Our file format is static enough that we don't need to use push/pop, we know
exactly which line is going where. So let's replace it with a static
indent instead.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is prep work to be more consistent with the use of tempfile later for
individual devices.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
With a new helper function strv_from_argv we can re-use the device opening
loop for all the use-cases we have.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
EditorConfig helps maintain consistent coding styles for multiple developers
working on the same project across various editors and IDEs:
https://editorconfig.org/
Signed-off-by: José Expósito <jose.exposito89@gmail.com>
The current fallback_dispatch wheel struct, a device_coords, doesn't allow to
save extra information.
The new anonymous struct will allow to add a is_inhibited field to disable mouse
scroll while the middle button is pressed and, potentially, any required extra
state in the future.
Signed-off-by: José Expósito <jose.exposito89@gmail.com>
The FreeBSD HID stack adds the device type to the evdev name,
so we get e.g. "ACPI0C50:00 18D1:5028 TouchPad".
(Maybe this shouldn't be matched by name at all though...)
Signed-off-by: Greg V <greg@unrelenting.technology>
FreeBSD does not use Linux modaliases, so we have to generate these strings.
Unfortunately for us, the data in kenv has the chassis type pre-parsed into
a nice string, so we have to match these strings back into numbers.
Only relevant types are included to avoid bloating the code.
Signed-off-by: Greg V <greg@unrelenting.technology>
The Linux applespi driver currently uses the Synaptics vendor ID
on the trackpad for some reason (even though, at least from bcm5974
we only know that Broadcom is involved..) but my upcoming FreeBSD driver
uses the Apple vendor ID everywhere, so add two quirks.
Signed-off-by: Greg V <greg@unrelenting.technology>
Apple MacBooks (Broadwell/Skylake/Kaby Lake and Apple Silicon)
use SPI to communicate with the keyboard and trackpad.
Signed-off-by: Greg V <greg@unrelenting.technology>
Make sure the unaccelerated deltas are comparable to scroll deltas.
edit by whot:
The original intention of the unaccelerated motion data here was to provide
both accelerated and unaccelerated motion for gestures so it was possible to
have 1:1 mapping from gesture motion to screen activity.
Normalizing to 1000dpi this way would've worked for mice but touchpad
acceleration also includes the TP_MAGIC_SLOWDOWN (amongst other tricks) which
slows down motion to around 27% *before* applying the acceleration function.
On a 1000dpi touchpad (~40 units/mm) simply normalizing touchpad motion to
1000dpi results in pointer motion that is way too fast, it's lacking that
slowdown to 27% of original speed.
This results in the accelerated and unaccelerated gesture data being in
effectively two different coordinate systems with the caller having no ability
to relate the two.
Switching to the special constant acceleration applies that slowdown and
matches the data to the part of the acceleration curve where no (additional)
acceleration is applied.
It makes the gesture unaccelerated data comparable to the accelerated data
and to scroll data which uses the same process.
Fixes#582
Signed-off-by: Alexander Mikhaylenko <alexm@gnome.org>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Without this, each finger movement happens in a different evdev event frame.
Since we average deltas for gestures, this messes with the expected data.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
There's no test case where we need to do something immediately after the last
event so we might as well do everything in the same loop.
This also fixes a bug where the first movement would usually get swallowed.
Test cases in general put the finger down at x/y, then move them to some other
position. We'd expect the first event in a loop to happen at x+n/y+n, not at
x/y again.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
See d6e5313497 for confirmation that the
threshold is intended to be in mm/s, the comment here is simply a leftover from
earlier times when the acceleration method was using device-units only.
Fixes#585
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The way touchpads (generally) work is that they get the position of each
finger on each scanout. The kernel filters touches that haven't moved to
reduce bandwidth so any touch that is logically down that we don't see an
update for is in the same position as during the last scanout.
Previously, touches that didn't sent events were effectively ignored, causing
our jump detection to fail:
- time t0: touch moves to position x/y, motion history time is set to t0
- time t1..t5: touch remains at position for several frames, no updates to the
motion history
- time t6: touch jumps to position x+a/y+b
- tp_detect_jumps() sees the last update time is t0 which is too long ago
and exits without detecting a jump
This is fixed by pushing to the motion history any time we have *any* update -
if the touchpad notices a state change on any touch update all touches with
their current position, whether it changed or not.
This obsoletes the `time` field in the tp_touch struct, most of this patch is
passing down the current time to the few users of t->time.
Fixes#578
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Incorrect comment, the purpose of this test was to ensure that an unused slot
doesn't affect how other touches are treated, see commit 928bad9.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Got papered over by bugs in the implementation and didn't trigger the jump
detection or movement detection otherwise.
Related to #578
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Add a second grey v-shaped (upside down triangle) pointer that moves around
with the unaccelerated deltas. This makes it easier to visualize how the
unaccelerated pointer moves around, the snake helps for some use-cases but not
all of them.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The assignment of zero is done to work around false-positives of
coverity about uninitialized variable usage. Getting rid of it inside
the macro will allow in later commit to declare a variable inside
`for-loop` rather than outside of it.
Do it by declaring a new list_first_entry_by_type helper which accepts a
type rather than a variable.
Signed-off-by: Konstantin Kharlamov <Hi-Angel@yandex.ru>
This makes it easier to visualize changes in various axes or key states that
should not be there, doubly so for long recordings.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Just like the other python-based tools it's just a basename copy, so let's be
consistent here and have all tools perform that way.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Using poll means more difficult fd management, epoll (together with am
modified version of the libinput_sources) makes this a lot easier by simply
using dispatch.
This means we are no longer reliant on a specific file descriptor order in the
poll array.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
So we don't need to worry about the libgen.h include game.
And we can switch trunkname over to that, making it a bit simpler.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If we have no events in any of the recorded devices, state that this is the
case and make Enter simply quit instead of a pointless while loop.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
