It does not make sense that the event "source" (aka the target) is
both content and recipient of a message. Not doing so, events become
largely independent of the actor that is receiving/handling an
event. This is small step toward making events opaque and immutable.
Every user of these API calls in our code have ported away from
them, but other users may remain in extensions, so make these
functions work on top of the alternative API without accessing the
soon to be removed event field.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2311>
This is just "necessary" for --nested stages, since the pointer is
allowed to leave the stage in that case. Since the only side effect
is that there is still a pointer focus somewhere inside the stage,
simply drop this.
This is a small leftover of commit b8f92a6ce4, since we stopped
handling the double ENTER event there.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2311>
In addition to the presented callback time, it shows the time to the
reported presentation time (which can be earlier or later than the
presented callback), as well as the GPU rendering duration.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1928>
The distribute_natural_allocation() function was copied over from Gtk to
Clutter 11 years ago with commit e636a0bbce.
Gtk only supports integers sizes in its layout machinery, while Clutter
does everything using floats.
Since this function sets the minimum_size (the size we allocate the
children in the end) to an integer, this means we're implicitly
typecasting floats to integers here, effectively floor()'ing all sizes
that we allocate the box children.
A bug this caused in gnome-shell was that a scrollView (like the one in
the endSessionDialog) was showing scrollbars even though the content
perfectly fit inside the view: Say the content and its scrollView parent
request a size of 63.9 px, but get allocated a size of 63 px by a box
layout. Now the scrollView notices that its allocated size is smaller
than the requested size and thus shows a scrollbar.
So fix that and use floats in distribute_natural_allocation() instead of
integers, as we do everywhere else in the layout machinery.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2284>
Without input device grabs in play, all functions that emit
pointer/key/crossing/touch events are pretty much the same. Remove this
duplication and use a common emit_event() function.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2100>
In case of misuse (e.g. passing NULL stage) this might result in crashes
before the precondition checks managed to kick in. Move this priv variable
initialization after these checks.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2099>
Wayland event processing and WM operations are themselves outside the
ClutterGrab loop so far. Until this is sorted out, these pieces of
event handling have got to learn to stay aside while there is a
ClutterGrab going on.
So, synchronize foci and other state when grabs come in or out, and
make it sure that Wayland event processing does not happen while
grabs happen.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2099>
Since we want these accessed from bindings this must be a boxed
type. This has the side effect of making ClutterGrab a refcounted
object, since we want to avoid JS from pointing to freed memory
and maybe causing crashes if misusing the object after dismiss.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2099>
Toggling the click action on when leaving the actor/action sounds weird,
this was presumably meant to toggle it off on leave, and back to in_held
on enter. This way, the CLUTTER_LEAVE handling also matches what we want
to do in case of grabs.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2099>
The lack of handling of regular crossing events here is dubious, perhaps
to be fixed later on. So far, ensure gestures are cancelled whenever
a grab-inducted crossing event would leave this action in the blue.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2099>
This is (luckily!) unused, and it's inconvenient to have a toggle to
break the input model we are striving towards. Drop this function
and stick to the default behavior.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2099>
This looks like a relic of glReadPixels-based picking, the pointer
might well be outside redrawn areas, yet still require a device
update (e.g. in order to reflect the actor layout changes in the
"clear area" info).
Instead, always update all devices that are inside the view after
relayouts, the tracking on the need for that update is now done
on each ClutterStageView, instead of globally in the ClutterStage.
