Send synthetic GestureTapDown to focus BrowserPlugin

content/browser/frame_host/render_widget_host_view_guest.cc

Index: content/browser/frame_host/render_widget_host_view_guest.cc
diff --git a/content/browser/frame_host/render_widget_host_view_guest.cc b/content/browser/frame_host/render_widget_host_view_guest.cc
index 798fe2d2761418a7bffc574c5eab23bee7b21b1d..9c941cdbd708ef68bccec20e5ee571e3007c0502 100644
--- a/content/browser/frame_host/render_widget_host_view_guest.cc
+++ b/content/browser/frame_host/render_widget_host_view_guest.cc
@@ -175,6 +175,32 @@ void RenderWidgetHostViewGuest::ProcessTouchEvent(
         guest_->GetOwnerRenderWidgetHostView()->GetRenderWidgetHost());
     if (!embedder->GetView()->HasFocus())
       embedder->GetView()->Focus();
+
+    // Since we now route GestureEvents directly to the guest renderer, we need
+    // a way to make sure that the BrowserPlugin in the embedder gets focused so
+    // that keyboard input (which still travels via BrowserPlugin) is routed to
+    // the plugin and thus onwards to the guest.
+    // TODO(wjmaclean): When we remove BrowserPlugin, delete this code.
+    // http://crbug.com/533069
+    if (!HasFocus()) {
+      auto offset_in_owner = GetViewBounds().origin();
+      auto owner_offset =
+          GetOwnerRenderWidgetHostView()->GetBoundsInRootWindow();

[tdresser, 2016/06/30 13:20:57]

I think using an explicit type here would be better.

Maybe just:
gfx::Vector2D offset = GetViewBounds().origin() -
GetOwnerRenderWidgetHostView()->GetBoundsInRootWindow();

You might want to add a comment describing why the offset is needed, or pick a
more descriptive name.

[wjmaclean, 2016/06/30 20:33:47]

Done.

+      blink::WebGestureEvent gesture_tap_event;
+      gesture_tap_event.sourceDevice = blink::WebGestureDeviceTouchscreen;
+      gesture_tap_event.type = blink::WebGestureEvent::GestureTapDown;
+      gesture_tap_event.x =
+          event.touches[0].position.x + offset_in_owner.x() - owner_offset.x();
+      gesture_tap_event.y =
+          event.touches[0].position.y + offset_in_owner.y() - owner_offset.y();

[tdresser, 2016/06/30 13:20:57]

Can you do this math directly on the Point/Vector objects, instead of handling
x/y separately? (If it isn't cleaner, don't bother)

[wjmaclean, 2016/06/30 20:33:47]

WebFloatPoint supports conversion to gfx::PointF but that won't add with
gfx::Vector2d, so we need to deal with individual coordinates somewhere along
the way. I've simplified it somewhat.

+      gesture_tap_event.globalX = event.touches[0].screenPosition.x;
+      gesture_tap_event.globalY = event.touches[0].screenPosition.y;
+      GetOwnerRenderWidgetHostView()->ProcessGestureEvent(gesture_tap_event,
+                                                          ui::LatencyInfo());
+      gesture_tap_event.type = blink::WebGestureEvent::GestureTapCancel;
+      GetOwnerRenderWidgetHostView()->ProcessGestureEvent(gesture_tap_event,
+                                                          ui::LatencyInfo());

[tdresser, 2016/06/30 13:20:57]

We should send a tap cancel here, to ensure the event stream is valid. This
seems correct.

[wjmaclean, 2016/06/30 20:33:47]

Acknowledged.

+    }
   }
 
   host_->ForwardTouchEventWithLatencyInfo(event, latency);