Re: Question about ABS_DISTANCE's intended usage.

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Wed, Feb 24, 2016 at 08:39:15AM +1000, Peter Hutterer wrote:
> On Tue, Feb 23, 2016 at 02:02:45PM -0800, Dmitry Torokhov wrote:
> > On Tue, Feb 23, 2016 at 08:04:32AM +1000, Peter Hutterer wrote:
> > > On Mon, Feb 22, 2016 at 08:35:31AM -0800, Charles Mooney wrote:
> > > > On Sun, Feb 21, 2016 at 2:04 PM Peter Hutterer <peter.hutterer@who-t.net> wrote:
> > > > >
> > > > > On Thu, Feb 18, 2016 at 10:19:30AM -0800, Dmitry Torokhov wrote:
> > > > > > On Tue, Feb 09, 2016 at 01:56:05PM -0800, Charles Mooney wrote:
> > > > > > > Hello all,
> > > > > > >
> > > > > > > I'm currently working with a touchpad vendor with a new device that
> > > > > > > supports a limited form of hover detection.  Their sensor is able to
> > > > > > > detect the presence or absence of a finger/hand/palm hovering over the
> > > > > > > sensor without touching it, but is unable to report any more details
> > > > > > > about it.  This is a more limited form of hover detection than some
> > > > > > > devices which attach a hover state to each finger they see, and can
> > > > > > > even report x/y coordinates to hovering finger.
> > > > > > >
> > > > > > > Instead of using ABS_MT_DISTANCE, it appears that the correct event to
> > > > > > > use would be ABS_DISTANCE, since the value is not tied to a specific
> > > > > > > finger.  I would like to check with you all about how this value is
> > > > > > > intended to be used, because it's not quite as obvious to me as I
> > > > > > > first thought.
> > > > > > >
> > > > > > > We need to handle three basic states:
> > > > > > >   1. At least one finger is touching the pad.
> > > > > > >   2. Something is hovering, but nothing is actually touching.
> > > > > > >   3. Nothing is touching the pad and nothing is detected hovering over it either
> > > > > > >
> > > > > > > It's seems clear to me that an ABS_DISTANCE of zero should indicate
> > > > > > > state #1 and that any other legal positive value should indicate state
> > > > > > > #2, but I'm less clear on what the best way to handle state #3 is.
> > > > > > > Currently, I think the best strategy would be to use a value of
> > > > > > > ABS_DISTANCE = -1 to indicate that there are no fingers seen (hovering
> > > > > > > or otherwise), does that make sense?
> > > > > > >
> > > > > > > If not this, how else would you suggest that this ought to be done?
> > > > > >
> > > > > > As we discussed in person, I believe that reporting an "out of bounds"
> > > > > > value for ABS_DISTANCE when we have to use single-touch mode and thus do
> > > > > > not have a good way to invalidate a contact, is the easiest option.
> > > > > > Alternative would be to invent another SYN event, which I'd rather not.
> > > > > >
> > > > > > So for devices that support hovering but can not report individual
> > > > > > hovering contacts we should declare 0..N as ABS_DISTANCE range and report
> > > > > > following values:
> > > > > >
> > > > > >  - 0 when a finger is actually touching
> > > > > >  - 1..N for hovering fingers
> > > > > >  - return X < 0 or X > N when no fingers are detected at all; in
> > > > > >    practice I think we should simply report -1 in this case.
> > > > > >
> > > > > > Benjamin, Peter, Henrik, any concerns?
> > > > >
> > > > > on the touchpads that support hovering we're already using BTN_TOOL_FINGER
> > > > > together with ABS_DISTANCE, without needing out-of-range reports.
> > > > > BTN_TOUCH is the signal when a finger is physically touching (or ABS_PRESSURE
> > > > > if it exists and clients care about it).
> > > > >
> > > > > So the sequence Charles should send is:
> > > > >
> > > > > 3) <nothing> :)
> > > > > 2)
> > > > >     EV_ABS ABS_DISTANCE <d> # for d > 0
> > > > >     EV_KEY BTN_TOOL_FINGER 1
> > > > >     EV_SYN SYN_REPORT 0
> > > > > 1)
> > > > >     EV_ABS ABS_DISTANCE 0
> > > > >     EV_ABS ABS_X <x>
> > > > >     EV_ABS ABS_Y <y>
> > > > >     ...
> > > > >     EV_KEY BTN_TOUCH 1
> > > > >     EV_SYN SYN_REPORT 0
> > > > > 2)
> > > > >     EV_ABS ABS_DISTANCE <d> # for d > 0
> > > > >     EV_KEY BTN_TOUCH 0
> > > > >     EV_SYN SYN_REPORT 0
> > > > > 3)
> > > > >     EV_KEY BTN_TOOL_FINGER 0
> > > > >     EV_SYN SYN_REPORT 0
> > > > >
> > > > > This should work with at least libinput, though I have to check what happens
> > > > > when you don't send x/y on the first event. I think this would need a patch
> > > > > in libinput, but that's doable. And it's the same sequence we also use for
> > > > > e.g. pen tools that support hovering as well.
> > > > >
> > > > > Cheers,
> > > > >    Peter
> > > > 
> > > > Hi Peter,
> > > > 
> > > > It looks like you're suggesting to use "BTN_TOOL_FINGER" as a signal
> > > > as to weather or not the value in "ABS_DISTANCE" is valid or not.
> > > > 
> > > > 1. No finger detected anywhere:
> > > >     BTN_TOOL_FINGER = 0
> > > >     ABS_DISTANCE = n/a
> > > > 2. Finger seen hovering but not touching:
> > > >     BTN_TOOL_FINGER = 1
> > > >     ABS_DISTANCE > 0
> > > > 2. Finger touching:
> > > >     BTN_TOOL_FINGER = 1
> > > >     ABS_DISTANCE = 0
> > > 
> > > you should set BTN_TOUCH here. This can be done based on some magic pressure
> > > threshold if you have pressure, otherwise just unconditionally set it
> > > whenever distance is 0.
> > > 
> > > > 
> > > > Am I understanding that correctly?
> > > 
> > > yes, but IMO better to think it this way: BTN_TOOL_FINGER signals whether a
> > > finger is detected by the device. everything else is just axis information
> > > if and when it becomes available.
> > > likewise, BTN_TOUCH signals whether the current tool (could be something
> > > other than finger) logically touches the surface.
> > 
> > I wonder if this will not confuse clients that do not pay attention to
> > ABS_DISTANCE though... I take it that libinput and x synaptics drivers
> > won't be confused, mousedev in kernel relies on BTN_TOUCH, what about
> > others?
> 
> if they get confused, they'd already be confused by a set of current
> devices. This isn't new behaviour, we've been doing this for quite a while.
> and as I said above, it matches the behaviour we use for BTN_TOOL_PEN, it's
> IMO hard to justify that the behaviour should be different between the two
> tools.

OK, fair enough. Any chance I could get a patch for ABS_DISTANCE section
of Documentation/input/event-codes.txt to mention that we expect
to see BTN_TOOL_* as well?

Thanks.

-- 
Dmitry