21 replies to this topic

[Evil Steve]

I'm trying to start a timer when the user starts to drag the window, so I can render a frame on the WM_TIMER message, to prevent the stall that happens when the user drags the window. I added some logging and found that DefWindowProc() for WM_SYSCOMMAND (With a wParam of 0xf012, which doesn't seem to be a recognised SC_* value) and WM_NCLBUTTONDOWN takes a considerable time to return (Over 500ms). With some more logging, it seems that WM_NCLBUTTONDOWN comes first, and calling DefWindowProc causes a WM_SYSCOMMAND message to be sent (So it's recursing into the window proc, hence the delay on both messages). The call stack at the time if I break with F12 isn't very exciting:

ntdll.dll!_KiFastSystemCallRet@0() 	
user32.dll!_NtUserMessageCall@28()  + 0xc bytes	
user32.dll!_RealDefWindowProcWorker@20()  - 0x79d6 bytes	
user32.dll!_RealDefWindowProcW@16()  + 0x27 bytes	
uxtheme.dll!DoMsgDefault()  + 0x29 bytes	
uxtheme.dll!OnDwpSysCommand()  + 0x29 bytes	
uxtheme.dll!_ThemeDefWindowProc()  + 0x61a8 bytes	
uxtheme.dll!_ThemeDefWindowProcW@16()  + 0x18 bytes	
user32.dll!_DefWindowProcW@16()  + 0x815 bytes	
>	Tutorial02.exe!D3DWindow::WndProc(unsigned int uMsg=274, unsigned int wParam=61458, long lParam=5636612)  Line 227 + 0x1b bytes	C++
Tutorial02.exe!D3DWindow::StaticWndProc(HWND__ * hWnd=0x000a06fa, unsigned int uMsg=274, unsigned int wParam=61458, long lParam=5636612)  Line 169 + 0x1a bytes	C++
user32.dll!_InternalCallWinProc@20()  + 0x23 bytes	
user32.dll!_UserCallWinProcCheckWow@32()  - 0xddcf bytes	
user32.dll!_DispatchClientMessage@20()  + 0x4b bytes	
user32.dll!___fnDWORD@4()  + 0x24 bytes	
ntdll.dll!_KiUserCallbackDispatcher@12()  + 0x2e bytes	
user32.dll!_NtUserMessageCall@28()  + 0xc bytes	
user32.dll!_RealDefWindowProcWorker@20()  - 0x79d6 bytes	
user32.dll!_RealDefWindowProcW@16()  + 0x27 bytes	
uxtheme.dll!DoMsgDefault()  + 0x29 bytes	
uxtheme.dll!OnDwpNcLButtonDown()  + 0x32 bytes	
uxtheme.dll!_ThemeDefWindowProc()  + 0x61a8 bytes	
uxtheme.dll!_ThemeDefWindowProcW@16()  + 0x18 bytes	
user32.dll!_DefWindowProcW@16()  + 0x815 bytes	
Tutorial02.exe!D3DWindow::WndProc(unsigned int uMsg=161, unsigned int wParam=2, long lParam=5636612)  Line 227 + 0x1b bytes	C++
Tutorial02.exe!D3DWindow::StaticWndProc(HWND__ * hWnd=0x000a06fa, unsigned int uMsg=161, unsigned int wParam=2, long lParam=5636612)  Line 169 + 0x1a bytes	C++
user32.dll!_InternalCallWinProc@20()  + 0x23 bytes	
user32.dll!_UserCallWinProcCheckWow@32()  + 0xb3 bytes	
user32.dll!_DispatchMessageWorker@8()  + 0xe6 bytes	
user32.dll!_DispatchMessageW@4()  + 0xf bytes	
Tutorial02.exe!WinMain(HINSTANCE__ * hInstance=0x009f0000, HINSTANCE__ * __formal=0x00000000, HINSTANCE__ * __formal=0x00000000, HINSTANCE__ * __formal=0x00000000)  Line 22 + 0xf bytes	C++
Tutorial02.exe!__tmainCRTStartup()  Line 578 + 0x35 bytes	C
Tutorial02.exe!WinMainCRTStartup()  Line 403	C
kernel32.dll!@BaseThreadInitThunk@12()  + 0x12 bytes	
ntdll.dll!___RtlUserThreadStart@8()  + 0x27 bytes	
ntdll.dll!__RtlUserThreadStart@8()  + 0x1b bytes	

