The app grid itself now is horizontal, and is displayed beneath
workspaces, above the dash. This makes the indicator animations
out of place, as they're not coming from the edge anymore.
Use PageIndicators for both FolderView and AppDisplay.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1629>
The new window preview overlay requires getting the app icon for a
window from the window tracker when it gets initialized. The window
tracker listens for the same 'window-added' signal on the MetaWorkspace
that the gnome-shell Workspace listens for to add the window preview.
The window tracker however reconnects all its signal handlers whenever
the number of workspaces changes, which means that its signal handlers
get called after the ones in Workspace ones. So by the time the
'window-added' handler in Workspace is called, the window tracker does
not have an app associated with the window.
To fix this ensure that all window related signal handlers in Workspace
are run after the ones in the window tracker.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3656
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1625>
Currently, there's one animation for the whole canvas. While it looks fine
with just one screen, it causes windows to move between screens when
switching workspaces. Instead, have a separate animation on each screen,
and sync their progress so that at any given time the progress "fraction"
is the same between all screens. Clip all animations to their screens so
that the windows don't leak to other screens.
If a window is placed between every screen, can end up in multiple
animations, in that case each part is still animated separately.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1213
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1326>
Currently, the workspace swipe transition only has one workspace in each
direction. This works until you try to do multiple swipes in quick
succession. The second swipe would continue the existing transition, which
only has 2 or 3 workspaces in it, and will hit a wall.
To prevent this, take all workspaces and arrange them into a column or row,
depending on the layout, and use that as a transition.
For the transition that happens when focusing a window on another workspace
(for example, via Alt+Tab), still use only two workspaces instead of all of
them.
Since we don't support layouts other than single rows/columns anymore,
diagonal transitions aren't supported anymore, and will be shown as
horizontal or vertical instead.
Since nw alt-tab and gesture transitions are different, don't allow to do
both at once, that is, disable swipe tracker when a programmatic transition
is going. This will also conveniently cancel a gesture transition if a
programmatic one is initiated while a gesture is in progress.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/2612
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1326>
In future we will need to use window clones to better support multiple
monitors. To avoid having to hide every window, show wallpapers behind
the workspace transition: one per monitor.
Put the wallpaper into a separate class right away, later it will be
useful to make the animation per-monitor.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1326>
Simplify the code a bit. The workspace group is relatively self-contained,
so split it from the general animation. Reimplement _syncStacking().
This will help a lot later, with workspace strip and multi-monitor support.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1326>
We always request a natural width based on the maximum thumbnail scale,
but may very well use a smaller scale when allocating. This currently
results in thumbnails being off center, fix this by distributing any
extra space evenly before allocating thumbnails.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1620>
Accessing GObject properties from JS has proven to be quite slow because
of the JS->C->JS roundtrip involved. With the WindowPreview this
actually has an impact since we're accessing those properties very often
while creating new layouts.
So cache the boundingBox and the windowCenter properties of the
WindowPreview using a this._cachedBoundingBox JS object. This might
speed up opening the overview with lots of open windows significantly.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1617>
Make computeScaleAndSpace() return an array including scale and space so
we no longer have to access the layout object from outside.
With this we also no longer need to set layout.space, since only the
scale property is needed in computeWindowSlots().
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1617>
Start cleaning up the whole mess around the layout object a bit and
return a new object in the LayoutStrategies computeLayout()
implementation. This object is supposed to be opaque to the API user and
will only be passed to the layout strategy.
For now, keep setting a few things on that object from outside, we'll
clean that up later.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1617>
Only keep computeLayout, computeWindowSlots, computeScaleAndSpace and
the contructor in the superclass, the rest is actually layout specific
and won't apply anymore when we introduce the new vertical layout
strategy.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1617>
Move the background to the Workspace class by introducing a new container
called WorkspaceBackground, which handles clipping the background to the
workarea.
Move the click action from WorkspaceDisplay into each workspace.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1599>
When the original animation was implemented, workspaces would only
ever be added at the end. We therefore got away with not having a
separate EXPANDING stage corresponding to the existing COLLAPSING
one when animating out.
Since support for creating in-between workspaces via DND was added,
this is no longer the case. And now that the thumbnails are centered,
the jump is quite noticeable.
Address this by adding new transitional states, so that we can
expand new thumbnails before scaling them in.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1609>
Make the slide property control the workspace scale, so that new workspaces
scale up, and destroyed workspaces scale down. The scale is done horizontally,
and only slightly vertically, as per design direction.
Rework the state tracking mechanism to remove the COLLAPSING state, since there's
no split between sliding out and collapsing anymore. Also remove the corresponding
'collapse-fraction' property from WorkspaceThumbnail.
Make ThumbnailsBox.vfunc_get_preferred_width() consider the slide-position property.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
There are situations where MetaWorkspaceManager and ThumbnailsBox disagree
on the number of workspaces, for example when animating them out. It's more
important to follow the visible number of workspaces while they're updated.