This theoretically fixes situations where pointers might miss
updating their "clear area" after the actor tree changed.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2117
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2257>
The stage window is an interface, that added properties, that were only
then actually managed by MetaStageImpl. Shuffle things slightly, and let
the MetaStageImpl object deal with these things itself.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2014>
As ClutterGrab is a stack, the backend only cares about some grab
existing currently or not. Make it sure that we grab whenever we
go to >=1 grabs, and ungrab whenever we go to <1.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2068>
Dissociate clutter_stage_set_key_focus() from the actors focused
state, so that it obeys stage grabs. The key focus actor state may
also change due to grab changes, add the code to notify about this.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2068>
Emit crossing events whenever a grab coming or going would cause a
pointer/touchpoint to become inactive on their position. Depending
on whether the pointer lies inside the old or new grab widgets,
enter or leave events would be generated.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2068>
We will want to be more specific about the portions of the actor
hierarchy that receive this event, separate creation and emission
so each place does what is relevant.
However, this commit brings no functional changes.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2068>
These events are not meant to be ever silenced away, every actor
that is meant to receive one should do so. Make it sure that those
events cannot be stopped, despite the event signal handlers return
values.
This opens the debate about whether crossing events should be
ClutterEvents, since they are more and more uncommon at being one,
maybe this notification mechanism should be taken away from the
event machinery, but that's something for future refactors.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2068>
Hopefully, the one to make them all converge. This new ClutterGrab
represents a handle on a created grab. These are stacked, so grabs
can be overridden and remain inactive until there is a time that
they become active again, although undoing these early is optional.
These grabs are global, they do apply to all pointer, touchpoint
and keyboard foci.
At the moment, only the API to create and stack those is added,
the actual functionality is added in future commits.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2068>
A property for reversing the visible order of children is a bit odd.
It has also been unused by actual gnome-shell code since 2010, and the
somewhat related pack_start()/pack_end() API in GtkBox(Layout) is gone
in GTK4.
With that in mind, turn the property into a no-op and deprecate it,
so that it can be dropped next cycle.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2206>
This ensures they remain perfectly smooth regardless of how the
dispatch time has been adjusted/optimized/delayed/jittered.
Idea by Ivan Molodetskikh <yalterz@gmail.com>
For example, dragging a window on a 60Hz monitor:
BEFORE
delta(time_us) = 17014μs
delta(time_us) = 15998μs
delta(time_us) = 17006μs
delta(time_us) = 16975μs
delta(time_us) = 16001μs
delta(time_us) = 17002μs
delta(time_us) = 17006μs
delta(time_us) = 16004μs
AFTER
delta(time_us) = 16667μs
delta(time_us) = 16667μs
delta(time_us) = 16670μs
delta(time_us) = 16667μs
delta(time_us) = 16669μs
delta(time_us) = 16668μs
delta(time_us) = 16664μs
delta(time_us) = 16674μs
Caveat 1: Because we don't know a "next presentation time" on the first
frame, the interval between the first and second frame will usually be
different to the subsequent steady interval. So this change increases the
jitter of just frame 2, but eliminates jitter thereafter.
Caveat 2: `clutter_frame_clock_schedule_update_now` schedules updates
earlier than `clutter_frame_clock_schedule_update`. This means potentially
you could get multiple frames targeting the same "next presentation time".
That doesn't really change here though - we're dispatching at the same
times as we used to and just giving timelines a better vsync-aligned
timestamp now.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/25
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2161>
This won't change anything for 60Hz displays but higher refresh rate
users will benefit.
Using Nvidia EGLStreams on a 240Hz monitor for example (refresh interval
~4.1ms), the maximum render time allowed before dropping to 120Hz is now
3.6ms whereas it was previously 2.1ms.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2158>
This is notably necessary with transformations, since these don't
trigger allocation machinery, but may affect the actor under the
pointer.
Visible e.g. with GNOME Shell's "Application does not respond"
dialogs.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
With Wayland handling all events as they come, this code now just
performs motion compression for events that will be handled by Clutter
widgetry.
