Do not filter IPC messages until the new RVH commits, to avoid a race.

content/browser/renderer_host/render_view_host.cc

Index: content/browser/renderer_host/render_view_host.cc
diff --git a/content/browser/renderer_host/render_view_host.cc b/content/browser/renderer_host/render_view_host.cc
index 921bd94fd3a8b73eb57230fe4795b34964b2de2d..4efd32d0795135638dc1c0fbe08ed0dc7e144357 100644
--- a/content/browser/renderer_host/render_view_host.cc
+++ b/content/browser/renderer_host/render_view_host.cc
@@ -333,12 +333,6 @@ void RenderViewHost::FirePageBeforeUnload(bool for_cross_site_transition) {
 
 void RenderViewHost::SwapOut(int new_render_process_host_id,
                              int new_request_id) {
-  // Start filtering IPC messages to avoid confusing the delegate.  This will
-  // prevent any dialogs from appearing during unload handlers, but we've
-  // already decided to silence them in crbug.com/68780.  We will set it back
-  // to false in SetNavigationsSuspended if we swap back in.

[battre, 2012/01/26 16:49:40]

Has the situation changed for crbug.com/68780 ("Disallow
alert/confirm/prompt/showModalDialog calls issued from within an unload
handler")? In the diagram you sent me, we would still receive IPCs between
RHV::OnSwapOut and RDH::OnSwapOutAck.

You introduced these lines in http://codereview.chromium.org/6319001 but I don't
see a specific unit test for these kind of race conditions.

[Charlie Reis, 2012/01/26 17:03:16]

That situation has not changed, but I view this as an improvement.  Before
removing this code, we couldn't allow dialogs during unload even if we wanted
to.  Now, it's a policy decision for bug 68780.  I'm glad that we still disallow
them, but now it's at least possible to change that decision.

As mentioned above, the test for this race is in my other CL for 93427.

-  is_swapped_out_ = true;
-
   // This will be set back to false in OnSwapOutACK, just before we replace
   // this RVH with the pending RVH.
   is_waiting_for_unload_ack_ = true;
@@ -371,6 +365,14 @@ void RenderViewHost::WasSwappedOut() {
   // Don't bother reporting hung state anymore.
   StopHangMonitorTimeout();
 
+  // Now that we're no longer the active RVH in the tab, start filtering out
+  // most IPC messages.  Usually the renderer will have stopped sending
+  // messages as of OnSwapOutACK.  However, we may have timed out waiting
+  // for that message, and additional IPC messages may keep streaming in.
+  // We filter them out, as long as that won't cause problems (e.g., we
+  // still allow synchronous messages through).
+  is_swapped_out_ = true;
+
   // Inform the renderer that it can exit if no one else is using it.
   Send(new ViewMsg_WasSwappedOut(routing_id()));
 }