Resizing effects are more finicky as other effects, as the actual
animation is delayed until we receive the ::size-changed signal.
However that signal may never be emitted if the window is destroyed
just after starting the size-change effect, in which case the effect
is never completed, blocking mutter from destroying the corresponding
window actor.
Address this by tracking when a resize effect is pending, and complete
the effect when appropriate.
https://gitlab.gnome.org/GNOME/mutter/issues/655https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/815
In the xwayland-on-demand scenario, it may happen that Xwayland is
shutdown (causing a restart of ibus-daemon to drop ibus-x11) while
we are typing.
If we have a bit of bad luck, this will cause the IBusInputContext
to be disposed (due to its bus "closing") at a time when we have
an ibus_input_context_process_key_event_async() request on the fly.
As the object is disposed in between this would tickle JS (rightfully
complaining that it's been disposed under its feet) and make us pass
an actually NULL IBusInputContext to the corresponding _finish()
function (despite the IBusInputContext being still held alive by some
other refs). This will assert and abort in
ibus_input_context_process_key_event_async_finish() then.
To handle this, listen for IBusInputContext::destroy, and reset our
internal state, this way we can compare on the JS side that the
IBusInputContext is indeed an up-to-date one.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/813
I've observed that UPower can occasionally report a charge level of 100%
while the state is still "charging". This usually doesn't last very long
but it is noticeable because the power icon changes to a "missing icon"
icon. This will handle that rare case correctly.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/814
Clutter originally cluttered its namespace with key symbols, before
prefixing all symbols with KEY. We still use the unprefixed symbols
occasionally, replace them so mutter can drop the deprecated symbols.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/808
Since polkit takes a few milliseconds from initiating the session to
emitting the "request" signal, don't introduce visual distraction by
hiding the password entry and showing it again a few ms later.
So start a timeout of 200 ms when we destroy a session and if no session
request (i.e. a request for a password-authentication) happened during
this timeout, hide the password entry.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/788
Since we don't know if polkit/PAM will request a password (emitting the
"request" signal) or use another authentication method like a
fingerprint after the current authentication failed, hide the password
field and make the "Authenticate" button insensitive after cancelling
the session, just like we do when creating the dialog.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/788
Set the key focus to the password field only after we got a request
(and therefore know that a password is requested) instead of using
`setInitialKeyFocus()`. This way we don't try to focus the password
field by default if we aren't showing it (e.g. in case the user has no
password or is using fingerprint login).
Also we have to move the call to `grab_key_focus()` to happen after
`_ensureOpen()`, because otherwise the ModalDialog will set the focus to
one of the buttons while opening itself.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/788
Show the user avatar for all users, including the root user. The root
user will always have the generic avatar, but it looks more consistent
than showing no avatar at all.
This way we also don't have to worry about the spacing introduced by the
polkit-dialog-user-layout CSS class, which would give the
"Administrator" label a small offset to the left.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/788
xgettext got better at recognizing template strings, so we can
replace more string concatenations. Alas xgettext is still buggy
(surprise, regular expressions are hard), so there are still a
handful of holdouts that prevent us from making a complete switch.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/792
Every since commit aa394754, StBoxLayout has supported ClutterActor's
expand/align properties in addition to the container-specific child
properties. Given that that's the only container left with a special
child meta, it's time to fully embrace the generic properties (and
eventually remove the child meta).
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/780
This wasn't added on the GObject-ification in commit c4c5c4fd5c. This
introduced warnings and misbehaviors when closing the dialog. (Eg.
pressing the "show OSD" pad action would show stack different dialogs
on top each other).
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/800
It seems like some recent change (maybe the move to a ClutterActor
subclass for AppIcon) broke the check whether the drag source is an
instance of AppIcon. While the drag source indeed is an AppIcon and
everything else works correctly, the check still returns false, which
breaks the creation of new folders using DnD.
Theoretically it makes sense that this doesn't work, because we're
assigning AppIcon using `var AppIcon =` and that will only get set after
`GObject.register_class()` finished, so accessing `AppIcon` inside that
function seems risky and is probably wrong.
Fix this by comparing to `this.constructor` instead of `AppIcon`, which
works fine and we know for sure exists at this point.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/794
I have observed a client in the wild (Chromium again) set CanPlay to
false momentarily while it is loading the next song. Previously, the
code would close the player proxy in that case, meaning that after
playing one track, the MPRIS message would disappear and never come
back.
However, I think this use of CanPlay, while apparently not usual, is not
incorrect according to the spec. We should hide the message instead of
closing the player proxy.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1362
In the wild we have buggy clients (notably Chromium 77 and earlier) that
send metadata with the wrong types. Previously, this would throw an
exception and prevent the MPRIS information from showing up in the
message list.
This changes the code to check if any incoming metadata is of the type
it is expected to be, and logs a warning if not, then continues on with
a default value.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1362
Make gjs to compute the GType name for registered GObject-derived
classes using the file basename and the first directory name, so that we
can avoid name clashing, and ensure that no extension will break the
shell by registering a name that is already used (by the shell or by any
other extension).
This requires gjs commit 02568304 [1] that will be part of release 3.35.2,
so bump the required version as gjs does post-release version bumps.
