[PATCH v2 wayland] protocol: add wl_pointer.axis_frame events

Peter Hutterer peter.hutterer at who-t.net
Sun Jul 12 23:01:20 PDT 2015


On Sun, Jul 12, 2015 at 03:58:47PM +0800, Jonas Ã…dahl wrote:
> On Thu, Jun 25, 2015 at 04:01:48PM +1000, Peter Hutterer wrote:
> > To group separate vertical/horizontal scroll events together. Likewise it
> > enables axis_stop events to be grouped so that the final vector for kinetic
> > scrolling may be calculated correctly.
> > 
> > Signed-off-by: Peter Hutterer <peter.hutterer at who-t.net>
> > ---
> > Changes to v1:
> > - drop the option of having multiple axis sources within the same frame
> >   (and reword the documentation accordingly)
> > 
> > Two comments here:
> > - the protocol feels a bit unbalanced now when discrete events are in use,
> >   these must still be sent before the matching axis events. i.e. we have 4
> >   events that commit with the axis_frame, but one event (discrete) that
> >   commit with the axis event:
> >       axis_source
> >       axis
> >       axis_discrete
> >       axis
> >       axis_frame
> >   though sending the source just before the frame may make this a bit easier
> >   on the eye
> 
> The "unbalancedness" I feel is that "axis_source" is a bit "let-loose"
> as it doesn't extend some other event, but kind of belong to axis_frame.
> So ...
> 
> > - we could now skip the source and merge it into the axis_frame event,
> >   though that would require the return of the "unknown" source.
> 
> .. now that we always have a single source per frame, I think this would
> make sense to do. An 'unknown' source I guess is not that big of a
> change, semantically. IIRC the issue with it was that a device
> which previously would have source "unknown" would suddenly change if we
> learned about it, but as long as we say that might happen, maybe its not
> an issue really.
> 
> Otherwise, maybe it'd be best to make axis_source extend axis_frame
> like all other such events, i.e. use the common way of adding
> information to an already existing event. That'd will also make it a bit
> less unbalanced IMHO.

looking at this again with fresher eyes I think the unbalancedness simply
doesn't matter. if we visualise it as a tree, then the sequence above looks
like this:

axis_frame
  +----> axis
  +----> axis
  |        +----> axis_discrete
  +----> axis_source 

i.e. the source is at the same level that axis/axis_stop are. that makes it
less unbalanced than it initially looked. and it avoids the "unknown" mess.
so I think we should leave it as-is.

> >  
> >  protocol/wayland.xml | 30 +++++++++++++++++++++++++++---
> >  1 file changed, 27 insertions(+), 3 deletions(-)
> > 
> > diff --git a/protocol/wayland.xml b/protocol/wayland.xml
> > index 48d0332..db9b164 100644
> > --- a/protocol/wayland.xml
> > +++ b/protocol/wayland.xml
> > @@ -1579,6 +1579,30 @@
> >      </request>
> >  
> >      <!-- Version 5 additions -->
> > +    <event name="axis_frame" since="5">
> > +      <description summary="end of axis set event">
> > +	Indicates the end of a set of wl_pointer.axis events that logically
> > +	belong together.
> > +
> > +	All wl_pointer.axis, wl_pointer.axis_stop, and
> > +	wl_pointer.axis_source before a wl_pointer.axis_frame event belong
> > +	logically together. For example, in a diagonal scroll motion the
> > +	compositor will send an optional wl_pointer.axis_source event, two
> > +	wl_pointer.axis events (horizontal and vertical) and finally a
> > +	wl_pointer.axis_frame event. The client may use this information to
> > +	calculate a diagonal vector for scrolling.
> > +
> > +	When multiple wl_pointer.axis events occur within the same frame,
> > +	the motion vector is the combined motion of all events.
> > +	When a wl_pointer.axis and a wl_pointer.axis_stop event occur within
> > +	the same frame, this indicates that axis movement in one axis has
> > +	stopped but continues in the other axis.
> > +	When multiple wl_pointer.axis_stop events occur within in the same
> > +	frame, this indicates that these axes stopped in the same instance.
> > +
> > +	Only one wl_pointer.axis_source event is permitted per axis frame.
> 
> Should we maybe point out that axis_frame is guaranteed to come for
> every axis event group, event though such a group might only be a single
> axis event? Meaning, you'd never get an axis without an axis_frame.

yep, easy enough. added:

        Only one wl_pointer.axis_source event is permitted per axis frame.
+
+       A wl_pointer.axis_frame event is sent for every logical axis event
+       group, even if the group only contains a single wl_pointer.axis or
+       wl_pointer.axis_stop event. In other words, a client may get the
+       sequence: axis, axis_frame, axis, axis_frame, ...

Cheers,
   Peter
 
> 
> > +      </description>
> > +    </event>
> >  
> >      <enum name="axis_source">
> >        <description summary="axis source types">
> > @@ -1605,9 +1629,9 @@
> >  	Source information for scroll and other axes.
> >  
> >  	This event does not occur on its own. It is sent before a
> > -	wl_pointer.axis or wl_pointer.axis_stop event and carries the source
> > -	information for that event. A client is expected to accumulate the
> > -	data in both events before proceeding.
> > +	wl_pointer.axis_frame event and carries the source information for
> > +	all events within that frame. A client is expected to accumulate the
> > +	data in all events events within the frame before proceeding.
> >  
> >  	The source specifies how this event was generated. If the source is
> >  	wl_pointer.axis_source.finger, a wl_pointer.axis_stop event will be
> > -- 
> > 2.4.3
> > 


More information about the wayland-devel mailing list