The intent to opt tablets and styli out of motion compression was
early and fast client handling, since that is now covered in a generic
manner, this code is superfluous. We don't really need the extra events
for these devices in compositor widgetry either.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
We essentially create those at the time they need to be handled, and
use shortcuts that avoid the event from being queued up. It's too much
of a short cut though, these events are also of interest to the Wayland
event handlers, e.g. to handle pointer state changes (e.g. repicks due
to the pick actor being destroyed) immediately, instead of at the next
event.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
If we are still under the "clear area" of the pick actor, we forget
to update the coordinates. This is usually not needed, unless we
need to repick again for non-event circumstances (e.g. pick actor
is destroyed). This will ensure the right pointer coordinates are
used afterwards in those situations.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
Traditionally, the next repaint would also involve picking, which
would correct the actor under the pointer. This now does not happen
out of the box, so we really are waiting for the next pointer event
here.
To avoid the pointer/cursor to lag behind, trigger an immediate
repick here, that will look up the new actor under the pointer
coordinates.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
And resort to it first, unless we are told to ignore the cache
(e.g. after relayouts). This avoids further pick context operations
while the pointer is on the current actor.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
This safe area is the region (in stage coordinates) where the pointer
is ensured to stay within the current actor. This is not used yet, but
will be used for optimizations in pointer picking.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
These may be used for optimizations once we find the pick actor,
so picking can be avoided in areas we know didn't cross into
other actors. Nothing makes use of it yet though, just log these
so far.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
Add a clutter_stage_pick_and_update_device() method that is the only
single entry point for updating a device position as seen by the
stage.
Also, update all callers to use it.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
The clutter_stage_get_actor_at_pos() calls it almost 1:1 underneath
and is public API, we can have all callers use this, and stop using
this function outside of clutter-stage.c.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
As event handling goes:
1) Events get generated and queued by the seat (from another thread in
native, in the same thread in X11)
2) The MetaBackend gets those events and forwards them to Clutter
via clutter_do_event()
3) The events get queued in the ClutterStage
4) At the time of processing a frame, the input events are processed,
5) Motion events are throttled, only the last is effectively handled
6) Events are filtered, wayland and WM handling happens here
7) Events maybe reach to clutter
This commit moves 6 to happen between 2 and 3. The end result is that:
- Throttling only applies to Clutter event handling, The wayland event
forwarding bits will handle the event stream as soon as it comes, as
timely as possible.
- WM event handling is also unthrottled, but that's more of a side
effect.
- This all still happens on the main thread, so there's the possibility
that other busy areas (e.g. relayout) temporarily block this event
forwarding.
- Sending events unthrottled inherently means more CPU, probably
dependent on input devices' frequency. The impact is not measured.
This should bring the best of both worlds with e.g. 1000Hz mice, wayland
clients get unthrottled events, while GNOME Shell UI still behaves like
it used to do.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
If we wait till finalize, dispose will destroy the actor hierarchy
and cause untimely repicks. Ensure to free the pointer/touch info
first, so the hooked signal callbacks are gone when destroying the
actors.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1915>
In the case a11y is required, the screen reader is very much
interested in getting an uninterrupted flow of key events. It attempts
so by setting a ::captured-event callback on the ClutterStage, but
that falls short with our MetaDisplay event handler, as clutter events
can be stopped before a11y gets a chance to see them.
This kind of selective amnesia wrt key events is not new, in X11 those
go unheard of by the WM as long as a client is focused and no grabs hold,
so it is clients' responsibility to talk with AT bridge.
This commit doesn't yet change that for X11, but we can do this right
away from the compositor on Wayland, and without any chance to be
tampered by clients.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1328>
If one would end up with an actor attached to mapped actor, where the
attached actor doesn't itself have an up to date stage view list while
listening on the stage for updating, when clearing the stage views of
the list, anything that would query the stage views list at this time
would end up accessing freed memory.
This could happen if
1) An actor was added to a newly created container actor attached to
the stage
2) The actor got a timeline attached to it
3) The actor was moved to a container that already was mapped
4) A hotplug happened
After (1) both the container and actor would not have any stage views.