Obviously returning 0 from WM_NCLBUTTONDOWN doesn't cause WM_SYSCOMMAND to be sent, and no delay, but it also prevents the window from being moved. So, does anyone know what's going on here and if there's any way of stopping it? Cheers, Steve

[SiCrane]

How complex are your renders and what kind of delay are you putting on your timer? If you have a relatively simple render and a relatively low update frequency, then you might want to just start the timer up when the window is created and let it run.

[Evil Steve]

Quote:

Original post by SiCrane How complex are your renders and what kind of delay are you putting on your timer? If you have a relatively simple render and a relatively low update frequency, then you might want to just start the timer up when the window is created and let it run.

Currently the render is very basic (A single triangle), and the timer is set to 0ms, so just "as fast as possible".

However, this is for a tutorial / article on D3D rendering, and I'd like to do things as "correctly" as possible, even if it means leaving the 500ms lag.

Also, this is D3D (In case that's not obvious from the call stack function names [smile]), so running rendering in another thread is out of the question.

EDIT: And I'm curious about what exactly is going on under the hood here.

[Edited by - Evil Steve on January 14, 2009 3:33:36 PM]

[kittycat768]

What kind of stall are you referring to? To my knowledge the window will NEVER update while it's being dragged. Your program will keep executing though. Have you considered throwing your timer code into the main loop instead of relying on your window procedure? I recommend QueryPerformanceCounter() if you do. 64-bit calculations... Nummy. Also, what is your definition of "correctly?"

[lordikon]

There are some other things that cause about a 500ms stall, such as grabbing a window title bar. In fact, now that I think about it, grabbing a window title bar causes WM_SYSCOMMAND with a wparam value of 0xf012 which is in fact (SC_MOVE + HTCAPTION).

EDIT: You may find Spy++ will give you useful information about what messages are occuring. To me it is worlds easier than a call stack for windows message monitoring.

[Codeka]

This is a guess, but I'm thinking DefWindowProc is waiting to see whether you're about to double-click or not. Try setting your double-click delay really slow or really fast to see if that affects things.

If that's the case, then there's probably not a lot you can do about it (but maybe make a note in your tutorial or something [smile]).

[Evil Steve]

Thanks for the replies.

Quote:

Original post by kittycat768 What kind of stall are you referring to? To my knowledge the window will NEVER update while it's being dragged. Your program will keep executing though. Have you considered throwing your timer code into the main loop instead of relying on your window procedure? I recommend QueryPerformanceCounter() if you do. 64-bit calculations... Nummy. Also, what is your definition of "correctly?"

For the stall, DefWindowProc doesn't return for over 500ms. It's as if it's an extremely expensive function call. Putting a timer in the main loop wouldn't help, and for now GetTickCount() is accurate enough (You don't need to render at full speed when the user is moving the window).
My definition of "correctly" is to render in the main loop, not from a timer handler - the timer handler isn't called all that frequently as far as I'm aware.

Quote:

Original post by lordikon There are some other things that cause about a 500ms stall, such as grabbing a window title bar. In fact, now that I think about it, grabbing a window title bar causes WM_SYSCOMMAND with a wparam value of 0xf012 which is in fact (SC_MOVE + HTCAPTION). EDIT: You may find Spy++ will give you useful information about what messages are occuring. To me it is worlds easier than a call stack for windows message monitoring.

Ah, I didn't notice that the wParam can include HTCAPTION, thanks. It's the WM_SYSCOMMAND handling in DefWindowProc that causes the 500ms stall - that's what I'm trying to fix or work around.

