Refactor the buffer-type check slightly so we can handle solid-color
buffers, which we do exactly nothing with.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Currently solid-colour displays (e.g. the background for fullscreen
views) is implemented by a special-case weston_surface which has no
buffer attached, with a special punch-through renderer callback to set
the colour.
Replace this with a weston_buffer type explicitly specifying the solid
colour, which helps us eliminate yet more special cases in renderers and
backends.
This is not handled yet in any renderer or backend, however it is also
not used by anything yet. Following commits add support to the renderers
and backends.
Signed-off-by: Daniel Stone <daniels@collabora.com>
When the renderer/backend indicate that they do not need a surface's
buffer content to be preserved, most often because they have copied it,
simply downgrade the buffer reference to 'will not access', rather than
drop the buffer reference altogether.
Signed-off-by: Daniel Stone <daniels@collabora.com>
In the original conception, a weston_buffer_reference indicated that the
underlying contents of the wl_buffer would or could be accessed, so
wl_buffer.release must not be sent until the last reference was
released, as the compositor may still use it.
This meant that renderers or backends which copied the buffer content -
such as the GL renderer with SHM buffers - could only send a buffer
release event to the client by 'losing' the buffer reference altogether.
The main side effect is that `weston-debug scene-graph` could not show
any information at all about SHM buffers when using the GL renderer, but
it also meant that renderers and backends grew increasingly exotic
structures to cache information about the buffer.
Now that we have an additional buffer-reference mode (still referring to
the weston_buffer/wl_buffer, but not going to access its content), we
can allow the weston_buffer_reference and weston_buffer to live as long
as the buffer itself, even if we do send a release event.
This will enable a bunch of backend and renderer deduplication, as well
as finally making scene-graph more useful.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Add a mode argument to weston_buffer_reference which indicates whether a
buffer's storage may/will be accessed, or whether the underlying storage
will no longer be accessed, e.g. because it has been copied. This will
be used to retain a pointer to the weston_buffer whilst being able to
send a release event to the client.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Keep the weston_buffer alive for as long as at least one of the
underlying wl_buffer or a backend usage exists.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Make sure we don't die if we're asked to flush the damage on a SHM
buffer which has subsequently been destroyed.
Signed-off-by: Daniel Stone <daniels@collabora.com>
We currently allow a weston_buffer to outlive the underlying wl_buffer
iff the renderer/backend has cached it. Currently the 'is this buffer
valid?' test relies on looking for the validity of the weston_buffer
itself; shift these tests to looking at the validity of the underlying
resource.
Signed-off-by: Daniel Stone <daniels@collabora.com>
y_inverted meant that the buffer's origin was (0,0), and non-inverted
meant that the buffer's origin was (0,height). In practice, every buffer
was 'inverted' into our natural co-ordinate space that we use
everywhere.
Switch to using an explicit origin enum to make this more clear.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Rather than calling accessors (wl_shm_buffer_get etc) to figure out
which type our buffer is, just look in the structure.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Rather than open-coding various resource -> type accessors, just stick a
type enum in the buffer struct.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Rather than only filling weston_buffer information when we first come to
use it, add an explicit hook so we can fill the dimensions the first
time the buffer's attached.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Make sure we only import dmabufs where the underlying pixel_format is
known: if we can't reason about the buffer content, we're not entirely
likely to be able to display it well.
Signed-off-by: Daniel Stone <daniels@collabora.com>
When we first see a buffer attached, we create a weston_buffer for it.
The weston_buffer doesn't contain any useful information in and of
itself; that's left to renderers to populate later.
Switch this to doing it in the core at the first opportunity, at least
for SHM and dmabuf buffers; EGL buffers will follow in the next commit.
Signed-off-by: Daniel Stone <daniels@collabora.com>
We already have the buffer in the caller, and every no-op implementation
will want to access the buffer. So might as well pass it.
Signed-off-by: Daniel Stone <daniels@collabora.com>
The comment about needing to have destroyed images is somewhat less true
now that we actively avoid doing so.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Fixes: 0b51b02c5e ("gl-renderer: Don't re-import dmabufs")
Add some logging helper functions along with two log scopes for debug
and extremely verbose debugging information.
Also add tangentially related logging for the synchronize event, so
the debug stream isn't empty right now. The vast majority of verbose
usage will come later.
