Last night just as I was readying to demo a controller subsystem Direct Input 8 begins having a tendency to never come back from Acquire...
Anyway dinput8d.dll is no longer in any of the DX SDKs and seems to have even been backwards removed from the download links of SDKs from years past.
I am not convertible downloading it from a .dll repository website as they all look like the seediest things on earth.
But I am wondering if some of my usage patterns might be error prone and would feel better if I had a debug build either way.
Background: I have at this point a single thread that manages multiple devices all self contained. The outer framework is very asynchronous but the DirectInputDevice stuff is all self contained within the single thread itself. So I just don't see any excuse for the method to deadlock. The coop mode is background nonexclusive.
The controls seem to work in the control panel even when the thread is waiting for Acquire to return. Not sure if that would work if it was a driver bug. Thanks.
FYI: I found passing a 0 for the window handle to SetCooperativeLevel seems to be less sticky (or rather may actually work) than when using a hidden window. The docs say a valid window handle is required, but I think it's possible that 0 is interpreted as the desktop or something??? Otherwise this is undocumented and therefore unreliable behavior.
I was using a hidden window because I've found that one is required for Direct3D9 to function on windows XP and under different circumstances. It may be Direct3D needs a window for messaging purposes but it seems like bad design either way. Post XP a 0 window seems to work with Direct3D. Not that it matters.
Edited by Mick is Stumped, 18 October 2012 - 07:07 PM.
^Thanks MikeBMcL. I think what I really wanted was windowless Direct Input which is not documented. But I know some win32 APIs treat 0 as a desktop window. I wonder about side effects. But I also wonder if it would be legit to pass the desktop to Direct3D if what you want is just a windowless but nonetheless windowed device instance (you can pass windows directly to Present) for instance when the renderer is inside a DLL and doesn't care about the Present targets.
Ontopic: For anyone looking for info on this matter here (http://xboxforums.create.msdn.com/forums/p/108720/641925.aspx#641925) is a detailed explanation, but because microsoft.com links tend to go stale in short order what I found was Sleep indeed causes a deadlock but assuming you are using SetEventNotification and [Msg]WaitForMultipleObjects what seems to work is manipulating the timeout parameter of MsgWaitForMultipleObjects.
The examples in the documentation are pretty confusing as they seem to make a number of assumptions that are not explained and in and of themselves appear to be slapped together as there are contradictions right there in the code/explanations if you look for them.
Tip: if you are trying to be nice setting the timeout to even 1ms will take your thread's CPU usage down from ~50% to ~3-1% or less. The docs imply that 0 is a good timeout but that is probably assuming that your loop has something to do during on timeout (eg. draw a frame) that will by Direct Input enough time to not deadlock itself. I would say more but there is the above link which pretty much covers the bases.