Chrome Version : beginning of time?!

The following code will never return :

  void SubPumpFunc(OnceClosure on_done) {
    MessageLoopCurrent::ScopedNestableTaskAllower allow_nestable_tasks;
    MSG msg;
    while (::GetMessage(&msg, NULL, 0, 0)) {
      ::TranslateMessage(&msg);
      ::DispatchMessage(&msg);
    }
    std::move(on_done).Run();
  }

  void Test() {
    MessageLoop message_loop(MessageLoop::TYPE_UI);

    RunLoop run_loop;

    // Enter multiple levels of nested subpumps.
    message_loop.task_runner()->PostTask(
        FROM_HERE, BindOnce(&SubPumpFunc, run_loop.QuitClosure()));
    message_loop.task_runner()->PostTask(
        FROM_HERE, BindOnce(&SubPumpFunc, DoNothing::Once()));

    // Quit the subsubpump (extra tasks ensure enough cycles to fully handle).
    message_loop.task_runner()->PostTask(FROM_HERE,
                                         BindOnce(&::PostQuitMessage, 0));
    message_loop.task_runner()->PostTask(FROM_HERE, DoNothing());
    message_loop.task_runner()->PostTask(FROM_HERE, DoNothing());


    // Quit the subpump.
    message_loop.task_runner()->PostTask(FROM_HERE,
                                         BindOnce(&::PostQuitMessage, 0));

    run_loop.Run();
  }

the reason is that the first quit eats the kMsgHaveWork message when it reposts the WM_QUIT (ref. https://chromium-review.googlesource.com/c/chromium/src/+/1240158). As such, the subpump is no longer processing application tasks when it un-nests (and hence fails to post the second quit).

This was worse before https://chromium-review.googlesource.com/c/chromium/src/+/1240158 (the first quit would be taken away from either subpump and we would stay stuck forever); now at least the quits are handled, it's just that application tasks may be starved on resume if there were multiple layers of nesting subpumps.