I think it's fine given that this code is still in the experimental bucket.  We
should make sure we fire it only once before turning the flag on by default,
however.
Yeah, that does seem a bit tricky.  Hopefully we can manager it.
Can we add those tests in this CL?

[mlamouri (slow - plz ping), 2014/04/18 16:42:17]

That will be the case.
Should be fine. ScreenOrientation flag will be enabled on all platforms while
the current window.orientation/window.onorientationchange is only enabled on
Android. onorientationchange will be enabled with ScreenOrientation and
window.orientation will stay behind the other flag. Hopefully, we will be able
to get ride of it at some point too.
I wrote a CL to improve the infra to write layouttests. I can add tests checking
that the event is sent.

 }
 
 const char* ScreenOrientationController::supplementName()