The raise_on_focus_delay is meant to dampen z-order churn from
focus-follows-mouse cursor passes. Applying it to every focus
change meant explicit actions (alt-tab cycle finish, Focus action,
xdg/xwayland activation, view map, etc.) also waited for the delay
before raising, which felt laggy.
Route all non-sloppy-focus callers through an immediate raise and
keep the timer-based raise only for desktop_focus_view_or_surface(),
which is the sloppy-focus entry point.
If the user disables raiseOnFocus or lowers raiseOnFocusDelay while
a raise is queued, the queued raise should not fire against the new
config. Cancel it in reload_config_and_theme() before rereading the
rc.xml.
The pending_auto_raise_view pointer would become dangling if the
view it references is destroyed before the timer fires. Clear it
in view_destroy() alongside the existing active_view cleanup.
When raiseOnFocus is enabled and raiseOnFocusDelay is > 0, defer
view_move_to_front() until a wl_event_loop timer elapses instead of
raising immediately. The pending view/timer pair lives on struct
server so that at most one raise can be pending per compositor.
Semantics:
delay_ms == 0 : immediate raise (preserves prior behavior)
delay_ms > 0, hover A : schedule raise of A after delay_ms
another hover before : timer is reset and the new view becomes
expiry the pending one (brief flyby hovers do
not stack up raises)
focus change, raise=false: pending raise is cancelled
This is most useful with followMouse to avoid brief passes of the
cursor stacking up z-order changes.
Add a new <focus><raiseOnFocusDelay> element accepting an integer in
milliseconds. The default of 0 preserves the current behavior (raise
immediately when raiseOnFocus is enabled).
The new field is carried as uint32_t raise_on_focus_delay_ms on
struct rcxml. This commit only adds the parser and default; the
actual delay logic follows in a subsequent commit.
Modifier+right-drag resize was silently ignored on fully maximized views
because of an early-return guard in interactive_begin(). The axis-specific
un-maximization logic introduced in #3043 already handles partial
maximization correctly; extend that to the VIEW_AXIS_BOTH case so both
axes are cleared while keeping the current geometry as the starting point
of the resize.
Move already permits dragging maximized views, so this also removes an
asymmetry between the move and resize paths.
Fixes: #3524
Use wlr_client_buffer->texture directly instead of wlr_texture_from_buffer.
All buffers in content_tree are wlr_client_buffer.
wlr_texture_from_buffer calls client_buffer_begin_data_ptr_access which
fails when client_buffer->source == NULL (client released buffer early,
e.g. wl_shm foot-terminal).
Add action `DebugToggleKeyStateIndicator` to show in the top-left corner
the pressed, bound, pressed-sent keys as well as modifiers.
This should help fault find issues like #2771 and #3238
Based-on: #3262
Co-authored-by: @tokyo4j
Android's bionic libc implements secure_getenv() as a function that
always returns NULL because app processes don't have AT_SECURE set.
This prevents xkbcommon from reading XKB_DEFAULT_LAYOUT and other
environment variables when resolving keyboard layouts.
Use XKB_CONTEXT_NO_SECURE_GETENV on Android so xkbcommon falls back
to regular getenv(), which works correctly in the Android app
environment.
It was used to get window icon via _NET_WM_ICON, which is now
implemented by wlroots 0.20. Anyone who needs another atom can revert
this commit and add atoms in the `atoms` array.
Ref: 515275ee7214bf91f8a758b660093eb4b932195a
(wlr_scene: Introduce wlr_scene_set_gamma_control_manager_v1)
This wlroots change eliminates the need for separate event tracking for gamma
control application.
v2: Fix code style
v3: Rebase now that 0.20 is merged
Ref: 84d603acc06a45dd3c3a4b2cf1fd08b2933ca2b5
(xwayland: take wlr_buffer in wlr_xwayland_set_cursor())
Ref: 6ae54dca23064e897b393283887986e5719a747f
(xwayland: lock new buffer instead of the old one)
Co-Authored-By: Consolatis
This wlroots change fixes a potential UAF which we dealt with in labwc.