After (2) the timeline would listen on the stage for stage views
updates. After (3) the actor would still listen on the stage for stage
views updates. When (4) happened, the actor would be signalled when the
stage got its stage view cleared, at which point it would traverse up
its actor's tree finding an appropriate stage view to base its animation
on. The problem here would be that it'd query the already mapped
container and its yet-to-be-cleared stage view list, resulting in
use-after free, resulting in for example the following backtrace:
0) g_type_check_instance_cast ()
1) CLUTTER_STAGE_VIEW ()
2) clutter_actor_pick_frame_clock ()
3) clutter_actor_pick_frame_clock ()
4) update_frame_clock ()
5) on_frame_clock_actor_stage_views_changed ()
6) g_closure_invoke ()
7) signal_emit_unlocked_R ()
8) g_signal_emit_valist ()
9) g_signal_emit ()
10) clear_stage_views_cb ()
11) _clutter_actor_traverse_depth ()
12) _clutter_actor_traverse ()
13) clutter_actor_clear_stage_views_recursive ()
14) clutter_stage_clear_stage_views ()
...
Avoid this issue by making sure that we don't emit 'stage-views-changed'
signals while the actor tree is in an invalid state. While we now end up
traversing tree twice, it doesn't change the Big-O notation. It has not
been measured whether this has any noticible performance impact.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1950
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2025>
This does two things to frown upon:
- Modifies ClutterEvent structs, while the effort is to have those
completely opaque, and readonly after creation from the input
thread side.
- Stores state in the ClutterInputDevice struct, event though those
are also considered static after creation, managed by the input
thread, etc.
Stop doing that. This makes all events just forwarded as-is in
the ClutterStage/clutter-main.c code.
Handling of click count sounds like material for a ClutterGestureAction
(or perhaps ClutterClickAction), all of both callers now do it in place
at the moment, while gestures lack a better state tracking and management.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2024>
This will not try the captured-event shenanigans to emulate grab
behavior, instead relying on event delivery being influenced by
other grab mechanisms.
While at it, improve handling of additional touchpoints by
cancelling the click action right away, as the differences in
event handling make this unwanted behavior surface.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2024>
By default, the pan action performs matrix translations on the
child widget. Nobody wants that (or, nobody wants *just* that).
It's cleaner not to mix mechanism and effect in ClutterGestureAction
subclasses, so drop this base implementation, and change the signal
accumulator so it's more similar to event signals (not that it's
used any longer, anyway).
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2024>
This warning is actually dead code, since should_be_mapped and
must_be_realized are always set to the same value, so it does not
make sense to check for "a && !b".
Turn this into an assert so we avoid the dead branch, but do not
remove the variable duplication so the more aptly named variable
is used where it belongs, for clarity.
CID: #1506254
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2061>
It was a feature relevant for when Clutter was an application toolkit
that wanted the application window to communicate a minimum size to the
windowing system.
Now, clutter is part of the windowing system component, so this feature
doesn't make any sense, so remove it.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2002>
This feature was configured depending on whether the Cogl backend
reported COGL_WINSYS_FEATURE_MULTIPLE_ONSCREEN or not. All cogl backends
do report this, so any code handled the 'static' case were never used.
While we only ever use one stage, it's arguable more correct to
consilidate on the single stage case, but multiple stages is something
that might be desirable for e.g. a remote lock screen, so lets keep this
logic intact.
This has the side effect of completely removing backend features, as
this was the only left-over feature detection that they handled.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2002>
This changes the setup phase of clutter to not be result of calling an
init function that sets up a few global singletons, via global singleton
setup vfuncs.
The way it worked was that mutter first did some initial setup
(connecting to the X11 server), then set a "custom backend" setup vfunc
global, before calling clutter_init().
During the clutter_init() call, the context and backend was setup by
calling the global singleton getters, which implicitly created the
backend and context on-demand.
This has now changed to mutter explicitly creating a `ClutterContext`
(which is actually a `ClutterMainContext`, but with the name shortened to
be consistent with `CoglContext` and `MetaContext`), calling it with a
backend constructor vfunc and user data pointer.
