gtk scales the cursor size itself since its CSD so if we scale the size
its gonna get double scaled. incorporate the scale into xcursormanager
to keep track of it.
after commit 4b4971c it uses m_iFbAllocated and deletes if upon calling
release() but Renderbuffer generates directly on m_iFb without calling
alloc() meaning it wont be deleted on release(), set m_iFbAllocated to
true after generating the buffer.
this avoids the usage of the unique_ptr PROTO::protocol before it has
been constructed incase one wants to log something inside the
constructor itself, move the logging to macros and print file:linenumber
on ERR,CRIT,WARN and classname on the rest of the levels.
instead of potentially causing wonky behaviour from destructing in the
constructor add the unique_ptr reset to doLater and dont use the not
done constructed protolog in the constructor, call Debug::log directly.
see issue #7240
* Update fullscreen state dispatcher behaviour
* Change syncFullscreen default to false
* Revert all changes
* Modify fullscreenstate dispatcher toggle behaviour
* Update syncFullscreen according to state
* Update syncFullscreen before setting fullscreen state
* keybindmgr: avoid uint rollover on mouse keycode
mouse keycode is 0, and the switch case checks for 0 - 8 and rolls over,
just return early if keycode is 0.
* watchdog: avoid data races in watchdog
asan thread sanitizer reported data races in the watchdog from reading
and setting the bool variables make them std::atomic bools. also add a
atomic bool for the main thread to wait for to avoid data race when
reading the config values.
* hyprdebug: change non unicode character to name
asan created false positives and didnt like this bit, so for the sake of
easier debugging rename it to something unicode.
change the hook to monitorAdded instead of newMonitor so its finalized
in the compositor and added to vMonitors, move the
checkDefaultCursorWarp to PointerManager and check for it upon mode
change. and also ensure it doesnt go out of bounds by replacing it in
the middle again on resolution changes.
cursor:sync_gsettings_theme is set to default true and if enabled it
will now sync xcursor theme loading with gsettings if it can, meaning
CSD clients will now also change to the appropiate theme upon start and
hyprctl setcursor THEME SIZE .
* framebuffer: avoid gluint overflow
GLuint was being initialized to -1 and rolling over to unsigned int max,
its defined behaviour but very unnecessery. add a bool and use it for
checking if allocated or not.
* opengl: avoid gluint rollover
-1 rolls over to unsigned int max, use 0xFF instead.
* core: big uint64_t to int type conversion
there were a few uint64_t to int implicit conversions overflowing int
and causing UB, make all monitor/workspaces/windows use the new
typedefs. also fix the various related 64 to 32 implicit conversions
going around found with -Wshorten-64-to-32
avoid adding ill formed Inherit lines to inherit vector and later
scanning them, it wont change anything in practice but makes the inherit
theme parsing more in line with what its supposed todo. also check for
return values of the various string functions so we dont end up erasing
the wrong thing.
When both flip directions use more space than is available, pick the
direction that has more space available instead of just the opposite
of what was initially requested.
* cursormgr: reduce duplicated code
add a few functions such as setCursorBuffer and setAnimationTimer to
reduce duplicated code and also avoid future mishaps of forgetting to
clear buffer or disarm timer. and generally reduce spaghetti even tho
pasta can be delicious.
* xcursormgr: implent inherited themes
implent index.theme parsing and inherited themes.
* cursormgr: ensure a fallback xcursor exist
ensure a xcursor fallback exist otherwise it wont load the proper theme
if we at launch have hyprcursor enabled and then set it to false in
config and reload. also use the env var when using hyprctl setcursor
incase its empty.
* xcursor: bootleg xcursors into its own manager
implent XCursorManager and load themes based on librarypath and its
dir, now we catch all supplied theme files. and also implent animated
cursors. also refactor a bit of spaghetti regarding xcursors in
CursorManager.
* hyprcursor: fix buffer leak
animated cursors are creating a new buffer for each image, ensure we
drop the buffers so it continously doesnt build up in infinity.
* cursormgr: use eventloopmgr for animation
use EvenloopManager for timers instead of adding it directly to
m_sWLEventLoop and using its related wl_* functions.
src/Compositor.cpp:2295:74: error: no member named 'bit_floor' in namespace 'std'
2295 | const eFullscreenMode CURRENT_EFFECTIVE_MODE = (eFullscreenMode)std::bit_floor((uint8_t)PWINDOW->m_sFullscreenState.internal);
| ~~~~~^
src/Compositor.cpp:2296:74: error: no member named 'bit_floor' in namespace 'std'
2296 | const eFullscreenMode EFFECTIVE_MODE = (eFullscreenMode)std::bit_floor((uint8_t)state.internal);
| ~~~~~^
src/desktop/Window.cpp:1242:34: error: no member named 'bit_floor' in namespace 'std'
1242 | return (eFullscreenMode)std::bit_floor((uint8_t)m_sFullscreenState.internal) == MODE;
| ~~~~~^
there were a bunch of missing cursors, rework the shape loading add a
function to get legacyname from new wayland names. also bootleg add a
cursor if no theme can be found and no shape. to atleast show
something.
