We need to store all touchpoint positions so that if we just get an
ABS_MT_POSITION_X or Y event, we can pull the other coordinate from the
cache. And we need this across invocations of evdev_input_device_data(),
so the accumulator approach doesn't work.
Instead, we go back to the approach of storing all this state in the
evdev device struct and we might as well just move the rel and abs state
there too.
This adds ABS_MT_* support for direct touch devices and notifies
the compositor. The compositor has a stub for now.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
fade_output() is strange in that it manufactures a wlsc_surface object
by hand, and then calls wlsc_surface_draw() on it.
Valgrind complained, that wlsc_surface_draw() accesses uninitialised
data: wlsc_surface::alpha. fade_output() forgets to set it.
Initialise surface.alpha in fade_output(). Specifically, set it to
compositor->current_alpha to deliberatly avoid the gluniform1f() call in
wlsc_surface_draw().
fade_output() binds a different GL shader program than
wlsc_surface_draw() expects. This program does not have a uniform called
"alpha", and the uniform location given in glUniform1f() is not for
this program anyway. A hint of that is the runtime error:
Mesa: User error: GL_INVALID_OPERATION in glUniform(type mismatch)
Fixing this seems to get rid of half a thousand of Valgrind errors, and
of course the Mesa user error.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Not sure why we get these, but it happens for Alt-click to move a window
(metacity binding) and messes up the idle inhibit counter.
FocusOut event as a result of ungrabbing doesn't really make sense and
fortunately we can safely ignore them.
Besides the new header file, there's also a change in the main evdev creation
procedure for a more suggestive name (evdev_input_add_devices ->
evdev_input_create). There's no real functional changes in this commit.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
I caught this when an evdev device fd was trying to trigger the main event
loop, which was already free'd and causing an invalid read.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
Was causing an invalid read when the output is in fact destroyed. That's only
visible (segfault on my machine) on drm compositor because it's the only
backend trying to finish correct the compositor.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
Activate the toplevel, fullscrren and menu surfaces during mapping,
so that the launched applications can get the keyboard focus without
clicking on that window.
Move idle_time variable to struct wlsc_compositor, so that a shell
plugin can change it. Also store the original value from the command
line.
Add "duration" option to the desktop-shell screensaver config. This is
the time the screensaver will be visible, after idle timeout triggers
another time and blanks the screen.
Now you can have different delays to lock the screen, and switch off the
screen while a screensaver is running.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Read the same configuration file in the shell plugin (desktop-shell) as
the desktop-shell client does.
Add a new section "screensaver", where "path" defines the path of the
idle animation client to be executed. Not defining "path" disables the
animation.
Idle animations are not in use by default. It must be configured in
wayland-desktop-shell.ini or launched manually.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add WLSC_COMPOSITOR_IDLE state to the possible compositor internal
states, and fix the drm backend to restore the previous state instead of
forcing ACTIVE.
Normally, the compositor only uses the ACTIVE and SLEEPING states. The
IDLE state is another active state, reserved for the shell, when the
shell wants to have unlock() calls on activity, but the compositor cannot
be SLEEPING.
Use the IDLE state to fix exposing the unlock dialog while a screensaver
is animating. Without this fix, is it impossible to activate the unlock
dialog without waiting for a second idle timeout that really puts the
compositor into SLEEPING.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Reorder code in fade_frame() to that if shell->lock() calls
wlsc_compositor_wake(), the fade animation will run again.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Screensavers become visible the first time only after the compositor has
gone to sleep state. Therefore, to see screensaver in the start, wake up
the compositor. After a second idle timeout, the compositor will stay
sleeping.
We could also not apply this patch. It would mean the screensavers would
be visible only with the unlock dialog, and not become visible
automatically.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Implement the basics of screensaver surface management. Exec'ing and
killing the screensaver client is punted for later.
When a surface registered as a screensaver is mapped, it stays hidden
if the screen is not locked, or it is added to the compositor visible
surfaces list if the screen is locked.
The map() is restructured to set initial position first, and stacking
next. This allows SHELL_SURFACE_SCREENSAVER share positioning with
SHELL_SURFACE_FULLSCREEN, while show_screensaver() does its own
wlsc_surface_configure() call.
Also fix centering to the given fullscreen output, not the first output.
