SwapBuffers will implicitly flush the query along with the other GPU
work for the frame.
The comment was about this function being called for direct scanout,
which hasn't been the case since 56580ea7c941 ("backends/native: Assume
zero rendering time for direct scanout buffers").
This eliminates one GPU work flush to the kernel per frame with the
Mesa radeonsi driver, leaving only a single flush per frame.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4158>
Indirectly via cogl_framebuffer_flush, since we also need to call
glFlush now. Need to do it in cogl_onscreen_egl_swap_* because
meta_onscreen_native_swap_buffers_with_damage uses
cogl_context_get_latest_sync_fd.
Doing it before the swap was problematic because the swap may do GPU
work of its own, which wasn't covered by the EGL sync object created in
_cogl_context_update_sync. This could result in visual artifacts. See
the discussion starting at
https://gitlab.freedesktop.org/mesa/mesa/-/issues/11996#note_2678104 for
details.
For similar reasons, move the cogl_framebuffer_finish calls after the
swap as well.
As a bonus, this eliminates one of 3 GPU work flushes to the kernel per
frame with the Mesa radeonsi driver, because the glFlush/glFinish call
in cogl_framebuffer_flush/finish doesn't have any GPU work to flush
after SwapBuffers.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4158>
The buffer resource doesn't need to exist for a timeline to exist so we
still have to send the timeline release. Only guard the
wl_buffer_send_relase call.
Fixes: e8b890ab53 ("wayland: Implement linux-drm-syncobj-v1")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4157>
Even though the touch moved to being set up in CLUTTER_ENTER, it
is still meant to lock onto the pressed surface, we however will
receive crossing events if the touch moves between actors/surfaces,
triggering warnings when the touch info registration is attempted
on the already existing touch sequence.
The reasons to handle this in CLUTTER_ENTER still apply, so ensure
the touch info for the sequence does not previously exist to avoid
the warning.
Fixes: 2e82a2049f ("wayland: Register touchpoint info on CLUTTER_ENTER")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4166>
We try to emit wl_touch.cancel in the situations where grabbing input
would take input away from the wl_touch interface. This however does
not play fine with grabs induced by wayland interfaces (e.g. xdg popups).
A more natural place to handle this is the MetaWaylandSeats' default
input handler, specifically the focus() vmethod. Here, we may know if the
focus surface matches the current surface as picked by MetaWaylandTouch,
and if it does not, either a grab or another input handler stole input
away from the default event handler.
A case where this will notably not happen anymore is in the transition
to a xdg_popup grab, since the grabbing client will still be handling
input, and touch input will not be unfocused away despite the transition
to a grab.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3752
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4166>
The logic to allow src/wayland to update the pointer cursor changed
to bypassing grabs in commit e69e4fa6, since it is nowadays the
responsibility of the DnD source to update the pointer cursor in reaction
to the negotiated action.
This is not entirely correct, and was done at the expense of regressions.
Change the logic so we explicitly check for a DnD grab existing, or use
the active focus instead (i.e. grab-aware). This fixes the regression, and
keep DnD cursor icons working.
This is a partial revert of commit e69e4fa6dbbd62677de61bcd2f5f3bec515e79ce.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3460
Fixes: e69e4fa6db ("Revert "wayland: Check focus surface to set a pointer cursor"")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4166>
Make create_transform_snippet method more consistent.
Abstract TransferFunction struct: convert it to ColorOpSnippet.
This transform snippet will be defined by these ColorOpSnippets.
These ColorOpSnippets are similar to the prescriptive DRM API for
color transformation.
A standard transform snippet would have as ColorOpSnippets:
1. eotf
2. luminance_mapping
3. color_space_mapping
4. inv_eotf
Update uniforms the same way the transform snippet is defined.
Update color transform key to consider how the transform snippet is
generated.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4144>
When creating a new color state from the primitives Colorimetry, EOTF
and Luminance; it is needed to previously check their tags to properly get
their values and avoid UB.
