- check for valid scene_output in output_is_usable()
- change many "output != NULL" checks to use output_is_usable()
- remove one now-redundant separate check for valid scene_output
Fixes a crash at startup (with autoEnableOutputs=no) due to
dereferencing null scene_output in create_output_config() since:
7d7ece21d9
("output: suppress error when output position is unavailable")
Fixes: #3357
This mitigates a race where the XWayland server may generate an unwanted
FocusOut event for the newly activated window, if it receives mouse/
pointer events over the parallel wayland connection first.
In particular, this fixes an issue with certain fullscreen applications
(such as Minecraft) that self-minimize when receiving FocusOut.
Also limit a previous similar workaround to apply only to XWayland views.
Fixes: #3344
See also: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/4044
f58b532 implemented output-relative position saving/restoring on output
un-plugging/re-plugging. It worked as follows:
1. Store the output-relative view geometry in `view->last_placement`
(if not set) before adding/removing an output from the layout.
2. After adding/removing an output, call `view_adjust_for_layout_change()`
after the layout change to restore the output-relative view geometry
based on `view->last_placement`.
However, it didn't consider `view_adjust_for_layout_change()` being
called from other places such as `regions_reconfigure()` and
`output_update_all_usable_areas()`, causing an error message "view has
no last placement info". This can happen when a panel is mapped or
unmapped, or on Reconfigure.
This commit fixes it by changing the life cycle of
`view->last_placement`. It used to be set only before output layout
changes and cleared on user-initiated moves/resizes, but now it is set
and updated on user-initiated moves/resizes. I think this is more
intuitive, too.
Fixes a regression in 83b619c2 that the bottom-most view is focused
when when an exclusive layer surface (e.g. fuzzel) is unfocused.
Also, added some comments to clarify the order.
Currently, with enough dexterity, you can minimize a window via hotkey
with one hand while continuing to move it (now invisible) with the
mouse in your other hand. The same can probably happen with XWayland
windows that self-unmap at an inconvenient time.
Probably not a common occurrence, but it's trivial to handle.
As described in the `FIXME` comment in `cycle.c`, we had spurious focus
changes where the keyboard focus is momentarily given to the previously
focused window when finishing the window switcher, an then it is given
to the selected window.
This commit fixes this by adding a parameter in
`seat_focus_override_end()` to avoid restoring the focus to the
previously focused window.
I also removed the check for `!seat->seat->keyboard_state.focused_surface`
in `seat_focus_override_end()`. I thought it was necessary to avoid
updating the keyboard focus if the focus was given to a session-lock
surface before e.g. finishing window switching, but `seat_focus()` is
no-op in that case anyway.
This commit does not change any behaviors.
We don't need to call `desktop_update_top_layer_visibility()` in
`view_discover_output()` since it's called in `view_update_outputs()`.
Before this commit, the window switcher skipped all the child windows.
However, as child windows not marked as modal dialogs can lose focus
(ref. `desktop_focus_view()`), it will make sense to include them in the
window switcher so that users can refocus them with keyboard.
This behavior follows KWin.
Currently, every click within a sub-view results in first restacking
the parent view in front, and then the sub-view. This is unnecessary
and has caused issues in the past, such as with Xaw popups (which
we've worked around in d748dc78bc by adding an xcb_flush()).
It would be better not to do the unnecessary restacking at all.
When unminimizing a group of N parent/child views, we currently end up
triggering N focus changes (as well as O(N^2) surface restackings) due
to calling desktop_focus_view() for each view in turn.
Since desktop_focus_view() already raises all sibling views together
via view_move_to_front(), let's make view_minimize() call it only once
at the very end, once all views are visible.
Test cases:
- Audacious with floating plugin views (XWayland)
- xfce4-terminal with About dialog (xdg-shell)
v2: also avoid repeated focus changes when minimizing
This commit removes the "failed to get output layout box" error printed
when disabling an output with `wlr-randr`.
Also, use `output->scene_output->{x,y}` instead of calling
`wlr_output_layout_get_box()` for simplicity. The positions of
scene-outputs are synced with the output-layout-outputs.
