[Adding linux-input and specifucally Jiri Kosina]

On Tue, Mar 11, 2008 at 01:52:47PM +0100, Nicolas Mailhot wrote:

Le Mar 11 mars 2008 13:28, Wolfgang Draxinger a ??crit :

quoted

Am Dienstag, 11. M??rz 2008 schrieb Daniel Stone:

quoted

Quirk tables are all through the USB code[0], IDE, etc, already, so
this would be the accepted practice.

I can understand it for single chipsets, or other specific HW. But
USBHID is generic to the bone, and it's device neutral. Having the
quirks table in the kernel would mean, that for every awkward device
you'd have to get the latest kernel, or if it's not in there yet
patch it.

I'd be very careful not to assume what can or can not be done at
kernel level, and what should and should not belong in the kernel.

The kernel HID people are the ones who see all the hardware diversity
and what people need. They're the ones you should consult first before
adding a quirks table somewhere else. Because if they have to
implement a quirks table in the kernel anyway (for some need you've
not envisionned today) having two layers of quirks that fight each
other for device control is going to make life miserable for everyone.

I think that if EV_ABS for a device is the one that really makes sense
and the device sends EV_REL in error we should just adjust the type
of events right in USBHID driver so incorrect events don't propagate
any further. If hovewer preference of EV_ABS vs EV_REL is just something
xf86-evdev wants just because of its internal architecture then the
quirk belongs to xf86-evdev itself.

Jiri, apparently some 3Dconnexion devices send EV_REL which makes
xf86-evdev not too happy. What do you think?

-- 
Dmitry