[1] https://gitlab.gnome.org/GNOME/gjs/merge_requests/337https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/790
Similar to the previous work-around, xgettext gets thrown off by
embedded quotes in template strings, in particular where a template
"breaks up" a pair of quotes.
Throw in some more comments to make xgettext happy, but whoever makes
the gettext toolchain not depend on fragile regular expressions will
be drowned in beverages of their choice ...
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/792
Xgettext learned about template strings now, which is good. However
it's still buggy, so instead of the "classic" xgettext issue with
backticks, we now have exciting new issues to find work-arounds for.
One issue is that it doesn't detect backticks inside string constants
as regular characters, so having an odd number of backticks throws off
its regex.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/792
This corrects weird-looking blending visible as it fades out when the
overview closes. Previously the entry's dark background would drown out
the text as it fades out, but now they maintain a consistent contrast ratio
during the fade.
There's no noticeable change in performance, but in theory it should be
faster as text entries don't change at full frame rate. So stage redraws
will usually have a cached searchEntry drawn and require less effort.
Though the main purpose here is to correct the appearance.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/778
The close dialog is added as a child to MetaWindowActor, and, in Wayland
sessions, since commit [1] MetaWindowActor applies a transformation
matrix which scales all it's children using the geometry scale factor.
Now because the dialog actor is not a window (i.e. a MetaSurfaceActor),
but a subclass of StWidget, the scale factor is also applied to the
properties of the dialog by StThemeNode, so we end up applying the
geometry scale twice to the close dialog.
Fix this by applying the inverted scale to the dialog, which leaves the
scaling only to MetaWindowActor. This means we also can't apply a pivot
point other than 0 to the dialog actor, so apply the 0.5-pivot point to
the `_dialog` child of the Dialog class (the actual visible dialog box)
and also perform scaling animations on this child.
[1] https://gitlab.gnome.org/GNOME/mutter/commit/fb9e8768https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/783
Remove all transformations from window actors after a window animation
was either cancelled or finished. Right now we only do that if the
transition finished successfully, which seems kind of pointless (it can
probably be historically explained because the callbacks inside the
"kill-window-effects" signal handler are connected to those
`*WindowDone()` functions though and the `*WindowOverwritten()`
functions were only added later in [1]).
This fixes a recent regression where a window animation would get
cancelled and remain stuck by switching workspaces. The regression
probably happened due to different behaviour of the `onOverwritten`
callback of Tweener and the `onStopped` callback of Clutter transitions:
For the workspace-switching animation the window actors get reparented
to a temporary container, which makes Clutter transititons emit
"stopped" (`clutter_actor_remove_child_internal()` stops transitions on
its children), while Tweener would continue the animation.
[1] 6dd302e5cehttps://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/784
Some vfuncs like `button_press_event`, `button_release_event` and
`touch_event` don't have handlers in the parent classes of
PopupBaseMenuItem. So don't call those handlers and return the default
Clutter.EVENT_PROPAGATE there.
This fixes a crash of the shell that happens when pressing a mouse
button inside the system popup menu and releasing it above a slider like
the volume slider again.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/787
We currently only update the windows section when either the focus app changes,
or when the app's windows change (that is, a window is opened or closed). This
allows the menu item labels to become stale if the window title changes after
one of those events (for example when switching tabs).
Fix this by updating menu items when the corresponding window title changes.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1830
g_desktop_app_info_get_categories() may return null. In that case, the
previous code would fail to create a folder when dragging an app with
no categories onto another app. Instead, simply continue with the next
app info.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/782
In the case of bugs in a drag target's acceptDrop() function, it may
throw an exception. In the previous code, this would break out of the
loop entirely and never cancel the drag, so the mouse button release
event would be ignored and you would have to press Esc to get out of the
drag.
In this change, if acceptDrop() throws an exception, we log it and move
on to the next parent target instead.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/777
Those are two useful ClutterTimeline properties and
will be needed for wiggling the search entry when
failing the password.
Add support for passing repeat-count and auto-reverse
to ClutterActor.ease and ClutterActor.ease_property.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/768
When accessing properties on ClutterActor for size and position there is
a notable access time overhead. This overhead adds considerable user lag
when opening the overview if many windows are open.
This is primarily due to these properties being accessed while sorting
WindowClone instances by their window's center for placement in the
overview. By pre-computing this center value only once when
initializing WindowClone, the induced lag can be significantly reduced.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/763
Overview's animationInProgress, visible and visibleTarget properties are not
meant to be modified from others, but be read only.
So make this clearer using properties getters and private values.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
Instead of manually updating properties on change, use native properties
bindings to keep the them synchronized.
Disable hover-tracking and focus-ability when the avatar is not sensitive.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
Now that the calendar message list and the message sections are actors, there's
no need to keep track of the sections in a different Map as we can just use the
native clutter functions to manage the children and access to their properties.
Also cleanup the signal connection/disconnection.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
When messages are added to the message list, we create a container for those,
however since now the messages are actor themselves we can just create a list
item actor that holds the message actor and refer to the message parent in order
to get their container.
This allows to remove the obj container map we used, using the native clutter
parent-child hierarchy and handle signal connections cleanly.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
Since all the search result classes are now GObject classes, we can enforce
the methods we want to have in there (just activate() for now) using an
interface, to make sure they are implementing what we require and to easily
group all the classes that can be used as search results, even though they
are not extending SearchResult.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
Remove the `this.actor = ...` and `this.actor._delegate = this` patterns in most
of classes, by inheriting all the actor container classes.
