In some cases, reparenting a window with its frame may fail; this seems
to happen especially during initialization of a window that may be
unmapped and re-mapped quickly and multiple times.
If this happens, we're never going to receive a remove event on the
stack tracker and so we may end up adding it twice to the list of the
windows to synchronize with the compositor, breaking its assumption that
the stack list is unique, and eventually leading to a crash because we
do not end up removing all the instances of a window on its destruction.
In particular we may end up in this situation:
Syncing Window 10485927: 0x555558863540 (actual xid is 10485927),
user time is 10485928 frame is 0x5555588715c0, frame xid 6291591
Syncing Window 14680081: 0x5555588664b0 (actual xid is 14680081),
user time is 14680082 frame is 0x555558871d80, frame xid 6291595
Syncing Window 6291460: 0x55555796dc80 (actual xid is 10485763),
user time is 10485764 frame is 0x555557a6f630, frame xid 6291460
Syncing Window 6291465: 0x555557a68af0 (actual xid is 14680067),
user time is 14680068 frame is 0x555557a73e80, frame xid 6291465
Syncing Window 6291509: 0x555557f9d830 (actual xid is 8388623),
user time is 0 frame is 0x555557fac780, frame xid 6291509
Syncing Window 6291586: 0x5555586e1690 (actual xid is 4194363),
user time is 0 frame is 0x55555886e550, frame xid 6291586
Syncing Window 6291591: 0x555558863540 (actual xid is 10485927),
user time is 10485928 frame is 0x5555588715c0, frame xid 6291591
Where the same meta window 0x555558863540 is added twice because that's
both mapped by the window itself (10485927) and by its frame (6291591).
This happens because for historical reasons the xids hash table managed
by the x11-display maps both the X11 windows, their frames and their
user time windows as the meta-window, and so if we don't filter out them
properly we end up duplicating the entries in the compositor list.
Such duplicates finally end up making mutter to crash in
meta_compositor_sync_stack() because we could end up trying to access
to an invalid window, given its actor has been destroyed but not all the
instances have been removed from the compositor windows list:
0x00007ffff71059 in meta_compositor_sync_stack (compositor=0x555555b8,
stack=0x555558701b80) at ../../mutter/src/compositor/compositor.c:773
773 if ((old_window->hidden || old_window->unmanaging) &&
(gdb) print old_window
$1 = (MetaWindow *) 0x0
So, in order to prevent this, check that XReparentWindow does not fail,
and in case of failure, reset the window state to the one it had before
we failed and more importantly, remove the association between the frame
X11 window and the MetaWindow, since this is not true anymore and so
that at the next stack synchronization there won't be any meta window
associated to that frame XID (unless there aren't further stack changes
impacting on that). Without this we would have instead waited for the
remove event that we predicted, but that could never happen because no
ReparentNotify is emitted in such case.
Since we listen the X events in the same thread, and they are delivered
through the main loop, there's not any need to set the frame details on
windows before the reparent operation, because such action could fail.
So move the code order, as preparation for handling the error.
In the future, the module will automate uploading the release
tarball. We already use the CI pipeline to generate the tarball,
so it's easy to hook up the module and provide some testing
before the module goes into production.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3775>
(cherry picked from commit a66dab9273)
We currently assume that the `CI_COMMIT_TAG` variable matches the
version component of the generated dist tarball.
That is usually correct, but sometimes errors happen and a wrong
tag is pushed, and the real release uses something like "46.0-real".
Account for that by building the artifact path from `meson introspect`
and exporting it as environment variable.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3775>
(cherry picked from commit bcfbc7fa17)
When meta_wayland_pointer_constraint_remove() is called, it can trigger
a meta_wayland_event_handler_invalidate_focus() via:
meta_wayland_pointer_constraint_destroy()
meta_wayland_pointer_constraint_disable()
meta_wayland_input_detach_event_handler()
meta_wayland_input_invalidate_all_focus()
meta_wayland_event_handler_invalidate_focus()
Which then would result in a "focus-surface-changed" signal which would
call meta_wayland_pointer_constraint_remove() a second time. This
happens after surface_remove_pointer_constraints() has already been
called in the first meta_wayland_pointer_constraint_remove() call,
leading to "data" being NULL.