Another bug fix: previously configure() would call
wlsc_surface_configure() unconditionally, which assigns an output to
the surface. While the compositor is locked, if an application resizes
its window, we hit configure() and assign an output while the surface is
not in compositor->surface_list. This leads to invalid memory access on
the next call to wlsc_surface_damage_below(). Fix this by not calling
wlsc_surface_configure() for surfaces that are not visible.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
So far nothing prevented a client for registering a surface as one type
and then as another type. With some special types, this would lead to
corrupted wl_lists.
Add a function, that either resets the surface type or posts an error to
the client. In case of an error, the set type operation must be aborted.
Change the type name SHELL_SURFACE_NORMAL to SHELL_SURFACE_NONE, as
there is nothing normal in the "none" type which just means uninitialised.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add the screensaver interface to the desktop-shell protocol file. Also
add stubs for it in the compositor, and make wscreensaver to bind to the
screensaver interface. Wscreensaver gets a new option --demo to retain
the current behaviour as a regular wayland client.
When a screensaver application starts, it should bind to the screensaver
interface, enumerate all outputs, create a surface per output, and
register those surfaces via screensaver::set_surface request. Then it
continues with the usual animation loop, waiting for frame events. The
compositor will decide, when the given screensaver surfaces are
displayed. A screensaver application should respond to outputs coming
and going away by creating and destroying surfaces.
The compositor is supposed to activate a screensaver by exec'ing it, and
stop the screensaver by killing the client process. Only one client may
be bound to the screensaver interface at a time. If there already is a
client, the compositor could either kill it first, or not exec a new
one.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
when a motion is being performed on ts device, only one axis can be sent
through the evdev bytestream whereas the other could be omitted. For instance:
-------------- SYN_REPORT ------------
type 3 (EV_ABS), code 53 (ABS_MT_POSITION_X), value 22208
type 3 (EV_ABS), code 58 (ABS_MT_PRESSURE), value 631
type 3 (EV_ABS), code 0 (ABS_X), value 22208
-------------- SYN_REPORT ------------
on such case we'd have to send the compositor the old value of Y. Commit
f547bd36 introduced this bug cause it was sending zeroed coordinate and not
the old one.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
We do not handle errors of gbm-buffer-creation and drm-mode-setting in
create_output_for_connectors. Correctly catch these now and free memory on error
to avoid memory leaks.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We currently simply return -1 on error in create_output_for_connector. This
correctly frees the output and all modes when we fail to avoid memory leaks.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
We need to correctly free every connector we retrieve. We currently loose them
if they are not connected.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
In the wl_shell_surface migration, I forgot to correct one cast in
shell_surface_set_transient(). 'parent_resource' is not a (struct
wlsc_surface *) but (struct shell_surface *).
This bug corrupts a wlsc_surface::output field, which later (in my
experiments) leads to a segmentation fault in surface_frame().
Fix the casts.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
One less syscall and error path to check, and feels like a cleaner approach.
The commit adds two lines, but that's because we actually handle the
potential error now.
The signal mask is inherited over fork() and exec(), we need to
explicitly reset it.
This allows the children to receive the signals the compositor itself
has blocked, for example SIGINT and SIGTERM.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
shell.c and tablet-shell.c had almost the same code for forking their
special shell client. Generalise this code and put it into
wlsc_client_launch() in compositor.c.
Improve error cleanup and reporting in wlsc_client_launch().
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
(!x < 0) is always false and doesn't make sense here. Looks like a typo so
remove the negation.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Do not allow multiple wl_shell_surface objects to be created for a
wl_surface object.
Multiple shell_surface objects would confuse the compositor as they
contain separate instances of the shell-private data.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Leftovers from an intermediate patch set, the proper function name is
shell_get_shell_surface. Cosmetic change.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
get_shell_surface() returns NULL, if the client has not created or has
destroyed the wl_shell_surface object.
All but one use of get_shell_surface() just retrieve the surface type,
so just fall back to SHELL_SURFACE_NORMAL there.
Resize hot-key binding really needs the wl_shell_surface object, as that
is the only way to send configure events. For surfaces without a
wl_shell_surface, simply do not resize them.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Remove shell_priv member from wlsc_surface, and replace it by a search
through the wl_surface destroy_listener_list.
This technique avoids any "extension" members in the wlsc_surface
structure.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Change desktop-shell protocol to use wl_shell_surface instead of
wl_surface.