Make vfunc_get_preferred_width() and vfunc_get_preferred_height() use the
current number of workspace thumbnails to calculate their sizes, instead of
MetaWorkspaceManager's n-thumbnails property.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
ThumbnailsBox currently allocates each workspace thumbnail using their
porthole size, and scales them down using scale-x and scale-y. This is
slightly problematic since it doesn't allow for properly styling these
thumbnails through CSS.
Rework ThumbnailsBox to allocate workspace thumbnails at their actual
sizes, and scale down the '_contents' actor inside WorkspaceThumbnail.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
Allocate workspace thumbnails horizontally. This requires introducing code
to handle the RTL direction. Do a small rewrite of the DnD hover method to
be simultaneously simpler and easier to follow, and work correctly on RTL.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
This is now all centralized in the apps page, so move the workspaces
thumbnails to ViewSelector's apps page. This allows us to remove
all the slider controls too, since they're now unused.
The transition between showing the workspaces, and the app grid, is
based on the most recent mockups: scale and move it down, and fade it
out.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
Now that there's only a ACTIVITIES and a SEARCH page, the old method of handling
keyboard tabbing (extra parameters to ViewSelector._addPage()) limits what we can
do.
Manually set up the Ctrl+Alt+Tab support for each element.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
Add them both in a StBoxLayout subclass with a vertical layout. This
new ActivitiesContainer class already contains an adjustment controlling
the transition between workspaces and app grid states, and althought it
is internal to it, it'll be easy to integrate with gestures in the
future.
Notice that AppDisplay is added before WorkspacesDisplay. That's because
we want the paint order to paint WorkspacesDisplay on top of AppDisplay.
Switch the ViewsPage enum to call this page ACTIVITIES, and adjust the
only caller in OverviewControls to it. At last, rename '_appsPage' to
'_activitiesPage' to also reflect the name change.
The usefulness of organizing this code in pages is lost here, but this
is a transitional state, and pages will be removed in future changes.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1593>
Only the area used by favorite apps can be used as drop targets, it
is not possible to add new favorites between the running apps at the
end. While that behavior makes sense, it is currently impossible to
distinguish the two areas with confusing results.
Address this by adding a visual separator between favorites and
running apps.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1606>
WokspacesDisplay simply remaps the scroll direction into the next
workspace, but that doesn't account for the new horizontal layout.
Scroll horizontally on horizontal layouts when scroll direction is
on the vertical axis.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1603>
The scroll adjustment's upper value corresponds to the number of
workspaces, not to the last workspace index. We want the latter
when mirroring the layout in RTL locales, so subtract 1.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1598>
Now that apps either appear in the dash or the app grid, it makes
sense to allow DND between the two components to add and remove
favorites.
Currently this only works for adding items to the dash, update the
app grid code to also accept drops from the dash.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1594>
We use a custom actor to make sure that the show-apps button remains
visible even when there's not enough space to show all icons.
We can achieve the same result with much less code, by using a custom
BoxLayout layout manager for the icons to override the minimum width.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1592>
With the transparent top bar in the overview, constraining to the
primary monitor's workarea causes the (now visible) area beneath
the top bar to not have the darker background, which causes a visual
discontinuity in the layout.
Don't constrain to the primary monitor workarea.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1590>
The code previously was using CSS to define row/column spacing and
padding which was combined with a subicon size computed in code relative
to the requested icon size.
In smaller icon sizes it was possible for the CSS spacing+padding + the
size of the two subicons to exceed the requested icon size. This then
would lead to the label being pushed down for app folders compared to
other icons.
Another more severe issue caused by this would happen if the first item
in an icon grid was an app folder. Then the calculation for the maximum
allowed icon size could be off, leading to all icons in the grid
becoming smaller than actually necessary.
This commit changes this to use homogeneous row and column layouts to
evenly distribute the remaining spacing instead of using a fixed CSS
value.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3069
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1581>
This is some leftover from code that was used to keep track of volumes
added/removed while the screen was locked before the move to a
components system in 2a800e4c. All that the remaining code does is
filter devices from an empty list.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1579>
Now that the Dash is horizontal, the popup menu of the Dash icons must
show up, instead of left/right.
Make AppIcon.popupMenu() receive an optional parameter with the side
to show the menu, using St.Side.LEFT as default. Override this method
in DashIcon to always pass St.Side.TOP.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1559>
And move it to the bottom of the overview. Change the height-based calculation
of the icon sizes to be width-based. Put the DashFader in a vertical box, and
make all corners of the Dash equally rounded.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1559>
Between the GTK4 port and the latest GTK4 version, calling realize()
on a newly created window to force its surface to be created stopped
working.
So instead, wait for the window to get realized regularly to set its
parent.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1574>
With the previous preparations in place, it is time to take the plunge.
As both the app and the portal use the same small library for handling
external windows, port everything at once to avoid the hassle of building
and installing two versions of the library.
