Chromium plumbing to use the selection root bounds.

content/renderer/render_widget.cc

Index: content/renderer/render_widget.cc
diff --git a/content/renderer/render_widget.cc b/content/renderer/render_widget.cc
index 198d7d4cdfeef0e149fe91ac495c13cadc3f0f95..7325e41f8d93e136f13d005d3c3f9eded1621f56 100644
--- a/content/renderer/render_widget.cc
+++ b/content/renderer/render_widget.cc
@@ -1499,6 +1499,7 @@ void RenderWidget::DoDeferredUpdate() {
   // enable GPU acceleration so they need to be called before any painting
   // is done.
   UpdateTextInputType();
+  UpdateSelectionRootBounds();
   UpdateSelectionBounds();
 
   // Suppress painting if nothing is dirty.  This has to be done after updating
@@ -1885,6 +1886,7 @@ void RenderWidget::willBeginCompositorFrame() {
 #if defined(OS_ANDROID)
   UpdateTextInputState(false, true);
 #endif
+  UpdateSelectionRootBounds();
   UpdateSelectionBounds();
 }
 
@@ -2387,6 +2389,7 @@ void RenderWidget::FinishHandlingImeEvent() {
   // While handling an ime event, text input state and selection bounds updates
   // are ignored. These must explicitly be updated once finished handling the
   // ime event.
+  UpdateSelectionRootBounds();

[timvolodine, 2014/03/05 19:50:46]

Should this be android-only? Not sure if the selection issue applies to
platforms other than Android.

[mlamouri (slow - plz ping), 2014/03/10 15:56:45]

I was considering making it Android only but I then decided to keep it as
::UpdateSelectionBounds. I do not know what are the consumers of
::UpdateSelectionBounds so maybe that analogy was a bad idea? Also, it's not
clear to me how sending those IPC could have a bad performance impact. I would
love to hear more about this.

   UpdateSelectionBounds();
 #if defined(OS_ANDROID)
   UpdateTextInputState(false, false);
@@ -2506,6 +2509,25 @@ void RenderWidget::UpdateSelectionBounds() {
 #endif
 }
 
+void RenderWidget::GetSelectionRootBounds(gfx::Rect* bounds) {
+  WebRect bounds_webrect;
+  webwidget_->selectionRootBounds(bounds_webrect);
+  *bounds = bounds_webrect;
+}
+
+void RenderWidget::UpdateSelectionRootBounds() {

[timvolodine, 2014/03/05 19:50:46]

I think we need a better name for "UpdateSelectionRootBounds" and the
corresponding IPC. The actual use cases are bounds for clipping
insertion/selection handles. So something like UpdateVisibleBoundsForIMEHandles
seems more appropriate. wdyt?

[mlamouri (slow - plz ping), 2014/03/10 15:56:45]

I would prefer to have the name reflecting what the function actually do and it
sends selection root bounds to the browser process. I would prefer to consider
the fact that this is used by Chrome Android to know whether the selection
handlers are visible as a detail and some other consumers could use it for other
things.

+  if (!webwidget_ || handling_ime_event_) {
+    return;
+  }
+
+  gfx::Rect bounds;
+  GetSelectionRootBounds(&bounds);
+  if (selection_root_rect_ != bounds) {
+    selection_root_rect_ = bounds;
+    Send(new ViewHostMsg_SelectionRootBoundsChanged(routing_id_, bounds));
+  }
+}
+
 // Check blink::WebTextInputType and ui::TextInputType is kept in sync.
 COMPILE_ASSERT(int(blink::WebTextInputTypeNone) == \
                int(ui::TEXT_INPUT_TYPE_NONE), mismatching_enums);