Co-authored-by: Steve Pronovost <spronovo@microsoft.com>
Co-authored-by: Brenton DeGeer <brdegeer@microsoft.com>
Signed-off-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Steve Pronovost <spronovo@microsoft.com>
Signed-off-by: Brenton DeGeer <brdegeer@microsoft.com>
There are currently compatibility issues between FreeRDP's implementation
of the RemoteFX codec and Microsoft's implementation.
Perhaps this will be fixed in the future and this option can go away,
but for now it's necessary to have a way to disable the codec if the
windows client is going to be connecting to a weston server.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
This allows easily co-developing a Wayland protocol and Weston.
Example setup:
ln -s subprojects/wayland-protocols /path/to/wayland-protocols
meson configure build/ --force-fallback-for=wayland-protocols
Signed-off-by: Simon Ser <contact@emersion.fr>
With commit 'screen-share: Add option to start screen sharing on weston
star' an section option has been added to start screen-sharing by
default on all outputs. This has the side-effect of attempting to start
screen-share on the same (RDP) output performing the screen-share. That
happens due to re-loading the screen-share module once more.
This patch recommends users to use --no-config option or alternatively,
use a different configuration file to avoid that.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Found while running with b_sanitize=undefined, which yields:
runtime error: shift exponent 909199186 is too large for 32-bit type 'int'
Converts the shm_formats to a boolean and checks for the correct pixel
format it directly, instead of trying to gather them all in an array and
then later on to do the check.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Trivial fix to clean itself on compositor tear-down. While at it
properly free the command string.
Fixes#298
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
With commit e825fe38, we no longer display the pointer if no movement is
detected, which will cause screen share to fail to start if that is the
case. There could be also legitimate cases where there's no pointer, so
let's allow screen share to function in those cases as well. Makes uses
of previous helper methods to find a proper output to share in case we
don't have an pointer.
Re-uses the shell utils functions.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Seems that we're still missing layer clean-ups, with the touch
calibrator being one of them. Call the appropriate function when
shutting down the compositor instance.
Fixes: #603
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Toy toolkit apps already do this since commit 807cd2e589
Co-authored-by: Steve Pronovost <spronovo@microsoft.com>
Co-authored-by: Brenton DeGeer <brdegeer@microsoft.com>
Signed-off-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Steve Pronovost <spronovo@microsoft.com>
Signed-off-by: Brenton DeGeer <brdegeer@microsoft.com>
Input method process may also be a XIM server which requires the correct
DISPLAY to be set after xwayland start up. This helps input method to
work for both wayland and Xwayland.
Signed-off-by: Weng Xuetian <wengxt@gmail.com>
Server generated key repeats are ignored - and they don't generate
matching release events. We early return to avoid generating events
for them.
We also need to push the idle inhibition counting after that early
return to prevent breaking idle inhibition with unmatched events.
Co-authored-by: Steve Pronovost <spronovo@microsoft.com>
Co-authored-by: Brenton DeGeer <brdegeer@microsoft.com>
Signed-off-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Steve Pronovost <spronovo@microsoft.com>
Signed-off-by: Brenton DeGeer <brdegeer@microsoft.com>
We want to wait for Xwayland to be ready before issuing it blocking
requests, but relying on USR1 is a bit unsafe:
- we can't ascertain the signal originated from Xwayland
- if weston is started as PID1 (e.g. in its own container), then
Xwayland will not send SIGUSR1 and X11 connections will be stuck
forever: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1312
Creating a pipe and using -displayfd, even if we don't care about the
display value itself, is safe and works for all cases
Signed-off-by: Dominique Martinet <dominique.martinet@atmark-techno.com>
These added lines are comments (do not affect output) that make it
easier to browse this file and find the section headings.
Removal of the preceding empty lines on two of the section headers
remove a blank line from PDF output. The blank line was kind of nice,
but presumably .SH should add any necessary blanks before a new section
heading. Now all the section headings have the same vertical space to
the preceding content in PDF.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Try to reduce the cargo-culted directives that do not seem to have any
effect on the output.
I verified that are no changes by doing before this patch
$ man -Tpdf man/weston.ini.5 > ref.pdf
and then after this patch
$ ninja && man -Tpdf man/weston.ini.5 > test.pdf
and looking at
$ diffpdf ref.pdf test.pdf
I used PDF as the format for comparisons, because it can express much
more typesetting features than plain text.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Without this, the earliest signal the ivi controller receives for a new surface
is configure_desktop_changed signal. And this is not emitted until the surface
has the first buffer attached.
