Message box as thread to stop main thread

[Yasir_Malik]

Hello,

My main thread does some computation indefinetly. To stop the computation, I created a thread that displays a message box, and when pressed, sets a global variable that stops the main program loop. However, my program just hangs when I get the start the thread. The message box does not show up either. Here's the code:

static bool keep_sending;

static UINT thread_func(LPVOID pParam)
{
  AfxMessageBox("Please OK to stop");
  keep_sending = false
  
  return 0;
}

void A::func(void)
{
  CWinThread *pthread;

  keep_sending = false;

  if(some_condition)
  {
    keep_sending = true;
    pthread = AfxBeginThread(thread_func, NULL, THREAD_PRIORITY_NORMAL);
  }
  
  do
  {
    // do some stuff
  }while(keep_sending);
}

Thanks.

[anonytmouse]

AfxMessageBox needs the main GUI thread to be pumping messages. You should reverse your scheme. Launch a worker thread to do the computations and run the message box in the main thread.

[IfYouSaySo]

maybe also declare your global bool to be volatile, this may be your problem.

[Yasir_Malik]

AfxMessageBox needs the main GUI thread to be pumping messages. You should reverse your scheme. Launch a worker thread to do the computations and run the message box in the main thread.

I tried doing that, but since the function has to be a non member function, I can't access any of the class members. I'll make the function friend function and see if that'll help.

maybe also declare your global bool to be volatile, this may be your problem.

How would turning off the caching help?

[IfYouSaySo]

My understanding is that is what volatile is for, by definition. You are checking the value of keep_sending in one thread, and updating it in another. The problem might be that the thread in the while loop doesn't realize the value has changed, precisely because the value changed in a different thread. Anyway, it's really easy to try, so why not test the theory?

[okinrus]

My understanding is that is what volatile is for, by definition. You are checking the value of keep_sending in one thread, and updating it in another. The problem might be that the thread in the while loop doesn't realize the value has changed, precisely because the value changed in a different thread. Anyway, it's really easy to try, so why not test the theory?

if the variable is cached in a register, then one thread could write to the register(but not the same variable in static memory), context switch, and then the other thread could read from the variable. But instead of reading the new value, the other thread would read the old value. .

Yasir_Malik, your gui thread must spawn a thread and then show the message box. (You'd want to avoid the situation of spawning the thread, have the thread exit, and then show the message box.) Then the worker thread continues processing, but when the user pressses a button on the message box, the static variable is set. This occurs in the gui thread. (The gui thread will probably either poll or wait for the worker thread to exit. ) Then, the worker thread eventually read the variable and knows to exit.

[Yasir_Malik]

Thanks. If by "gui thread" you mean the main thread, that's exactly what I did. The gotcha was that the thread function has to be a global function or static function member function, which in either case does not allow access to the member variables. A work around at http://www.codeproject.com/threads/memberthread.asp fixed that.
I don't think making keep_sending volatile would make a difference (even though I added it anyway) in my code because the worker thread is not changing keep_sending. Instead, it's just reading the value that the main thread sets after the message box is clicked.

[okinrus]

I don't think making keep_sending volatile would make a difference (even though I added it anyway) in my code because the worker thread is not changing keep_sending. Instead, it's just reading the value that the main thread sets after the message box is clicked.

The worker thread might be reading a register that is caching the value stored statically. Say, for instance, I wrote

static int staticVar = 1;

void f()
{
       while(staticVar <= 100)
              staticVar++;
}

This isn't how the code looks compiled. I doubt a good compiler would access memory more than twice, but instead do something like

static int staticVar = 1;

void f()
{
       reg = staticVar;

       while(reg <= 100)
              reg++;

       staticVar = reg;
}

reg representing any register, This then runs faster as the time to access a register is faster than to access memory. But say while function f is executing staticVar changes, due to a another thread accessing staticVar, then function f has no way of knowing it's supposed to change reg. Without volatile you could run into a situation where the main thread sets keep_sending to false but the worker never reads from the variable in memory, but only in the register, resulting in the worker thread never stopping.

[Yasir_Malik]

I doubt a good compiler would access memory more than twice

That explains everything. Thanks.