Adapt the desktop-shell client and the shell plugin.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Protocol changes in Wayland core introduced a new interface
wl_shell_surface, and moved all wl_shell surface methods into it. Adapt
the compositor and its Wayland backend, shell plugin, and all clients to
the new interface.
Depends on the Wayland core commit "protocol: introduce wl_shell_surface"
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Fix two bugs:
- if there are no backgrounds at all, the background pointer would have
been bogus. Lead to a segfault.
- if the hidden_surface_list is empty, wl_list_insert_list() would
corrupt the list. Lead to a hang in pick_surface().
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Enumerate the different surface purposes for the shell: background, panel,
lock surface, and normal (other) surfaces.
Instead of testing wlsc_surface pointers against known pointers, check
the purpose field.
This change will ease implementing per-output background, panel, etc.
when the number of "known" surface pointers grows dynamically.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add a pointer to wlsc_surface for shell-private data. This is a
temporary solution.
Add struct shell_surface, where you can add any shell-private data
members related to a wlsc_surface. The getter function takes care of
creating the private data if it does not exist yet.
Not used anywhere yet.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
We'll want to enhance later the driver regarding the tool being used, but for
now just remove unused bits.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
Very likely that 2.4 kernels won't be used with Wayland compositor so the
check for signal value is pretty much useless.
It's okay to change e->value inside evdev_process_absolute_motion_touchpad
given it's not used later on, and I also rather not touch this snip because it
will be changed when multi-touch support arrives.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
A copy & paste bug, that resulted setting to NULL something else than
shell->lock_surface when that surface was destroyed.
The symptom: let compositor lock down, unlock it, let it lock down
again, and the unlock dialog is never requested again. This bug was
triggered by the previous fix "shell: fix compositor wakeup while
locked".
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Compositor is locked, woken up, unlock dialog is shown; if the
compositor does to sleep again, before being unlocked, it will never
wake up again, because unlock() becomes a no-op, yet it should wake the
compositor up.
Fix this by letting unlock() to wake up the compositor, if lock surface
is present.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
When the lock surface was map()'d while the compositor was locked,
wlsc_surface_configure() was never called for the lock surface. Hence,
the surface->output was NULL, and the 'frame' event was never sent,
causing desktop-shell to loop in dri2_swap_buffers().
Fix this by calling wlsc_surface_configure() for the lock surface
always in map().
Additionally, adjust the comments in map() to make it more readable.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
This way we can still use surface->link when a surface is not in
the main compositor surface list and don't need the hidden_surface
wrapper object. Also, setting surface->output to NULL will block
the surface frame callback until we put the surface back into the
main list. This has the effect of blocking animations while a surface
isn't visible.
When the compositor is locked, all surfaces are moved from the
compositor's list to a private list in the shell plugin. This prevents
any of those surfaces from being visible or receiving input. All new
surfaces will be moved to the private list, too.
The background surface is an exception, it is left to the compositor's
list, so the background will be painted. It is assumed that the
background surface does not allow any actions while being locked.
When desktop-shell announces a lock surface (an unlock dialog), it is
added to the compositor's list, so the user can interact with it.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Check that wlsc_surface::link is part of a list before assuming it is
part of the compositor->surface_list list.
The shell plugin may want to remove a surface from the compositor's
surface list to hide it. Note, that the shell plugin cannot use
wlsc_surface::link for its own purposes.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
wlsc_compositor_fade() ends up in wlsc_compositor_schedule_repaint(),
which is a no-op if compositor is SLEEPING.
On wakeup, first set status to ACTIVE, then call wlsc_compositor_fade()
to start and actually show the animation.
Before, fade was called first, which reset the animation, but did not
cause a repaint. The following wakeup (any input event) would then cause
a repaint, showing the animation from the middle or end of it.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
With the idle_inhibit optimization, wlsc_compositor_wake() is a no-op if
idle_inhibit > 0.
When the shell is waking up the compositor from SLEEPING state as an
indirect response to input activity, it does not work. The call path is:
notify_key() / notify_button()
wlsc_compositor_idle_inhibit()
wlsc_compositor_activity()
shell->unlock()
send prepare_lock_surface event
idle_inhibit++
and when the desktop-shell client responds to the event:
desktop_shell_set_lock_surface() / desktop_shell_unlock()
wlsc_compositor_wake()
no-op, because idle_inhibit > 0
Fix this by removing the idle_inhibit check from wlsc_compositor_wake().
