Use a separate window to notify the user that permissions are required.

remoting/webapp/app_remoting/js/app_remoting.js

Index: remoting/webapp/app_remoting/js/app_remoting.js
diff --git a/remoting/webapp/app_remoting/js/app_remoting.js b/remoting/webapp/app_remoting/js/app_remoting.js
index 83493ebf5e370678727c054bb3d800840b2ba9c3..2eb83775e7522820da048781865fe5e18b59ff11 100644
--- a/remoting/webapp/app_remoting/js/app_remoting.js
+++ b/remoting/webapp/app_remoting/js/app_remoting.js
@@ -91,8 +91,6 @@ remoting.AppRemoting.prototype.init = function() {
   this.contextMenu_ = new remoting.ApplicationContextMenu(adapter);
   this.keyboardLayoutsMenu_ = new remoting.KeyboardLayoutsMenu(adapter);
   this.windowActivationMenu_ = new remoting.WindowActivationMenu(adapter);
-
-  remoting.LoadingWindow.show();

[Jamie, 2015/03/21 01:18:23]

Showing this dialog in init() is not correct, as we haven't started connecting
yet. It also means that we either have to close the loading window explicitly
when we show the auth window (which introduces an awkward dependency) or have
two dialogs on screen at the same time. The only situation in which there is an
appreciable lag between init() and start() is when the user needs to approve
permissions, in which case the identity API is showing its own UI. In all cases,
it's safe to defer showing the loading window until start().

[garykac, 2015/03/23 23:25:33]

Sounds good.

 };
 
 /**
@@ -105,6 +103,8 @@ remoting.AppRemoting.prototype.init = function() {
  *     start-up.
  */
 remoting.AppRemoting.prototype.start = function(connector, token) {
+  remoting.LoadingWindow.show();
+
   /** @type {remoting.AppRemoting} */
   var that = this;