Uses interfaces when needed for making sure that multiple classes will implement
some required methods or to avoid redefining the same code multiple times.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
Dispose the Source Object when dispose() is called, avoiding that it could be
called twice on a destroyed Source.
So, notify count changes before destroying the object, and don't emit this
twice on destroyNonResidentNotifications (as if a notification is destroyed
the property notify will happen in the notification destroy callback anyways).
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
When the notification is destroyed we should also dispose the underneath GLib
object, and ensure that we don't dispose this twice.
In order to avoid this, don't destroy transient notifications that have been
already been removed and only destroy the resident notifications on activation
if they have not been destroyed earlier.
Thus connect after to the 'activated' signal and once the default handler has
been called destroy the notification if not requested earlier.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
The Keyboard class used to be both a view and controller class, however in order
to make the keyboard a native Clutter.Actor, we need to separate the widget from
the controller class, so that we can manage the actor lifetime from the JS side.
Thus, initialize the keyboard actor on the Keyboard constructor and create a
KeyboardManager class to manage its state and lifetime.
Add proxy methods for the public functions that were used by other shell
components
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
Meta.Background has already a 'changed' signal and not to confuse the source
signal with the wrapper one, rename the wrapper class signal into 'bg-changed'.
This will be relevant when we'll inherit from Meta.Background, as signal
emissions from the base class could interfere with the wanted derived class
behavior and with the the grouping of successive changes into a single ::change
emission.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/559
When the screen is marked as idle, we normally start a fading animation and
a timeout to finally lock the screen. This timeout is configured using the
fade time if no longer delay is set in settings.
However if animations are disabled or slowed-down/up, the fade time is
different from the STANDARD_FADE_TIME and so we might end up showing the
lock shield without actually locking for STANDARD_FADE_TIME in the disabled
or slowed-up animations case, or locking too early in case of slowed-down
animations.
So, just adjust the timeout time using the same logic of animations so that
this value is matching all the times.
Related to https://gitlab.gnome.org/GNOME/gnome-shell/issues/1744https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/749
Devices like cameras and microphones are privacy sensitive, as they can
be used to spy on the user. We cannot prevent non-sandboxed apps from
doing that, but as we already track when the microphone is recording,
we can at least show an indicator to make sure it doesn't happen behind
the user's back.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/729
Search result views can include also objects that are not inheriting from
SearchResults (such as the AppIcon) that has not any 'activate' signal, to
connect to. Since we want to enforce a more formal interface, we want to
have just a simpler requirement as an activate() method.
So, instead using the 'activate' signal in SearchResult to activate a result
via SearchResultsBase just implement activate() in the result.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/765
When a transition is set up with a delay, it may be removed before it
actually started. We won't get a ::stopped signal in that case, with
the result that we currently end up with a mismatched unredirection
disabling.
Address this by only disable unredirection once the transition has
actually started.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1788
gnome-session used to show a dialog in this case, but a
notification is more natural nowadays. Doing it in gnome-shell
avoids complicated synchronization between gnome-session and
gnome-shell.
https://bugzilla.gnome.org/show_bug.cgi?id=701212
We share this actor with other shell menus, which arguably track a different
"cursor" as we care of the caret/anchor text positions, and menus care about
pointer click coordinates.
Use a standalone actor for this, so popups/IM are entirely decoupled.
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/1571
If graphical applications want to start from systemd units, they need to
start after we're properly ready to display them. This is particularly
important under X where `_GTK_FRAME_EXTENTS` and other xprops are needed
to have the right theming.
We're doing this in an idle callback so that the dynamic starting of
`gnome-session-x11-service.target` (which launches `gsd-xsettings`) as
the result of a signal emission happens before us signalling we're ready
for later things to start.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/750
As per clutter optimizations in should_skip_implicit_transition() any
transition will be ignored if applied to an actor with unmapped clones.
Since we initialize the lightbox as hidden, when we use it standalone (as it
happens for the long fade in screenShield) the transition will be ignored.
This causes the lockscreen fade-out after the idle delay not to work, but
instead to have an apparently locked system that is instead not locked at
all.
So, just ensure that the lightbox actor is visible before applying to it any
transition.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1683
The shell tries to spawn the ibus daemon on startup if unavailable, however
as per commit 8adfc5b1 we also force restarting it once the X11 server is
available.
Unfortunately this could cause a race if we disconnect while we were already
connected to an ibus daemon, but still in the process of going through the
various nested calls.
In fact the ::disconnect callback didn't stop any further async ibus call
that, even if failing, would have eventually triggered the emission of a
'ready' signal and to the Keyboard's callback, leading under X11 to a full
grab owned by ibus daemon.
In order to avoid this and keep control of the calls order, use in both
IbusManager and InputMethod a cancellable that is setup before connecting to
the bus, and that is cancelled on disconnection.
Then handle the finish() calls properly, using try/catch to validate the
returned value, taking in account the potential error and just not
proceeding in case of cancellation.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1712
gnome-shell calls ibus_input_context_focus_in() in InputMethod.focus_in()
but the event is not actually forwarded to panels and engines in GNOME
Wayland because gnome-shell changes IBus.Capabilite by focus events and
disables IBus.Capabilite.FOCUS when ibus_input_context_focus_in() is called.