This function now explicitly creates the backend, without having to go
via the previously set global vfunc.
This changes the behavior of some "get_default()" like functions, which
will now fail if called after mutter has shut down, as when it does so,
it now destroys the backends and contexts, not only its own, but the
clutter ones too.
The "ownership" of the clutter backend is also moved to
`ClutterContext`, and MetaBackend is changed to fetch it via the clutter
context.
This also removed the unused option parsing that existed in clutter.
In some places, NULL checks for fetching the clutter context, or
backend, and fetching the cogl context from the clutter backend, had to
be added.
The reason for this is that some code that handles EGL contexts attempts
to restore the cogl EGL context tracking so that the right EGL context
is used by cogl the next time. This makes no sense to do before Cogl and
Clutter are even initialized, which was the case. It wasn't noticed
because the relevant singletons were initialized on demand via their
"getters".
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2002>
This one is a trivial wrapper around clutter_actor_get_children(), so just
use that in the two places where clutter_container_get_children() is used,
and remove clutter_container_get_children().
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2057>
Right now we damage the stage even if an actor is not mapped, for
example in the overview.
Stop doing so, reducing over-paint significantly in some situations.
Clones will still do stage damage on their own.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2035>
ClutterText implements its own get_paint_volume() with its own cache,
but was not invalidating the actor paint volume when when it has
changed. This sometimes could result in labels, especially quickly
changing ones, using the old paint volume which either would cut off the
label or leave parts of the old label on screen.
Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1943
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2006>
This mode is passed along by the ClutterInputMethod, the
ClutterInputFocus will preserve it and ensure it is honored
whenever the IM is being reset.
This mode is immediate. The ClutterInputFocus commits the
text directly without queueing a CLUTTER_IM_COMMIT event.
This is important so events are serialized in the right order
in the wayland implementations (i.e. commit before wl_pointer.press).
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1940>
In line with GTK, the input method context should be reset when clicks
are handled by the ClutterInputFocus user. The reset action can then
either clear or commit the preedit text, as configured by the IM module.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1940>
Make sure that when we've recreated views that we'll actually paint a
new frame for it. This was very rarely a problem, as views tend to
result in getting damage etc being queued as side effects of various
things, like layout, but e.g. when running certain tests, this might not
happen. There is no situation where we want to create a new view that
should remain unpainted, so just make sure we initialize it to become up
to date.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1947>
This code sneaked unconditionally, even though we can disable
tracing code with -Dprofiler=false. Add some COGL_HAS_TRACING
checks so that this code is also optionally built.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1951>
Will be used to trace a lot more, and with more details, and thus may
have a larger impact on what is actually measured. This potential impact
is the reason for enabling only when needed.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1700>
The failure to allocate was not properly handled, causing crashes later
on due to the offscreen being NULL.
#0 cogl_gl_framebuffer_bind (target=36160, gl_framebuffer=0x0)
#1 _cogl_driver_gl_flush_framebuffer_state (...)
#2 cogl_context_flush_framebuffer_state (read_buffer=0x55f48f386780, draw_buffer=0x55f48f386780, ...)
#3 cogl_framebuffer_clear4f (framebuffer=0x55f48f386780, ...)
#4 clutter_layer_node_pre_draw (...)
#5 clutter_paint_node_paint (...)
...