Before this patch, we always tried to preserve the global positions of
floating windows across layout changes, which meant that windows could
jump to different outputs if the output coordinates changed.
Instead, this patch adds the output name and output-relative position in
`view->last_placement` to keep them across layout changes, like KDE and
GNOME do.
This also allows us to remove `view->lost_output_due_to_layout_change`,
which was required to keep the output of fullscreen/maximized windows,
since we now always try to keep `view->output` whether or not the window
is floating.
I will add output name and relative view position in
`view->last_placement` later, which needs to be saved before layout
changes unlike the global position.
After several iterations, this is basically a complete re-work. The old
implementation was difficult to follow and sometimes failed to restore
fullscreen/maximized/tiled geometry correctly, since it was based
entirely on natural (floating) geometry.
The new implementation:
- always saves the actual (pending) geometry at first layout change
- explicitly tracks whether a view has moved between outputs
- consolidates invalidating the saved geometry into one place, rather
than having lots of invalidate() calls sprinkled everywhere
Reduce the overhead of fork/execve/perl startup time by not
doing those for every files that are checked.
This also makes the check execution complete faster.
On similar systems and similar background load, the execution time
varies based on how find(1) outputs the (*.[ch]) files it sees on
filesystem -- the filenames are not sorted but are written from
directories src/ include/ clients/ t/ in that order -- more than
80% of the time goes checking files in src/, and how the 16-file
batches from that dir (108 files in src/, 208 total, as of 2026-01)
are distributed the checkpatch.pl processes affect mostly to the
total run time.
Void made the foot package depend on the font.
Running this test now always fails with
> `dejavu-fonts-ttf-2.37_3` in transaction breaks installed pkg `foot-1.25.0_2'
As this test uses both and the feature itself
seems to work well lets just skip it for now.
These were added to fix handling of natural geometry for snap-to-edge
behavior back in 9021020f6e and seemed like a good idea at the time.
Since then, the number of call sites has exploded, so it seems more
maintainable to put explicit checks for interactive move within the
three functions affected.
Before this commit, the OSD could overflow the screen when displaying
many window items. In this commit, we hide the overflowed items and
show a scrollbar to fit the OSD within the screen.
There was an invalid memory access (since introduction of thumbnail
style in 2e9292b) with following steps:
1. Press Alt-Tab
2. Update `<windowSwitcher><osd><style>` from `classic` to `thumbnail`
3. Run `Reconfigure`
4. Press Alt-Tab again
...because `cycle_osd_thumbnail_update()` is called even though
`cycle_osd_output->items` holds `cycle_osd_classic_item`.
This commit halts window switcher on `Reconfigure` to clear
`cycle_osd_output->items` and avoid that invalid memory access.
We declared `cycle_state` struct in `labwc.h` and `cycle_osd_scene`
struct in `output.h`, which was unclean in terms of separation of
concerns.
So this commit firstly moves `cycle_state` to `cycle.h`, then replaces
`cycle_osd_scene` in `output.h` with `cycle_osd_output` in `cycle.h`
which is dynamically allocated in a similar manner to
`session_lock_output`. This ensures that all states about alt-tabbing
are stored in `server->cycle`.
Also, this commit fixes a rare memory leak in `output->cycle_osd.items`
when an output is destroyed while alt-tabbing, by freeing it when the
osd tree is destroyed.
Libera started requiring an account for the IP ranges of GH CI.
This currently results in the IRC notification job running for
more than one hour because the IRC client it uses doesn't
terminate properly after giving up and thus eats our CI minutes.
The patch disables the notifications for now until we figure out
a better way, either by registering a libera account for the bot
and supplying the credentials via secret or by changing to GH
webhooks to a publicly available service like pipe.pico.sh which
then can be listened to by our existing IRC bot.
This fixes - among others - cases like `R.E.P.O..desktop`.
Due to the duplicated dot at the end we were matching against
a 0 sized string which was always true and thus would always
match the desktop file, regardless of what the app-id was.
A related issue is that an app-id of `osomething` would match
a desktop file name called `R.E.P.O.desktop` due to the string
being size of 1.
Fix this by requiring the partial desktop filename string to
be at least 3 characters wide.
Fixes#3257