IBus.Capabilite is assumed a fixed value per input context in the
first place and it should not be changed by focus events.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/573
Unset the signal IDs we connected to when starting the drag. Otherwise
we get error messages if a touch drag is ended after a mouse drag
happened because the signal IDs are still set but no signals are
connected.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/740
The screen shield creates the unlock dialog based on the session mode.
However since commit 0c0d76f7d6 turned LoginDialog into an actor
subclass (while UnlockDialog kept using the delegate pattern), it is
no longer possible to handle both objects the same way without warnings.
Allow this again by turning UnlockDialog into an actor subclass as well.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/736
When easing, we need the transition of one of the involved properties
to connect our callbacks. Currently we simply get the transition for
the first property, however as Clutter optimizes the case where a
property doesn't actually change, that transition may be NULL even
though we still animate other properties.
So instead of only looking at the transition of the first property,
try to find a transition for any of the involved properties.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1572
The RenameFolderMenu uses the internal box as a menu item, while PopupMenu
expects to have PopupBaseMenuItem based children with a delegate set.
Instead of using a custom menu with a customized box acting as menu
item,just add a RenameFolderMenuItem that inherits from the parent,
adjusting the features as we need them. In fact, the rename folder menu item
doesn't need any label, padding or default styling so we can reuse
PopupMenuBaseItem after we use our styling properties and we set the
Ornament to HIDDEN.
To get the proper style in place, define rename-folder-popup and
rename-folder-popup-item to override the default popup-menu-item rule
padding instead of using margins.
Pass the menu item as menu's focusActor as this will key-focus it on pop-up,
by overriding the key_focus_in() vfunc we can then delegate the focus
handling to the entry's clutter-text.
Also override the map() vfunc in order to update the entry's content before
mapping the entry.
Finally, use the item's activate method in order to tell the parent menu
we're done with it and that the menu can be closed.
As consequence we can also remove the menu's popup() method, and just use
the default open().
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/720
When the rename folder menu is opened the text entry is expected to be
focused and selected for a quick editing.
While this is required it doesn't actually happens since PopupMenu by
default gives the key focus to the source actor, that is then free to pass
the key focus to the menu if there's an user interaction.
In this case however, we want the text entry to be focused once we prompt
the menu, so just use the PopupMenu's focusActor property to ensure it will
handle it for us.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1604https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/720
The PopupMenuManager is supposed to grab and focus the menu actors, with
normal menus we always need to grab the actual menu but set the key focus to
the source actor so that it will be able to move the focus to the menu
child, if requested.
However there are menus such as the RenameFolderMenu that requires the
key-focus once prompted, so provide a focusActor property (defaulting to the
sourceActor) that can be set in order to define the actor to give the
keyboard focus to, when the menu is popped-up.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1604https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/720
The menu item ornament is used to put dots or checks in menus or otherwise
to define a padding for a label.
However in some cases we want to create a menu item with no left (in ltr)
padding.
In order to do that, define a HIDDEN Ornament mode that completely hides the
ornament actor.
The naming here might be confusing as this should probably be called NONE,
while the default mode is the invisible one, but it's too late to change it
now.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/720
While commit 3094f863 was intended to cancel the ongoing idle hide
timeout before we start a new one, a mistake slipped in there while
rebasing: Obviously we should check if the signal id is NOT 0 here.
This didn't prevent timeouts being started while old ones are still
running and did override `this._idleHideOverlayId`, which caused the old
timeouts to run indefinitely after an overlay actor was destroyed
because we fail early (and don't return TRUE) in `_idleHideOverlay()`.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/735
Indicate whether dropping an app icon was successful or not by using the
newly added `animateLaunchAtPos()` API of AppIcon which starts a zoom
out animation of the icon at the position the drop happened.
To get the position of the drag actor, we have to forward the arguments
passed to `acceptDrop()` and `handleDragOver()` to the internal drag
handlers of the WorkspaceThumbnails. We can use this position directly
without transforming it to stage coordinates because the actor is a
child of `Main.uiGroup` and the animation actor will also be a child of
this container.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/121
Add a `animateLaunchAtPos()` method to the AppIcon class to animate the
launch of an app at a given position. This allows for a visual
indication of whether dropping an app icon using DnD was successful at
the position the drop happened in a later commit.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/121
Return the results of calls to acceptDrop that we forwarded to the
Workspace object.
This fixes a bug where app icons that were dragged and released above a
window clone would get animated back to their original position
(indicating that nothing happened) even though they opened correctly.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/121
`AppIcon.shellWorkspaceLaunch()` can easily be replaced by checking for
`AppIcon.app` and calling `AppIcon.app.open_new_window()` directly.
For compatibility and to prevent breaking extensions implementing the
function, keep supporting the `shellWorkspaceLaunch` API in AppIcon
while logging a deprecation warning. Also keep supporting the API on
drag sources (without deprecating it) to allow extensions to define
custom actions on their drag sources.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/121
If the search entry does not have any text typed in and a button press
happens outside of the search entry, we set key focus to NULL to make
the search entry appear unfocused.
