Fixes rounding error when calculating MOVE event

components/exo/pointer.cc

Index: components/exo/pointer.cc
diff --git a/components/exo/pointer.cc b/components/exo/pointer.cc
index 073c750192f374b967e3c25061542d8f88136460..6d6db757793e3f96485ec00e4300076c57c7edc1 100644
--- a/components/exo/pointer.cc
+++ b/components/exo/pointer.cc
@@ -41,7 +41,16 @@ bool SameLocation(const ui::LocatedEvent* event, const gfx::PointF& location) {
   if (event->flags() & ui::EF_IS_SYNTHESIZED)
     return event->location() == gfx::ToFlooredPoint(location);
 
-  return event->location_f() == location;
+  // In general, it is good practice to compare floats using an Epsilon.
+  // In particular, a mouse location_f() could differ between the MOUSE_PRESSED
+  // and MOUSE_RELEASED events. At MOUSE_RELEASED, it will have a targeter()
+  // already cached, while at MOUSE_PRESSED, it will have to calculate it
+  // passing through all  windows, and that could generate rounding error.
+  // std::numeric_limits<float>::epsilon() is not big enough to catch the
+  // rounding error generated by protation.
+  static const float epsilon_scale = 0.0005f;

[Luis Héctor Chávez, 2017/05/17 23:57:11]

nit: constants are usually declared at the very top of the file and with the
convention kMyConstantName (see kLargeCursorScale in L35).

Also, I wonder if it's worth exposing
https://cs.chromium.org/chromium/src/ui/gfx/geometry/quad_f.cc?type=cs&q=eps+...
and addint the epsilon value as template parameter.

[reveman, 2017/05/18 04:21:10]

Yes, kLocatedEventEpsilon above is better. Also, maybe write this as 1.0f /
2000.0f;
Sg, keeping in this file is fine with me as well.

+  return std::abs(event->location_f().x() - location.x()) <= epsilon_scale &&
+         std::abs(event->location_f().y() - location.y()) <= epsilon_scale;
 }
 
 }  // namespace