UpdateAllViews crash Problem

extern CMainApp theApp;

UINT __cdecl MyThread( LPVOID pParam );



the thread is started with AfxBeginThread(......)





in the thread functions


UINT __cdecl MyThread( LPVOID pParam )
{


 while(done == FALSE)
 {
        result = WaitForMultipleObjects(numHandles, pWaitHandles, FALSE,
           THREAD_WAIT_TIMEOUT);        //used to be INFINITE
 
       handleIndex = result - WAIT_OBJECT_0;

      switch(handleIndex)
        {
            case 0:
            //Happens every 1 second
                bResult = theApp.PostThreadMessage(WM_UPDATE_VIEWSETA, 0, 0);
                break;
            case 1:
            //Happens every 2 Seconds
                bResult = theApp.PostThreadMessage(WM_UPDATE_VIEWSETB, 0, 0);
            break;
            //Happens every 10 milliSeconds
            case 2:
                bResult = theApp.PostThreadMessage(WM_UPDATE_VIEWSETC, 0, 0);
                break;
              :
            :
      }
}


On recieving each of these messages UpdateAllViews is called with different hints hence OnUpdate for all views are called and will update if matching hint is recieved.
I am not passing any window handles across threads.


ViewA1::OnUpdate()
{
}
ViewA2::OnUpdate()
{
}
ViewA3::OnUpdate()
{
}


ViewB1::OnUpdate()
{
}
ViewB2::OnUpdate()
{
}
ViewB3::OnUpdate()
{
}


ViewC1::OnUpdate()
{
}
ViewC2::OnUpdate()
{
}
ViewC3::OnUpdate()
{
}






My Problem is

When the messages WM_UPDATE_VIEWSETA,WM_UPDATE_VIEWSETB and WM_UPDATE_VIEWSETC are being at 1,2 secs and 10 milisecs respectively, all views update fine.  
Now if I close a open View say ViewB1, then the program crashes.

Looking at the call stack, it crashes at

BOOL CObject::IsKindOf(const CRuntimeClass* pClass) const
{
      ENSURE(this != NULL);
      // it better be in valid memory, at least for CObject size
->>      ASSERT(AfxIsValidAddress(this, sizeof(CObject)));

      // simple SI case
      CRuntimeClass* pClassThis = GetRuntimeClass();

      ENSURE(pClassThis);
      return pClassThis->IsDerivedFrom(pClass);
}

This is being called by

CView* CDocument::GetNextView(POSITION& rPosition) const
{
      ASSERT(rPosition != BEFORE_START_POSITION);
            // use CDocument::GetFirstViewPosition instead !
      if (rPosition == NULL)
            return NULL;    // nothing left
      CView* pView = (CView*)m_viewList.GetNext(rPosition);
-->      ASSERT_KINDOF(CView, pView);
      return pView;
}


And the view is the ViewB1...Which has already been closed.

I checked step executing at the destructor , the CDocument::RemoveView() is being called for the view being removed.

Closing the view interrupts UpdateAllViews at the middle of its calls, nulls the window pointers for that view. When execution resumes back it UpdateAllViews application crashes. Wierd that MFC is supposed to synchronize this operation.

How can this happen?

Thanks



LVL 3
ghimirenirajAsked:
Who is Participating?
 
DanRollinsCommented:
Without knowing the exact cause of the problem, I can describe a solution I found to a (possibly) similar problem I faced in a multi-threaded app.

Rather than do the update in the OnUpdate call, all I did was set an m_fNeedsUpdate flag in each View.  Then in the FrameWindow, I created a timer to and it would periodically check the flag of each active view and do the actual update during that timer cycle.

There are two advantages:  
1) Multiple update requests now collapse down to a single update action; this eliminates the chance that your program will do two time-consuming operations (such as db access, window redraw, etc.) when there is no need.

2) All of the update action runs on the main U/I thread.  MFC gets into strange error conditions when a worker thread does anything whatsoever with the U/I.  I understand that your PostThreadMessage() logic should avoid that, but the timer technique insulates the workers so completely that I never had a similar problem again.
0
 
DanRollinsCommented:
The posted messages are probably piled up in the queue, and one of the OnUpdate() calls is being executed during a moment in which the CView-derived object still exists, but the window it encapsulates does not.

In your WM_UPDATE_VIEWSETA handler, I suggest you verify that that all of the objects that are in your "viewset A" are still valid (and not in the process of being shut down).

BTW, a view update can take a looooong time (I'm thinking of one I wrote that involved database access).   In that case your worker thread will be thowing lots of update requests into the event queue, and they will just pile up and your program might never catch up.
0
 
ghimirenirajAuthor Commented:
Thanks Dan for your suggestions. Sorry I was out for few days and could not respond immediately.

>>>The posted messages are probably piled up in the queue, and one of the OnUpdate() calls is being executed during a moment in which the CView-derived object still exists, but the window it encapsulates does not.

How would I prevent this?


>>>>In your WM_UPDATE_VIEWSETA handler, I suggest you verify that that all of the objects that are in your "viewset A" are still valid (and not in the process of being shut down).

I wrote my own version of UpdateAllViews and I iterate through all view and check if the handles are null, (it could be bad pointers instead of nulls too) but this does not solve the problem since the window could even be destroyed right after it passes the invalid handle check, and before it makes the actual update call. SO I would think checking for valid window object didnot solve the problem although I see enough evidence that the crash happens due to bad handle to the window object.

>>>>>BTW, a view update can take a looooong time (I'm thinking of one I wrote that involved database access).   In that case your worker thread will be throwing lots of update requests into the event queue, and they will just pile up and your program might never catch up.

How would I prevent this?

Synchronizing the destroy and Update would not help either coz that would cause a deadlock.

It has been a hard problem to get around.

Any more ideas?









0
 
ghimirenirajAuthor Commented:
Thanks a Lot Dan...
0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.