C / C++ / MFC

4-Jul-13 21:32

I am creating a modeless dialog from an ActiveX which loads in the Internet Explorer 9/Windows 7 using following code:

Where CMyActiveXCtrl class is of COleControl type.

C++

void CMyActiveXCtrl::OnBnClickedTest()
{
   m_MainDialog.bClosed = false;
   m_MainDialog.ShowWindow( SW_SHOW );
   
   MSG Msg;

   BOOL bRet;

   while (bRet = GetMessage(&Msg, NULL, 0, 0))
   {
         if (bRet == -1)
         {
            MessageBoxA(this->m_hWnd, "Message loop error", "My ActiveX", MB_OK);
            break;
         }

         if (PreTranslateMessage( &Msg ) == 0 )
         {
              if (!::IsDialogMessage(m_MainDialog.m_hWnd, &Msg)) 
              { 
                  TranslateMessage( &Msg );
                  DispatchMessage(&Msg);
              }
         }

         if (m_MainDialog.bClosed == true)
         {
             break;
         }
   }// while
}

m_MainDialog.bClosed set true on modeless dialog close/destroy. Now when my modeless dialog is opened in ActiveX and the user closes the browser window using IE > File > Exit or clicks on top right "X" button, IE main windows closes (no longer visible) but it leaves an orphan IE process in the task manager.

The same code working fine on Windows XP/ IE 8. Can anyone suggest me what could be the problem with above message loop in OnBnClickedTest() else without that message loop, it works fine.

Manish Agarwal
manish.k.agarwal @ gmail DOT com

modified 5-Jul-13 8:03am.

pasztorpisti5-Jul-13 14:50

5-Jul-13 14:50

Here is my guess:
Message order can be different with different window managers. Even on the same windows installation the window manager and the message order can be different if you change between aero/oldschool/whatever look.
One possible reason can be the following: when you press X the window receives WM_SYSCOMMAND/SC_CLOSE that triggers a WM_CLOSE that triggers WM_DESTROY that finally executes a PostQuitMessage() in a usual windows program. Pressing the Exit menuitem usually skips the first step of the chain and sends WM_CLOSE directly to the window. I think here the problem can be that your loop consumes the WM_QUIT message placed by PostQuitMessage() to the message queue (when your GetMessage call returns FALSE) so when you return from your loop the message loop of IE doesn't have any quit message to process. I'm not exactly sure about this (haven't tried it) because WM_QUIT isn't really a message in the message queue of your thread, its rather a bit that tells GetMessage()/PeekMessage() to return WM_QUIT when there are no more messages left in the queue, maybe this flag remains set but who knows... If the flag doesn't remain set (that can be windows version specific) then the problem can be what I described. When your main loop ends (GetMessage returns false) try to repost the quit message by calling PostQuitMessage((int)Msg.wParam) a the end of your OnBnClickedTest() function to place a WM_QUIT message (flag) in the queue for the main loop of IE. Try it and good luck with this.

Of course this disaster can happen only if the WM_QUIT is consumed and if WM_QUIT comes before the bClosed flag of your dialog becomes true. You can check that by placing a log into your bClosed checking if block and after the end of your loop.

EDIT: Another even better solution if you know the HWND of the main window: after checking the m_MainDialog.bClosed flag of your dialog put in another if block that checks whether this message is a WM_CLOSE message sent to the main window and in that case break out of your loop. An even better version of this: in your loop you should call PeekMessage() before every GetMessage() call just to look ahead in the queue (without removing the message) and if its a suspicious message sent to the main window (WM_SYSCOMMAND/SC_CLOSE or WM_CLOSE) then you break out your loop immediately without processing the message and you let the message loop of IE to process it. What is nice about this solution is that it would work even if the main loop of IE wanted to act on the WM_SYSCOMMAND/SC_CLOSE or WM_CLOSE message of the main window - putting such handlers into the main loop is not a good practice (you should handle that in the windowproc of the specified hwnd) but maybe MS guys committed that hack...

briefly:

C++

for (;;)
{
   WaitMessage();
   PeekMessage();
   if (WM_SYSCOMMAND/SC_CLOSE or WM_CLOSE sent to the main window)
      break;

   BOOL bRet = !GetMessage();
   if (!bRet)
      break;

   // original code from your while loop
}

EDIT: of course you know the HWND of the main window because you just have to call GetParent() on the HWND of the activex control or its window in a loop (or recursively) until it returns NULL HWND and the HWND before the NULL result is the top level window.

modified 7-Jul-13 21:13pm.

Manish K. Agarwal5-Jul-13 18:42

5-Jul-13 18:42

Thanks a lot.

PeekMessage approach is working with a small correction if (!bRet) to if (bRet) i.e. removed NOT operator.

