meta_window_update_monitor() can emit "::highest-scale-monitor-changed",
and we connected to that signal right before. Let's avoid calling
meta_wayland_surface_notify_highest_scale_monitor() twice and move the
g_signal_connect() for that signal and the initial call to
meta_wayland_surface_notify_highest_scale_monitor() to happen after
meta_window_update_monitor().
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3598>
Turns out there is a better solution: Almost always, MetaWindow already has
an idea on which monitor it will be, even if it isn't positioned yet. Since
the last commit we're now using that monitor for setting the
highest-output-scale of the window, so this fallback is no longer necessary.
While we could keep this fallback around and also return a valid scale in
case the surface is not even mapped yet, this means we report fractional
scale twice for new surfaces: Once from
wp_fractional_scale_manager::get_fractional_scale() (here we'll enter the
fallback), and a second time (this time with correct scale) right after
creating the MetaWindow.
Note that wp_fractional_scale_v1 doesn't specify that a preferred_scale
event must be sent immediately after
wp_fractional_scale_manager::get_fractional_scale(), so we can safely remove
the fallback.
This reverts commit 8cfbdb4313.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3598>
MetaWindow always tries to have a main monitor: If the frame rect is empty
and the window has not been positioned, in meta_window_constructed() we fall
back to asking the backend for the current monitor, and in
meta_window_wayland_update_main_monitor() we fall back to
meta_window_find_monitor_from_id(), which then falls back to the primary
monitor.
In general this means that window->monitor is always set as long as there is
a monitor around.
For getting the highest-scale-monitor the window is on, we currently rely
completely on the frame rect. If the frame rect is empty, we set the
highest-scale-monitor to NULL. Since we usually know though which monitor
the window is, or will be on, and window->monitor is even set to that, we
can just fall back to window->monitor for the highest-scale-monitor.
This makes sure ::highest-scale-monitor-changed is emitted right after the
window is created, and it's set to the correct monitor that the window will
be on. This in turn means that we can send a correct wp_fractional_scale
fraction_scale event to clients right away.
https://gitlab.gnome.org/GNOME/mutter/-/issues/3262
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3598>
The existing comment tells us this is necessary:
> there may be drawing between the last damage event and the
> XDamageSubtract() that needs to be flushed as well.
But the commit message for 551101c65c also tells us that
synchronization is necessary before-update. Assuming both are correct
then it needs to be done in both places.
I did try optimizing out the second sync to only do it if damage
arrived during the update, but that doesn't seem to be the issue.
The damage event is arriving before the update starts and it's some
secondary changes within the damage region running late that need
flushing. So this means the client is reporting damage more frequently
than the frame rate and we're ignoring the secondary damage reports
for efficiency (XDamageReportBoundingBox), which is still a good thing.
Fixes: 551101c65c ("compositor-x11: Move synchronization to before-update")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2880
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3590>
Do not include it at header side as it is not part of the installed headers.
Only keep it in cogl-gl-headers.h as it is a private header.
Add it to all the source files that depend on it.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3593>
This change adds modifier-aware screencasting support to Mutter.
Implicit modifier support is kept for backward compatibility and the
code fallbacks to implicit modifiers in case any new functionality added
for explicit modifier support fails.
The advertised modifiers are retrieved by a call to
eglQueryDmaBufModifiersEXT() function. The "external only" modifiers are
excluded as Mutter uses the buffers created with the explicit modifiers
as renderbuffers. Support for implicit modifiers is checked with a test
allocation since there are drivers that do not support them.
This change also removes various implicit modifier support checks that
disable DMA-BUF screen casting support globally as they are no longer
needed. DMA-BUF support for screencasting is determined by the available
formats and modifiers case-by-case now.
It also effectively enables DMA-BUF screencasting on NVIDIA hardware as
well since GBM buffer objects with linear modifiers are no longer used
by default to create a renderbuffer object for screencasting.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3537>
meta_render_device_gbm_allocate_dma_buf() function is updated to take a
list of modifiers. If no modifiers are specified, the modifier is
selected by the allocator, and implicit modifiers are used to import the
created DMA-BUF.
Co-authored-by: Jonas Ådahl <jadahl@gmail.com>
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3537>
Currently, we blindly apply the transformation matrices of all parent
actors when calculating the absolute coordinates. This means if this
function is called while the window actor containing the surface is in
the middle of a transition (e.g. window open animation), it may return
incorrect values. As this function is used for calculating pointer
confinement bounds for a specific surface, this will result in incorrect
bounds value being used if pointer constraints are applied by the
application at the same time the window is created and the mouse is
inside the surface's bounds when it's created.