To prevent this issue disconnect the signal handler before calling
meta_wayland_pointer_constraint_disable() when destroying a constraint.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3476
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3749>
(cherry picked from commit c3e626405f)
In case of empty regions (e.g. when locking the pointer) the pointer
was only forced to stay within the boundaries of its current pixel
(i.e. culling subpixel position), instead of the position where the
pointer lock did start.
Fixes: 07d24fe50 ("backends/native: Allow infinitely small pointer constraint regions")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3749>
(cherry picked from commit d686865918)
Since 07d24fe50 regions are not translated to their on-screen
coordinates anymore, but are relative to the origin stored in the
constraint. This origin however was not considered when checking whether
the pointer was within the constraint region. This meant that the
constraint region would appear to always be placed at 0,0 instead of on
the surface.
Fix this by using the cursor position relative to the origin.
Fixes: 07d24fe50 ("backends/native: Allow infinitely small pointer constraint regions")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3409
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3749>
(cherry picked from commit c0537096c2)
We close wayland popups when a button or touch release happens outside
of the grab, except we don't want to close them when that button release is
actually the release of the press that was opening the grab in the first
place.
We never see the press event that opened the grab, so the first event we
see is actually always a release. Make sure to not close the popup on that
event, and instead only close the popup if we see the press count drop from
1 to 0.
This fixes a bug where popup would close right after they open. To
reproduce, click to open a popup, hold pressed and move the cursor over
shell chrome, then release. Or alternatively test with a popup that gets
opened with a long-press gesture (eg. long touch long press on libadwaita
tabs), just doing the touch long-press and then release.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3631>
(cherry picked from commit 79a79b3450)
because we want to switch between two workspaces. In some configurations
there is only a single workspace at this point so trying to get current
workspace + 1 gets us a NULL pointer.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3730>
(cherry picked from commit 301c154f02)
The internal representation of the min/max width of windows include what
is outside of the window geometry, so when the window geometry changes,
but the min/max size did not in the same commit, we'd be left with an
out of date min/max size, potentially causing windows to shrink when
configured.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3755>
(cherry picked from commit 97af5f8705)
local_error in meta_egl_query_device_string() is using g_autoptr,
meaning that it was getting freed after g_propagate_error(). This then
would result in error->message becoming invalid, causing crashes when
logging the error message later on.
Fixes: 8234f5bc7 ("egl: Return success status from meta_egl_query_device_string")
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3758>
(cherry picked from commit 7a38e12ed0)
Just shovel the data through our own stdin/stdout, which will end up at
the right place (e.g. /dev/null).
This should hopefully solve `mutter-dist` failing due to a D-Bus method
call timeout in CI.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3757>
(cherry picked from commit e2e687c9db)
When a window with an input shape on its decoration window becomes
undecorated and meta_window_x11_update_input_region() gets called via
notify::decorated, the buffer_rect of the window has not been updated
yet while the decorated property has. This would lead to us comparing
the input shape of the client window to the buffer_rect which still
includes the decoration window. This would fail to detect the common
case when the client window has no input shape set, leading to the input
region being set to the size of the client window rather than NULL. If
the window is then resized later, the input shape would remain at the
previous size.
This was not a problem before 6bd920b35, because then we were (wrongly)
always comparing to the client_rect.
Fix this by choosing the correct rect for comparison depending on
whether the window is decorated.
Fixes: 6bd920b35 ("x11/window: Use correct bounding rect to determine NULL input region")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3451
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3720>
(cherry picked from commit 1f1538be76)
There is nothing to allocate for a 0-sized files, and indeed
posix_fallocate() will error out if the passed len isn't greater
than 0.
