Remove ui::TouchEvent -> blink::WebTouchEvent conversion methods.

ui/events/gestures/motion_event_aura.cc

Index: ui/events/gestures/motion_event_aura.cc
diff --git a/ui/events/gestures/motion_event_aura.cc b/ui/events/gestures/motion_event_aura.cc
index 11f8ba14c73a48a5d1f713f23055736cf01e29af..1164cd0af589c31f60b568ce2143092eb1555cad 100644
--- a/ui/events/gestures/motion_event_aura.cc
+++ b/ui/events/gestures/motion_event_aura.cc
@@ -69,7 +69,33 @@ MotionEventAura::MotionEventAura() {
 MotionEventAura::~MotionEventAura() {
 }
 
-void MotionEventAura::OnTouch(const TouchEvent& touch) {
+bool MotionEventAura::OnTouch(const TouchEvent& touch) {
+  int index = FindPointerIndexOfId(touch.touch_id());
+  bool pointer_id_is_active = index != -1;
+
+  if (touch.type() == ET_TOUCH_PRESSED && pointer_id_is_active) {
+    // Ignore touch press events if we already believe the pointer is down.
+
+    // FIXME(tdresser): this should return false (or NOTREACHED());

[jdduke (slow), 2015/01/05 22:35:14]

Nit: TODO is more common in chromium (vs blink).

[tdresser, 2015/01/06 14:20:16]

Done.

+    // however, there is at least one case where we need to allow a
+    // touch press from a currently used touch id. See

[jdduke (slow), 2015/01/05 22:35:14]

Will the drag/drop case only occur when the touch id is already active?

[tdresser, 2015/01/06 14:20:16]

In the drag and drop case, there are two windows involved, the source window
(from which the drag is started), and the target window (which pops up and is
dragged around).

We get:
- touch press on source
- target pops up, and we transfer the current GestureProvider to the target
window
- touch release on target
- touch press on source

The second touch press on source will fail, because the RWHVAura's
pointer_state_ received the first touch press, but not the touch release.

[jdduke (slow), 2015/01/06 16:49:54]

OK, I just want to make sure we don't get into a situation where the new touch
press has a *different* id than the current touch press (the latter never having
received a touch end), resulting in two logical pointers when we really only
have one.

[tdresser, 2015/01/07 17:01:15]

That can happen, both with this patch, and on master. It's another symptom of
crbug.com/373125.

+    // crbug.com/373125 for details.
+

[jdduke (slow), 2015/01/05 22:35:14]

No need to include the commented out code.

[tdresser, 2015/01/06 14:20:16]

Done.

+    // return false;
+  } else if (touch.type() != ET_TOUCH_PRESSED && !pointer_id_is_active) {
+    // We could have an active touch stream transfered to us, resulting in touch
+    // move or touch up events without associated touch down events. Ignore
+    // them.
+    return false;
+  }
+
+  // If this is a touchmove event, and it isn't different from the last

[jdduke (slow), 2015/01/06 16:49:54]

Nit: This comment is somewhat redundant. Or maybe say that we can safely ignore
stationary events?

[tdresser, 2015/01/07 17:01:15]

Removed comment.

+  // event, ignore it.
+  if (touch.type() == ET_TOUCH_MOVED && touch.x() == GetX(index) &&
+      touch.y() == GetY(index)) {
+    return false;
+  }
+
   switch (touch.type()) {
     case ET_TOUCH_PRESSED:
       AddTouch(touch);
@@ -86,12 +112,13 @@ void MotionEventAura::OnTouch(const TouchEvent& touch) {
       break;
     default:
       NOTREACHED();
-      break;
+      return false;
   }
 
   UpdateCachedAction(touch);
   set_flags(touch.flags());
   set_event_time(touch.time_stamp() + base::TimeTicks());
+  return true;
 }
 
 int MotionEventAura::GetId() const {