libinput Acer c720 Observations

[Peter Hutterer]

On Fri, Jun 05, 2015 at 04:22:51PM -0600, Wade Berrier wrote:
> Hello,
> 
> I'm running libinput 0.17 on Fedora 22 (compiled and installed the
> rawhide src.rpm) on an Acer c720 chromebook.  The observations below
> also apply to 0.15 that shipped with the distro, but I upgraded to see
> if it changed the behavior.

just fwiw, I'm pushing out updated packages on an almost daily basis (f22
and rawhide) and ptraccel is the main things being tweaked at the moment.
 
> The trackpad doesn't quite "feel right".  Here's some of the main
> observations compared to when using the trackpad with ChromeOS and/or
> the cmt driver:
> 
> 1. the pointer is very "fidgety".  Relatively small movements move the
>    pointer around quite a bit.  If I adjust the pointer speed to the minimum
>    (using "Mouse & Touchpad" gui) these small controlled movements seem
>    much more accurate.  But, then it takes several "swipes" to move the
>    pointer to the other side of the screen.
> 
> 2. at first it seemed that no matter what "pointer speed" setting I
>    tried, finger distance always covered the same amount of screen
>    distance, no matter the finger velocity.  ie: no acceleration
>    handling.  I came across [1] and realized that wasn't the case.
>    After additional tinkering, I was able to affect screen distance
>    with finger speed, but it seemed very difficult to get into that
>    state/window.

Note: the below is gut feeling at this point:
I think our biggest problem with touchpad acceleration is that we get too
quickly into full acceleleration, so the gap between no acceleration and
full acceleration is too narrow.
if you always hit full acceleration, then you wouldn't notice the difference
and it would feel like constant acceleration.

> In short, the lowest pointer speed feels natural in localized
> scenarios requiring high precision (ie: clicking on tiny stuff).  But,
> when swiping faster to travel to the other side of the screen, a 40%
> setting seems to "feel" about right.  It's interesting that ChromeOS
> gets away with not even having a pointer speed adjustment.

note that chromeos afaik only needs to support two touchpads, which makes
the job a bit easier.
 
> 3. the pointer moves around when doing "hard clicks".  Both on the way
>    down and the way up.  It seems like ChromeOS "locks" the pointer in
>    place during this case.  Without this, my 5 year old son is unable
>    to hard click a button, whereas he's able to on ChromeOS.  Also
>    interesting, is that in ChromeOS I can roll my fingertip in circles
>    and the pointer barely, if at all, moves.  So, I guess this is an
>    issue whether I'm clicking or not.

file a bug for this please, with an evemu recording for one of these moving
clicks. we have that feature in libinput, but you're clearly getting outside
of it's boundaries.

> 4. I get spurious pointer movement and clicks from my right palm while my
>    hands are in typing position.  The feature of disabling the mouse
>    while typing partially helps. Also, the trackpad isn't centered
>    between my hands on the c720, possibly making it only happen with my
>    right hand.  On ChromeOS, I can tap my palm all over the pad and
>    will get an occasional click but no pointer movement.

same here, please file a bug with a couple of emu recordings. there's a plan
for thumb detection (which would also detect palms) but it's not there yet.

> As an aside, it sounds like the mouse team at google pressure
> calibrates [2] their hardware.  Is libinput normalizing pressure in
> addition to dpi?  If so, does the hardware database also include
> information for device specific pressure calibration?

libinput doesn't use pressure on touchpads yet. it's somewhere on the 
list, but so far we've gotten away without having to worry about it :)

> 
> I realize this is all highly subjective to muscle memory and personal
> preference, but I think it can be useful to compare to the excellent
> experience that the ChromeOS driver offers.
> 
> So, where do I go from here?  Are my observations specific to the
> c720?  Is there something I can try out or tweak?  If it were easy to
> use the xf86-cmt driver on fedora [3] I probably would, but given
> wayland will be using libinput it would be nice to get libinput
> working nicely on this trackpad.

as above, 3) is definitely a bug, 4) is a todo, 2 is being worked on, 1) as
well but that too would likely benefit from a bug+evemu recording of such a
small movement. these touchpads have a comparatively low resolution, so I
wonder if we have a simple bug there.

Cheers,
   Peter