This is quite intrusive and can easily cause unwanted focus changes, so
change the captured-event handler to only call `reset()` if the search
entry actually is focused.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/683
Instead of returning and waiting until the old timeout is finished,
start a new idle hide timeout for the overlay when the pointer enters a
window clone. This makes sure the timeout for hiding the overlay after
the pointer left the clone mostly stays the same (except when leaving
the clone via the title or the close button).
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/683
As arrow functions have an implicit return value, an assignment of
this.foo = bar could have been intended as a this.foo === bar
comparison. To catch those errors, we will disallow these kinds
of assignments unless they are marked explicitly by an extra pair
of parentheses.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/731
Calling await in a loop means the asynchronous operations are
run sequentially instead of in-parallel. Usually that's not
what's wanted, so eslint has a rule to warn about this.
However here we use async/await to handle control back to the
mainloop between steps, so running operations sequentially is
actually intended.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/731
The intention of the code is clearly to operate on a copy, but that's
not how the Object constructor works. While it doesn't matter in
practice that we modify the passed-in object parameter, it's still
a good idea to fix the code.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/731
The legacy indent rule currently ignores arrow functions in parameters
to allow callbacks to not align with the other arguments:
this._someFunctionWithFairlyLongishName(arg1, arg2, arg3,
() => {
this._someOtherFunctionWithLongName(arg1);
});
But as ignoring entire nodes means we can end up with arbitrary
indentation, we should drop the exception. While this would make
the above "illegal" under the legacy config, it conforms with the
non-legacy style, so everything should be fine ...
... except that eslint starts to complain about some function args
that should be fine under the legacy config. Maybe it's thrown off
by the function-arg-in-arrow-function-in-function-arg structure, but
rather than figuring it out, let's just move those to the new style.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/725
We are currently inconsistent whether to put the operators in front
of the corresponding line or at the end of the preceding one. The
most dominant style for now is to put condition and first branch on
the same line, and then align the second branch:
let foo = condition ? fooValue
: notFooValue;
Unfortunately that's a style that eslint doesn't support, so to account
for it, our legacy configuration currently plainly ignores all indentation
in conditionals.
In order to drop that exception and not let messed up indentation slip
through, change all ternary operators to the non-legacy style.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/725
Some more places where the indentation doesn't comply with either
the old or new style. They slipped through because the legacy eslint
configuration accounts for some patterns by plainly ignoring certain
nodes. We'll address that later, first fix up the indentation errors.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/725
When there is a transition, it's likely that we are animating some part
of the desktop, and in such situations we don't want to unredirect
fullscreen windows.
This fixes unwanted unredirection when e.g. hiding a modal dialog by
re-enabling the unredirection after the animation has finished, instead
of when it starts.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/721
Updating the window list in the Looking Glass is a costly
operation: it destroys a whole lot of actors, and recreates
them. This triggers CSS changes, repaints, and allocations.
It is specially bad when paired with Wayland's big number
of window creations and deletions when showing Builder's
and Epiphany's popup window.
Only update the window list in the Looking Glass when it is
visible.
Related: https://gitlab.gnome.org/GNOME/mutter/issues/556https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/719
Since ES5, it is possible to create objects with no prototype at all:
let foo = Object.create(null);
Those object won't have any builtin properties like hasOwnProperty(),
which is why eslint added a corresponding rule to its default rule set.
While this isn't an issue that affects our code, there's no harm in fol-
lowing the recommendation and call the method through Object.prototype.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/716
We currently use no less than three different ways of indenting
object literals:
let obj1 = {
foo: 42,
bar: 23,
};
let obj2 = { foo: 42,
bar: 23 };
let obj3 = { foo: 42,
bar: 23
};
The first is the one we want to use everywhere eventually, while the
second is the most commonly used "legacy" style.
It is the third one that is most problematic, as it throws off eslint
fairly badly: It violates both the rule to have consistent line breaks
in braces as well as the indentation style of both regular and legacy
configurations.
Fortunately the third style was mostly used for tween parameters, so
is quite rare after the Tweener purge. Get rid of the remaining ones
to cut down on pre-existing eslint errors.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/716
Since disabling an extension will lead to disabling and re-enabling all
following extensions in the list, always disable multiple extensions by
looping through the list in reverse order.
This lowers the execution time of the event handlers quite a bit if many
extensions are installed.
Thanks to Philippe Troin for identifying the problem and proposing the
initial patch to change the extension order when reloading.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/177https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96
Only push uuids of newly enabled extensions to the `_extensionOrder`
array if enabling them was successful.
Otherwise, since `_callExtensionDisable()` doesn't remove uuids that
weren't successfully enabled from the array, those extensions get added
to the array multiple times when they're disabled and enabled.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96
It's currently possible to circumvent the `sessionMode.allowExtensions`
property: For already enabled extensions one can call reloadExtension
via DBus, for new extensions it's possible by adding the extension to
the enabled-extensions gsettings key and setting the
disable-extension-version-validation key (which triggers a reload of
`this._enabledExtensions`) and then calling reloadExtension via DBus.
So to enforce `allowExtensions` while still allowing to update
extensions and keeping the extensionSystem synced with various gsettings
keys, replace the checks for `this._enabled` with simple checks for
`Main.sessionMode.allowExtensions` inside `_callExtensionInit()` and
`_callExtensionEnable()`.
