multi-touch-dev team mailing list archive

[Martin Wawro <wawro@xxxxxxxxxxxxxxxx>]

Hi,

Not sure if this posting is helpful, but you may just decide for yourselves :)

I happen to have patched a more or less complete MT event chain from the 
kernel-driver up to Qt 4.7.0 to work with XInput-based multitouch. The code
"below" the Qt layer is meant to be rather low-level, meaning that it does not
perform any gesture recognition at all, but leaves this task to the 
application layer that has the proper logic to do so (see the Qt patches for 
that). I think the topic of which layer should do the high-level recognition
has been discussed extensively, so I do not want to add to that.

The changes I made in the Xorg evdev driver mainly follow along the line that 
Stephane and his team already did (I even copied parts of their code). 
Because I had a special target device in mind that also runs "legacy" X 
applications (i.e. w/o multitouch support of any kind), I also cared about 
not breaking things with these apps, ending up in only sending button events
for the primary finger on the screen as well as always sending the
coordinates of the primary finger in the "non-MT" Abs fields.
I also added a little convenience in the kernel's input.h which basically is a 
bitfield that keeps track of the fingers on the screen by setting the bits of 
the corresponding tracking IDs to 1 during touchdown. This is exported as 
just another button in the event device. As I avoided button events for all 
non-primary fingers, this bitfield makes it easy to keep track 
of "non-primary" touches and releases and also comes in handy when doing 
event compression (see the Qt patches for that). It turns out that 
multitouch-aware Qt applications work like a charm and even the legacy 
applications do (no jumping or locking "mouse-pointer" when (accidentally) 
hitting the screen with the second finger).

Another useful thing when using the actual Xorg driver to relay the multitouch 
events (instead of directly working on /dev/input/eventXX from within the 
application layer) is the fact that software can very easily be tested on the 
actual multitouch device (in our case a slate) without even compiling on the 
device, just forward the X screen to the target device. With that you can do 
funny things like running the test application on your (fast-compiling) 
64-bit workstation while having the output and the input on your 32-bit 
slate.

Anyway, here is a URL to the patches as well as a small description:
http://www.digitalmedics.de/projects/wetab
Note that these patches currently only work for a certain (hid-egalax.c) 
device, as they were targeted at a special hardware platform. However, the 
code is written in a quite general way, so that adaptions to other hardware 
should be easy, in fact it suffices to work on the kernel HID layer.
As the manufacturer of the slate device is already in the process of 
incorporating these patches into their system, I will take down the page in a 
couple of weeks. I felt that before I take it down, maybe those patches are 
of use to someone else, if not I apologize for wasting your time :)

With best regards,

Martin