The `last_presentation_time` is usually a little in the past (although
sometimes in the future depending on the driver). When it's over 2ms
(`sync_delay`) in the past that would trigger the while loop to count up so
that the next `update_time` is in the future.
The problem with that is for common values of `last_presentation_time`
which are only a few milliseconds ago, incrementing `update_time` by
`refresh_interval` also means counting past the next physical frame that
we haven't rendered yet. And so mutter would skip that frame.
**Example**
Given:
```
last_presentation_time = now - 3ms
sync_delay = 2ms
refresh_interval = 16ms
next_presentation_time = last_presentation_time + refresh_interval
= now + 13ms
-3ms now +13ms +29ms +45ms
----|--+------------|---------------|---------------|----
: :
last_presentation_time next_presentation_time
```
Old algorithm:
```
update_time = last_presentation_time + sync_delay
= now - 1ms
while (update_time < now)
(now - 1ms < now)
update_time = now - 1ms + 16ms
update_time = now + 15ms
next_presentation_time = now + 13ms
available_render_time = next_presentation_time - max(now, update_time)
= (now + 13ms) - (now + 15ms)
= -2ms so the next frame will be skipped.
-3ms now +13ms +29ms +45ms
----|--+------------|-+-------------|---------------|----
: : :
: : update_time (too late)
: :
last_presentation_time next_presentation_time (a missed frame)
```
New algorithm:
```
min_render_time_allowed = refresh_interval / 2
= 8ms
max_render_time_allowed = refresh_interval - sync_delay
= 14ms
target_presentation_time = last_presentation_time + refresh_interval
= now - 3ms + 16ms
= now + 13ms
while (target_presentation_time - min_render_time_allowed < now)
(now + 13ms - 8ms < now)
(5ms < 0ms)
# loop is never entered
update_time = target_presentation_time - max_render_time_allowed
= now + 13ms - 14ms
= now - 1ms
next_presentation_time = now + 13ms
available_render_time = next_presentation_time - max(now, update_time)
= (now + 13ms) - now
= 13ms which is plenty of render time.
-3ms now +13ms +29ms +45ms
----|-++------------|---------------|---------------|----
: : :
: update_time :
: :
last_presentation_time next_presentation_time
```
The reason nobody noticed these missed frames very often was because
mutter has some accidental workarounds built-in:
* Prior to 3.32, the offending code was only reachable in Xorg sessions.
It was never reached in Wayland sessions because it hadn't been
implemented yet (till e9e4b2b72).
* Even though Wayland support is now implemented the native backend
provides a `last_presentation_time` much faster than Xorg sessions
(being in the same process) and so is less likely to spuriously enter
the while loop to miss a frame.
* For Xorg sessions we are accidentally triple buffering (#334). This
is a good way to avoid the missed frames, but is also an accident.
* `sync_delay` is presently just high enough (2ms by coincidence is very
close to common values of `now - last_presentation_time`) to push the
`update_time` into the future in some cases, which avoids entering the
while loop. This is why the same missed frames problem was also noticed
when experimenting with `sync_delay = 0`.
v2: adjust variable names and code style.
Fixes: https://bugzilla.gnome.org/show_bug.cgi?id=789186
and most of https://gitlab.gnome.org/GNOME/mutter/issues/571https://gitlab.gnome.org/GNOME/mutter/merge_requests/520
We only call _clutter_input_device_update with devices that are not
Keyboard devices. Also passing a Keyboard device to a function whose
primary purpose is picking should be considered a bug in the caller.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/547
We're bailing out of clutter_stage_cogl_add_redraw_clip() early without
doing anything if we're ignoring redraw clips, so no need to call it if
we already know that will be the case.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/547
GList's used in legacy code were free'd using a g_slist_foreach + g_slist_free,
while we can just use g_slist_free_full as per GLib 2.28.
So replace code where we were using this legacy codepath.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/576
GList's used in legacy code were free'd using a g_list_foreach + g_list_free,
while we can just use g_list_free_full as per GLib 2.28.
So replace code where we were using this legacy codepath.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/576
Moving an actor with a ClutterDeformEffect applied flickers because
the depth_testing, setting the depth testing test function to
COGL_DEPTH_TEST_FUNCTION_LEQUAL fixes the problem.
Fixes https://gitlab.gnome.org/GNOME/mutter/issues/507
When the pointer is grabbed, we send the crossing events that are
initiated by this pointer only to the actor that has the grab. For
grabbed touch sequences, we always capture and bubble the crossing
events right now.
Fix this and make grabbed pointers and touch sequences behave the same
by sending touch crossing events only to the grab actor.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/422
It's important to cancel click actions when we get a touch cancel event,
otherwise the long press event might get emitted after the compositor
took over the touches because it detected a gesture.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/396
Clutter does the nicety of connecting just created PangoContexts to
ClutterBackend signals in order to update it on resolution/font changes.
However the way the signals are disconnected (automatically via
g_signal_connect_object() auto-disconnect feature) may incur into
performance issues with a high enough number of ClutterActors with a
PangoContext (eg. ClutterText) as the lookup by closure is linear across
all signals and handlers.
Keep the handler IDs around, and disconnect them specifically on dispose
so it is more O(1)-ish.
Related: https://gitlab.gnome.org/GNOME/mutter/issues/556
The clutter/evdev implementation of mousekeys is designed after the
current implementation in X11, and works when the setting is enabled
regardless of the status of NumLock.
The GNOME documentation on accessibility features states however that
mousekeys work only when NumLock is OFF:
https://help.gnome.org/users/gnome-help/stable/mouse-mousekeys.html
Change the clutter/evdev implementation to match the documentation, i.e.
disable mousekeys when NumLock in ON so that switching NumLock ON
restores the numeric keypad behaviour.
