The backend destroy signal is emitted before the output_remove
signal is. When the destroy signal is emitted listeners remove
their output_remove listener, so the output_remove signal is never
received and listeners have an invalid output pointer.
The correct way to solve this would be to remove the output_remove
signal completely and use the wlr_output.events.destroy signal
instead. This isn't yet possible because wl_signal_emit is unsafe
and listeners cannot be removed in listeners.
We now use doubles until the last minute, which makes it so we can move
the pointer more precisely. This also includes a fix for tablet tools,
which move absolutely and sometimes do not update the X or Y axis.
In pointer.c, some axis event was emitted even if the event pointer did not have
current axis.
In X11 backend pointer scroll events seem to be composed of both BUTTON_PRESS
and BUTTON_RELEASE. Therefore we should skip one of them (RELEASE) to avoid
event duplication.
This runs through events pending at init on initialization so we can
tell if some devices are available.
Note that with the way wlr_device_lists is managed, this checks that
there is at least one device we handle - it doesn't have to be a
keyboard, but there is at least a mouse or tablet_pad or something
that we care about.
Instead of failing inconditionally it might be better to leave the
decision to the user, e.g. add a "backend_has_devices" function to
call later.
(Tested by moving /dev/input off)
Fixes#24.
- 'libinput' (backend's) to libinput_context
- 'device' (libinput_device) to libinput_dev
- 'dev' (wlr_device) to wlr_dev
- 'devices' lists tangling of libinput devices to wlr_devices
- 'devices' list of wlr_devices in backend state to wlr_device_lists
