Browser Side Text Input State Tracking for OOPIF.

content/browser/frame_host/render_widget_host_view_child_frame.cc

Index: content/browser/frame_host/render_widget_host_view_child_frame.cc
diff --git a/content/browser/frame_host/render_widget_host_view_child_frame.cc b/content/browser/frame_host/render_widget_host_view_child_frame.cc
index c27af7e68bc4d35ae014a653597a531ed0f451cb..3b44173619bfeb6d597c72e9cf7050e3c76a1301 100644
--- a/content/browser/frame_host/render_widget_host_view_child_frame.cc
+++ b/content/browser/frame_host/render_widget_host_view_child_frame.cc
@@ -196,11 +196,6 @@ void RenderWidgetHostViewChildFrame::SetIsLoading(bool is_loading) {
   NOTREACHED();
 }
 
-void RenderWidgetHostViewChildFrame::TextInputStateChanged(
-    const ViewHostMsg_TextInputState_Params& params) {
-  // TODO(kenrb): Implement.
-}
-
 void RenderWidgetHostViewChildFrame::RenderProcessGone(
     base::TerminationStatus status,
     int error_code) {
@@ -223,6 +218,11 @@ void RenderWidgetHostViewChildFrame::Destroy() {
   // RenderWidgetHostInputEventRouter afterwards.
   NotifyObserversAboutShutdown();
 
+  // If this view has an active text input, the RenderWidgetHostDelegate should
+  // be notified.
+  if (text_input_state()->type != ui::TEXT_INPUT_TYPE_NONE)
+    TextInputStateChanged(TextInputState());

[Charlie Reis, 2016/03/15 18:32:00]

Why are we passing in an empty one rather than text_input_state()?

[EhsanK, 2016/03/15 23:51:17]

This line will reset the WebContents knowledge of text input state, if it had
come from this RWHV. The WebContents currently holds a pointer the RWHV as well
as the last reported value of TextInputState. If this RWHV goes out of
existence, those information will be invalid and it will cause errors in IME
workings (also cursor stuff).

This line will hit when the renderer crashes.

However, as I see it now, it might be best to separate this task into another
method. I have done that in the new patch. Please let me know if it makes sense.

This new method is called NotifyHostDelegateAboutShutdown() and should be called
before host_->SetView(NULL). I think this method should also be called in
top-level RWHV, otherwise, the same issue might arise in cases with nested
WebContents.

[Charlie Reis, 2016/03/16 16:55:13]

I see-- I didn't notice that this was during Destroy before.  Yes, the new
approach makes sense.

[EhsanK, 2016/03/30 20:46:03]

Done.

+
   host_->SetView(NULL);
   host_ = NULL;
   base::MessageLoop::current()->DeleteSoon(FROM_HERE, this);