GitLab

It was not the lack of forcing the shadow fb that caused slowness, but
rather due to the method the shadow fb content was copied onto the
scanout fb. With 'clutter: Use cogl_blit_framebuffer() for shadow FB'
we'll use a path that shouldn't be slow when copying onto the scanout
fb.

Also 437f6b3d accidentally enabled
shadow fb when using hw accelerated contexts, due to the cap being set
to 1 in majority of drivers. While the kernel documentation for the
related field says "hint to userspace to prefer shadow-fb rendering",
the name of the hint when exposed to userspace is
DRM_CAP_DUMB_PREFER_SHADOW, thus should only be taken into consideration
for dumb buffers, not rendering in general.

This reverts commit 437f6b3d.

!818

!814

The commit 'renderer/native: Use shadow fb on software GL if preferred'
attempted to force using a shadow fb when using llvmpipe in order to
speed up blending, but instead only did so when llvmpipe AND the drm
device explicityl asked for it.

Now instead always force it for llvmpipe and other software rendering
backends, and otherwise just query the drm device (i.e.
DRM_CAP_DUMB_PREFER_SHADOW).

!807

If there is no transformation, use `cogl_blit_framebuffer()` as a
shortcut in `clutter_stage_view_blit_offscreen()`, that dramatically
improves performance when using a shadow framebuffer.

!809

Since the recent clutter-content work, legacy scaling (in contrast
to the new stage-view-scaling) only applies to surfaces that belong
to a window. This broke scaling of DnD surfaces.

As a workaround, apply the same scaling on DnD-surface-actors until
we use stage-view-scaling by default and can remove this again.

Also: small corrections of geometry calculation

!780

This allows us to implement more sophisticated logic for the different
cases. For DnD surfaces, use the geometry scale of the monitor where
the pointer is, instead of incorrectly assuming '1' as it was before.

!780

Otherwise we won't get core dumps if the launching binary has
capabilities set.

!811

The meta_display_update_focus_window() call has indirect dependencies
on the X11 focus window, in order to determine the correct focus window
on the Wayland side (i.e. may turn out NULL with certain X windows).

In order to have the right x11_display->focus_xwindow there, we should
perform first the focus update on the X11 display.

Fixes focusing of Java applications, as those don't seem to go through
_NET_ACTIVE_WINDOW.

Closes: #819

When tearing down, gnome-shell may call various methods a bit late,
specifically while MetaDisplay is closing, after MetaCompositor is
freed.

Handle calls to the fullscreen unredirect inhibitation counters
happening after MetaCompositor tear down by ignoring them. We're closing
anyway, so it's not a problem.

Fixes: gnome-shell#1710

!808

With the addition of the locate-pointer special keybinding (defaults to
the [Control] key), we have now two separate special modifier keys which
can be triggered separately, one for the locate-pointer action and
another one for overlay.

When processing those special modifier keys, mutter must ensure that the
key was pressed alone, being a modifier, the key could otherwise be part
of another key combo.

As result, if both special modifiers keys are pressed simultaneously,
mutter will try to trigger the function for the second key being
pressed, and since those special modifier keys have no default handler
function set, that will crash mutter.

Check if the handler has a function associated and treat the keybinding
as not found if no handler function is set, as with the special modifier
keys.

#823

The `process_event()` would check for a existing keybinding handler and
abort if there is none, however the test is done after the handler had
been accessed, hence defeating the purpose of the check.

Move the check to verify there is an existing keybinding handler before
actually using it.

#823

shaped_texture requires the clip to be in surface coordinates. Scale
it accordingly.

!758

There were multiple bugs present after the ClutterContent transition.
Refactor `get_image` to:

- always assume surface coordinates for the clip
- return a cairo_surface in buffer size
- make the offscreen path take size arguments, so we can
easily change the assumption in get_image
- fix some clipping bugs on the way

!758

As per commit 75cffd0e MetaShapedTexture is a GObject implementing the Clutter
Content interface.

Reflect this on the type declaration.

!796

Redefine the docstring for functions using shaped texture, setting the clip
as nullable.

!79

As per the code definition the clip can be a null rectangle, and we can also
return a null surface, so mark them as nullable in the introspection.

