[PATCH libinput 0/24] Tablet support

[Peter Hutterer]

On Tue, May 27, 2014 at 01:11:31PM -0700, Jason Gerecke wrote:
> I've been away from my computer for most of the (long) weekend up
> here, so apologies for being a bit quiet :)
> 
> 
> On Sun, May 25, 2014 at 7:13 PM, Peter Hutterer
> <peter.hutterer at who-t.net> wrote:
> > On Fri, May 23, 2014 at 10:34:10PM -0400, Chandler Paul wrote:
> >> On Fri, 2014-05-23 at 17:00 +1000, Peter Hutterer wrote:
> >> > I'm almost sold on normalization since it does reduce the likelihood of
> >> > things going wrong. We need to provide the axis resolution to convert back
> >> > to the real data though where needed.
> >> >
> >> > once you provide the axis resolution, it doesn't matter if you provide raw
> >> > data unless you also want to provide "raw" resolution, which is excessive..
> >> >
> >> > so, given that this would be sent down the protocol (and for the limited
> >> > resolution) the range should be normalised to uint16_t or uint32_t max, with
> >> > the resolution in units/mm or canonicalized where more appropriate. This
> >> > would be what goes on the wayland protocol as well then.
> >> >
> >> > helper functions to convert that back to doubles, or elbows per square ounce
> >> > would be part of the wl-client package that parses that protcol, not
> >> > libinput.
> >> >
> >> > that doesn't seem like the worst architecture for both libinput and the
> >> > protocol, any comments?
> >>
> 
> There's a subtlety on the protocol side of things that can't be
> ignored. When normalizing data, you want to be careful to preserve
> information about the zero point. Without that, you can't meaningfully
> pass the data along. Lets imagine that we have some sensor that will
> report values between 10 and 100, with a resolution of 1 unit = 1
> elbow per square ounce. If we normalize that to the range [0,
> UINT32_MAX] we've lost information about where "zero" is. A normalized
> value of zero does not correspond to zero elbows per square ounce as
> you might expect, and the resolution info is insufficient to correct
> the offset.

good point, keeping the zero point is important.

> Now, if we've done our jobs properly in libinput, that shouldn't be a
> problem. We would have normalized that sensor's values to [0.1, 1] and
> announced the axis to have a resolution of 1 unit = 100 elbows per
> square ounce. Because the zero point is offset like it originally was,
> it's preserved through the scaling done for the protocol and so the
> original 10-100 range can be recovered. The only amendment I'd make is
> to use a signed integer type rather than an unsigned one, since we may
> have negative normalized values that need to be sent through the
> protocol.
> 
> >> Seems fine to me. As for normalizing values to units/mm or the like, is
> >> there any known conversion for the units the tablet returns for distance
> >> to metric?
> >
> > Benjamin answered that on IRC, but for the archives: the distance is in mm,
> > though in reality the data is inprecise.
> >
> > Cheers,
> >    Peter
> >
> 
> I would avoid attaching units or resolution to axes which do not
> already declare them. The distance values on our pens do roughly
> correspond to millimeters from the sensor (which itself is usually a
> few mm below the surface) but we should be reporting a non-zero
> resolution through evdev if the data were reasonably accurate :D
> 
> Also, libinput shouldn't generally be "normalizing values to units/mm
> or the like." Data should be normalized to some range within [-1, 1]
> so that the zero point is preserved. Resolution data should be
> provided through another means which relates normalized values to
> real-world units (and should probably be documented to be zero if the
> resolution is unknown). The only exception to this /might/ be
> something like tilt or rotation (though the more I think about it, the
> less I believe it to be worthy of exception given how apps actually
> use the data).

looking at linux/input.h, the axes where we should attach resolution in a
defined unit (units/mm or radians) are:
x/y/z, rx/ry/rz, distance, tilt x/y and their equivalents in the ABS_MT
range. The kernel already does that anyway if the device supports it, we're
just passing this on here. I'm not suggesting attaching a unit to e.g.
ABS_RUDDER, who knows what that is :)

so just in case this isn't clear what I'm suggesting is to export the range
normalised in [LI_FIXED_MIN, LI_FIXED_MAX] or [0, LI_FIXED_MAX], with extra
information to map min/max into a physical dimension if possible. which is
what you're suggesting too afaict.

Cheers,
   Peter