* bezier: dont loop on float values
Using a floating-point loop variable with a fixed increment can cause precision
errors over time due to the nature of floating-point arithmetic.
and cause undesired effects.
ex
iteration 1 = 0.10000000149011611938
iteration 2 = 0.20000000298023223877
eventually..
iteration 8 = 0.80000001192092895508
iteration 9 = 0.89999997615814208984
* hyprctl: close sockets on destruction
store socketpath and close the fd and unlink the socket path on exit.
* eventloopmgr: close the timerfd
close the timerfd on exit.
* debug: make logging thread safe
instead of opening and closing the logfile on each write open it on init
and close it on compositor exit. also add a mutex so accidently using
logging from a thread like the watchdog or similiar doesnt cause issues.
* xwl: clean up fd logic
check if the fd is actually opened before closing, and close the
pipesource FD on exit.
* protocols: avoid undefined behaviour in C macro
to safely use wl_container_of with a class the class has to be no
virtual functions, no inheritance, and uniform access control (e.g all
public)
work around this by putting this into a destroywrapper struct.
* opengl: clean memory on destruction
add a destructor and free the allocated memory and close the fd
Emits `SModifiersEvent` in `updateModifiers()`
Before the patch:
Changing layout with `hyprctl switchxkblayout ...` results in:
* active keymap in `hyprctl devices` is changed
* no event
* no layout is actually changed UNTIL you press one of the mod keys
(Alt | Shift | Super | Ctrl)
After:
* active keymap in `hyprctl devices` changed
* activelayout IPC event emitted
* layout is changed
This fixes https://github.com/hyprwm/Hyprland/issues/7044
XWayland does not use the regular commit(null) method to unmap, which results in buffers never being released.
release the buffers if present and un-released in the unmap() handler
ref #6584
fixes#6754
This will break if the client uses a transform that is not equal to the display, reverting to old behavior. Combining transforms is left as a todo for the future.
* compositor: implement wayland socket handover
This commit implements the compositor side of the Wayland socket
handover protocol as described in the [KDE Wiki]. The CLI options are
chosen so that they are compatible with Kwin.
[KDE Wiki]: https://invent.kde.org/plasma/kwin/-/wikis/Restarting
* main: verify that --wayland-fd is a valid file descriptor
* main: fail if only one of --socket and --wayland-fd is passed
The protocol says:
> The locked event "must not be sent until a new "locked" frame (either from a
> session lock surface or the compositor blanking the output) has been presented
> on all outputs and no security sensitive normal/unlocked content is possibly
> visible".
This helps users ensure the screen is properly locked before suspending
the machine. (e.g. with swaylock --ready-fd)
* add functions to hyprctl header
* refactor monitor json into own function and add it to header
* format hyprctl.hpp
* move functions to namespace
* move helper functions to class
* fix: change min cursor padding to 0
* chore: set default hotspot padding to 0
* fix: adjusting clamp after getting closest point to fix getting off limit point
* fix: deal with floating point to clamp to prev value
* refactor: max coords to vector
* fix: remove box closestPoint adjustment due its fix on hyprutils
in X11 some surfaces is a parent of itself and creates a cyclic loop
when trying to find its parent. check for old parent and break if its
beginning to roll over.
some weird combination of scrolling/nesting hyprland and closing a
window i managed to divide by zero here, reported by ubsan. add a check
to ensure we dont hit UB.
* watchdog: dont detach and cause race condition
instead of detaching and causing a race condition on destruction where
the thread is alive and watchdog has been destroyed, check if its
joinable and join it on destruction.
causes heap use after free on exit of compositor.
* render: add checks for compositor shutting down
avoid member call on null pointer, if the g_pHyprRenderer is destroyed
we can call the member makeEGLCurrent on it, causes undefined behaviour
on destruction of the compositor/hyprrenderer.
found with ubsan.
* surfacerole: add virtual destructor
all classes that will be derived from should have a virtual destructor
otherwise deleting an instance via pointer to a base class is undefined
behaviour, layershell/xdgshell hits this with std::default_delete in the
new sharedptr implentation.
* includes: fix missing includes
fix missing includes for no precompiled headers builds, and remove a
redefiniton of a macro already defined in macros.hpp
if we are blurring, we cannot be sure whether the occluded region won't be included in the expanded damage. If it is, we'd get dark shimmers.
fixes#6547
* pointer_manager: add lock/unlock software wrappers that receive the raw pointer
* monitor: lock/unlock software pointer rendering when adding/removing mirrored screens
* use relative path in includes
* window: use const references instead of copies
use const references instead of wasteful copies and make the = operator
check for self assignment and return early. also use const in all the
other operators.