Fix this by only applying transformation matrices up to the window actor
of the surface and then calculating the absolute coordinates by adding
the position of the window actor.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3585>
They are float in libdisplay-info and our own EDID parsing also returns
a float but when then converted both to an integer. Especially the min
luminance can be <1.
We also don't need a variable for indicating presence of a CTA Static
Metadata block. The values are all zero if it is absent.
Found by Dor Askayo.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3574>
The function was used only once so just move it content where it is
called. It allows us to drop more cairo paths from the API surface even
if it is not part of a public api
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3581>
If we don't have a monitor for a surface - e.g. because the surface is
not mapped yet - return the highest scale of all outputs. This makes us
send a preferred scale before a client draws its first frame. The highest
scale is always correct in single monitor cases and arguably a good
option otherwise as scaling down usually looks better than scaling up.
Note that this is currently only used by the fractional scale protocol,
but will also be used for the core `send_preferred_scale()` once we
implement it.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3217>
With all early "goto out" paths bypassing wayland, we can pretty
much avoid the goto and use early returns in this function. This
will hopefully improve readability.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
We've so far returned FALSE (i.e. PROPAGATE) here, somehow
oblivious of the fact that the core event handler would stop
all non-gesture events directed to windows.
Incorporate this knowledge there, in order to be able to
streamline this piece of event handling in core/ code.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
This is Wayland specific code, handle it directly in MetaWaylandPointer.
This also fixes issues with the crossing event itself managing to reach
the window occluded by modals.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
With Wayland popups and drag-and-drop using grabs, we
should let window cursors prevail when there is one
in effect.
Also, resort always to the actor as known by the
stage. This fixes the cursor lookup right after crossing
events induced by grabs, e.g. right clicking on the
gtk4-demo textview without motion would keep the I-beam
cursor, now results on the right actor/cursor for the
menu being picked.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
Enable grabbing input for popups, and drag-and-drop. Since the very
switch to using ClutterGrab underneath Wayland grabs will challenge
assumptions in existing code, these had to change in one go. A notable
one is that meta_display_windows_are_interactable() is not 100% true
anymore for xdg_popups, at least not the same.
Another change happening in lockstep is MetaDnD no longer having
to funnel events to Wayland, since the grab triggered by Wayland DnD
is now a cause of "compositor grabs", and will naturally receive events
as long as it hold. while "modal".
A number of ad-hoc checks for grabbing state has also been dropped
from src/wayland/ internals, since again Wayland grabs are a reason
for Clutter grabs, plus the mechanism itself will already take care
of focus loss and restoration.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
Add the mechanism to integrate MetaWaylandEventInterface with grabs,
callers may now specify whether a grab is required, in which case
one is created, shared by all the event interface stack.
ClutterStage grab state is also tracked, so the MetaWaylandEventInterface
in charge will focus or unfocus depending on whether input should be
handled (if ungrabbed, or grabbed by the MetaWaylandInput itself), or
not (if grabbed by something else).
At the moment nothing uses this mechanism yet, later commits will add
the first users.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
Now that the backend handles 0-size regions naturally and MetaWaylandPointer
avoids sending wl_pointer.motion on unchanged coordinates, we can use the
default motion handler for the locked pointer constraint.
And since that is the only difference with the pointer constraint event
interface, we can unify them both into a single MetaWaylandEventInterface
handling focus for them both.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
The small catch is that MtkRegion (and pixman regions) "optimize away"
0-size rectangles, so a 0-sized region will always be seen as having
a 0,0 origin. We don't want that, so transfer the origin separately from
the region.
While at it, make the Wayland pointer lock use one such 0-size region,
to avoid the 1x1px wiggle room that it currently has (accounting for subpixel
motion).
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
While every kind of input is seen as coming from the Virtual
Core Pointer in the X11 case, we can largely abstract away from
that fact, and lock XDnD pointer input to the most plausible
source (e.g. a device with a pressed button), instead of only
working with pointer input.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
This collection of event handlers is the most special of them all, as
they want to unset any pointer/touch/stylus/keyboard/pad/etc focus,
and handle events from a selected device/sequence combination through
the MetaWaylandDragDest interfaces.
The same interfaces also replace the MetaWaylandKeyboardGrabInterface
in effect that handled DnD action changes.
On the XDnD special grab side, we mainly need to let the current
client (i.e. the drag source) keep receiving input events, as they
drive the DnD operation from the X11 realm.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
This is again a grab interface that mostly wants to meddle with focus,
logically setting a NULL surface if the surface client does not match
the popup client.