This check is duplicated and is a bit unreadable.
Using this new function helps keeping readability.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4144>
Most of the implementation at color state was specific to a color state
generated from parameters so move it to a new class Params.
In the next commits a new color state ICC class will be added.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4144>
The EOTFs snippets were only added when color states were not equal.
It makes more sense to only add the whole color transformation pipeline
in that case.
This makes the assumption to always append EOTFs snippets.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4144>
This makes DMA buffer allocation in all layers take an array of possible
modifiers to allocate using, or zero modifiers if implicit modifiers
should be used.
The format hard coding previously used is moved to the screen cast code,
or in case of the (unused by default) shadow buffer buffers use the same
format as the the CoglOnscreen.
This also means the CoglDmaBufHandle and MetaDrmBuffer got taught how to
distinguish between planes. It's mostly unused in practice, so rather
untested.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3900>
It's only implemented by the GBM render device backend, and uses EGL,
thus does not distinguish scanout capable modifiers.
A filter enum is added to Cogl, since it'll be used via the Cogl API,
but the actual Cogl API isn't added yet.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3900>
clutter_stage_schedule_update() sets the field `update_scheduled` to
`TRUE` as an optimization to make redundant updates a no-op. This failed
if there was a pending event and if the stage was not yet mapped.
What happened is:
* clutter_stage_schedule_update() is called
- ClutterStage::update_scheduled is set to TRUE
- frame clock scheduled
* frame clock dispatches
- frame is discarded early, no actual stage update happens
* device is created (e.g. virtual device from remote desktop session)
- `device-added` event reaches ClutterStage::event_queue
* stage is shown
- clutter_stage_schedule_update() is called
- ClutterStage::update_scheduled is TRUE
- ClutterStage::event_queue has events in it
- These two conditions means clutter_schedule_update() becomes a
no-op
At this point, no more updates will happen from
clutter_stage_schedule_update().
Fix this by resetting `ClutterStage::update_scheduled` to `FALSE` even
if the frame was discarded due to the stage not yet being mapped.
A test case is added that replicates the above descibed events.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3804
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4152>
To allow tests to emulate GNOME Shell behavior, where the stage showing
is delayed. Do so by adding the `"options"` property to the plugin
object type. The property expects a vararg GVariant, and supports
the `"show-stage"` (boolean) vararg entry.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4152>
This is needed to allow tests to manipulate the behavior of the test
shell plugin during startup. Since the plugin is created and started
when the MetaDisplay is created, it needs to be handled via MetaContext,
by setting the options after creating the context, but before starting.
For simplicity reasons, make the options an opaque GVariant, passed via
a an `"options"` property when the plugin object is created, if the
passed options is non-NULL. Only passing the options when non-NULL
allows for backward compatibility.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4152>
This disables the visibility change effects for the dragged window,
which makes it feel a bit smoother.
TODO: Double-check whether this is indeed desirable. Note: ChromeOS imp
does this and Chromium-X11 used to do this at client side.
Status:
- [x] Basic window drag triggering
- [x] Exclude the dragged window from event targets
- [x] Event forwarding (window drag vs wayland grabs)
- [x] Offset calc relative to toplevel geometry
- [x] Attach already mapped windows
- [x] Properly support not-yet-mapped windows
- [x] Disable visibility change animations
- [ ] Dnd events stream adaptations
Signed-off-by: Nick Diego Yamane <nickdiego@igalia.com>
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4107>
- Event stream adaptations. When there is a toplevel-drag in place, do:
- Send wl_data_source.dnd_finished and end the session successfully
- Send wl_data_source.cancelled and end the MetaWindowDrag when ESC key
is pressed.
- Modify MetaWaylandDataDevice such that, when a toplevel-drag is
running, it does:
- Propagate motion events, so that they can be processed further by
MetaWindowDrag.
- Ends the associated MetaWindowDrag upon release event.