Quote:

Original post by Codeka This is a guess, but I'm thinking DefWindowProc is waiting to see whether you're about to double-click or not. Try setting your double-click delay really slow or really fast to see if that affects things. If that's the case, then there's probably not a lot you can do about it (but maybe make a note in your tutorial or something [smile]).

I don't think that's it, I always have my double click delay set to as fast as possible, and that should be less than 500ms. I'll give that a try when I get a chance though.

Cheers,
Steve

[SiCrane]

Quote:

Original post by Evil Steve My definition of "correctly" is to render in the main loop, not from a timer handler - the timer handler isn't called all that frequently as far as I'm aware.

You could try starting a timer at window creation and calling InvalidateRect() in the timer handler, which will cause WM_PAINT messages to be generated even when your window is being dragged.

[Endurion]

Could this be because you've got a Direct3d device open inside the window?

Does that stall also happen without D3D? It might be because you're actually losing the device.

If it is because of D3D, could it be the debug runtimes?

[Evil Steve]

Quote:

Original post by Codeka This is a guess, but I'm thinking DefWindowProc is waiting to see whether you're about to double-click or not. Try setting your double-click delay really slow or really fast to see if that affects things.

Setting my double click speed to extremely slow still takes the same time for DefWindowProc to return, so that doesn't seem to be related.

Quote:

Original post by SiCrane You could try starting a timer at window creation and calling InvalidateRect() in the timer handler, which will cause WM_PAINT messages to be generated even when your window is being dragged.

Hmm, it's possible - but still seems a bit ugly to me. This is beginning to look like the only option though...

Quote:

Original post by Endurion Could this be because you've got a Direct3d device open inside the window?

Disabling D3D completely (Commenting out any reference to any D3D interface) still yields the same problem.

[Erik Rufelt]

This problem happens even if you create a timer at window-creation. I've always wondered why..
It doesn't happen if you start dragging the window right away, moving the mouse already when you click. It also pauses if you hold down the button on the minimize button for example. I think all such things are handled in some loop that doesn't return, except for that once the window starts moving it returns to allow for interactive moving.
Don't know why it waits 500 ms before returning unless the mouse moves.. the reply about double-clicking sounds plausible to me..
It also freezes if you hold down the right mouse-button in the title bar.

[lordikon]

Quote:

Original post by Evil Steve Hmm, it's possible - but still seems a bit ugly to me. This is beginning to look like the only option though...

I create the timer the first time WM_MOVING is reached, and remove the timer when WM_EXITSIZEMOVE is reached. This doesn't help with the delay though. There is about a 500ms delay between the time the user clicks the title bar and I receive SC_MOVE + HTCAPTION. During this time the window stops updating.

[Evil Steve]

Quote:

Original post by lordikon I create the timer the first time WM_MOVING is reached, and remove the timer when WM_EXITSIZEMOVE is reached. This doesn't help with the delay though. There is about a 500ms delay between the time the user clicks the title bar and I receive SC_MOVE + HTCAPTION. During this time the window stops updating.

Yep, it's that 500ms delay that I'm trying to avoid.

[lordikon]

I'm also fairly interested in this solution. I wish I could help, but after some research I found no answers either.

[lordikon]

So, it appears you've been having this problem for awhile. :D

http://www.gamedev.net/community/forums/topic.asp?topic_id=440341

[Evil Steve]

Quote:

Original post by lordikon So, it appears you've been having this problem for awhile. :D http://www.gamedev.net/community/forums/topic.asp?topic_id=440341

Yeah, I found that post while Googling, but decided not to necro the thread [smile]

[Amr0]

Here is a quick workaround. Note that this has the little quirk of setting the cursor position to the middle of the window's caption. Also, if the app has other windows and he starts dragging them, the problem remains. Also, the delay will remain if the window is sizable and the user [EDIT: referred to in the previous sentence as "he"] clicks on the borders to resize it. Also, this will effectively disable double-clicking the title bar to maximize the window. Plus possibly a few more things I haven't thought about yet.

case WM_NCLBUTTONDOWN:
		if( SendMessage( hWnd, WM_NCHITTEST, wParam, lParam ) == HTCAPTION )
		{
			SetTimer( hWnd, 1000, 0, NULL );
			SendMessage( hWnd, WM_SYSCOMMAND, SC_MOVE, 0 );
			return 0;
		}
		break;

	case WM_TIMER:
		Render();
		break;

	case WM_EXITSIZEMOVE:
		KillTimer( hWnd, 1000 );
		break;

But personally I would recommend against messing with this because, well, you're just nitpicking one of Windows' insignificant quirks. That's not patriotic!