* listener: pass std::function as const reference
instead of copies pass the std::functions as const references.
* config: dont unnecessarily convert to c_str
getHyprlangConfigValuePtr wants an std::string and we already have an
std::string, dont convert it to a c_str only for it to be converted back
to an std::string.
* buffer: pass attributes as const reference
pass attributes as const reference instead of copies.
sometimes there is no focused device (e.g. when dnd'ing on nothing or xwayland) in which case abort would fail to send cancelled to the source.
ref #6543
* input: add cursor:warp_on_changeworkspace
If enabled, warp the cursor to the last focused window on the workspace in the `changeworkspace' dispatcher, except if the cursor is currently on the WLR top layer.
Respect persistent warps.
* warp_on_change_workspace: check if focused layer is a window.
* ## Open window relative to active window
`new_on_active`:
- `none` (default):
- `before`: above of the focused window
- `after`: below the focused window
If the focused window is the solo master window, or the new window replaces master, this option has no effect and new_on_top are respected.
## Refine new window status control
**BREAKING CHANGE**: new_is_master removed in favour of new variable
`new_status`:
- `slave` (default): new window open as slave
- `master`: new window open as master
- `inherit`: new window inherit status from active window, i.e. when the focused window is master, new window will become new master, otherwise new window are added to slaves
* refactor: rename a few variables
* layerSurface: fix layer being refocused every commit with on_demand
Fixes#6477
The surface will now only receive focus when its keyboard
interactivity is more than the previous keyboard interactivity in the
order none -> on_demand -> exclusive.
* layerSurface: only kb focus if becoming exclusive
* renderer: add destructor and destroy event source
add destructor and destroy the event source.
one less leak on exit of compositor reported by asan.
* compositor: cleanup eventloop on exit
destruct hyprctl to release the event sources, and properly cleanup the
event loop on exit of compositor. less leaks on exit reported by asan
* threadmgr: destroy event source on destruction
destroy the event source on destruction.
* eventloopmgr: reset eventloopmgr on exit aswell
reset the eventloopmanager on exit of compositor and free the leaking
last idle frame on monitor destroy.
* core: move to hyprutils for utils
Nix: add hyprutils dep
* Meson: add hyprutils dep
* flake.lock: update
---------
Co-authored-by: Mihai Fufezan <mihai@fufexan.net>
synchronous buffers are read instantly and we can release them, but asynchronous ones have to be locked until they are unref'd from .current to avoid reading from a buffer after .release()
When gestures:workspace_swipe_min_fingers is enabled,
gestures:workspace_swipe_fingers is considered to be the minimum
number of fingers required to swipe.
This behavior is more similar to sway and macOS's default behavior.
For example, this allows you to set workspace_swipe_fingers to 3,
but swipe with 4 or more fingers instead of 3.
Allows the cursor to return to its last relative position within a window when the window is refocused.
Allows the cursor to retain its relative position within a window when the window is swapped, moved, changed workspace, added to or removed from groups.
controlled with cursor:persistent_warps
* pointermgr: add destructor to state and free buf
if the pointer has a buffer set it wont be freed upon destruction, make
asan more happy by adding a destructor and wlr_buf_unlock it on exit.
* cursormgr: free the animation timer event source
properly free the animation timer event source on destruction.
* compositor: free the critsig event source on exit
properly free the critical signal event source on exit.
* popup: clang format style
clang format.
* Discrete scrolling round away from zero
e.deltaDiscrete can be multiples of 30 instead of the usual 120 causing
the rounded value to be 0 when too small causing erratic scrolling.
* Send value120 alongside discrete
Fixes sensitivity issues for clients that support value120 axis events
* xwayland: add destructor to CXWM and free resource
the wl_event_resource was running upon destruction of the compositor
causing a null pointer segfault in onX11Event so ensure the event is
removed upon destruction, also free the memory allocated by
xcb_errors_context_new and finally call xcb_disconnect on the connection
to free the fd and its memory.
* hyprctl: dont leak the fd on destruction
add a destructor and properly free the fd on destruction
* eventloop: add destructor and free event source
properly free the wl_event_source upon destruction.
Clients using zxdg_decoration_manager_v1::get_toplevel_decoration may
expect a receiving a zxdg_toplevel_decoration_v1::configure event to
determine the initial decoration mode, without having to go through a
zxdg_toplevel_decoration_v1::set_mode request. Hyprland was not sending
this event, resulting in unwanted decorations being drawn.
Specifically, clients using libdecor, e.g. applications using recent
GLFW, would draw GTK decorations with artefacts. This change fixes
these.