Since popups are meant to naturally work with any input device, the
code has been refactored to not involve the MetaWaylandPointer directly
in MetaWaylandPopup creation or getting the top popup surface (memory
management was shuffled), or compressing multiple grabbing xdg_popups
together (the existing grab maintains a single MetaWaylandEventHandler
for all).
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
Besides the pointer locking/constraining mechanism, these grab interfaces
were more of a focus tracking mechanism, revoking the constraints when
the conditions didn't meet.
This can be handled pretty similarly to keyboard grabs with the new
interface, with the added bonus that we can chain up to let the
parent/default handler handle the events themselves, without poking at
MetaWaylandPointer API.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
This is implemented at the MetaWaylandSeat level, and it governs
focus and event delivery for all devices, falling through each
of the MetaWaylandPointer/MetaWaylandKeyboard/etc components.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
MetaWaylandInput is an object that will become in charge of handling
input events on their way to the Wayland socket. It keeps a stack
of event handlers, and propagates events and changes across them in
order to have them emit Wayland events, or change focus.
Each of these event handlers has a MetaWaylandEventInterface, this
is a vtable meant to replace MetaWaylandPointerGrabInterface and
MetaWaylandKeyboardGrabInterface in an unified manner, with the
following methods:
- get_focus_surface: to return the focus surface for a device/sequence.
Since several handlers will want to delegate logic on previous
handlers, it is optional to chain up with
meta_wayland_event_handler_chain_up_get_focus_surface().
- focus: To trigger a focus change for a device/sequence, since
event handlers are daisy chained by default, it is mandatory to
chain up with meta_wayland_event_handler_chain_up_focus(), either
with the given surface, or passing NULL to let later handlers
unset their state.
- press/motion/release: Unified handlers for pointer/touch/stylus
input, they chain up like event handlers do.
- key: Key event handler, propagates like event handlers do.
- other: Fallthrough for other events (pad, scroll, ...), propagates
like event handlers do.
Since there is a variety of expected behaviors, and the possibility
of stacking for some of the existing Wayland "grabs", this provides
the mechanism for that to happen.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
That would be the surface under the tool that is being currently used
on the device. This will be used by MetaWaylandSeat to implement the
default MetaWaylandEventInterface.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
The MetaWaylandPointer used to put this together through
MetaCursorTracker cursor visibility, and ClutterSeat-level
inhibition API, applying the pointer focus changes due to
visibility logically to Wayland clients.
In order to make this work over all Clutter widgetry
instead of just Wayland clients, make the ClutterSeat-level
inhibition API control this feature at the ClutterStage picking
level, and leave/enter the seat pointer as appropriate.
By default, the seat pointer has (un)focus inhibited. The
MetaCursorTracker has been made another player in unfocus
inhibition, simply asking for the pointer to get its focus
while the cursor is visible.
This in practice means that picking code may return a NULL
actor, some asserts and preconditions had to be changed to
handle this, plus some test code slightly.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
Do not jump at MetaWaylandSeat across the MetaWaylandTabletSeat to
poke at the tablet tools, and chain up the checks through a
MetaWaylandTabletSeat method instead.
While at it, use g_autoptr to manage the tool list, and fix a leak.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3420>
After negotiation of DMABUF transport mutter will silently allocate SHM
buffers if the allocation in the add_buffer callback fails. It's cleaner
to renegotiate the supported formats without announcing DMABUF
capabilities in this case.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2557>
To fixate the format or renegotiate after a DMABUF allocation failed we
need to rebuild the EnumFormat params.
The function meta_screen_cast_query_modifiers will return false if no
modifiers are supported, thouse we can drop the check and remove the
macro guard.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2557>
This function contains a stub, which returns support for implicit
modifiers, if modifiers are supported preserving the current
capabilities. The stub has to be replaced with a query to the cogl
renderer to support explicit modifiers.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2557>
This check was originally added because `window` was actually used.
While technically correct, there's no reason to keep it around.
Fixes: 4736f873f2 ("compositor/native: Add support for direct scanout per view")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3177>
Until now we only supported direct scanout to the primary plane if the
buffer size perfectly matched the display size.
Since display controllers usually support scaling and cropping buffers
highly efficiently, try to let them do the job. This is usually helpful
if wp_viewporter is used by the client or Mutter uses fractional
scaling.
This has several advantages:
- Games (e.g. SDL2 based ones) can almost always hit direct scanout
paths in fullscreen mode. Notably when fractional scaling is used or
the game renders in a non-native resolution (or both).