The remaining checks for `this._enabled` are only small optimizations to
prevent running code on irrelevant sessionMode updates.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96
Right now we're only handling added sessionMode extensions correctly on
sessionMode updates, also handle the other case and disable removed
sessionMode extensions on sessionMode updates.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96
Instead of only logging a message that loading the extension stylesheet
failed and silently returning we should use `logExtensionError` for that
instead. This also sets the extension state to ERROR and makes sure we
don't try to enable it again.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96
If the extension doesn't exist in the `this._extensions` Map, we'd try
to access `extension.dir` on undefined/null. So set the `dir` variable
after checking if `extension` is defined.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96
Code and comment were based on the old get_input_rect() and get_outer_rect()
method names that were changed to the more appropriate get_buffer_rect() and
get_frame_rect() a long time ago.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/713
The animation was removed in commit 6a00a504d4 for consistency with
other menus. However commit a9b12d5d73 then *added* animations to
those just four minutes later.
So add back the original animations for consistency, both with menu
closing and with other menus.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1595
Remove transitions of the `slide-x` property of the layout manager
before we set the property to a fixed value, otherwise the transitions
might still be running and change the value after we set it.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/707
When selecting an area for screenshot we monitor the events while we've valid
coordinates in order to redraw the rubber band.
However, we don't stop ignore the motion events after button release and so
while animating. This might cause an unwanted effect if moving the mouse away
during fade out that is way more visible slowing-down the animations.
To fix this ignore any motion event once we've set the results.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/711
Add a new popover with a regular entry + button to rename
folders. The layout is similar to other GNOME applications.
The popup is implemented as a PopupMenu subclass, leaving
the grab management to PopupMenuManager.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/675
We disable and enable extensions inside the `notify::active` signal
handler, but we shouldn't do that in case the change didn't come from
the user but because something else changed the state of the extension.
This causes an issue when the extensionPrefs window is open and the
session gets locked: The extensions are temporarily disabled by the
shell, extensionPrefs updates its switches on the state change and adds
those extensions to the `disabled-extensions` gsettings key inside the
signal handler. Now when the session is unlocked again, the extensions
won't be enabled again since they're forced-disabled.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/705
We're using a vfunc override for `get_paint_volume` to exclude children
with an opacity of 0 from the paint volume and thus decrease the size of
the area we need to paint.
Now if the paint volume is requested during the spring animation (the
real icons are hidden using an opacity of 0 and clones are used for the
animation), `get_paint_volume` returns a paint volume with a height of
0. After that, the spring animation finishes and the icon-opacities are
set to 255 in `_resetAnimationActors`, and since we cache paint volumes
and there's no reason for Clutter to assume it got invalid, the icons
end up not being painted.
Fix this by queuing a relayout of the grid when the opacity of a child
is changed from or to 0, which manually invalidates the paint volume.
The reason why this is not an issue with the paginated icon grid
(all-apps view) is probably because StScrollView invalidates the paint
volume a lot more often than regular containers.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1502
Extensions might emit JS errors explicitly or implicitly, however GNOME
Shell doesn't present any stack trace for those making them quite hard
to debug.
Make this easier by logging errors with logError() whichs includes the
stack dump.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/700
When the GridSearchBase actor is destroyed we should remove the
ongoing later that might try to access to invalid resources.
To do this, add an _onDestroy() callback function to SearchResultsBase
and override it in GridSearchBase.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/700
Easing calls on show/hide functions have some parameters in common whether the
radial effect is enabled or not.
So instead of doing repeated calls with similar parameters, initialize common
values in params objects.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/700
If the resource scale or the scale factor changes while the animation
is playing, we need to stop the animation and start it again once the
texture is loaded, as the idle might try to access an invalidated
animation child otherwise.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/700
Commit 21e14bd46f fixed this for the
brightness slider, but we have the same problem for volume too. When the
volume is muted - for example in Settings or via a media key, we update
the slider to '0' to indicate this visually. But we also actually invoke
the slider's callback to *set* the volume to zero. That means that the
previous level is overwritten so it can't be restored when unmuting.
The fix is the same - when we update the slider internally ourselves,
don't call the signal handler.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1557
We now do 2 things along Xwayland startup/shutdown:
- Start or stop the gnome-session-x11-services target, that will
pull all X11 related services that the session might depend on.
- As we start ibus-daemon manually, trigger a restart in order to
toggle the XIM daemon on and off along with Xwayland presence.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/680
We may need to restart it with different arguments, so make it
possible to do that. Also, avoid to just restart it on _clear(),
this is now most likely through our --replace call than it is
through ibus-daemon eg. dying, avoids some noise in logs as
there is already an ongoing ibus-daemon.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/680
An Endless OS system was found in the wild with a malformed
.local/share/gnome-shell/notifications which causes _loadNotifications()
to raise an exception. This exception was not previously handled and
bubbles all the way out to gnome_shell_plugin_start(), whereupon the
shell exit(1)s. The user could no longer log into their computer.