- Hook up the window mapping process in MetaWindowWayand, such that:
- the initial position of the window attached to the ongoing
toplevel-drag instance can be calculated and set.
- the appropriate gravity and flags can be set when calling MetaWindow's
meta_window_move_resize_internal, which allows it for example to be
moved freely (unconstrained) as per current dragging cursor.
Status:
- [x] Basic window drag triggering
- [x] Exclude the dragged window from event targets
- [x] Event forwarding (window drag vs wayland grabs)
- [x] Offset calc relative to toplevel geometry
- [x] Attach already mapped windows
- [x] Properly support not-yet-mapped windows
- [x] Disable visibility change animations
- [x] Dnd events stream adaptations
Signed-off-by: Nick Diego Yamane <nickdiego@igalia.com>
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4107>
- For already mapped windows, the window drag session is started
straight away;
- For about-to-be-mapped window (ie: undocking window use case):
- The "shown" signal for the dragged window triggers the actual
MetaWindowDrag once it's mapped.
- MetaWindowWayland now handles the case of toplevel-drag and position
the window about to be mapped according to the toplevel-drag
parameters.
- While attached to a toplevel-drag, the window state is updated to:
- Actor's "reactive" state is set to false, which in practice excludes
it from the possible drag target list;
- WindowActor's "tied to drag" state is set to true, which results in
initial placement constraints to be skipped, so newly created
(detached) windows can be freely dragged around.
- Toplevel drag session ends upon:
- dnd drop and cancellation.
- xdg_toplevel_drag_v1 object destruction (client-side).
- data source destruction.
Status:
- [x] Basic window drag triggering
- [x] Exclude the dragged window from event targets
- [x] Event forwarding (window drag vs wayland grabs)
- [x] Offset calc relative to toplevel geometry
- [x] Attach already mapped windows
- [x] Properly support not-yet-mapped windows
- [ ] Disable visibility change animations
- [ ] Dnd events stream adaptations
Signed-off-by: Nick Diego Yamane <nickdiego@igalia.com>
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4107>
- This adds a new ClutterActor parameter to meta_compositor_drag_window
and meta_window_drag_begin such that it's possible to inject an
external grab actor, which is then used to implement a "non-grabbing"
mode that allow a MetaWindowDrag to be used in conjunction, for
example, with a Wayland drag-and-drop session;
- This new API is going to be used to implement the toplevel drag
protocol in a subsequent commit.
Signed-off-by: Nick Diego Yamane <nickdiego@igalia.com>
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4107>
- Skip placement constraints when it is set.
- Required to properly support about-to-be-mapped windows, which get
"detached" when dragged out of its original owning window.
Signed-off-by: Nick Diego Yamane <nickdiego@igalia.com>
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4107>
The cogl_scanout_get_dst_rect() fell back on the buffer dimensions as
the destination rectangle when nothing was explicitly set. This,
however, is not necessarily correct. For example, if a buffer is larger
the CRTC resolution, but the surface is scaled to exactly match the CRTC view,
the expected destination size should match the CRTC resolution, not the
buffer dimension, which would be the case if no explicit destination was
set.
In meta_wayland_try_aquire_scanout() we're in a good position to
determine the destination rect in the CRTC primary plane, since we have
all the prerequisits, i.e. that the surface effectively covers the whole
CRTC, the actor allocation box (the non-black border part), the scale
and transform of the view.
This tweaks the CoglScanout API a bit to make it explicit that the
dst_rect must be unconditionally provided, and removes the fallback to
the buffer dimension as the destination rectangle, which sometimes
resulted in a destination rectangle being larger than the primary plane
itself, resulting in clipping and incorrect scaling.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3773
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4147>
The "dst_rect" calculated is in "CRTC space", meaning the bounding box
used for calculating it should be the view layout dimension, scaled by
the view scale and transform by the view transform. Previously it was
only transformed, not scaled. While fixing this, rename the variables to
make it a bit more clear what coordinate space they are expected to be
in.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/4147>
