QEventDispatcherWin32 does not handle WM_INPUT messages properly - BUG?

Uwe Kindler

Thank you for this hint. Unfortunately the "http://spacenav.sourceforge.net/spnav-win32.html":http://spacenav.sourceforge.net/spnav-win32.html driver is no option for us for two reasons:

It is not in production quality.
It requires "http://sourceforge.net/projects/hidlibrary":http://sourceforge.net/projects/hidlibrary/ which is also in beta state and requires .NET Framework 2.0. We would like to avoid any dependency to .NET Framework.

I will try to work around the issue by creating a second message pump in a different native Win32 thread that has its own message window and its own windows procedure. But this still does not answer the question why QEventDispatcherWin32 does not handle WM_INPUT messages properly.

dibben

I have been having the same problem trying to get the 3D Connexion mouse working and I think I have discovered where the problem is in QEventDispatcherWin32.

I am using Qt 4.6.2 with VisualStudio 2005 on Windows XP. The problem I was having was that the WM_INPUT messages were not being received by window/event filter the until some other event such as a mouse or timer event occurred. I could move the 3D mouse and nothing would happen then move the normal mouse and all the 3D mouse events would arrive.

I traced the problem to
@MsgWaitForMultipleObjectsEX(nCount, pHandles, INFINITE, QS_ALLINPUT, MWMO_ALERTABLE);@

In the processEvents function of QEventDispatcherWin32. This was blocking and not waking on WM_INPUT events.

The windows documentation for this notes that QS_ALLINPUT, which includes QS_INPUT, "does not include QS_RAWINPUT in Windows2000." Therefore I think the problem is that the binary Qt installer is built to work on all windows platforms, so during the compile _WIN32_WINNT is defined to something less that WinXP (0x501) so that raw-input messages are not handled.

I re-compiled Qt passing -D _WIN32_WINNT=0x501 to configure and now the WM_INPUT messages are being passed to the event filter when the 3D mouse is moved.

So the options seem to be

Recompile Qt with _WIN32_WINNT=0x501 so that XP (and later) functions are enabled
or
Use a timer to unblock the event loop at regular intervals so that the messages get processed. I am not sure yet what effect this will have on the smoothness of the mouse operation.

I also found that just using the winId() function of the main window for the HWND to use for registering the device also worked fine.

Uwe Kindler

Hi David,

hey thank you very much for this solution. We use a self-compiled Qt installation here so it should be no problem to recompile it with _WIN32_WINNT=0×501.

We worked around the problem by creating a native Win32 DLL (without Qt) that creates an internal win32 message window in its own thread that receives the messages from Space Navigator device. Then we wrote a Qt wrapper for this DLL that translates the messages into Qt events an created a QTDx device.

But your solution makes it much easier to use Space Navigator from Qt - thank you :O)

Uwe

BennM

Hi,

can anybody of you post an example code which shows how to implement the spacenavigator in qt?

Thank you very much

dibben

I wrote an slightly updated version of my comments above and added some sample code of a Qt class to get the data from the mouse on Windows at:

"http://www.codegardening.com/2011/02/using-3dconnexion-mouse-with-qt.html":http://www.codegardening.com/2011/02/using-3dconnexion-mouse-with-qt.html

With Qt 4.7.1 it seems to work without having to re-complile the Qt.

BennM

@David Dibben: Thank you very much for the sample code. I tried to compile it with the qt eclipse integration but it stops during compiling the file "Mouse3DInput.cpp" in line around 345 "pri = NEXTRAWINPUTBLOCK(pri);" complaining that "NEXTRAWINPUTBLOCK was not declared in this scope".

I added the line in 3DMouse.pro

@LIBS += -lUser32 -LC:\programs\Microsoft SDKs\Windows\v6.0A\Lib@

added user32.lib and the paths to the lib und include directories of Microsofts SDKs.

Is there anything missing?

[EDIT: code formatting, Volker]

dibben

NEXTRAWINPUTBLOCK is a windows macro, defined in winuser.h which should be included with windows.h
See - http://msdn.microsoft.com/en-us/library/ms645593(v=vs.85).aspx

Which compiler are you using? I am guessing that it is mingw. I don't have experience using g++ on Windows - I compiled the sample using MS VC++ 2008 Express.

The windows headers should be available with mingw as far as I know, but the windows headers are full of conditional statements so it is possible that something has to be defined for the macros to show up correctly.

BennM

Yes I use mingw and did the following changes in your sourcecode:

I added this in Mouse3DInput.cpp

@#define RAWINPUT_ALIGN(x) (((x) + sizeof(DWORD) - 1) & ~(sizeof(DWORD) - 1))
#define NEXTRAWINPUTBLOCK(ptr) ((PRAWINPUT)RAWINPUT_ALIGN((ULONG_PTR)((PBYTE)(ptr) + (ptr)->header.dwSize)))@

and this

@#include "C:\Qt\2010.05\mingw\include\windows.h"
#include "C:\Qt\2010.05\mingw\include\winuser.h"@

Then I found a difference between Microsoft's WinUser.h and mingw's winuser.h (see following):
@
typedef struct tagRAWHID {
DWORD dwSizeHid;
DWORD dwCount;
BYTE bRawData;
} RAWHID,*PRAWHID,*LPRAWHID;
@
I changed it to @bRawData[1]@ like in Microsoft's WinUser.h. I don't know why it works with the range of one, because in the sourcecode you'll find the command
@pRawInput->data.hid.bRawData[1] == 0x01@
so it must be at least 2, right?

Additionally I added
@
CONFIG += console

DEFINES += _WIN32_WINNT="0x0501"
DEFINES += _WIN32_WINDOWS="0x0501"

INCLUDEPATH += "C:\Qt\2010.05\mingw\include" \

LIBS += -luser32 -L"C:\Qt\2010.05\mingw\lib"
@

to 3DMouse.pro, but I am not shure if I need all of that.

After that I was able to compile it. At the moment I am not sure if the translation works correctly, because the data appears too fast in the textboxes, so I will change the program that it saves the data in a textfile.

BennM

Hi,

I recorded some data which my program (original written by David Dibben) received from the spacenavigator. I made translational movements at first - followed by rotations. There is only noise on the x-axis and no reaction on y- and z-axis. But you can see the translational movements in the last three plots (plots of the rotations). Does anybody has any idea what's going wrong?

!http://img6.imagebanana.com/img/ab52tne4/temp.png(plot)!

The horizontal axis shows the number of received events - it's not the time!

BennM

Hi,

I switched to Visual Studio and your code worked out of the box.

Thank you for the sample code, David Dibben!