Handle exceptions from _loadNotifications(), log them, and attempt to
continue. Ensure that this._isLoading is set to 'false' even on error,
so that future calls to _saveNotifications() can overwrite the (corrupt)
state file.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1552
The updatesPermission is currently initialized synchronously, which
blocks the Mainloop for quite some time and therefore slows down startup
of the shell, let's do it asynchronously instead.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/689
Since we now put a short timeout in before the start of the actual pie
timer we don't start the timer as often as we used to. This allows us to
create a new PieTimer object each time a timeout is started and
therefore play a finish animation independently of other (new) timeouts.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/688
If the circle is complete and the pie timeout finished, we don't need
the lines to the center point indicating the ends of the pie anymore.
We just draw a clean circle instead, which allows for a zoom-out and
fade animation of the circle when we're done.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/688
If the pie timeout has finished successfully there's no need to cancel
the pie animation, instead we can just wait for that animation to finish
and show some visual feedback like a zoom-out animation to indicate the
click afterwards.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/688
Fade the pie timer in using a duration of 1/4 of the timeout and a
EASE_IN_QUAD animation. This significantly reduces flickering of the pie
timer while moving the cursor and makes the timer less distracting.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/688
gjs no longer has an implicit dependency on GTK 3.0, so without
requesting an explicit version, we will get the highest available.
Our code isn't GTK-4 ready, so request 3.0 explicitly.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/686
Before the move to Clutters implicit animations with 0846238f6 and
bf497ed64, we used Tweener to do a fake-animation of the opacity and
Tweeners `onUpdate` signal to queue a repaint of the PieTimer everytime
Tweener tries to update the animation.
Now, with Clutters implicit animations, there is no `onUpdate` signal
anymore and also `notify::opacity` no longer gets emitted since the
value doesn't actually change. This lead to the PieTimer no longer being
repainted, which broke the animation.
Fix this by implementing the current angle of the pie using a custom
GObject property `angle` and animating this property using the new
`actor.ease_property` method.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1533
When cancelling the animations of the icon grid, right now we simply
destroy all the clones without resetting the opacity and making the
actor reactive again. So if the spring animation to show the grid is
cancelled by pressing a key to start a search, the icon clones would be
destroyed, but the icon-opacity would still be set to 0. Now if the
Escape key is pressed, viewSelector will show the last active page (ie.
the iconGrid) without a custom animation and only fade in the page, and
because the icons still have an opacity of 0, they will be invisible.
Fix this by always restoring the opacity and reactive property of the
original actors if the animation is cancelled instead of only destroying
the clones.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/678
Since we set the proxy value when the slider changes and set the slider
value on proxy property changes, we run into a cycle.
Before commit 3d3dca4aa this was addressed by not notifying on all slider
changes, but only in reaction to direct user action. Given that since the
splitting out of the BarLevel class those events are handled in a subclass,
that approach is at least unconvential and fairly fragile.
Instead, make the brightness indicator ignore any changes to the slider it
initiated itself.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1500
With 8b368d010 we fixed a bug where the onComplete callback was always
called no matter whether the transition was interrupted before or not.
This exposed another bug: viewSelector depends on this behaviour when
fading out pages: After fading out a page, we call `this._animateIn` to
show the new page. Now if the fade-out animation gets interrupted, with
the correct behaviour of onComplete we end up not showing a new page and
the viewSelector remains empty instead. One case where this happens is
when pressing a key to start a search during the overview-animation.
Obviously we also want to show the new page in case the fade-out
animation was interrupted, so use the onStopped callback instead of the
onComplete callback here.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/674
An actor ease callback could destroy the actor, in such case we should do not
touch the actor anymore.
So, before calling the callback, reset restore the easing state and don't
perform any further action with it.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1507
When a drag is cancelled and the source actor
is visible, the drag actor is animated back to
the source position. The scale that the drag
actor will become is calculated as:
scale = this._dragActor.width / sourceScaledWidth
However, this is wrong; what we wanted to do
is the opposite:
scale = sourceScaledWidth / this._dragActor.width
Fix the scale calculation to match the math
above.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/671
Create a new folder when dropping an icon over another
icon. Try and find a good folder name by looking into
the categories of the applications.
Delete the folder when removing the last icon.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/671
Because the Dash icons are not drop targets themselves,
add a tiny DashIcon class, which is an AppDisplay.AppIcon
subclass, and disable all DND drop code from it.
Show a folder preview when dragging an app icon over another
app icon.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/671
As per design direction, scale and fade the app icon
when starting dragging it, and show it again if the
drop is accepted. Clutter takes care of animating the
rest of icon positions through implicit animations.
Scale and fade the dragged icon while it's being dragged.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/671
Since the `notify::allocation` signal will obviously get emitted while
the actor is inside an allocation cycle and we might end up doing
changes to its allocation inside `updateSearch` by hiding or showing the
actor (which queues a relayout), we get a warning from Clutter.
Fix this by delaying the call to the parent method until the next
redraw, which should happen a few moments after the current relayout.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/672
Since commit 1a27ff6130 we use the
allocation width of the GridSearchResults actor to calculate the max
number of results that can be shown for this search provider.
On the first run of the search, when no previous (cached) allocation is
available for the actor of GridSearchResults, the allocation width used
in `_getMaxDisplayedResults` will be 0, which in turn will make
`updateSearch` filter out all results returned by the search provider.
Now if this search provider is the only search provider that's enabled,
after calling `updateSearch`, the `SearchResults` class will call
`_updateSearchProgress` to check if any results are visible, assume
nothing was found and therefore hide the scrollView. This in turn causes
the GridSearchResults actor to not get a new allocation, which prevents
our code to fixup the max number of results on `notify::allocation` from
working: The number will continue to be 0 and we'll never show any
results.
