[PATCH] libinput device capability modification regarding combo input devices

[Peter Hutterer]

Sorry about the delay, I was on vacation.

On Fri, Sep 25, 2015 at 03:32:43PM +0900, 박성진 wrote:
> Dear Peter, thanks.
> That's right. If we modify the enum with the new one, ABI break will be made.
> Then how about add new capabilities ?

we can add new capabilities - gestures was added fairly recently and we have
two more capabilities in the pipe for graphics tablets/button set devices.

> IMO, it's better to get capabilities by calling once an API than calling libinput_device_has_capability twice.

"better" is hard to quantify :) There's no real technical difference between
passing a mask in or calling the function twice.

> Otherwise, we can enhance wayland protocol to get keyboard/pointer name from the compositor by event(s).
> Please share your opinion.

The wayland protocol is quite a different beast to libinput though. libinput
gives you per-device handling (since it aims to be for the compositor), the
wayland protocol hides all input devices behind the virtual
pointer/keyboard/touch interface. Two very different things, maybe you
should specify the problem you're trying to solve in more detail.
Right now, even if you send the device name for each input event down the
protocol, it still wouldn't gain you much (or anything?).

Cheers,
   Peter

> 
> Thanks and regards,
> Sung-Jin Park
> 
> -----Original Message-----
> From: Peter Hutterer [mailto:peter.hutterer at who-t.net] 
> Sent: Friday, September 25, 2015 3:13 PM
> To: �ڼ���
> Cc: wayland-devel at lists.freedesktop.org
> Subject: Re: [PATCH] libinput device capability modification regarding combo input devices
> 
> On Wed, Sep 23, 2015 at 11:29:13AM +0900,  ڼ    wrote:
> > Dear all,
> > 
> > as you guys know, there are some combo input devices which support 
> > both keyboard functionality and pointer functionality.
> > 
> > We can also see some of input devices supports both keyboard 
> > functionality and touch functionality.
> > 
> >  
> > 
> > In theory, each input device functionality needs to be provided 
> > through each kernel device node.
> > 
> > By the way, there are many combo input devices whose input events is 
> > coming from one device node.
> > 
> > Now and in near future, many vendors can make many combo input devices.
> > 
> >  
> > 
> > For these kinds of input devices, how about adding new enums for combo 
> > devices?
> > 
> > Otherewise, how about defining enums with masks ?
> > 
> >  
> > 
> > For example, we can modify the definition of 
> > libinput_device_capability enum like the following:
> > 
> >  
> > 
> > diff --git a/src/libinput.h b/src/libinput.h
> > 
> > index 9057446..458f256 100644
> > 
> > --- a/src/libinput.h
> > 
> > +++ b/src/libinput.h
> > 
> > @@ -51,14 +51,22 @@ enum libinput_log_priority {
> > 
> > /**
> > 
> >   * @ingroup device
> > 
> >   *
> > 
> > - * Capabilities on a device. A device may have one or more 
> > capabilities
> > 
> > + * Capability masks and capabilities on a device. A device may have 
> > + one or
> > more capabilities
> > 
> >   * at a time, capabilities remain static for the lifetime of the device.
> > 
> >   */
> > 
> > +#define LIBINPUT_DEVICE_CAP_KEYBOARD_MASK (1L<<0)
> > 
> > +#define LIBINPUT_DEVICE_CAP_POINTER_MASK (1L<<1)
> > 
> > +#define LIBINPUT_DEVICE_CAP_TOUCH_MASK (1L<<2)
> > 
> > +#define LIBINPUT_DEVICE_CAP_GESTURE_MASK (1L<<3)
> > 
> > +
> > 
> > enum libinput_device_capability {
> > 
> > -       LIBINPUT_DEVICE_CAP_KEYBOARD = 0,
> > 
> > -       LIBINPUT_DEVICE_CAP_POINTER = 1,
> > 
> > -       LIBINPUT_DEVICE_CAP_TOUCH = 2,
> > 
> > -       LIBINPUT_DEVICE_CAP_GESTURE = 5,
> > 
> > +       LIBINPUT_DEVICE_CAP_KEYBOARD = 
> > + LIBINPUT_DEVICE_CAP_KEYBOARD_MASK,
> > 
> > +       LIBINPUT_DEVICE_CAP_POINTER = 
> > + LIBINPUT_DEVICE_CAP_POINTER_MASK,
> > 
> > +       LIBINPUT_DEVICE_CAP_TOUCH = LIBINPUT_DEVICE_CAP_TOUCH_MASK,
> > 
> > +       LIBINPUT_DEVICE_CAP_GESTURE = 
> > + LIBINPUT_DEVICE_CAP_GESTURE_MASK,
> > 
> > +       LIBINPUT_DEVICE_CAP_KEYBOARD_POINTER =
> > LIBINPUT_DEVICE_CAP_KEYBOARD_MASK | LIBINPUT_DEVICE_CAP_POINTER_MASK,
> > 
> > +       LIBINPUT_DEVICE_CAP_KEYBOARD_TOUCH =
> > LIBINPUT_DEVICE_CAP_KEYBOARD_MASK | LIBINPUT_DEVICE_CAP_TOUCH_MASK,
> > 
> > +       LIBINPUT_DEVICE_CAP_TOUCH_GESTURE = 
> > + LIBINPUT_DEVICE_CAP_TOUCH_MASK
> > | LIBINPUT_DEVICE_CAP_GESTURE_MASK,
> > 
> > };
> > 
> >  
> > 
> > /**
> > 
> > With this kinds of modification,
> > 
> > A wayland compositor is able to recognize the actual input device 
> > capabilities and able to interpret events properly.
> > 
> >  
> > 
> > Any ideas ?
> 
> Two things: libinput's API has been stable for a while now and cannot be changed, you can only add to it in backwards-compatible manner. The above isn't backwards-compatible, so it's a non-starter.
> 
> The mask-like behaviour you ask for is provided by these enums. if you want to check if a device is both pointer and keyboard, you call libinput_device_has_capability twice, once with CAP_POINTER, once with CAP_KEYBOARD. There's no need for specific combination defines or masks.
> 
> Cheers,
>    Peter
> 
>