[PATCH libinput 6/6] touchpad: pretend the jumpy semi-mt touchpad is a single-touch touchpad

[Hans de Goede]

Hi,

On 31-07-15 01:36, Peter Hutterer wrote:
> On Thu, Jul 30, 2015 at 04:45:04PM +0200, Hans de Goede wrote:
>> Hi,
>>
>> On 30-07-15 14:33, Hans de Goede wrote:
>>> Hi,
>>>
>>> On 30-07-15 08:11, Peter Hutterer wrote:
>>>> The first finger is accurate, it's just the second finger that is imprecise,
>>>> so we can't handle it as a true touch. Instead, revert the device back to
>>>> being a single-touch touchpad and use the fake touch bits for second finger
>>>> handling.
>>>>
>>>> Two-finger scrolling thus becomes usable though we will lose out on
>>>> other features like thumb detection. Useful scrolling trumps that though.
>>>
>>> I think we should wait with doing this until the scroll situation is more
>>> clear, see my last comment here:
>>>
>>> https://bugzilla.redhat.com/show_bug.cgi?id=1235175
>>
>> Ok so Benjamin responded in the bug report, these touchpads
>> report accurate single touch data in the old-style single-touch
>> events, and inaccurate data in the mt events when 2 fingers are down.
>>
>> So this patch which switches things over to using st coordinates
>> is the right thing todo. Review inline.
>>
>>
>>>> Signed-off-by: Peter Hutterer <peter.hutterer at who-t.net>
>>>> ---
>>>>   src/evdev-mt-touchpad.c | 19 ++++++++++++++-----
>>>>   test/touchpad-tap.c     |  3 +++
>>>>   test/touchpad.c         |  9 +++------
>>>>   3 files changed, 20 insertions(+), 11 deletions(-)
>>>>
>>>> diff --git a/src/evdev-mt-touchpad.c b/src/evdev-mt-touchpad.c
>>>> index af1cd47..54ea819 100644
>>>> --- a/src/evdev-mt-touchpad.c
>>>> +++ b/src/evdev-mt-touchpad.c
>>>> @@ -1465,6 +1465,19 @@ tp_init_slots(struct tp_dispatch *tp,
>>>>
>>>>       tp->semi_mt = libevdev_has_property(device->evdev, INPUT_PROP_SEMI_MT);
>>>>
>>>> +    /* This device has a terrible 2fg resolution, but only for the
>>>> +     * second finger, causing scroll jumps when we use the touch points
>>>> +     * properly. The first finger resolution is accurate though, so
>>>> +     * we simply pretend it's a single touch touchpad with the BTN_TOOL
>>>> +     * bits.
>>>> +     */
>>
>> This comment is wrong, the old-style st coordinates report accurate
>> data where as the mt data reports inaccurate data for both fingers
>> when 2 fingers are down. What is happening is that when a single finger
>> is down the st data is copied over to the mt data slot 0, so as to
>> have something to report to user-space which only listens to mt events,
>> and as soon as a second finger comes down the driver starts reporting
>> the bounding box limits in the mt data, making *both* slots inaccurate.
>>
>> So it is not "only for the second finger" nor is "The first finger
>> resolution is accurate" really accurate to say :) The first finger as
>> reported in the st events is accurate, the first finger reported in the
>> mt events is still no good, otherwise we would not need this commit at
>> all.
>
> ok, I've amended the comment to:
>
> +       /* This device has a terrible resolution when two fingers are down,
> +        * causing scroll jumps. The single-touch emulation ABS_X/Y is
> +        * accurate but the ABS_MT_POSITION touchpoints report the bounding
> +        * box and that causes jumps.  So we simply pretend it's a single
> +        * touch touchpad with the BTN_TOOL bits.
> +        * See https://bugzilla.redhat.com/show_bug.cgi?id=1235175 for an
> +        * explanation.
>
> I hope that's close enough now, let me know if you want me to amend this
> further. Worst case, the bugzilla link has the more detailed comments :)

That is fine, thanks.

Regards,

Hans