When window.close() is called, we should just ask the compositor to exit

normally.

The old code made the mistake of attempting to shutdown the associated
pipelines itself, which caused race conditions with the constellation and
compositor, as they expect to be able to drain their message queues
before exiting.

src/components/script/script_task.rs

@@ -531,10 +531,7 @@ impl ScriptTask {
                 ReflowCompleteMsg(id, reflow_id) => self.handle_reflow_complete_msg(id, reflow_id),
                 ResizeInactiveMsg(id, new_size) => self.handle_resize_inactive_msg(id, new_size),
                 ExitPipelineMsg(id) => if self.handle_exit_pipeline_msg(id) { return false },
-                ExitWindowMsg(id) => {
-                    self.handle_exit_window_msg(id);
-                    return false
-                },
+                ExitWindowMsg(id) => self.handle_exit_window_msg(id),
                 ResizeMsg(..) => fail!("should have handled ResizeMsg already"),
             }
         }
@@ -614,9 +611,13 @@ impl ScriptTask {
         }
     }
 
-    fn handle_exit_window_msg(&mut self, id: PipelineId) {
+    /// We have gotten a window.close from script, which we pass on to the compositor.
+    /// We do not shut down the script task now, because the compositor will ask the
+    /// constellation to shut down the pipeline, which will clean everything up
+    /// normally. If we do exit, we will tear down the DOM nodes, possibly at a point
+    /// where layout is still accessing them.
+    fn handle_exit_window_msg(&mut self, _: PipelineId) {
         debug!("script task handling exit window msg");
-        self.handle_exit_pipeline_msg(id);
 
         // TODO(tkuehn): currently there is only one window,
         // so this can afford to be naive and just shut down the