To fix this regression, return -1 in `_getMaxDisplayedResults` if the
allocation width is 0 to inform `updateSearch` that we can't calculate
the maximum number yet and interpret a return value of -1 as "show all
results" in `updateSearch`. The same problem would probably also appear
if the allocation width is anything between 0 and the width of the
iconGrid with one icon in it, although this might as well be a valid
width in case a very small screen is used or with very large icons. So
let's only check for a width of 0 and hope the GridSearchResults actor
won't get weird temporary allocations in that range.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/672
Just as we animate the apps launch using the zoom out animation if the
'new-window' action provided by the app is launched, we should also show
this animation if the 'activate-discrete-gpu' action provided by the app
via its AppInfo is launched.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/673
For the "New Window" entry we add to the AppIcons popup menu we should
always animate the app icon if the menu entry is activated as it was
intended by commit 62786c09a8.
For the "Launch using Dedicated Graphics Card" entry we can also always
show the animation if the entry is activated since the entry should only
be visible if the app is stopped.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/673
Clocks has exactly the same issue as Weather: Its integration currently
relies on accessing its settings directly, which isn't possible when
the app is sandboxed.
Fix this the same way we did for Weather, by adding our own setting
and syncing it with the app via a custom D-Bus interface.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1158
We add our own "New Window" menu entry if the app doesn't already
provide a 'new-window' action. For this menu entry, we show the zoom out
animation on the app icon when the user clicks the entry.
To be consistent in case the app already provides its own 'new-window'
action via its AppInfo, also show the zoom out animation when this
action is activated.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/662
As pointed out by designers, fading it signals that the
icon grid is not a drop target, when now it actually is.
Remove the fade effect applied to the icon grid when
dragging. Since this is the only caller of fadeIn() and
fadeHalf(), also remove these methods.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/664
App icons inside folders are already animated when the folder is
opened, but moving an app icon from a folder doesn't, making the
transition abrupt.
Fortunately, it's easy to detect icons that were previously hidden
but are not anymore.
Add an animation to these icons when showing.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/664
This is necessary for being able to drag application icons
to folders in different pages.
Add a drag motion handler to AllView and handle overshoots
when dragging. Only handle it when dragging from AllView.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/664
After dropping an application into the folder icon, the
list of applications is updated but the folder icon itself
is not.
Introduce BaseIcon.update() and call it from FolderIcon
when redisplaying.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/664
The event blocker is an actor that is added in between the
icon grid and the app folder popup in order to guarantee
that clicking outside the app folder will collapse it.
However, the next patch will require allowing dragging events
to be passed to folder icons, and the event blocker gets in
our way here, preventing drag n' drop to work properly.
Add an API to inhibit the event blocker. This API will be
used by the app folders while an item is dragged.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/664
When plugging in a device with sensors that are unsupported by
iio-sensor-proxy, the proxy may quit so fast that the name disappears
from the bus before we get to construct the SensorProxy in response
to the name-appeared handler, resulting in the following warning:
JS ERROR: TypeError: this._sensorProxy is null
_sensorProxyAppeared/this._sensorProxy<@resource:///org/gnome/shell/misc/systemActions.js:217:17
_makeProxyWrapper/</<@resource:///org/gnome/gjs/modules/overrides/Gio.js:243:21
Address this by creating the proxy unconditionally instead of monitoring
the bus name, and using the g-name-owner property to determine whether
iio-sensor-proxy is active.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1357
When transitioning to or from the overview, windows travel
a certain distance between their real desktop position and
their place in the overview window grid. The less this travel
distance is, the smoother, more polished, and less jarring
the overall transition looks. This is why it makes sense to
try reordering and repositioning windows to minimize their
travel distance. That being said, there are other factors
that impact the quality of the overview layout, such as how
much the windows get scaled and what portion of the overall
available space they take up.
The existing code tries to minimize the travel distance by
sorting the windows in each row by their horizontal position.
There are, however, two problems with this implementation.
First, it compares the coordinates of windows' left edges as
opposed to their centers, which means it yields unexpected
results when a small window is positioned next to the left
edge of a large window. Second, it completely disregards
vertical coordinates, instead assigning windows to the grid
rows using their monotonically increasing window numbers,
effectively vertically sorting them by the order they were
created in.
This commit changes both vertical and horizontal ordering
to work based on the coordinates of the geometric centers
of the windows. That is to say, windows are first assigned
to grid rows based on the vertical coordinates of their
centers, and subsequently sorted inside each row based on
the horizontal coordinates of said centers. In my testing,
this leads to a much more intuitive and visually pleasing
window placement.
Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/267
GDBusProxy::g-properties-changed is also emitted when the name drops from
the bus, at which point any properties will be null. That's not a valid
gsettings value, so to avoid the corresponding warning, move the g-name-owner
check accordingly.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1158
Those checks were carried over from the very first DND implementation;
if they were ever actually required at all, this is no longer the case
as we moved away from Tweener for all our animations.
The number of cases where an extension is still using Tweener, creates
draggable actors, *AND* requires the checks for proper functioning
should be indistinguishable from zero, so drop the code.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/669