Now that anonymous files are used to back the memory selection
source, this fixes unsetting the selection when the screen is
locked.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3752>
(cherry picked from commit da0bd303ad)
When a key binding is removed, and a trigger key sequence is dispatched
before the idle callback that resolves and updates the actual binding,
we should handle that gracefully.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3711>
(cherry picked from commit 22f67d107d)
Two new fields: ref_count and removed, are added to MetaKeyHandler, and
it would be freed only if the ref count has reached 0. When handler is
removed from key_handlers GHashTable, key_handler_destroy() would mark
removed as TRUE, and do an unref. handler->removed is checked in
get_keybinding, and binding with handler removed would not be used.
Also in MetaKeyBinding, it now has the ownership of the name field, to
avoid it being freed before logging. Create or copy a binding would
do a ref inc for handler, and free one would unref handler.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/1870.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3711>
(cherry picked from commit 8c39a25459)
When a client resizes on its own, make sure the new size is passed
through the window constraints machinery directly, to trigger any
potential window management rule that might apply.
Fix a couple of tests to make use of this behavior by introducing a new
'wait_size' command that waits until a window has been resized to a
expected size.
This replaces the fix introduced in 0e736af301 ("window: Ensure
constraints after a Wayland client resize").
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3700>
(cherry picked from commit 57e16cf010)
Starting the timeout to move from hidden to suspended before the window
is mapped means we don't have a previous window configufration which we
need to get the new window configuration with the suspended state.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3731>
(cherry picked from commit e509fc7f00)
Sometimes the test runners are saturated with other work. Bump the test
timeouts by a multiplier of 5 with the hope that they now will be much
more likely to have time to finish in time.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3735>
(cherry picked from commit 3dbe3668e6)
Prior to commit 5dfed8a431, the MetaWaylandKeyboard would always remember
the last key press serial, and consider it valid after the key was released,
as long as no other key presses/releases happened in between.
That commit improved things so that MetaWaylandKeyboard can track multiple
keys being pressed simultaneously, but also changed so that the serial for
a key press is immediately forgotten after the key press event was received.
This may break in situations like testing or keyboard macros where key
press and release is handled in a quick sucession, so the client reaction
to the key press (e.g. popping up a menu) might arrive too late.
Add a sort of spiritual successor to this handling, and make keyboard
press serials corresponding to the last key up forgotten at the next
key press/release received.
Fixes: 5dfed8a431 ("wayland: Preserve serial for all pressed keys")
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3458
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3721>
(cherry picked from commit 243890a688)
Fix an obvious copy paste error that slipped through the cracks.
Fortunately it doesn't have a visual impact for well behaving clients
but only makes us not hit direct-scanout paths, assuming no other bugs
in the stack.
Fixes: f21762ea6e (wayland: Add support for preferred_buffer_scale/transform)
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3717>
(cherry picked from commit 6a81d5f0bb)
Until this surface or its parent is finalized.
This makes sure that any `MetaWaylandSubsurfacePlacementOp` referencing
this surface for sibling will be applied as intended.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3737>
(cherry picked from commit be4bf8da9c)
This reverts commit 35d92e0fac.
This turned out to cause trouble, because it can prevent
MetaWaylandSurface::applied_state.subsurface_branch_node from ever
getting linked up for a sub-surface.
It shouldn't be necessary anyway, since permanently_unmap_subsurface /
wl_subcompositor_get_subsurface reset the sub-surface state as defined
by the protocol.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3737>
(cherry picked from commit 5749810ddb)
Mutter is currently enabling the protocol when the fence can't be
duplicated and it never signals the release of it (which is problematic
in some Mesa drivers like nouveau when used together with the nouveau
Vulkan driver because of the recent explicit sync support merge)
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3744>
These functions are specific for X11 windows only so we should check
if the passed window is an X11 one, not just a MetaWindow since we're
casting to the actual type at later point.
Fixes changes part of commit e1e6534eb
If a window sends a configure stacking request, we were comparing the
active window with the event window even though they were different
client types (e.g. wayland and x11).
This was leading to a critical error, so let's handle this by ensuring
that the active window is of the same kind of the event window before
doing x11-specific checks. Behaving as different applications in case.
