Chrome requires WebTouchPoint to store WebFloatPoint, instead of WebPoint.

content/renderer/input/input_handler_proxy.cc

Index: content/renderer/input/input_handler_proxy.cc
diff --git a/content/renderer/input/input_handler_proxy.cc b/content/renderer/input/input_handler_proxy.cc
index 95809a857511b889237d2e7337be62ace6c0be43..13c985091c5e02560ff7000308d33d16b619724a 100644
--- a/content/renderer/input/input_handler_proxy.cc
+++ b/content/renderer/input/input_handler_proxy.cc
@@ -12,6 +12,7 @@
 #include "third_party/WebKit/public/web/WebInputEvent.h"
 #include "ui/events/latency_info.h"
 #include "ui/gfx/frame_time.h"
+#include "ui/gfx/geometry/point_conversions.h"
 
 using blink::WebFloatPoint;
 using blink::WebFloatSize;
@@ -228,8 +229,7 @@ InputHandlerProxy::EventDisposition InputHandlerProxy::HandleInputEvent(
       if (touch_event.touches[i].state != WebTouchPoint::StatePressed)
         continue;
       if (input_handler_->HaveTouchEventHandlersAt(
-              blink::WebPoint(touch_event.touches[i].position.x,
-                              touch_event.touches[i].position.y))) {
+              gfx::ToFlooredPoint(touch_event.touches[i].position))) {

[jdduke (slow), 2014/03/10 22:03:23]

What about changing InputHandler::HaveTouchEventHandlersAt() to take a
gfx::PointF? Or would that be a follow-up? Looking at LayerTreeHostImpl, it
should be able to operate on a gfx::PointF without any meaningful change other
than the method signature.

[tdresser, 2014/03/12 13:54:40]

I did that in patch set #4. It felt like it involved changing a bunch of code
with no clear benefit. Why do you feel it would be preferable? Just to reduce
the number of ToFlooredPoint calls?

         return DID_NOT_HANDLE;
       }
     }