If a device has multiple capabilities, has_button is imprecise. A device with
tablet and pointer capability for example may have BTN_LEFT on the pointer
interface but not on the tablet interface.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Use ID_INPUT_FOO to assume a device is a FOO, don't decide ourselves based on
whatever bits are available. This moves the categorization out to udev's
input_id builtin by default and other bits that tag the device. libwacom tags
all known devices as ID_INPUT_TABLET and (for touch-enabled ones)
ID_INPUT_TOUCH - we can re-use that knowledge then.
Ignore anything that doesn't have ID_INPUT set, this provides for an easy way
of making devices "invisible" to libinput.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
udev already tags the devices by opening each of them and analyzing their
features. We are basically re-doing this in libinput.
The advantage of udev tags over the plain heuristic from libinput is that
users (or driver writers) can force some tags that are not detected by
common rules. For instance, the pad part of the Wacom tablets is difficult
to discriminate from a joystick or a pointer.
For now we tread INPUT_ID_KEY and INPUT_ID_KEYBOARD as equivalent. It may
become necessary to separate them later.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
Devices like Wacom tablets have multiple event nodes (touch, pad and stylus).
This requires some logical grouping, e.g. setting an Intuos 5 tablet
left-handed effectively turns it upside down. That then applies to both the
stylus and the touch device.
Merging the devices into one struct libinput_device is not feasable, it
complicates the API for little benefit. A caller would still need access to
all subdevices to get udev handles, etc. Some configuration options apply to
the whole device (left-handed) but some (may) only apply to a single subdevice
(calibration, natural scrolling).
Addressing this would make the libinput API unwieldly and hard to use.
Instead, add a device group concept. Each device is a member of a device
group - a singleton for most devices. Wacom tablets will have a single group
across multiple devices, allowing the caller to associate the devices together
if needed.
The API is intentionally very simple and requires the caller to keep track of
groups and which/how many devices are in it. The caller has more powerful
libraries available to do that than we have.
This patch does not address the actual merging of devices into the same
device group, it simply creates a new group for each new device.
[rebased on top of 0.10]
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
When using libinput with xf86-input-libinput, the device speed is
represented as a float passed via X properties.
If a buggy client gives a broken value, the conversions that occur
can cause the value of speed to be NaN (not a number), aka infinity.
In C, any comparison with NaN always gives false, whatever the value.
So that test in libinput_device_config_accel_set_speed():
(speed < 1.0 || speed > 1.0)
will necessarily return FALSE, defeating the test of range.
However, since since any comparison with NaN is false, the
opposite assert() in accelerator_set_speed():
(speed >= 1.0 && speed <= 1.0)
will be false as well, thus triggering the abort() and the crash of
the entire X server along with it.
The solution is to use the same construct in both routines, so that
it fails gracefully in libinput_device_config_accel_set_speed().
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Coverity pointed these out, they're false positives but mark them with
comments to make it obvious to the reader.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
When creating uinput devices, we get the devnode from the kernel directly
rather than through udev. When we add this to the path backend too quickly the
udev_device we get may not be fully initialized and properties may be missing.
This causes false test results.
Avoid this by making sure the handle we have is initialized. This should never
trigger on a real device anyway, even creating a device through litest is slow
enough to avoid this issue. Only affected are the tests in misc.c where we
create the uinput device directly.
Nonetheless, handle this for the generic case so we don't run into heisenbugs
later.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Note: touchpads have a different backend, we never get here in that case. This
only applies to true absolute pointer devices.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
Don't rely on a magic version tag, instead let a device define a udev rule and
drop that into the udev runtime directory before the device is created.
There are a couple of caveats with this approach: first, since this changes
system-wide state it may cause issues on the device the test suite is run on.
This can be avoided if the udev rules have filter patterns that ensure only
test devices are affected.
Second, the check test suite aborts but it doesn't run the teardown() function
if a test fails. So far this wasn't a problem since uinput devices disappear
whenever we exit. The rules files will hang around though, so an unchecked
fixture was added to delete all litest-foo.rules files before and after a test
case starts. Unchecked fixtures are run regardless of the exit status of the
test but run in the same address space - i.e. no ck_assert() usage.
Also unchecked fixtures are only run once per test-case, not once per test
function. For us, that means they're only run once per device (we use the
devices as test case), i.e. if a test fails and the udev rule isn't tidied up,
the next test may be unpredictable. This shouldn't matter too much though.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
The laptops on this series have the physical trackpoint buttons back but
wired them up to the touchpad instead of the trackpoint device and they appear
as BTN_0, BTN_1 and BTN_2 for left, right, middle.
The udev hwdb marks these for us with the TOUCHPAD_HAS_TRACKPOINT_BUTTONS tag
[1]. Use that tag to identify them and re-route the events through the
trackstick device after mangling the event codes to represent the actual
buttons.
[1] http://cgit.freedesktop.org/systemd/systemd/tree/hwdb/70-touchpad.hwdb
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
After switching my main workstation over to using xf86-input-libinput, I
noticed that the multi-media keys like play/pause on my keyboard no longer
worked.
It turns out that the second hid interface on my keyboard which has the
multimedia-keys, also declares having: BTN_BASE6 and BTN_MODE which both
fell into the range we were using to test for something being a joystick.
The commit makes our joystick test mode strict, restoring functionality
of the multi-media keys on the keyboard in question.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Allow switching between softbuttons and clickfinger on any mt-capable
clickpad.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
[hdegoede@redhat.com] Keep top softbuttons working when enabling clickfinger
[hdegoede@redhat.com] Simply touchpad click method switching
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Two methods are provided:
* button area - used on most clickpads, a click with a touch within a given
area generates left/middle/right clicks
* clickfinger - used on apple touchpads, a click with 1/2/3 fingers on the
touchpad generates a left, right, middle click
Both methods already exist in the touchpad code, this is just the
configuration interface.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Allow the center of pinned fingers to drift over time, to avoid accidentally
unpinning fingers.
