Our core test structure is 36 bytes wide. Declaring it with a 32-bit
alignment should thus stripe it to 64 bytes. For some reason, clang+lld
lays them out with a 96-byte stride within the section (does it want an
entire 32-bit word when building with ASan?), getting the code wildly
confused when it tries to step through the structures.
So we could fix all our tests to avoid the fragile section dance, or we
could just waste another 4 bytes per test definition by bumping the
alignment up to 64 bytes, which seems to do enough to magically accord
with what clang+lld+ASan expect.
Signed-off-by: Daniel Stone <daniels@collabora.com>
EGL's native display type is not particularly well defined; in
gl-renderer we get around this by always treating it as a void *, since
for all the platforms we care about it's a pointer - gbm_device,
wl_display, or Display *.
The surfaceless platform doesn't care what the native display is (since
it doesn't have one by definition), so just use NULL instead of what may
be either NULL or 0 depending on environmental factors.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Give ourselves a bit more separation between the different job types, so
it's easier to see what's running and/or failing.
Signed-off-by: Daniel Stone <daniels@collabora.com>
ARMv7 is still an important architecture for us to run on, and running
on ARMv7 also gives us 32-bit build coverage. As distros are deprecating
their non-64-bit-x86 support, this may be our only realistic chance of
ensuring that our build is also 32-bit-clean.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Build a kernel for AArch64 and run it under virtme just like we do for
x86-64.
This requires adding support for the AArch64 defconfig variant, and
accommodating for the fact that it builds DRM as a module by default
rather than built in. The virtme branch we are using has also been
rebased on top of newer virtme upstream which unbreaks AArch64.
Signed-off-by: Daniel Stone <daniels@collabora.com>
This only runs a single build job, to build without GL and not run any
tests, as KVM support is not yet included.
Signed-off-by: Daniel Stone <daniels@collabora.com>
GitLab CI has two execution scheduling models. The original model is to
split jobs into stages; jobs within a single stage may execute in
parallel, but execution is serialised between stages.
As we move to supporting multiple OSes and architectures, there is no
need to serialise, e.g. the AArch64 Linux build against the x86-64
FreeBSD container preparation.
Declare our dependencies explicitly using `needs`.
Signed-off-by: Daniel Stone <daniels@collabora.com>
0.57.0 has a bug where the whole test harness crashes when using TAP and
failing tests, cf. https://github.com/mesonbuild/meson/pull/8385
Signed-off-by: Daniel Stone <daniels@collabora.com>
When we build our base-OS container, we run debian-install.sh to install
packages and compile our build dependencies. Since the latter is mostly
OS-independent, split debian-install.sh into two scripts: one to install
and cleanup packages, and another just to compile stuff.
Signed-off-by: Daniel Stone <daniels@collabora.com>
No functional change to the test runs, apart from changing the job
names. This will allow us to test along more axes without more
duplication.
Signed-off-by: Daniel Stone <daniels@collabora.com>
GitLab already groups our various stages (container, build, etc) into
separate UI elements. Within those stages, the important information is
the parameterisation (architecture, OS, toolchain). We don't want that
to get ellipsised, so put that first in the job names.
Signed-off-by: Daniel Stone <daniels@collabora.com>
The shared runners export ${FDO_CI_CONCURRENT} for the appropriate
number of CPUs we should use during our builds.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Because we weren't properly pinning the wayland-protocols version, and I
can't read, we missed updating this in !563.
Signed-off-by: Daniel Stone <daniels@collabora.com>
The API expects uintptr_t (good!), but we're passing an unsigned long
here. Make the conversion explicit.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Follow-up from commit 'desktop-shell: don't run fade animation if
compositor is inactive' where the reference was dropped directly,
instead of using weston_surface_destroy().
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
pointer/touch drag-n-drop operations could happen if there's no keyboard
hooked up or when it is unplugged.
Fixes: #235
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Updating PipeWire to the latest version lets us check that everything
still works well, and shows users that they are able to use it.
[daniels: Updated to 0.3.31, use symbolic ref tags rather than SHAs.
0.3.32 is released, but doesn't build in our Debian
environment; this is fixed upstream but there is no release
for it yet.]
Signed-off-by: James Hilliard <james.hilliard1@gmail.com>
This adds support for xdg-shell v3, which adds support for repositioning
popups. This adds support for explicit popup repositioning, as weston
doesn't yet apply any constraining to popups, thus cannot implicitly
reposition.