!796

Delayed clutter timelines might be removed while they are still in the
process of being executed, but if they are not playing yet their delay
timeout won't be stopped, causing them to be executed anyway, leading to a
potential crash.

In fact if something else keeps a reference on the timelines (i.e. gjs), the
dispose vfunc delay cancellation won't take effect, causing the timelines to
be started and added to the master clock.

To avoid this, expose clutter_timeline_cancel_delay() function and call it
if a timeline is not playing but has a delay set.

Fixes #815
!805

If a timeline is delayed and we request to stop or pause it, we are emitting
the "::paused" signal on it, however this has never been started, and so
nothing has really be paused.

So, just try to cancel the delay on pause and return if not playing.

No code in mutter or gnome-shell is affected by this, so it is safe to
change.

!805

Avoid repeating the same code for canceling the delay timeout, using a
function for later uses.

!805

Some drivers expose EGL_EXT_image_dma_buf_import_modifiers so you can
query supported formats, but don't support any modifiers. Handle this by
treating it like DRM_FORMAT_MOD_INVALID.

!782

Clutter stage used to compute the initial projection using a fixed z
translation which wasn't matching the one we computed in
calculate_z_translation().
This caused to have a wrong initial projection on startup which was then
correctly recomputed only at the first paint.

However, since this calculation doesn't depend on view, but only on viewport
size, perspective's fovy and z_near we can safely do this at startup and
only when any of those parameters change.

Then we can move the computation out _clutter_stage_maybe_setup_viewport()
since the cogl framebuffer viewport sizes aren't affecting this.

Fixes gnome-shell#1639
!803

GCC's manpage says that this flag does the following:

  Do not store floating-point variables in registers, and inhibit other
  options that might change whether a floating-point value is taken from
  a register or memory.

  This option prevents undesirable excess precision on machines such as
  the 68000 where the floating registers (of the 68881) keep more
  precision than a "double" is supposed to have.  Similarly for the x86
  architecture.  For most programs, the excess precision does only good,
  but a few programs rely on the precise definition of IEEE floating
  point.

We rely on this behaviour in our fork of clutter. When performing
floating point computations on x86, we are getting the wrong results
because of this architecture's use of the CPU's extended (x87, non-IEEE
confirming) precision by default. If we enable `-ffloat-store` here,
then we'll get the same results everywhere by storing into variables
instead of registers. This does not remove the need to be correct when
handling floats, but it does mean we don't need to be more correct than
the IEEE spec requires.

!785

On 64-bit systems uint64_t is %lu but on 32-bit systems it is %llu.

!778

On 64-bit systems uint64_t is %lu but on 32-bit systems it is %llu.

!778

On 64-bit systems int64_t is %lx but on 32-bit systems it is %llx.

!778

On 64-bit systems int64_t is %ld but on 32-bit systems it is %lld.

!778

To support 32-bit systems without compiler warnings.

!778

The watch might be removed during firing, meaning the source is
destroyed after returning. Avoid use-after-free by unsetting the timeout
before firing. Returning G_SOURCE_CONTINUE in that case is harmless, as
source is destroyed.

Fixes: #796

!799

No need to type cast a `MetaIdleMonitorWatch *` to a
`MetaIdleMonitorWatch *`.

!799

It wasn't used outside the file, so no reason to not have it static.

!799

free path in finalize

!775

Properly free kms-device in case of invalid initialization

!772

Instead of open coding the X11 focus management in display.c, expose
it as a single function with similar arguments to its MetaDisplay
counterpart. This just means less X11 specifics in display.c.

Closes: #751

MetaDisplay and MetaX11Display focus windows are slightly decoupled,
we cannot rely here on the MetaDisplay focus to be updated yet. We
however know the X Window that got focused, so lookup the corresponding
MetaWindow (and client X window) from it.

Closes: #751

Warning about it causes test case failures if there are not enough
privileges, which is likely when running the test suite locally.

!797

`_clutter_actor_get_transition` is almost identical to
`clutter_actor_get_transition` and isn't used anywhere in the code (we
stopped using it with ee00e37b), remove it!

!791

According to the definition of ClutterAnimationInfo in
clutter-actor-private.h, `transitions` is the third variable of the
struct, not the first one.

!791

They are missing the closing parenthesis. Add them.

!795