[HEADS UP] clutter-evdev is using libevdev

David Herrmann dh.herrmann at gmail.com
Mon Sep 9 06:14:05 PDT 2013


On Mon, Sep 9, 2013 at 2:28 PM, Benjamin Tissoires
<benjamin.tissoires at gmail.com> wrote:
> On Mon, Sep 9, 2013 at 2:17 PM, Giovanni Campagna
> <scampa.giovanni at gmail.com> wrote:
>> 2013/9/9 Benjamin Tissoires <benjamin.tissoires at gmail.com>:
>>> Hi Giovanni,
>>> On Mon, Sep 9, 2013 at 1:49 PM, Giovanni Campagna
>>> <scampa.giovanni at gmail.com> wrote:
>>>> Hello folks!
>>>> I just wanted to tell you that the evdev backend of Clutter, which is
>>>> at the heart of the gnome-shell input handling when running as a
>>>> wayland compositor, is now using libevdev for reading the events.
>>>> This also means that, if the API is broken, it would be helpful to
>>>> know in advance. I'm subscribed to the input-tools list, so that
>>>> should be enough.
>>> That's great. Thanks for sharing with us.
>>> I don't know if you saw, but the API is nearly ready to be marked as
>>> stable[1]. Could you please have a look at the patch and share with us
>>> if you like it or not?
>> Cool to see the API will be stable, and thank you for providing
>> compatibility through deprecations.
>> I think the new naming scheme makes sense, by the way.
> Ok, thanks :)
>> It should be said, clutter-evdev doesn't implement a lot of the
>> functionality evdev would allow for, so we're basically just calling
>> next_event() and kernel_set_leds(). Also, we're using udev properties
>> instead of testing the device bits, which I'm not sure is a good
>> thing, but it seems to work.
> I would say relying on the udev properties is currently just fine, and
> is the best you can do for now.
> I think that relying on the devices bits would infer both a great
> complexity in your code, and also a duplication of the effort
> regarding the discovery/handling of devices.
> Maybe the assignments of the type of a device will be handled later in
> libinputmapper (maintained by David Herrmann), but currently, udev is
> the only place to have a comprehensive way of type assignments for the
> devices.

We really need to discuss that at LPC or XDC this month. Everyone is
using different sources for device-detection which is a mess. I have
some ideas, but I'm not sure everyone is happy with it.

Some sketches can be found here:


More information about the Input-tools mailing list