google base 之MessagePumpForUI

base库中比较有意思就是这个类了，如同很多界面库一样，创建了一个隐藏窗口来处理需要在界面线程处理的消息，大体原理也就是需要执行task的时候发送一个自定义的消息，当窗口接收到task的时候调用保存起来的回调函数，还有的是通过把回调放在消息结构体里面

自下义的消息

// Message sent to get an additional time slice for pumping (processing) another
// task (a series of such messages creates a continuous task pump).
static const int kMsgHaveWork = WM_USER + 1;

值得注意的是，别自己定义个消息和这个消息重复了，所以比较好的习惯就是定义到wm_user+100，呵呵，不过如果遇上蛋筒的库估计还是会冲突。

void MessagePumpForUI::ScheduleWork() {
  if (InterlockedExchange(&have_work_, 1))
    return;  // Someone else continued the pumping.

  // Make sure the MessagePump does some work for us.
  BOOL ret = PostMessage(message_hwnd_, kMsgHaveWork,
                         reinterpret_cast<WPARAM>(this), 0);
  if (ret)
    return;  // There was room in the Window Message queue.

  // We have failed to insert a have-work message, so there is a chance that we
  // will starve tasks/timers while sitting in a nested message loop.  Nested
  // loops only look at Windows Message queues, and don't look at *our* task
  // queues, etc., so we might not get a time slice in such. :-(
  // We could abort here, but the fear is that this failure mode is plausibly
  // common (queue is full, of about 2000 messages), so we'll do a near-graceful
  // recovery.  Nested loops are pretty transient (we think), so this will
  // probably be recoverable.
  InterlockedExchange(&have_work_, 0);  // Clarify that we didn't really insert.
  UMA_HISTOGRAM_ENUMERATION("Chrome.MessageLoopProblem", MESSAGE_POST_ERROR,
                            MESSAGE_LOOP_PROBLEM_MAX);
}

跟其它的不同，这里的ScheduleWork只不过是发送了一个自定义的消息，然后接收么这个消息的时候处理回调以达到deal task的目的。具体看下面代码：

// static
LRESULT CALLBACK MessagePumpForUI::WndProcThunk(
    HWND hwnd, UINT message, WPARAM wparam, LPARAM lparam) {
  switch (message) {
    case kMsgHaveWork:
      reinterpret_cast<MessagePumpForUI*>(wparam)->HandleWorkMessage();
      break;
    case WM_TIMER:
      reinterpret_cast<MessagePumpForUI*>(wparam)->HandleTimerMessage();
      break;
  }
  return DefWindowProc(hwnd, message, wparam, lparam);
}

红色部份，传入的参数进行处理：

void MessagePumpForUI::HandleWorkMessage() {
  // If we are being called outside of the context of Run, then don't try to do
  // any work.  This could correspond to a MessageBox call or something of that
  // sort.
  if (!state_) {
    // Since we handled a kMsgHaveWork message, we must still update this flag.
    InterlockedExchange(&have_work_, 0);
    return;
  }

  // Let whatever would have run had we not been putting messages in the queue
  // run now.  This is an attempt to make our dummy message not starve other
  // messages that may be in the Windows message queue.
  ProcessPumpReplacementMessage();

  // Now give the delegate a chance to do some work.  He'll let us know if he
  // needs to do more work.
  if (state_->delegate->DoWork())
    ScheduleWork();
}

　红色部份进行了task的处理。只是这个函数现在还理解得不透彻，暂且放着

bool MessagePumpForUI::ProcessPumpReplacementMessage() {
  // When we encounter a kMsgHaveWork message, this method is called to peek
  // and process a replacement message, such as a WM_PAINT or WM_TIMER.  The
  // goal is to make the kMsgHaveWork as non-intrusive as possible, even though
  // a continuous stream of such messages are posted.  This method carefully
  // peeks a message while there is no chance for a kMsgHaveWork to be pending,
  // then resets the have_work_ flag (allowing a replacement kMsgHaveWork to
  // possibly be posted), and finally dispatches that peeked replacement.  Note
  // that the re-post of kMsgHaveWork may be asynchronous to this thread!!

  bool have_message = false;
  MSG msg;
  // We should not process all window messages if we are in the context of an
  // OS modal loop, i.e. in the context of a windows API call like MessageBox.
  // This is to ensure that these messages are peeked out by the OS modal loop.
  if (MessageLoop::current()->os_modal_loop()) {
    // We only peek out WM_PAINT and WM_TIMER here for reasons mentioned above.
    have_message = PeekMessage(&msg, NULL, WM_PAINT, WM_PAINT, PM_REMOVE) ||
                   PeekMessage(&msg, NULL, WM_TIMER, WM_TIMER, PM_REMOVE);
  } else {
    have_message = PeekMessage(&msg, NULL, 0, 0, PM_REMOVE) != FALSE;
  }

  DCHECK(!have_message || kMsgHaveWork != msg.message ||
         msg.hwnd != message_hwnd_);

  // Since we discarded a kMsgHaveWork message, we must update the flag.
  int old_have_work = InterlockedExchange(&have_work_, 0);
  DCHECK(old_have_work);

  // We don't need a special time slice if we didn't have_message to process.
  if (!have_message)
    return false;

  // Guarantee we'll get another time slice in the case where we go into native
  // windows code.   This ScheduleWork() may hurt performance a tiny bit when
  // tasks appear very infrequently, but when the event queue is busy, the
  // kMsgHaveWork events get (percentage wise) rarer and rarer.
  ScheduleWork();
  return ProcessMessageHelper(msg);
}