In the shell code we often use GLib.spawn_async to launch processes
with a GSpawnChildSetupFunc implementation in JavaScript to reset the
mutter nofile rlimit in the new child process.
However, this is highly unsafe to do because this implies that the child
setup function code is executed in gjs where a lot of allocations are
done and even more not-async-signal-safe code is executed, in fact
leading to dead-locks as reported in the past.
To prevent this, declare a new functions that do the same of the GLib
counterpart but without providing a GSpawnChildSetupFunc that is instead
implemented in the C-side doing the cleanup that mutter requires without
allocations or async-signal-unsafe code.
Helps: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/6698
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3293>
(cherry picked from commit 781010be66)
The child context setup function used as launch GSpawnChildSetupFunc
uses g_object_get that internally potentially allocates memory, making
it not async-signal-safe and so not something that is safe to use in
between fork and exec, so just use the simpler getter here.
Note that the current implementation of app_child_setup() is safe
already as sd_journal_stream_fd, dup2 and close are so.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3293>
(cherry picked from commit dcb5956dea)
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/gnome-shell/-/merge_requests/3340>
(cherry picked from commit 1fbfb93cbd)
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/gnome-shell/-/merge_requests/3340>
(cherry picked from commit 27445a1c98)
We currently compute the target scale from the allocated height,
which means that it only becomes available after thumbnails have
been allocated at least once at their expanded height.
As the minimap is initially hidden, this only happens after it is
expanded for the first time, which means the corresponding transition
is not animated.
In order to allow for a fix, compute the target height ourselves
to allow the scale computation to work independently from the
expand factor.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3331>
(cherry picked from commit 2694a64ebb)
The auth prompt currently propagates all key presses, even the
Escape press that is used to cancel it.
On the lock screen that means that the same event that cancels
the prompt (and switches back to the clock) is *also* propagated
to the handler that activates the prompt on key press.
That handler doesn't do anything when the prompt is already visible,
which is the case when the transition to the clock is animated.
However when animations are disabled, canceling the prompt will
result in a new prompt getting created immediately, and the login
screen is stuck on the prompt.
Fix this by not propagating key events that are used to cancel
the prompt.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3317>
(cherry picked from commit e7dc0de75e)
The prompt itself may get destroyed when canceled, in which
case it is no longer possible to chain up in the vfunc.
This is usually not an issue as the prompt is only destroyed
at the end of a transition, but it results in a warning if
animations are disabled.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3317>
(cherry picked from commit dac4f2cb86)
gsettings overrides can be in affect when XDG_CURRENT_DESKTOP is set. We
need predictable default values for predictable tests in mutter and thus
mutter will start asserting that GSETTINGS_BACKEND='memory' and
XDG_CURRENT_DESKTOP=''.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3329>
(cherry picked from commit beb3f120dd)
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/gnome-shell/-/merge_requests/3299>
(cherry picked from commit f31099cd85)
Currently we're using relative positioning when sending click events to
tray icon clients, and this leads to some apps (especially Qt ones) to
try to open the menus at such absolute coordinate under X11.
To prevent this to happen, let's get the root coordinate from the xembed
and let's use it to compute the synthetic event root x/y.
We could have even used the actual event position for this, but getting
it from the xembed makes this more consistent.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3283>
(cherry picked from commit 794acd65a8)
Override redirect windows manage their own positioning and size alone
and are always sticky, so we're not covering them either with the
animation MonitorsGroup, and thus there's no need to clone them or we'd
end up having two windows painted.
This was causing the shell tray icon window actors (that have no opacity
by default but that are override redirect) to show up during the
animation as their clone animation is not 0.
The other option would be hide them during the animation phase, but
there's no need for this.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3285>
(cherry picked from commit 3a34c16eca)
Simply looking at the return value of sd_pid_get_user_unit can be used
to determine if the process is in a systemd unit, but it doesn't say if
gnome-shell is managed by systemd.
For example, running toolbx on a host with systemd creates a libpod
unit, even if the gnome-shell that gets started in the toolbx is itself
not managed by systemd.
We can make sure that gnome-shell is managed by systemd simply by
checking if the unit we're running starts with org.gnome.Shell.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3267>
(cherry picked from commit b3580919e3)
We auto-close FDO notifications when the sender leaves the bus,
given that the protocol was created without persistency in mind
and any action will become invalid.
However that broke when moving the public-facing implementation
into a separate service, as we now track the (always running)
service instead of the original sender.
Fix that by forwarding the sender to the internal implementation
via a private hint, just like we already do for the PID.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3204>
(cherry picked from commit 9af81e44bc)
The hint is a private implementation detail between the public
and internal services, not something anybody else should set
(*cough* libnotify *cough*).
Prefix the name to hopefully make that clearer.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3204>
(cherry picked from commit f0b1cf5f8d)
commit c8bb45b41c introduced a new
verification state, VERIFICATION_IN_PROGRESS, to detect when the user
has already interacted with the authentication prompt, so the
prompt can rate limit the number of times the user can cancel
authentication attempts with the escape key (without also rate limiting
the number of times they can hit escape to go back to the clock without
interacting with the prompt).
That means there are now two states that represent the user actively
undergoing verification: VERIFYING and VERIFICATION_IN_PROGRESS.
It's inappropriate to reset the smartcard service if the user is
actively conversing with it. We try to check for that by looking at the
original verification state, VERIFYING, but we unfortunately, neglect
to account for the new VERIFICATION_IN_PROGRESS state.
The result is that if a user types their smartcard pin at the clock,
and then inserts their smartcard, the pin will get cleared instead of
used, and they have to retype it again.
This commit fixes the oversight, and allows users to again pre-type
their smartcard pin at the clock before inserting their smartcard.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2765>
(cherry picked from commit 3dd5dcd9bb)
Extensions like dash-to-dock use set_icon_geometry() to window.
This changes the dest and scale of ease animation of minimize and
makes it looks very strange. By setting dest opacity to 0 the animation
could be more natural.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2968>
(cherry picked from commit 5d1a0cc525)
Returning a GLib.Error from a method invocation will encode the
error when sent over the wire. In case the error itself is already
an encoded remote error, just passing it on will result in double
wrapping.
Avoid that by stripping any remote error information before
returning it.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3159>
(cherry picked from commit a5d33026f0)
Pipelines for non-protected branches are set to 'manual', and
thus cheap. However they may still get picked by `@marge-bot`,
meaning that the bot waits for the completion of a pipeline that
never starts.
Avoid that by not creating pipelines for branches with open
merge requests.
Credit to Jordan, who came up with this for gst.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3153>
(cherry picked from commit d3e96a36ce)
If something grabs the key focus while a modal is pushed, keeping
key-focus on that actor seems like the smarter thing to do than setting
it back to the last focus after the modal gets popped again. So check if
the key focus actor that we set when pushing the modal got changed when
popping that modal, and if it got changed, simply don't touch key focus.
This fixes a bug with the close dialog, where key focus isn't correctly
set to the dialog after alt-tabbing to a window showing a close dialog.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/786>
(cherry picked from commit 4d544d7b56)
When setting the key focus to `this._dialog`, the default button is not
automatically focused and no button has key focus.
Use the `initialKeyFocus` property of the dialog instead, and set focus
to the default button if the dialog is not already focused.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/786>
(cherry picked from commit de834fe307)
Now that StIcon updates its texture automatically on icon theme
changes, we only have to recreate icon actors that aren't StIcons.
(This probably only applies to the folder icon in the app grid
where the sub-icons do use St.Icon, but making that assumption
feels dodgy with an API as generic as `createIcon()`)
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/3141>
(cherry picked from commit 8a1f44f10a)
