IPC clients generally expect executed commands to have taken effect when
the command completes, while delayed modeset means that it can take
several milliseconds more before e.g. an output is enabled.
However, modesetting on every output command in the IPC call could on
systems with already slow modesetting behavior lead to an unresponsive
system for a not insignificant period of time.
To strike a balance, force modeset once all the commands of this IPC
call have executed if a modeset is pending.
Some commands require outputs to be enabled. These commands are deferred
to allow outputs to be discovered, but the delayed modeset might only
run some time later.
Force a modeset to occur before running deferred commands.
Fixes: https://github.com/swaywm/sway/issues/8433
Emit property change signals for the IsStatusNotifierHostRegistered and
RegisteredStatusNotifierItems properties in StatusNotifierWatcher,
so code relying on the PropertiesChanged signal, instead of signals
such as StatusNotifierHostRegistered, can work properly.
A library that is affected by this is the libappindicator-gtk3* library
and it can cause tray icons to be missing after starting swaybar due to
a race condition, as follows:
* An application using libappindicator-gtk3 starts, e.g. nm-applet.
* Some time later, swaybar starts.
* swaybar creates the StatusNotifierWatcher.
* libappindicator-gtk3 observes the new watcher, but it sees that
IsStatusNotifierHostRegistered=false, so it falls back to the
Freedesktop System tray protocol.
* swaybar creates the StatusNotifierHost.
At this point, libappindicator-gtk3 should "un-fallback" back to SNI.
However, since swaybar does not emit the PropertiesChange signal on
IsStatusNotifierHostRegistered, libappindicator-gtk3 doesn't get
notified, and stays in fallback state forever.
* As a result, nm-applet will not show in the swaybar tray.
This race can be made reliable by inserting a 1-second long sleep here:
03483ff370/swaybar/tray/tray.c (L57)
(*) Note that the libappindicator-gtk3 library has been mostly replaced
by libayatana-appindicator, which is not affected by this.
The affected version is still used by Arch Linux, source code at:
https://bazaar.launchpad.net/~indicator-applet-developers/libappindicator/trunk/files/298
The modeset logic iterates over all outputs at the end, sets their new
position in the layout and takes a copy of its geometry that is later
referenced by layout and scene management code.
If one output is auto configured, then a later output that is manually
configured can lead to the first output being moved without the stored
geometry being updated.
Split this into two passes: The first pass finalizes the output config
and makes updates to the layout, while the second pass updates the copy
of the geometry and arranges things as a result of it.
An output backend might request any change to an output state at any
time, although currently only this is currently only used for changing
window size on the wayland and x11 backend.
Applying the configuration directly means that the current output state
becomes inconsistent with the configured state, which can cause the new
state to be reverted later if apply_stored_output_configs is called.
Before 4f9ce4675c. the output geometry would be updated by
arrange_outputs, but this is only done by the modeset logic now,
resulting in the stored geometry never being updated on wayland backend
window resize. This was not discovered as the stored geometry is not
used particularly often.
Solve both by storing a new output configuration and relying on the
modeset logic to apply a new state.
Fixes: 4f9ce4675c ("tree/arrange: Remove redundant output geometry update")
The job of queue_output_config is now just to fill out a
wlr_output_state according to the output configuration, but it still has
a lot of logging from before we had wlr_output_state or the new modeset
logic, when queue_output_state instead touched the implicit pending
state of wlr_output.
Whatever debug logs it had would already be covered by the output state
debug logs or the command debug logs, so let's just remove it.
Starting by setting some special initial output config values settings
was something sway used to do when the config was initially being
processed. This was later changed to always happen, as there shouldn't
be differences in how output config is calculated during config load and
after.
Most of these values are redundant, as they are either the zero value or
a value that would be selected if the unset (-1) value was found.
For output transforms, the automatic panel orientation code would only
trigger if the final output config has an unset transform, which the
initial values set in find_output_config made impossible.
Remove these initial values and instead use a fresh output config as is.
queue_output_config had some remaining logic that would avoid setting
output states if they already appeared to be in effect. That is not what
most of the states did nor what is currently expected, so clean that up.
switching workspace directly to urgent window creates timer which delays
reset of urgent state so user is able to notice it. make sure state
change is reflected visually as well (border change) by triggering
container update
Fixes: #8377
Man sway(5) specifies that when tiling_drag is enable, the floating_mod
can be used to drag tiling, as well as floating containers. However the
current code indiscriminately assumes any button press to be intended
for moving the container, consequently causing an unintended call to
`seatop_move_tilting:handle_button` rather than
`seatop_default:handle_button` to pass
`state=WL_POINTER_BUTTON_STATE_RELEASED` to `get_active_mouse_binding`
My idea was to make 'Handle moving a tiling container' follow the same
path as 'Handle moving a floating container' because the initial call to
handle moving a floating correctly exits that branch and ends up passing
the RELEASED state to `get_active_mouse_binding`.
Fixes#8334
in pr https://github.com/swaywm/sway/pull/8196, when im_popup_surface is unmapped, author set the popup->relative to NULL, butt popup is still in popup groups, where assert the relative is not NULL, this cause the panic
Take the suggestion of Nefsen402, remove the line where set relative to
NULL, and add NULL check in scene_descriptor_destory
sway_root.outputs only include enabled outputs. We also need to re-init
the renderer for any disabled outputs, so use sway_root.all_outputs
instead.
Resolves the following heap-use-after-free accessing the render formats
when a disabled output is modeset after a GPU reset has occurred.
Currently, swaybar does not gracefully die if it detects
that the dbus connection was lost. Although it's not recommended
to restart dbus without restarting the compositor, it can very
easily happen. In the case it does, compositor's tray should
not consume 100% cpu until it has to be force killed.
apply suggestions
just setting the bar to not running will call teardown and unref the
dbus.
We batch modesets and input configuration performed during config reload
but commit for every command during the intial config load. There is no
need to perform commits during the initial config load as outputs have
not yet been created, but swaybg spawn should still be batched.
At the same time, replace direct calls to apply output configuration
with request_modeset to properly handle the modeset timer.
GUD devices uses RGB565 by default for performance reasons. Allow
specifying render_bit_depth 6 to pick this format. The definition works
out if you consider the maximum number of bits per channel instead of
the average.
Arranging lock surfaces rely on the sway_output width and height being
updated, but these are only updated after the commit has been completed
and all commit listeners have executed. This means that the lock
surfaces will not be appropriately scaled to match a change in output
dimensions, and may reveal what is under the lock background.
Replace the implicit arrange through the output commit listener with an
explicit arrange after the output configuration is finalized.
This might have regressed by other transition away from output commit
listeners for other arrange tasks, but even then it would have
erroneously relied on signalling order.