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
Tested-by: Emmanuel Gil Peyrot <linkmauve@linkmauve.fr>
When a window is closed, Weston will, by default, run a fade out animation and
defer destroying the underlying surface until it completes. However, if the
compositor is sleeping, and therefore not rendering any frames, this animation
will *never* complete. Therefore, if windows are repeatedly created and
destroyed while in sleep mode, these surfaces will keep accumulating, and since
the buffers attached to them may be backed by an fd, eventually the ulimit will
be reached resulting in a potential crash or other errors.
This can be demonstrated repeatedly launching and killing an X11 application
with Xwayland running.
while true; do xterm & pid=$!; sleep 0.5; kill $pid; done
As soon as the compositor goes to sleep, one can observe a steadily growing
list of dmabufs in the output of lsof.
As a fix, desktop_surface_removed should check whether the compositor is active
before kicking off the fade animation. If it is not, it should instead drop the
extra reference taken in desktop_surface_committed and then destroy the surface
immediately.
Signed-off-by: Erik Kurzinger <ekurzinger@nvidia.com>
And ultimately, fail to start when there are no input devices on the
system. Patchs adds consistency to touch/pointer initialization to
return -1 in case same thing happens.
Further more, when the device is not created we can't assume to retrieve
a valid one from a libinput_device so guard against it. This takes care of
hot-plugging situations when we couldn't create the (keyboard) device,
or when removing it.
Fixes: #117, #402, #485
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Suggested-by: Daniel Stone <daniel.stone@collabora.com>
Some graphics drivers (currently at least VMware and AMD) will give a 0
timestamp for the atomic mode flip completion event when turning off
the display. This causes us to trip an assertion in
weston_output_frame_finish() because the clock jumps backwards, which
isn't a condition the presentation feedback code should be dealing with.
This is a good assertion and we'd like to keep it. And there's some
expectation that this is buggy behaviour in the graphics drivers that will
be fixed at some point.
Pragmatically speaking though, there's nothing productive we can do with a
correct timestamp for the display shutdown. So let's just flag the
event sent for DPMS off as invalid so presentation feedback doesn't have
to worry about it, and the assert doesn't fire.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
weston_frame_callback is needed primarily to store the doubly-linked list link,
but it can be also retrieved by using the wl_resource_get_link() function.
This removes an extra heap allocation per every wl_callback object.
Signed-off-by: Vlad Zahorodnii <vlad.zahorodnii@kde.org>
Pull the container from a source without a rate limit
Suggested by Daniel Stone <daniels@collabora.com>
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
This is the first step towards removing wl_shell support.
Add an option so that users can toggle support for the deprecated
wl_shell protocol. This lets users test their clients to make sure
they work fine without wl_shell.
The option is set to false by default.
Signed-off-by: Simon Ser <contact@emersion.fr>
If a given wl_surface already has a role (e.g. cursor or subsurface),
there is nothing you can do with an xdg_surface which won't raise an
error, apart from destroying it.
As of wayland/wayland-protocols@11fecf080860 this is now explicitly
specified to be illegal, so disallow it within libweston-desktop. This
avoids us tying ourselves in knots with surface-private ownership.
Signed-off-by: Daniel Stone <daniels@collabora.com>
It's illegal to create an xdg_surface for a surface which already has a
buffer attached to it. We check for this, but only after we've created
our weston_desktop_surface; this simply avoids creating the internal
tracking structure when we're only going to destroy it after posting the
error.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Conditionally build support when libdrm is at least 2.4.107 to make use
of it. Plug it in when printing out the buffer information.
With this in, we add a hard dependecy for libweston to link against
libdrm.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
asprintf() has the problem that it leaves *strp undefined when it
fails. Here is a simple wrapper that ensures NULL if asprintf() fails,
which is much more convenient to use.
This will be useful in future patches, where one needs to return error
messages from maybe failing functions, and more.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Potential failures when creating the EGL image could cause an incorrect
number of num images (num_planes > num_images). With this change
egl_image_unref() requires an additional check to avoid any potential NULL
derefs when cleaning up. We do it straight in egl_image_unref() instead
of adding guards all over the necessary parts.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
As observed on some platforms, importing known DMA buffers can cause
failures, leading to an attempt of destroyng an EGL image not set. This patch
resets the num_images such that loop becomes inert when destroying the
DMA buffer, and avoids passing an egl image to it.
The initial import doesn't have this issue as it sets the num_images in
case it succeeds. This also corrects the assumption that the num_images
were 0 at that point which, if the initial import succeded, was actually set
to 1.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Daniel Stone
Marius Vlad
Sebastian Krzyszkowiak
Jonas Ådahl
Nicholas Niro
James Hilliard
Emmanuel Gil Peyrot
Link Mauve
Erik Kurzinger
Derek Foreman
Vlad Zahorodnii
John Good
Simon Ser
Pekka Paalanen