Op 25-01-12 14:26, Henrik Rydberg schreef:

quoted

+
+struct mstm_state {
+	bool advance_flag;
+	int x;
+	int y;
+	unsigned char last_timestamp;
+	unsigned char data[MSTM_DATA_WIDTH][MSTM_DATA_HEIGHT];
+};

The ability to simply send an "input screen" would be perfect
here. This device may be on the border of what can/should be handled
via input events. A memory mapped driver, uio-based or something
similar, could be an option.

Op 30-01-12 08:27, Dmitry Torokhov schreef:

On Wed, Jan 25, 2012 at 04:00:35PM +0100, Henrik Rydberg wrote:

quoted

quoted

quoted

One possible option could be to use the
slots, but only send ABS_MT_TOUCH_MAJOR or ABS_MT_PRESSURE, nothing
else. The device would (rightfully) not be recognized as MT since the
position is missing, all data would be available for processing in
userspace, and bandwidth would be minimized since there could only be
so many changes coming in per millisecond.

So how does userspace then finds out where these pressure points are
located?
Or do you mean to just dump all data to user space (15 * 13 *
sizeof(ABS_MT_PRESSURE value) + overhead)?

Having each pressure point represented by one slot id was the idea.
Userspace would have to know how the points are mapped, of
course. Still not overly happy about the general fit, though. Dmitry?

I am having doubts that this device, as it is, is suitable for input
interface; I really do not think that bastardizing slot IDs is good
idea. Unless we move all computation necessary to identify individual
contacts into the kernel (and then use standard MT protocol), I'd
recommend looking into hidraw + uinput solution.

Ok, so it's pretty clear the consensus is that there should be a 
different interface to push the data to user space.
My initial reaction is V4L, as the data is basically a stream of 
monochrome frames, this kind of fits the video description. Also, this 
could be of some benefit to all those emulate-multi-touch-using-webcam 
projects.
Other options I see are mmap on the input device, creating a separate 
(char?) device, UIO or even having the whole thing in userspace.

However, there's one tiny hack I'd like to have and I'm unsure whether 
this can be done from userspace: as the mouse has only one physical push 
button, it can't distinguish between left, middle or right click.
Obviously, it can do this based on the information it gathers from its 
touch surface, however the firmware has only the following dumb 
implementation: when only one finger is present on the right part of the 
surface, assume right click; otherwise assume left click. This means 
that you need to lift your left finger when performing a right click.

Now, the receiver is represented as 3 HID devices: one keyboard 
descriptor, one mouse descriptor and one miscellaneous/control 
descriptor (containing touch surface data). My plan was to intercept 
mouse clicks on the mouse descriptor and look at what part of touch 
surface (left/right) has the most pressure and, if needed, swap the 
click to a right click.

Any hints on how to do this in userspace? Of course I could hack up 
xf86-input-evdev, but I'd like to do this at the highest possible layer, 
ie the input subsystem.

-- 
Maurus Cuelenaere