The optimization did not work for pointer motion while no keys pressed,
anyway, so the performance hit is probably unobservable.
Now the compositor wakes up also on key or button press.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
When compositor enters SLEEPING state, the shell plugin goes locked. If
compositor wakes up itself, it will fade in while the shell may not yet
have a lock surface to show.
Fix this by assigning wake-up to be called from the shell, if the
compositor is SLEEPING. The shell may wait for the lock surface request,
and only then wake up the compositor. The compositor will fade in
directly to the lock screen.
krh: original patch for compositor.c
ppaalanen: integration and shell.c changes
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add protocol and functions for supporting screen locking, triggered by
activity timeout.
After activity timeout, compositor starts the fade to black, and then
enters SLEEPING state. At that point it calls lock() in the shell
plugin.
When input events trigger a wakeup, unlock() in the shell plugin is
called. This sends prepare_lock_surface event to the desktop-shell
client. The screen stays locked while the compositor starts fade-in.
At this point, desktop-shell client usually creates a surface for the
unlocking GUI (e.g. a password prompt), and sends it with the
set_lock_surface request. The compositor supposedly shows and allows
interaction only with the given lock surface (not yet implemented).
When desktop-shell has authenticated the user, or instead of issuing
set_lock_surface, it sends the unlock request. Upon receiving the unlock
request, the shell plugin unlocks the screen.
If desktop-shell client dies, the screen is unlocked automatically.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Pass the correct pointer to free().
This is just a cosmetic change, because 'resource' happens to be the
first member in wlsc_frame_callback.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
This is the same as the damage of the top-level surface so just use that.
There's a problem that if we change the stacking, the damage layering breaks,
but that's a problem we already have.
The shell module only needs to deal with attach when it's either the initial
attach or when the attach changes the size of the surface. In case of
initial attach, the shell needs to pick a position for the surface and a
place in the surface stack. We split this case out as a new shell->map
callback. The other case is split into the shell->configure callback,
where the shell may adjust the surface position or reject the new size.
The third parameter of open() is for file-creation modes. File flags are passed
in the second paramater.
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Check, that only the desktop-shell client spawned by the compositor
(desktop-shell plugin) is allowed to bind to desktop_shell interface.
Other clients will receive an error like:
wl_display@1.error(desktop_shell@20, 0,
"permission to bind desktop_shell denied")
The error has the proper object id and interface type.
Note: desktop-shell cannot be started manually anymore, it has to be
started by the compositor automatically.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Fork and exec desktop-shell in the compositor. This is a way to create
an authenticated client. Later, the desktop-shell interface will be
reserved for this client only.
For exec to work, the compositor should be started from the
wayland-demos' root directory.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
All the compositors are using GLES2 so check for the appropriate
surfaceless extension.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
The repaint logic breaks when finish_frame is called from the present
callback. Ideally we should throttle to vsync (or even better, the
compositor repaint cycle, but hey, X is X), but this goes a long way.
The drm compositor always creates a 64x64 bo for the cursor image
regardless of the size of the actual cursor. When the fade animation
kicks in it disables the hardware cursor so that it is rendered as a
regular surface. This surface is rendered to a 32x32 region but using
a 64x64 texture so the cursor gets scaled down.
Fix this by making create_cursor_image return the actual size of the
image created to the compositor.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
If the current or pending scanout buffer is destroyed, the client frame
will remain being displayed until something else causes a repaint to be
scheduled.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Without this fix, the dnd demo would make the demo compositor crash in
shell.c:drag_offer() because resource->data is NULL.
Initialise resource->data in shell_create_drag().
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
wlsc_output_repaint may call wlsc_surface_damage indirectly through
wlsc_output_set_cursor. If this happens in the call made from repaint,
one ends up with repaint being called from both idle_repaint and
wlsc_output_finish_frame.
Fix this by setting output->repaint_scheduled to 1 before calling
wlsc_output_repaint in the function repaint.
[krh] Edited to just only clear repaint_scheduled, when we no longer have
a repaint scheduled.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
On repaint, wlsc_output_repaint will replace output->scanout_buffer with
the new front buffer and then output->present() will cause this buffer
to be displayed. When wlsc_output_finish_frame is called, the
compositor will send a release buffer event for output->scanout_buffer
which is actually the front buffer now.
