m_alpha was freed by fullyDestroy, but was then being touched because setCallbackOnEnd is activated by tick, which is the same function that updates animating variables
---------
Co-authored-by: Vaxry <vaxry@vaxry.net>
Rewrites layouts to be much smaller, and deal with much less annoying
BS. Improves the overall architecture, unifies handling of pseudotiling,
and various other improvements.
* pointermgr: damage only the surface size
CWaylandOutput returns a vector2d with -1, -1 set as a "no limit",
passing that down into beginSimple and the renderer it hits pixman bug
of invalid sizes and wrong rectangles gets created.
causing bunch of
*** BUG ***
In pixman_region32_init_rect: Invalid rectangle passed
set the damage either to cursor plane size or fallback to 256x256.
* pointermgr: dedup if hw cursorsize checks
dedup a bit if else casing
* multigpu: fix multi gpu checking
drmFD() from allocators is not always equal, because we reopen them
inside AQ for refcounting, meaning they get duplicated and become their
own fds, so checking if fd1 == fd2 ends up wrong.
introduce sameGpu in MiscFunctions that checks the actual drmDevice
meaning we can now even check if a rendernode is the same gpu as a
display node if we want.
* multigpu: move sameGpu to DRM namespace
move sameGpu out of MiscFunctions to DRM namespace.
we cant unlock all states on empty commits, at best tryProcess them.
for example if a state is locked and waiting for a fence to become readable,
and another commit comes in we cant unlock it until the fence has actually signaled.
* directscanout: fix dgpu directscanout explicit sync
without setting an infence, AQ doesnt explicit sync, nor recreate the
dgpu fence for the blit work. and as such attemptdirectscanout path
artifacts and breaks. create a dummy CEGLSync even tho we dont really
have any pending glwork to get a proper fence, and set it.
* monitor: dont use new scheduling if direct scanout
using the new_render_scheduling makes no sense in the direct scanout
path, add a if guard against it.
* move commit timing fields to surface state
* fix toTimespec init
* update sendQueued api
* update onPresented api
* set zero copy flag
* send clock id
* move presented calcs inside condition
* use only CLOCK_MONOTONIC for commit/presentation timings
* fix setSetTimestamp
* do not wait for commit timing while tearing
* proto config
* fix config defaults
add debug:gl_debugging so we can disable gl debugging entirerly,
both glGetError and enabling EGL_KHR_debug has its cost, we still have
EXT_create_context_robustness and glGetGraphicsResetStatus that should
catch context loss, and is generally cheap to call it only checks a flag
set.
glGetError might cause a implicit flush to get any pending calls sent to
the gpu.
however to get EGL_KHR_debug back enabled we now require a restart of
the compositor after changing debug:gl_debugging
