The hysteresis cuts the first pointer motion by the hysteresis margin. On some
touchpads this causes the tests to fail when the motion history length is
reduced (future patch). Allow the first event to be smaller than the expected
minimum.
This doesn't trigger in current tests because the hysteresis is per-event and
by the time we get past the minimum 4 events to move the pointer, we're
already flying unaffected by the hysteresis.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
This doesn't have an effect in our current tests because the touchpad always
needs 4 motion events to get moving. But for the future, it simplifies the
case of "i want to move between x1/y1 and x2/y2", because it fills in only the
events in between rather than re-using the touch down coordinates and thus not
causing a motion on the first event.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Not all mice have a click angle with integer degrees. The new
MOUSE_WHEEL_CLICK_COUNT property specifies how many clicks per full rotation,
the angle can be calculated from that.
See https://github.com/systemd/systemd/pull/4440 for more information
CLICK_COUNT overrides CLICK_ANGLE, so we check for the former first and then
fall back to the angle if need be. No changes to the user-facing API.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Fixes the respective clang warnings
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Touch sequences are interrupted by TOUCH_FRAME events which makes them
annoying to handle event-by-event.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
This prevents any tests from being added but not run in the normal setup. But
as soon as filters are manually specified on the list proceed anyway.
Otherwise it's impossible to run specific sets of tests, e.g. things like
running all tests applicable to a specific device with
--filter-device=foo
Now that all tests are in the same binary we are guaranteed that at least some
tests don't apply, so the above was guaranteed to abort.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Don't fork by default if we're in gdb.
Note that is_debugger_attached() is now inside #ifndef LITEST_NO_MAIN, gdb for
the litest selftest will now require a manual CK_FORK=no.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Add a make-like -j/--jobs option to split the number of parallel test
processes. Defaults to 8 if not specified, future patches will default this to
1 for special cases where filters are specified or gdb is detected.
Each subprocess overwrites argv[0] to be easier identifiable in the ps
output when we're trying to figure out which tests are still running.
A -j1 is equivalent to the previous functionality, i.e. we don't fork.
One quirk needed for check: any test case not part of a test runner will not
be freed and thus triggers valgrind. We do test filtering by splitting
up the tests across multiple forks (i.e. each process has several tests that
are in the list but not added to the runner). Thus we need to mark those we
expect check to free as used.
Then on cleanup we traverse the test list, add all the unused one to a
test runner and free that test runner (without actually running it). This
cleans up both the filtered tests in each subprocess and the whole test list
in the parent process which doesn't run a test itself.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We can't call system() in the signal handler but we are allowed to fork. Do
that, update the hwdb and immediately exit the child again.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The udev hwdb takes about 200ms and we still trigger it on each device. The
udev rules don't actually change after compiling, so simply create them
once and remove them after the test run.
For multiple test binaries this needed to be synchronized (which is hard),
hence the previous merge into a single binary for all tests.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Call it a libinput-test-suite-runner, in subsequent patches we'll handle doing
parallel tests ourselves instead of relying on automake features.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Easier to clean up than knowing all the destination paths we'll install.
Only affects global udev rules so far.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
10s is not enough when running the test suite in parallel as any test may have
to wait longer than that to get access to the udev lock. Especially for
tests with multiple timeouts it was too easy to trigger timeouts.
Up the timeout to 30s, this seems reliable enough now.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
litest_add_device and litest_delete_device trigger a udev rule reload. This
messes with some test devices and when we run multiple tests in parallel we
get weird errors like "keyboard $BLAH failed the touchpad sanity test".
Still not 100% reliable to run tests in parallel, but it's vastly improved
now.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
If the first device we got didn't have the expected syspath we'd leak the
device and cause the valgrind tests to fail.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
Reviewed-by: Carlos Garnacho <carlosg@gnome.org>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
Reviewed-by: Carlos Garnacho <carlosg@gnome.org>
Move mode control to libinput. This reduces some flexibility on what we can do
with modes but makes it a lot easier for anyone to implement modes correctly
and have the LEDs apply appropriately, etc. Let's go with the option to make
the 95% use-case easy. Note: whether the mode is actually used is up to the
caller, e.g. under Windows and OS X the mode only applies to the
rings/strips, not the buttons.
A tablet pad has 1 or more mode groups, all buttons/ring/strips are assigned
to a mode group. That group has a numeric mode index and is hooked to the
LEDs. libinput will switch the LEDs accordingly.
The mode group is a separate object. This allows for better APIs when it comes
to:
* checking whether a button/ring/strip is part of a mode group
* checking whether a button will trigger a mode transition
and in the future potentially:
* checking which mode transition will happen
* setting which button should change the mode transition
* changing what type of mode transition should happen.