[Edited by - hikikomori-san on January 15, 2009 10:47:56 AM]

[lordikon]

Just found this: http://developer.popcap.com/forums/showthread.php?p=19636

They mention the same delay problem.

[Evil Steve]

Quote:

Original post by lordikon Just found this: http://developer.popcap.com/forums/showthread.php?p=19636 They mention the same delay problem.

Yes, I've been through Google. I've found plenty of reports of the problems, but no solutions.

Quote:

Original post by hikikomori-san Here is a quick workaround. Note that this has the little quirk of setting the cursor position to the middle of the window's caption. Also, if the app has other windows and he starts dragging them, the problem remains. Also, the delay will remain if the window is sizable and the user [EDIT: referred to in the previous sentence as "he"] clicks on the borders to resize it. Also, this will effectively disable double-clicking the title bar to maximize the window. Plus possibly a few more things I haven't thought about yet.

Hmm, I'll give that a go later tonight, thanks.

[Evil Steve]

Quote:

Original post by hikikomori-san Here is a quick workaround. Note that this has the little quirk of setting the cursor position to the middle of the window's caption. Also, if the app has other windows and he starts dragging them, the problem remains. Also, the delay will remain if the window is sizable and the user [EDIT: referred to in the previous sentence as "he"] clicks on the borders to resize it. Also, this will effectively disable double-clicking the title bar to maximize the window. Plus possibly a few more things I haven't thought about yet.

This seems to work pretty well, with a bit of an ajustment:

LRESULT D3DWindow::WndProc(UINT uMsg, WPARAM wParam, LPARAM lParam)
{
static bool sbMoving = false;
static POINT sptOffset;

	switch(uMsg)
	{
	case WM_NCLBUTTONDOWN:
		if(SendMessage(m_hWnd, WM_NCHITTEST, wParam, lParam) == HTCAPTION)
		{
			POINT ptCursor;
			RECT rcWnd;
			GetWindowRect(m_hWnd, &rcWnd);
			GetCursorPos(&ptCursor);
			sptOffset.x = ptCursor.x - rcWnd.left;
			sptOffset.y = ptCursor.y - rcWnd.top;

			SetCapture(m_hWnd);
			sbMoving = true;
			return 0;
		}
		break;

	case WM_NCLBUTTONUP:
	case WM_LBUTTONUP:
		sbMoving = false;
		ReleaseCapture();
		break;

	case WM_MOUSEMOVE:
	case WM_NCMOUSEMOVE:
		if(sbMoving)
		{
			POINT ptCursor;
			RECT rcWnd;
			GetWindowRect(m_hWnd, &rcWnd);
			GetCursorPos(&ptCursor);
			SetWindowPos(m_hWnd, NULL, rcWnd.left+(ptCursor.x-(rcWnd.left+sptOffset.x)),
				rcWnd.top+(ptCursor.y-(rcWnd.top+sptOffset.y)), 0, 0, SWP_NOZORDER | SWP_NOSIZE);
		}
		break;
	}

	return DefWindowProc(m_hWnd, uMsg, wParam, lParam);
}

I dread to think what horrible bugs this will cause though, so I think I'm just going to use the method with the 500ms delay for this, and add the above method as an alternative.
Another issue is that the user can still access the system menu of the window (Which will cause a stall), and select "Move" from there - which will stall. The second issue can be fixed with the timer as before, but probably not the stall when selecting the system menu.

Thanks for all the replies,
Steve