Closes: https://gitlab.gnome.org/GNOME/mutter/issues/530
The keyboard accessibility setting "enable" is actually even more
misleading that initially anticipated, as it does not control the
entire keyboard accessibility feature, but just the "enable by
keyboard" feature, i.e. being able to enable or disable stickykeys
or slowkeys using various keyboard actions.
Yet the accessibility features should still work even if the "enable"
setting is unset, those can be controlled by the accessibility menu in
GNOME Shell for example.
Change the clutter/evdev implementation to match that behavior as found
in the x11 backend, so both backends are now consistent.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/531
In a multi-monitor setup there is a separate paint run for each monitor.
If an actor doesn't intersect the first monitor painted then it is culled
out for that monitor to save time. Unfortunately this would mean
`clutter_actor_paint` was setting `is_dirty = FALSE` before the actor had
yet been painted on any monitor.
This meant that effects like `ClutterOffscreenEffect` were not receiving
the flag `CLUTTER_EFFECT_PAINT_ACTOR_DIRTY` when they should have, and
so would rightfully think they don't need to do a full internal
invalidation. So `ClutterOffscreenEffect`, and probably other effects,
did not repaint correctly unless on the first monitor in the list.
The fix is to simply avoid setting `is_dirty = FALSE` on those paint
runs where the actor has been culled out (`clutter_actor_continue_paint`
wasn't called). It is only safe to clear the flag after
`clutter_actor_continue_paint` has been called at least once per stage
paint.
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/1049https://gitlab.gnome.org/GNOME/mutter/merge_requests/511
clutter_input_device_get_physical_size was just used for device mapping
heuristics in MetaInputMapper. It now started using the info from udev
on for both backends, so this means this clutter API is no longer
necessary.
https://gitlab.gnome.org/GNOME/mutter/issues/514
Enabling keyboard accessibility features on Wayland from the keyboard
was wrongly assumed to be controlled by the "togglekeys" setting,
whereas it should be simply controlled by the "enable" setting.
As "togglekeys" is off by default and doesn't have a UI option to
enable, that would prevent turning on or off the keyboard accessibility
features using the keyboard.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/501
Meson 0.50.0 made passing an absolute path to install_headers()'
subdir keyword a fatal error. This means we have to track both
relative (to includedir) paths for header subdirs and absolute
paths for generated headers now :-(
https://gitlab.gnome.org/GNOME/mutter/merge_requests/492
The ClutterVirtualInputDevice API was fixed to use Clutter button
internal codes, whereas the mousekeys still uses evdev codes.
Change the mousekeys implementation to use the Clutter button code
instead to remain compatible with the ClutterVirtualInputDevice API.
Fixes: 24aef44b (Translate from button internal codes to evdev)
https://gitlab.gnome.org/GNOME/mutter/merge_requests/473
Since e3e933c4 a keyval can be temporarily remapped to an unused keycode. Due to
some limitations in XTestFakeKeyEvent, the remapping has to be done in the first
xkb group/layout. In case there are two or more keyboard layouts enabled and the
selected keyboard layout is not the first, clutter_keymap_x11_keycode_for_keyval
will fail to retrieve the correct keycode for a remapped keyval. Let's use the
reserved_keycodes map in order to retrieve the correct keycode if needed.
Fixes: https://gitlab.gnome.org/GNOME/mutter/issues/443
The clamped rectangle currently could not fully contain the original fractional
rectangle because it doesn't take care of the fact that the new width should
consider the fact that flooring we'd translate the rectangle, and thus to cover
the same area we need to take care of it.
So, to properly compute the width and height, calculate x2 and y2 first and then
use this ceiled value to compute the actual width using the floored x1 and y1.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/3
When we try to update the FB, we might face the case in which the effect target
framebuffer does not need any redraw, because it's already properly sized and
scaled, but the filter applied to the pipeline is not, because it has been
computed for a non-fractional scaling.
This is happens for example to clutter actors with a flattening effect (i.e.
override redirect mode set), that might have been generated properly for a
celied scaling level, but when we go fractional we need to ensure to use a
linear filter, as the 1:1 texel:pixel assumption is not true anymore.
https://bugzilla.gnome.org/show_bug.cgi?id=765011https://gitlab.gnome.org/GNOME/mutter/merge_requests/3
At this level we use ceiled resource-scale when painting fractional value
When using fractional scaling we still need to create an integer-sized
texture and then we should paint it using a size which is proportional
to the real actor size ratio, and only paint a subsample of it, but this
doesn't seem to work properly with some weird scaling values.
Then, it's just better to draw the texture ceiled and then we scale it
down to match the proper actor scaling at paint level.
https://bugzilla.gnome.org/show_bug.cgi?id=765011https://gitlab.gnome.org/GNOME/mutter/merge_requests/3
When resource scale is set we need to generate a scaled PangoLayout (by adding
a new scale attribute, or adjusting the one we already have according the
resource scale), then it has to be painted with proper scaling matrix.
So everything that has to do with PangoLayout has to be in real coordinates,
then clutter logical coords multiplied by resource scaling.
While the actual size of the layout is the one of the PangoLayout divided by
resource scale.
We map the text positions to logical coords by default, while using
the pixel coordinates when painting.
We fall back to scale 1 when calculating preferred size if no scale is
known. The pango layout will not have set a layout scale attribute,
meaning it'll be 1, thus we should just assume the layout scale is 1 here.
Not doing so might result in the preferred size being 0x0 meaning the
actor won't be laid out properly.
Fixes https://gitlab.gnome.org/GNOME/mutter/issues/135https://bugzilla.gnome.org/show_bug.cgi?id=765011https://gitlab.gnome.org/GNOME/mutter/merge_requests/3