We can thus remove the workaround completely.
Ref: 06275103f249cd2954630e59383342e102a6c1a3
(input-method-v2: Destroy keyboard grab before input method)
Background:
My MR in wlroots (!5107) stopped emitting `wlr_input_method_v2`
on its `commit`/`destroy` events, but didn't stop emitting
`wlr_input_method_keyboard_grab_v2` on its `destroy` event. That was
because `handle_keyboard_grab_destroy()` was called *after*
`handle_input_method_destroy()` for some reason, which caused segfault
when dereferencing `relay->input_method.keyboard_grab` in
`handle_keyboard_grab_destroy()`.
MR 5170 reversed this weired order of destroy handler calls, and finally
stopped emitting `wlr_input_method_keyboard_grab_v2` on its `destroy`
event.
Empirically, the associate event is always seen just after map_request
but before surface map. Window properties are also read by wlroots just
before emitting associate. So after some trial and error, this seems to
be the best place to set initial view states and compute initial
placement.
Fixes initial positioning of "urxvt -g -0-0".
urxvt placement regressed in:
9903331995
("xwayland: rework setting initial geometry from surface")
map_request handler was added ~2 years ago in:
c9018da4c4
("xwayland: set initial geometry in map_request handler")
I'm not sure if the map_request handler was incorrect from day one,
or if something changed in wlroots and/or Xwayland since then.
The basic idea is to set the initial geometry from the surface exactly
once, just before we need it, i.e. either (1) when mapping the view or
(2) right before processing an initial maximize/fullscreen request.
I've consolidated various parts of the initial geometry setup to take
place at this point (in ensure_initial_geometry_and_output()).
The main motivation is to have a valid, adjusted floating geometry for
the view *before* saving the natural geometry when processing an initial
maximize/fullscreen request. This reduces code duplication and addresses
a FIXME in set_initial_position(), as well as fixing an issue where the
natural geometry could exceed the usable output area.
Some other minor changes:
- The initial output is now set directly from the surface geometry if
the "application/user-set position" hint is given. This is unlikely
to matter in practice, but theoretically an initially maximized view
could now appear on a different (application-chosen) output.
- Floating view size is now constrained to the usable area even if a
position hint is set. It seemed inconsistent that `xterm -g 200x200`
was constrained but `xterm -g 200x200+0+0` was not.
wlr_scene_*_create() functions all allocate memory via calloc() and
return NULL if the allocation fails. Previously, the failures were
handled in any of 3 different ways:
- sending a wayland protocol error
- exiting labwc with an error
- segfault (no NULL check at all)
Since labwc does not attempt to survive heap exhaustion in other
allocation paths (such as `znew`), it seems more consistent to use the
same die_if_null() check used in those paths to exit with an error.
For the three most common create() functions (tree, rect, buffer),
add small lab_wlr_ wrappers to common/scene-helpers.
When an X11 client clears override-redirect on a window (i.e. it
becomes "managed") without an unmap/map cycle, wlroots doesn't
re-emit the set_title/class/icon events. We need to synthesize
them to re-read the values from the surface.
From what I can tell, only older/misbehaving X11 clients would
actually do this (toggle override-redirect without unmap/map), so
this is perhaps a bit pedantic, but an easy fix anyway.
Before this commit, ssd_update_title() returned early and the title was
not updated when the title gets empty. Furthermore, this caused the
title buffer to be misplaced outside of the titlebar when the window is
resized.
Reproducer:
1. Build and launch https://github.com/tokyo4j/wayland-demo/tree/repro-labwc-ssd-title-not-updated
2. One second later, the window title gets empty. But the title in
SSD is not updated.
3. Un-minimize the window. Now the title buffer is misplaced outside of
the titlebar.