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1942>
We only listen to it for 2 settings (drag threshold, double click
time), and we already have the stock ClutterSettings object tracking
the source of these. This code is redundant.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1862>
Not sure how to update the damage or redraw clip or something; at least
this works properly when under a constantly-redrawing window, which is
ok for debugging purposes.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1762>
Max render time shows how early the frame clock needs to be dispatched
to make it to the predicted next presentation time. Before this commit
it was set to refresh interval minus 2 ms. This means Mutter would
always start compositing 14.7 ms before a display refresh on a 60 Hz
screen or 4.9 ms before a display refresh on a 144 Hz screen. However,
Mutter frequently does not need as much time to finish compositing and
submit buffer to KMS:
max render time
/------------\
---|---------------|---------------|---> presentations
D----S D--S
D - frame clock dispatch
S - buffer submission
This commit aims to automatically compute a shorter max render time to
make Mutter start compositing as late as possible (but still making it
in time for the presentation):
max render time
/-----\
---|---------------|---------------|---> presentations
D----S D--S
Why is this better? First of all, Mutter gets application contents to
draw at the time when compositing starts. If new application buffer
arrives after the compositing has started, but before the next
presentation, it won't make it on screen:
---|---------------|---------------|---> presentations
D----S D--S
A-------------X----------->
^ doesn't make it for this presentation
A - application buffer commit
X - application buffer sampled by Mutter
Here the application committed just a few ms too late and didn't make on
screen until the next presentation. If compositing starts later in the
frame cycle, applications can commit buffers closer to the presentation.
These buffers will be more up-to-date thereby reducing input latency.
---|---------------|---------------|---> presentations
D----S D--S
A----X---->
^ made it!
Moreover, applications are recommended to render their frames on frame
callbacks, which Mutter sends right after compositing is done. Since
this commit delays the compositing, it also reduces the latency for
applications drawing on frame callbacks. Compare:
---|---------------|---------------|---> presentations
D----S D--S
F--A-------X----------->
\____________________/
latency
---|---------------|---------------|---> presentations
D----S D--S
F--A-------X---->
\_____________/
less latency
F - frame callback received, application starts rendering
So how do we actually estimate max render time? We want it to be as low
as possible, but still large enough so as not to miss any frames by
accident:
max render time
/-----\
---|---------------|---------------|---> presentations
D------S------------->
oops, took a little too long
For a successful presentation, the frame needs to be submitted to KMS
and the GPU work must be completed before the vblank. This deadline can
be computed by subtracting the vblank duration (calculated from display
mode) from the predicted next presentation time.
We don't know how long compositing will take, and we also don't know how
long the GPU work will take, since clients can submit buffers with
unfinished GPU work. So we measure and estimate these values.
The frame clock dispatch can be split into two phases:
1. From start of the dispatch to all GPU commands being submitted (but
not finished)—until the call to eglSwapBuffers().
2. From eglSwapBuffers() to submitting the buffer to KMS and to GPU
work completing. These happen in parallel, and we want the latest of
the two to be done before the vblank.
We measure these three durations and store them for the last 16 frames.
The estimate for each duration is a maximum of these last 16 durations.
Usually even taking just the last frame's durations as the estimates
works well enough, but I found that screen-capturing with OBS Studio
increases duration variability enough to cause frequent missed frames
when using that method. Taking a maximum of the last 16 frames smoothes
out this variability.
The durations are naturally quite variable and the estimates aren't
perfect. To take this into account, an additional constant 2 ms is added
to the max render time.
How does it perform in practice? On my desktop with 144 Hz monitors I
get a max render time of 4–5 ms instead of the default 4.9 ms (I had
1 ms manually configured in sway) and on my laptop with a 60 Hz screen I
get a max render time of 4.8–5.5 ms instead of the default 14.7 ms (I
had 5–6 ms manually configured in sway). Weston [1] went with a 7 ms
default.
The main downside is that if there's a sudden heavy batch of work in the
compositing, which would've made it in default 14.7 ms, but doesn't make
it in reduced 6 ms, there is a delayed frame which would otherwise not
be there. Arguably, this happens rarely enough to be a good trade-off
for reduced latency. One possible solution is a "next frame is expected
to be heavy" function which manually increases max render time for the
next frame. This would avoid this single dropped frame at the start of
complex animations.
[1]: https://www.collabora.com/about-us/blog/2015/02/12/weston-repaint-scheduling/
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1762>