- Video players using YUV buffer formats and wp_viewporter can easily
hit direct scanout paths, making displaying video very power
efficient as the 3D engine is not used at all.
Note that this still only uses the primary plane, no overlay or underlay
planes, making this change comparatively low risk.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3177>
In a following commit we will start supporting scaled and croped
surfaces, thus, in preparation, update the logic to three common cases:
1. only one surface, fullscreen (most apps)
2. a content surface and a black background surface which the client
does not want to unmap, fullscreen
3. top-level subsurface covers the whole window and is opaque (Firefox)
The remaining currently supported cases should be fairly uncommen and
and harder to compute.
Note that we already check that the window cover the stage view in
MetaCompositorView.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3177>
This allows us to pass on the related data from CoglScanouts.
If dst_rect does not match the mode, we assume that not covered areas
are opaque black - usually black bars around a centered surface.
While such driver behaviour does not appear to be documented (well) yet,
it seems to be followed by all known existing drivers and is used in a
similar way in ChromeOS.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3177>
We need an object to hold additional scanout related information, such
as scaling and positioning data. Turn CoglScanout into such an object,
moving the interface into CoglScanoutBuffer.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3177>
Except meta_window_x11_get_group, which is still used by GNOME Shell
and we can't make it a private API for now.
Will need further investigation and could be done as a future
step
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3555>
The macro used to call into a bunch of other macros so let us turn it
into a single function.
This would simplify things for the next commit that puts the MetaGroup
usage behind a X11 ifdef
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3555>
If such a failure is followed by a successful frame then the Cogl frame
queue would have size 2, leading to an assertion failure in
`meta_onscreen_native_notify_frame_complete`:
```
g_assert (!cogl_onscreen_peek_head_frame_info (onscreen));
```
Notifying on the failure however keeps the Cogl frame queue limited to
a size of 1 and we recover gracefully with only a missed frame and a
warning message.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3278
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3565>
Popups were missing the "input focus" unification in the pointer
seat, triggering MetaWaylandKeyboard focus changes underneath. On
one hand this missed moving all associated focus with it, on the
other hand this made keyboard and global input focus get out of
sync, and bring funky behavior like keyboard focus loss after
dismissing popups.
Fixes: 7b232d9f65 ("wayland: Keep track of the "input focus" on MetaWaylandSeat")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3256
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3568>
In some circumstances, we may end up with outputs with the same
vendor/product/serial, in which case we have a hard time finding the
right one to map tablets to, since configuration only has these 3
pieces of data.
Add the handling of a 4th argument containing the output name
based on the connector (e.g. HDMI-1), so that it can be used to
disambiguate the output if necessary.
This only kicks in if there actually are multiple outputs with the
same EDID data. A goal of the configuration as it was stored was to
remain useful if the user changed how the device is physically
connected to the computer, this remains true for the vast majority
of users having a single thing of each.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3556>
Our hashtable stores tools by the serial but our stylus tool and eraser
tool share the same serial - they only differ by the tool type.
This results in only one tool being created and this tool re-used for
the other type tool. Fun side-effects of this are that the stylus ends
up using the eraser pressure curve (or vice versa).
Hack around this by bit-flipping the serial for the eraser to
make it distinct - this is the only place we need to wrorry .
Closes#1884
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3542>
For barrier validation, check_barrier() would start from the
(presumably) left-most monitor and walk the neighbor monitors to the
right.
This is assuming that there is always a monitor at (0.0), which is not
necessarily the case. If the first monitor on the left is not aligned at
the top, there is no logical monitor at (0.0) causing a NULL pointer
derefence.
Instead of starting from the monitor at (0,0), start from the primary
logical monitor, as there is necessarily one.
Fixes: 85885c6 - Check barriers don't extend into nonexisting monitors
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3272
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3562>
There doesn't seem to be a good reason to keep this code in
`MetaWaylandSurface`. Moving it to `MetaWaylandBuffer` cleans things
up and will allow us to tread buffers differently depending on their
type.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3559>
For secondary GPU rendering contexts we currently might choose an EGL
config with a format which is not supported on all primary planes. The
renderer is created when a GPU is detected and lighting up outputs and
thus assigning CRTC and primary planes can happen at any point after
that. This means we have to make sure that all possible plane
assignments will work with the rendering context when we create it.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3235
Fixes: cc7bca073 ("crtc/kms: Dynamically assign primary and cursor planes")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3512>
It can be used to force a specific RGB range. Some monitors don't follow
the specification and expect a signal different from what we send. This
property allows to force a mode which hopefully then works correctly for
the sink.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3535>
A ring will naturally go from 355 degrees to 5 degrees (or vice versa),
giving us the illusion of a direction change. Avoid this by assuming
that any change larger than 180 degrees is actually the equivalent
smaller change in the other direction.