* moving a button from one mode group to the other
This patch adds the basic scaffolding, without any real implementation.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Proofread-by: Yong Bakos <ybakos@humanoriented.com>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
Reviewed-by: Carlos Garnacho <carlosg@gnome.org>
We will reinstate the hysteresis for all devices making the negative
pressure check unncessary.
This reverts commit ef48c07a96.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
This device has a touchpad on the mouse but it's labeled as mouse. For litest
we only label it as LITEST_MOUSE feature and test the touchpad directly on the
device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Otherwise the abs->value could lie outside the [min, max] range of the axis.
This isn't much of an issue for actual axes but in the case of ABS_MT_SLOT
(value 47) it causes errors when libevdev sanitises the event into the allowed
slot range.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
If we have libevdev 1.5 or later, the resolution is already set, no need to
change it again. Let's rely on libevdev to do the right thing and simply skip
the rest if we have one correct nonzero resolution already set on the device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
Reviewed-by: Carlos Garnacho <carlosg@gnome.org>
This interface handles the buttons on the physical tablet itself, including
the touch ring and the strip.
A notable difference to other libinput interfaces here is that we do not use
linux/input.h event codes for buttons. Instead, the buttons are merely
numbered sequentially, starting at button 1. This means:
* the API is different, instead of get_button() we have get_button_number() to
drive the point home
* there is no seat button count. pads are inherently different devices and
compositors should treat them as such. The seat button count makes sense
when you want to know how many devices have BTN_LEFT down, but it makes no
sense for buttons where all the semantics are handled by the compositor
anyway.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
Reviewed-by: Carlos Garnacho <carlosg@gnome.org>
If a touchscreen has a fuzz value use it for motion hysteresis similar to how
we do it for a touchpad. This stops pointer wobbles as seen in
https://bugs.freedesktop.org/show_bug.cgi?id=94918
It's up to the system to override or set the kernel's fuzz value correctly,
i.e. a udev hwdb entry is required where the kernel driver does not set it.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jonas Ådahl <jadahl@gmail.com>
This path is hit when we set up an abs device without setting specific axis
ranges. Usually because we only care that the axis is there, not the values,
see the tests of invalid devices.
0xffff is fine as max for most axes but setting it for ABS_MT_SLOT can cause
allocation errors. Reduce to 100 because we don't care about the range anyway
and 100 is still big enough for basic tests.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Resetting the motion history has the side-effect of swallowing movements, we
don't calculate deltas until we have 4 motion events. During a finger release,
we're likely to get a large pressure change between two events, resetting the
motion history prevents the cursor from jumping on release.
The value of 7 found by trial-and-error, tested on the T440 and T450 hardware.
The absolute value is highly variable but recordings show that the pressure
changes only by 1 or 2 units during normal interaction. Higher pressure
changes are during finger position changes but since those should not cause a
jump anyway, we tend to win there too.
Currently only enabled for negative pressure changes, let's see how we go with
that.
https://bugs.freedesktop.org/show_bug.cgi?id=94379
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
One test now gets close enough to 100% of the axis value that
ck_assert_double_lt() is not fine-grained enough. Switch to a direct x < 100.0
check.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If a USB keyboard like the YubiKey is found before the internal keyboard, it
will be paired with the touchpad when it is seen. The internal keyboard is
seen later bug ignored because we already have a keyboard paired with the
touchpad.
This is obviously wrong. For now, give priority to serio keyboards, and
override existing dwt pairings with the new keyboard.
https://bugs.freedesktop.org/show_bug.cgi?id=93983
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
This was the reason for the valgrind test case failures whenever we
accumulated too many tests (see 9c2afae1 and 2a110104). The cause was simply
that we ran out of fds which caused libevdev to fail the scandir() searching
for the event node. That resulted in a NULL devnode and an abort in litest.
Close the fd before freeing the evdev device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This button sends a release N, press N+1 on each press, cycling through the
three event codes supported. This causes a stuck button since the current mode
is never released.
Long-term this better served by a set of switches that toggle accordingly, for
now disable the button codes.
https://bugs.freedesktop.org/show_bug.cgi?id=92127
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
More accurate representation of what we actually want to do. Plus it avoids
weird test case failures in semi-mt where we always pick the t/l and b/r
touches for the bounding box. That is the proper behavior for semi-mt, but
it's not for the tests where we expect simultaneous finger movement.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
A fake MT device may have ABS_MT_POSITION_X but not Y. In this case we don't
care, because we don't handle those axes anyway.
http://bugs.freedesktop.org/show_bug.cgi?id=93474
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
