Unlike virtually everything else, the tablet tool was processed at the time
the event was read rather than when the subsequent EV_SYN came in. This causes
difficulties with tablets that send the wrong BTN_TOOL_PEN events.
Moving the tool change processing to tablet_flush() makes the injection of the
BTN_TOOL_PEN event a lot easier, simply flipping the matching bit does the
job. It also makes it easier to ignore duplicate tool updates like we've seen
in #259.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Recent Apple touchpads use a proper Bluetooth vendor ID assigned to Apple instead of the USB one,
so this code would have to check for two vendor IDs and their udev types. However, we already
have that matching done via models in quirks, so let's just use that.
Signed-off-by: Sebastian Krzyszkowiak <dos@dosowisko.net>
Also, set a default AttrTouchSizeRange for Apple touchpads via Bluetooth
to match the one from the USB rule.
Signed-off-by: Sebastian Krzyszkowiak <dos@dosowisko.net>
And fix the cases where the default value isn't filled in correctly
Issue found because of the following ubsan error:
../src/evdev-tablet.c:182:19: runtime error: signed integer overflow: 0 - -214783648 cannot be represented in type 'int'
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
A few leaks in the test code were found when running linput-test-suite
with the -fsanitize=address option enabled. Clean up these leaks so that
we can more clearly see real issues.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Running libinput-test-suite with -fsanitize=undefined highlights the two
following errors. Force C to realize we want an unsigned result by making
the '1' literal unsigned.
../src/evdev-fallback.c:314:22 runtime error: left shift of 1 by 31 places cannot be represented in type 'int'
../src/evdev-fallback.c:377:24 runtime error: left shift of 1 by 31 places cannot be represented in type 'int'
v2: use bit() instead of manual shift 1U<<1
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Make sure we check the expected sequence more stringent and change the x/y
coordinates on prox in so the kernel doesn't filter them.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The test device sent a serial of 0. That would end up creating a new tool in
libinput which is wrong. Let's hope this was just an error in creating the
test device, if the device really sends that sequence, we're in trouble.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
With the previous code we'd set both tools simultaneously which isn't allowed.
It only worked because the second tool set was the one we cared about.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Just use the wait() timeout directly instead of sleep and kill. This allows us
to have a longer timeout and still get fast handling where the tool
immediately exits, but less failure when running on busy machines.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Don't require a quirk update, just enable this by default for all tablets. If
we get a proximity out event at the right time, the quirk is disabled for that
tablet for the rest of its lifetime. And it's virtually impossible to have a
false positive here anyway - you cannot hold the pen still enough to not
trigger events for 50ms.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We expect the kernel to transition properly for us, e.g. BTN_TOOL_PEN goes to
0, BTN_TOOL_ERASER goes to 1. Two cases have surfaced recently where this
doesn't happen and debugging this takes time - so let's warn about it to make
it obvious.
Example 1: https://github.com/linuxwacom/libwacom/issues/70
Example 2: https://gitlab.freedesktop.org/libinput/libinput/issues/259
This is just a warning, nothing more. We should just handle that case
accordingly but that requires more effort.
Fixes#260
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reduces the size of the group name in the title.
Removes the background image from the navpath where it appears (file reference
page for example).
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
right now the check_if_older_than_a_week rule does (in pseudo-code):
- get timestamp of current image or 0
- get timestamp of upstream image or 0
- if upstream image is newer than current image
copy upstream image into current
- if we are in a scheduled pipeline, or if there is no current image
(timestamp of 0), rebuild the current image
The ci-templates if-not-exists rule does:
- if there is a current image, exit
- if there is an upstream image, copy it to current and exit
- rebuild
Having the following is equivalent to the current behaviour and
can be used instead of check_if_older_than_a_week:
- if there is an upstream image, copy it to current and exit
- if there is a current image, exit
- rebuild
Because what matters is:
forks should be running the upstream image if available
forks should be running the latest upstream image in the libinput case
forks should be able to rebuild the images if there is no upstream
(change of the image tag)
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
We better not rebuild the image in regular operations unless there is a
strong need for it.
We can however set up a scheduled pipeline to rebuild the images once
a week or once a month in the upstream repo, and the forks will fetch
those new images when they need.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
We should rely on the provided tag, not latest.
Move the clean stage at the end, there is no point in running it at
the beginning.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
Now everybody gets to rebuild their own containers if there is a change
(too old or change in the packages). This should allow the MR touching
the package list to succeed.
Removal of the container_check stage, we can just have this in a
before_script.
Removal of the manual creation of the containers, not needed as we better
just increment <DISTRO>_TAG.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
We want to have a common repo for the containers templates.
So we can reuse the produced image from this repo and remove our custom
bootstrap image.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
This triggers on Fedora 30, even though skip skip the tools options test when
running under valgrind.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Not sure how this ever worked correctly: a process terminated by a signal has
the negative signo as return code. This would apply to every debug-events and
debug-gui test because they have to get killed by a signal. This failed
occasionally, presumably a race with the GTK startup/signal handler/whatever.
Fix this by a) sending SIGQUIT because that won't get handled by the tools and
b) prending that if we get a -SIGQUIT exit code, everything is fine.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Pass arguments we don't handle directly through to the unittest module. This
way we can filter tests with -k testname etc.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Still not great and probably makes any professional designer's eyes bleed, but
at least it's more readable now.
Changes:
- spacing after param name so they don't cuddle up with the description
- color changes and background image removals to drop the doxygen default look
- font size changes to not make things overrun
- font family change to make the function prototypes readable
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Now that we're emulating everything correctly, let's mark it as proper touch
device.
Two test cases need to be excluded:
- double-down triggers an assert in the test device because this isn't
possible this way with protocol A devices
- the axisrange warning test can't be triggered, mtdev clips those axes
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