By emitting the signal during surface creation, the controller is able to set
the initial width and height for the first configure.
Signed-off-by: Michael Olbrich <m.olbrich@pengutronix.de>
The pixel format was hardcoded to PIXMAN_a8r8g8b8. All other
renderer->read_pixels() calls in weston use dynamic format selection via
the compositor->read_format instead.
The problem was spotted on the ARM devices with Mali-400 GPU. The visual
effect of the problem was black screen on the remote display, when using
screen-share with the VNC or RDP backends. Related wayland-devel thread:
https://lists.freedesktop.org/archives/wayland-devel/2020-September/041624.html
Signed-off-by: Maciej Pijanowski <maciej.pijanowski@3mdeb.com>
Refactor some of rdp.c into a header file.
Co-authored-by: Steve Pronovost <spronovo@microsoft.com>
Co-authored-by: Brenton DeGeer <brdegeer@microsoft.com>
Signed-off-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Steve Pronovost <spronovo@microsoft.com>
Signed-off-by: Brenton DeGeer <brdegeer@microsoft.com>
We've already allocated the listener by the time we hit this failure,
so we must exit through the path that frees it.
Co-authored-by: Steve Pronovost <spronovo@microsoft.com>
Co-authored-by: Brenton DeGeer <brdegeer@microsoft.com>
Signed-off-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Steve Pronovost <spronovo@microsoft.com>
Signed-off-by: Brenton DeGeer <brdegeer@microsoft.com>
By some dark magic this accidentally doesn't crash because there are
zeroes in the right places at startup, but black_surface_get_label()
very much expects surface_private to be a view.
Let's hand craft a new bespoke label function for this use.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
When we launch clients they currently stay in the same session as
weston, and share its controlling terminal. This means hitting ctrl-c
in weston's controlling terminal will send SIGINT to the clients as
well. It also means SIGHUP will be propagated to our launched clients
if weston's controlling terminal is closed.
While generally not harmful, this behaviour is not beneficial, and
is present by default and not by design.
Problems arise when launching weston in a debugger, as a ctrl-c sent to
the debugger will be propagated not only to the debugger, but all the child
processes sharing weston's session. This results in weston-desktop-shell
being killed by the ctrl-c that was intended to stop weston for debugging.
If weston-desktop-shell is killed within 30 second of startup, it will
result weston performing a clean shutdown. This clean shutdown can
make debugging a little too surprising.
Ostensibly, clients launched via weston_client_launch will be wayland
clients that terminate cleanly on their own if weston is killed, so
there should be no need for them to remain in weston's session to
catch ctrl-c from its controlling terminal. Nor should they need a
controlling terminal for their general operation.
Use setsid() to move them to their own session, devoid of controlling
terminal, to make using a debugger a little less confusing in some
cases.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
signalfd interacts badly with gdb's signal trapping - when hitting
ctrl-c in a debugger attached to weston, weston will receive the
signal. This results in weston exiting cleanly when the intent
was to use gdb to interfere with its operation.
Trapping SIGINT was introduced in commit 50dc6989 which ensured we
would call wl_display_terminate() on SIGINT or SIGTERM to clean
up our socket.
Killing weston with SIGINT is quite common for several developers,
so it's important to preserve this clean shutdown behaviour, so
we can't naively stop trapping SIGINT entirely.
Instead, use the sigaction() function to trap SIGINT, and have
the SIGINT handler send weston SIGUSR2 (SIGUSR1 is already
used by xwayland). SIGUSR2 can be trapped in the proper wayland
way via wl_event_loop_add_signal(). This way we can properly
break our event loop and clean up on SIGINT, but we can also
have gdb intercept SIGINT.
There are other ways around this, but I'm hoping this one allows
people to continue using ctrl-c to stop weston, and doesn't
require additional project specific gdb knowledge.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Daniel Stone
Hideyuki Nagase
Derek Foreman
Simon Ser
Marius Vlad
Weng Xuetian
Dominique Martinet
Pekka Paalanen
Michael Olbrich
Maciej Pijanowski
shierote
James Hilliard