Closes#1885
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3545>
BTN_STYLUS is the lower one and traditionally (read: in X) maps to
middle button (2), BTN_STYLUS2 is the upper one and traditionally maps
to right button (3).
This is also what GTK does and our desktop actions too map MIDDLE to
BTN_STYLUS and RIGHT to BTN_STYLUS2.
See also gtk!6168
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3541>
The code that maybe flushed IM state before processing a key event
became ineffective at commit 7716b62fa2, since the handle_event()
method on MetaWaylandTextInput won't handle key events, only IM
events and touch/button press events causing IM state to be
committed. Basically, the events that directly change the IM state.
Move this ineffective code to the the filter_event() method handling
the key presses in order to let the IM maybe filter them, and handle
them so that any key event that is let through (both key events
previously injected by the IM, and key events that the IM chooses to
ignore) will ensure that the pending IM state is flushed before the
key event is handled and emitted to the client.
This brings back lost guarantees of orderly event emission when IMs
alternate key events and IM actions.
Fixes: 7716b62fa2 ("clutter: Separate ClutterInputFocus event processing and filtering")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3090
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3536>
adwaita-icon-theme updated its cursor metaphors and changed all DnD
cursors to use arrows instead of hands, except for the grab related
ones. Mutter was using "grabbing" as default DnD cursor, which now
does not match the other DnD cursors ("copy" and "no-drop") anymore.
Change this to the "default" cursor.
Additionally, because the "no-drop" cursor now puts a stronger emphasis
on the crossed out symbol also prefer "default" for
META_CURSOR_DND_IN_DRAG and only use "no-drop" for things that
explicitly don't accept a drop.
Related: https://gitlab.gnome.org/GNOME/adwaita-icon-theme/-/merge_requests/63
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3532>
When Wayland clients send commits without a buffer attached ("empty"
commits), they may lead to stage updates that do not result in any
frame being submitted for presentation ("empty" updates).
Due to how frame scheduling is handled, there can be many such
"empty" updates in a single refresh cycle. If frame callbacks were
emitted after each of these "empty" updates, and if the client
sending "empty" commits was using frame callbacks to throttle the
same logic that results in these "empty" commits being sent, it would
result in a feedback loop between Mutter and the client where the
client would send "empty" commits and Mutter would reply almost
immediately with a frame callback causing the client to send "empty"
commits continuously.
As such, when an "empty" update is detected, frame callbacks are
scheduled to be emitted only once in every refresh cycle, avoiding the
feedback loop.
When a "non-empty" update is detected, frame callbacks are instead
emitted immediately to allow clients to draw their next frame as soon
as possible. It is safe to emit frame callbacks in this case because
the frame for the current refresh cycle is already "finalized" and
that any commit sent by the client at that point would only be handled
in a future refresh cycle.
To implement this, the previous logic had used
meta_frame_native_had_kms_update() to detect "non-empty" updates,
assuming that those would always result in a KMS presentation with the
native backend.
However, this approach misses the fact that virtual monitors do not
use KMS, and as such do not result in KMS presentation even for
"non-empty" updates. As a result, frame callbacks would not be emitted
immediately, resulting in unintended throttling of client rendering.
Instead, assume that it is safe to emit frame callbacks immediately
whenever an update results in the frame clock waiting to be notified
of presentation, since this is also when commits sent by clients are
scheduled to be handled in a future refresh cycle.
This issue was mostly hidden because frame callbacks would be sent
immediately when the target presentation time for the frame had
changed compared to the previous frame. However, this behavior was
removed in 26d8b9c69 ("wayland: Remove unnecessary dispatch of frame
callback source"), exposing the issue.
Fixes: a7a7933e0 ("wayland: Emit frame events in GSource after "empty" updates")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3263
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3549>
Calculate the frame deadline in ClutterFrameClock's
calculate_next_update_time_us() rather than in MetaWaylandCompositor's
on_after_update().
The specifics of the deadline calculation for a given frame should be
implementation detail of the frame clock and and remain internal to
allow extensibility.
This extensibility is specifically useful for scenarios where a
different deadline calculation is needed due to alternative frame
scheduling logic, such as for VRR.
No change in behavior.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3521>