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=86807
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Some touchpads provide touch information while the finger hovers over the
touchpad, i.e. before BTN_TOUCH. Add a touch state for those touchpads so we
can ignore the touches until they actually start.
The approach is now: instead of BEGIN we mark a new touch as HOVERING.
Use the BTN_TOOL_FINGER, BTN_TOOL_DOUBLETAP information during
tp_process_state() to mark any touches that are hovering as down or ended.
i.e. provided BTN_TOUCH is down: if BTN_TOOL_FINGER is down, one hovering
touch gets marked as down, if DOUBLETAP is down, two touches are marked as
down, etc.
When ending touches, switch them back into HOVERING if the BTN_TOOL_FINGER
is still set, otherwise end them properly.
https://bugs.freedesktop.org/show_bug.cgi?id=87197
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
BTN_TOOL_FINGER, DOUBLETAP, etc. are mutually exclusive in the kernel, so we
can use ffs here instead of manually counting.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
We need this for determining hovering touches on some semi-mt touchpads.
This makes the fake_touches mask use bit 0 for BTN_TOUCH, and the other bits
for BTN_TOOL_FINGER, BTN_TOOL_DOUBLETAP, etc. BTN_TOUCH is independent of the
rest, the others are mutually exclusive in the kernel.
Since the mask isn't a straightforward bitmask anymore, abstract it all
through helper functions.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Some devices require more than just flipping around the buttons, such as
tablets.
When it comes to devices like tablets, because the position of the palm rest is
on the right, the entire tablet has to be flipped around in order to be usable
by lefties. As such, this requires that we reverse the coordinates of the
tablets in addition to flipping the buttons on the tablet. As such, renaming
these functions so that they aren't specific to devices where only the buttons
are flipped seems appropriate.
Signed-off-by: Stephen Chandler Paul <thatslyude@gmail.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is merged on top of the wheel normalization patches. Those introduced an
axis source and an extra "discrete" value to the various internal and external
APIs. This branch changed from a single value to passing dx/dy into all scroll
events.
The conflicts are to change everything to take x, y, x_discrete, y_discrete as
values (and the source axis mask of course).
Conflicts:
src/evdev-mt-touchpad-edge-scroll.c
src/evdev.c
src/libinput-private.h
src/libinput.c
The recent normalization of wheel events means we get the angle in degrees but
we don't know how this corresponds to clicks. The M325 has a 20 degree click
angle, most other mice have 15 degrees. So an angle of 60 can be 3 or 4 click
events.
Most clients care more about the click count than the angle on a mouse wheel.
Provide that value when needed.
Adding a discrete value to the axis event leaves the possibility of defining
discrete units for finger/continuous scroll sources in the future. Right now,
these will always reuturn 0.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Sending separate axis events instead of one unified events is limiting,
especially when simultaneously scrolling in both directions and the caller
tries to implement kinetic scrolling.
Take a page from the tablet-support branch and instead implement the axis
event as a generic event that can contain multiple axes simultaneously.
Right now we only have two (scroll) axes and we could easily just check both
for non-zero values. If we want to allow further axes in the future, we need
a check whether an axis is set in an event, that's what
libinput_event_pointer_has_axis to scroll events() is for.
We also need the mask to notify of a scroll stop event, which could otherwise
be confused as a vertical-only or horizontal-only event.
This is an API and ABI break.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Similar to the mouse resolution, let's make the scroll distance a sensible
predictable value. Most mice use a 15 degree angle per scroll click, so let's
change to that. This will alter behaviour in clients that expect 10.
We return doubles for the axis value, so that leaves the option of
really fine-grained step sizes in the future.
We currently assume all mice have 15 degree angles. Like the DPI settings, it
will require a udev property to be set. Patch for that to follow.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
prop isn't the full property line, just the value.
And document the return value too while we're at it.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Discreet means to not draw attention.
Discrete means non-continuous.
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
For a caller to implement/provide kinetic scrolling ("inertial scrolling",
"fling scrolling"), it needs to know how the scrolling motion was implemented,
and what to expect in the future. Add this information to the pointer axis
event.
The three scroll sources we have are:
* wheels: scrolling is in discreet steps, you don't know when it ends, the
wheel will just stop sending events
* fingers: scrolling is continuous coordinate space, we know when it stops and
we can tell the caller
* continuous: scrolling is in continuous coordinate space but we may or may not
know when it stops. if scroll lock is used, the device may never technically
get out of scroll mode even if it doesn't send events at any given moment
Use case: trackpoint/trackball scroll emulation on button press
The stop event is now codified in the API documentation, so callers can use
that for kinetic scrolling. libinput does not implement kinetic scrolling
itself.
Not covered by this patch:
* The wheel event is currently defined as "typical mouse wheel step", this is
different to Qt where the step value is 1/8 of a degree. Some better
definition here may help.
* It is unclear how an absolute device would map into relative motion if the
device itself is not controlling absolute motion.
* For diagonal scrolling, the vertical/horizontal terminator events would come
in separately. The caller would have to deal with that somehow.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Original patch, before the rebase onto today's master:
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Previously, the pointer could only be passed into the context on creation
time and was immutable after that.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Add a boolean state machine for two-finger scrolling so we know when we're
currently scrolling. If we were scrolling and it stops, pick the active
touch as pointer touch so we can go back to pointer movement without having to
lift the finger off the touchpad.
https://bugs.freedesktop.org/show_bug.cgi?id=86807
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