This patch changes this code to release the previous scanout_buffer
instead of the front buffer on wlsc_output_finish_frame.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Move this to a separate function to better accommodate changes in the
following commit.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
The files in question are copyright Benjamin Franzke (who agrees),
Intel Corporation, Red Hat and myself. On behalf of Red Hat,
Richard Fontana says:
"Therefore, to the extent that Red Hat, Inc. has any copyright
interest in the files you cited as of this date (compositor-drm.c,
compositor.c, compositor.h, screenshooter.c in
http://cgit.freedesktop.org/wayland/wayland-demos/tree/compositor),
Red Hat hereby elects to apply the CC0 1.0 Universal Public Domain
Dedication to such copyrighted material. See:
http://creativecommons.org/publicdomain/zero/1.0/legalcode .
Thanks,
Richard E. Fontana
Open Source Licensing and Patent Counsel
Red Hat, Inc."
At initialization, if it fails in binding the socket or creating the
lock file then the pointer will be already freed and will result a
segfault when quiting the compositor.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
This isn't going to change over time, so just tracking it in the
evdev device is a little easier. Also, we need to adjust for the
output position when transforming the device events to screen space.
The compositor was never actually calling the output backend to turn off
the hardware cursor when the screen begins fading. This would result in
a stuck hardware cursor and movable software cursor for the duration of
the fade/unfade.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Adds a general wlsc_compositor_shutdown() function that all output
backends call when shutting down. wlsc_compositor_shutdown() will call
a new 'destroy' method of each output to perform backend-specific
cleanup (e.g., turning off the hardware cursor in the DRM compositor).
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
The DRM cursor ioctl requires DRM master. We shouldn't drop
master until after we're done turning off the hardware cursors
for a VT switch.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
We may want to adjust the protocol later for clients that care for
these devices only, generating a special event.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
texture_region was getting a trash value for computing. I don't
understand how we couldn't see any artifact on surface output in such
case.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
The bo for the cursor image is always created with size 64x64 even if
the actual cursor image is smaller than that. If this memory is not
initialized, random data can create artifacts near the cursor.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
- Fixes segfault, if first enumerated connector is not connected.
- Corrects x-offset where e.g. the 2. of 3 connectors is not connected.
(where width of 1. output would have taken twice as
offset for output at 3. connector)
With this the X server directory can become independent from the
installation prefix set by autoconf.
Signed-off-by: Tiago Vignatti <tiago.vignatti@intel.com>
This reverts cde9bfc805. We need to damage the
area covered by the old surface when attaching a new buffer. The new surface
area will be damaged by the client.
I may have missed something, but - since the Wayland compositor
already picks a platform backend, opens a connection and initializes the
backend specific display data structure it doesn't make sense
to let egl pick a platform. If it picks a different one the
display specific data structure will most likely not match.
Thus determine the platform in the Wayland rendering backend by setting
the EGL_PLATFORM env variable.
For the client any other platform than 'wayland' doesn't seem to make
sense.
I'm not sure if I've got the the platform ofr openfwd right.
Signed-off-by: Egbert Eich <eich@freedesktop.org>
The shell module is responsible for implementing the higher level
compositor behavior. We default to the desktop-lite shell built in to
the compositor.
Instead of having a separate check function, we just mark our x11 event
source as needing a post-dispatch check. The event loop will call our
dispatch function again after all dispatching is done, this time with
mask = 0. If we don't process any events, return 0, so the event loop
doesn't keep calling us.
If somebody else did an X11 round trip, that could leave events in the
XCB buffer that we wouldn't see until the next X event came in. The new
event source check function lets us check the XCB queue after dispatching
and this way we'll see events we need to deal with right away.
The event handling gets a little trickier this way but we need the
keymap sent immdiately after the focus_in event to determine which keys
are pressed as the compositor receives keyboard focus.
We need to look at the focus_in and keymap notify pair to correctly determine
the set of held down keys at focus in time, so generalize the handling
of event pairs in preparation for that.
Instead of calling XGetXCBConnection() blindly, check XOpenDisplay()'s
return value to avoid a possible segfault in the former. That happens if
$DISPLAY is set, but if that display isn't available.
Signed-off-by: Cyril Brulebois <kibi@debian.org>