Manish Agarwal
manish.k.agarwal @ gmail DOT com

pasztorpisti6-Jul-13 1:55

6-Jul-13 1:55

You are welcome! I haven't put in any checks on my PeekMessage() so you couldn't negate it! Smile | :)

Both peekmessage and getmessage are needed, with careful parametrization of PeekMessage(). Of course the PeekMessage doc says that it returns nozero if there is a message available so you have to check for (bRet). Can you please post the resulting code? We may be able to spot other errors and it can help other people in the future.

pasztorpisti6-Jul-13 2:08

6-Jul-13 2:08

I have already found a bug in my solution: If there is no message in the queue when PeekMessage() is executed then control is immediately transferred to the GetMessage() call so the next incoming message will be removed for sure by your loop no matter what the message is. To avoid this you should probably call WaitMessage() before PeekMessage().

Manish K. Agarwal7-Jul-13 21:17

7-Jul-13 21:17

My final code looks like as below and it does not require handling of WM_SYSCOMMAND/SC_CLOSE or any other message as my test shows that we always get WM_CLOSE irrespective of how we are closing it.

C++

for (;;)
{
   WaitMessage();
   PeekMessage(&Msg, NULL, 0, 0, PM_NOREMOVE);

   if (Msg.message == WM_CLOSE)
   {
      // Do not remove the WM_CLOSE from message Queue and leave it for IE.
      break;
   }

   BOOL bRet = GetMessage(&Msg, NULL, 0, 0);
   if (!bRet)
      break;

   // original code from your while loop
   if (PreTranslateMessage( &Msg ) == 0 )
   {
        if (m_MainDialog.m_hWnd == (HWND) NULL || !::IsDialogMessage(m_MainDialog.m_hWnd, &Msg))
        {
            TranslateMessage( &Msg );
            DispatchMessage(&Msg);
        }
   }

   if (m_MainDialog.bClosed == true)
   {
       break;
   }
}

Manish Agarwal
manish.k.agarwal @ gmail DOT com

pasztorpisti7-Jul-13 22:11

7-Jul-13 22:11

Thank you for posting, WM_CLOSE should be okay. You don't check the return value of PeekMessage (that should always be zero but its better to make sure...) and you should check Msg.hwnd whether its the main window hwnd or something else (it could be one of your dialog hwnds or any other HWND inside the IE process). I would write "if (PeekMessage()) { check for WM_CLOSE }". You can get the main window hwnd by performing the following operation on one of your activex hwnds that are placed directly/indirectly on your main window:

C++

HWND GetRootWindow(HWND hwnd)
{
    for (;;)
    {
        HWND parent = GetParent(hwnd);
        if (!parent)
            return hwnd;
        hwnd = parent;
    }
}

WM_SYSCOMMAND/SC_CLOSE is gui specific (it indicates that someone pressed X on the window), noone should put critical handling on this message as your window can easily get simulated WM_CLOSE messages without gui interaction/WM_SYSCOMMAND/SC_CLOSE.

Manish K. Agarwal7-Jul-13 22:53

7-Jul-13 22:53

Thanks again for your detailed reply. I will take care of this.

Manish Agarwal
manish.k.agarwal @ gmail DOT com

pasztorpisti8-Jul-13 2:24

8-Jul-13 2:24

You are welcome! Ignore my previous post, I posted something here and the deleted it later but you probably received a mail notification! Smile | :)

Erudite_Eric7-Jul-13 23:26

Erudite_Eric

7-Jul-13 23:26

There is one message pump and one message handler per process, why would you want to duplicate the message pump of IE in a dialog box onclick handler?

Its uterly stupid.

pasztorpisti8-Jul-13 8:40

8-Jul-13 8:40

For example because he may want dialogbox features on his dialog using IsDialogMessage() that wouldn't be done by the main loop of IE. Such dialogbox features include for example hotkeys like closing the dialog with ENTER/ESC, traversing the controls with TAB. Of course this own loop is a bit hacky but still can be a good balance between invested time and functionality. Sometimes its not you writing the code - it can be heritage from another guy who worked for the company years ago and your boss wont give you time to put together a better solution especially if the current one works in a more or less satisfactory way. The world is not perfect, and one day you will have to realize this.

Erudite_Eric9-Jul-13 0:20

Erudite_Eric

9-Jul-13 0:20

You never heard of windows hooks then?

Erudite_Eric10-Jul-13 2:14

Erudite_Eric

10-Jul-13 2:14

You still havent answered. Just because a dialog is interested in some messages and register for them why should it duplicate the message pump?

Thwere is one message pumop, one queue, one handler per process. Period.

pasztorpisti10-Jul-13 2:25