With the portal using GTK4 now, all extensions must port their preference
widgets as well.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1495>
We currently use separate frames for the details expander and the
expanded details. That layout works as long as frames are boxy (as
in the default GTK3 style), but breaks down with rounded corners
(as in the default GTK4 style).
In order to work with either style, adapt the layout to use a single
surrounding frame and appropriate borders as separator.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1495>
GTK4 will remove the GtkHeaderBar:title property, so stop using it
and set the window's title property instead, as that's what headbars
use in both GTK3 and GTK4 unless explicitly overridden.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1495>
There's little point in setting properties to their default value,
so stop doing that.
(GtkFrame:shadow-type actually defaults to "edged-in" rather than "in",
but all types other than "none" are treated the same nowadays)
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1495>
The SHADE_ANIMATION_TIME variable sets the duration of the animation of
the background shading that is done when showing the overview. As
explained in the code-comment, that value must be smaller than the
animation time of the overview.
Now since we're going to start animating the background color of the
panel when showing the overview and we're going to use the overviews
animation time for that, we want to make sure the shading of the
background image and the animation of the panel are kept "in sync",
otherwise the transitions would look bad.
So slightly increase the value of SHADE_ANIMATION_TIME to 240 (the
overviews animation time is 250) to make sure those happen in the same
timeframe.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1397>
The current way of indicating focus of elements in the panel does not
work very well with a fully-transparent panel, a line at the bottom of
the panel doesn't make too much sense if there is no real panel, but
only the text and icons.
To make the indicators look better in this case, switch to a pill-shaped
background color to indicate the focus of items in the panel.
For this to look good, there has to be a small black border above and
below the background, this also requires increasing the height of the
panel (from 1.86em to 2.2em) for visual purposes.
Also, since we now no longer need to color the lower bottom of the
panel, we can remove the custom drawing code for the border of the
panels corner, so do that.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1397>
Currently the hint reflects the `active` state, which effectively
corresponds to the screen blank. That's a bit surprising considering
the name, plus the `active` state is already exposed by the ScreenSaver
D-Bus interface for anyone interested.
It seems reasonable that the `LockedHint` property reflects the lock
state, so change the handling to do exactly that.
https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/351
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1561>
We want to either handle a scroll event ourselves, or delegate it
to the swipe tracker. What we never want is StScrollView's default
handler that doesn't have any knowledge of pages, so disable it.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1560>
Even if they're in the end of the list. So far we've managed to not be affected
by this bug because until GNOME 3.38, workspaces didn't have a background, and
there was no way to navigate to these leftover workspaces, but with the proposed
overview changes for GNOME 40 it'll be very much visible.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1560>
When handling all scroll directions, it is imperative to ignore emulated
events. Otherwise we may get the wrong scroll direction, e.g. when natural
scrolling is enabled.
Ignore pointer emulated events in WorkspaceDisplay._onScroll().
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1560>
All subclasses of BaseAppView now are horizontal, thus we
don't need to deal with the vertical case anymore.
Remove the corresponding parameter from the BaseAppView
constructor, and move the StBoxLayout that both AppDisplay
and FolderView have in common into the base class.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1554>
Recent commit [1] added a strong light border around user avatar
icons, in accordance with design mockups.
As a probably unintentional side-effect, the border was also added
around the symbolic fallback icon, which is displayed whenever the
user avatar is not available. This doesn't work well with the current
design, as the strong border makes the subtle fallback icon
background indistinguishable. Additionally, it doesn't match the
design mockups for the symbolic avatar icon [2].
Correct this by adding a style class for when avatar image is used,
and apply the border only for that case.
[1] 498710c2ec
[2] https://gitlab.gnome.org/Teams/Design/os-mockups/-/blob/master/lock-login/username-based-login.png
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1490>
When checking for a suitable icon size, Dash currently checks which
of the hardcoded icon sizes is smaller than the calculated available
size.
On some circumstances, however, when the calculated available size
is exactly equal to the hardcoded icon sizes, Dash selects a smaller
size. This cascades (the next icon size is exactly the smaller size,
etc) and ends up with always Dash selecting smallest size available,
even with plenty of available space.
Check if the calculated available size is smaller or equal to the
hardcoded icon sizes.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1533>
Updating scroll position may have significant side effects, e.g.
switching workspace; this should never happen during allocation, as
we're in the middle of painting a frame. So, put it in an idle callback
if we're doing it from an allocation to have the side effects happen the
right time.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1527>
Screen magnification is the compositor's business, not that of "random"
unprivileged tools. And for cases where a more specialised behavior is
wanted, an extension likely does a better job than a consumer of the
D-Bus API.
In addition to that, exporting the interface has been broken for an
unknown time, because the object that holds the implementation isn't
referenced and thus ends up being garbage collected, whoops.
And last but not least, this gets rid of the last public D-Bus name
that isn't clearly in the system namespace (org.gnome.Shell,
org.gnome.Mutter, org.gtk).
https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3452
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1523>
Simplify the opacity dance to simply setting it to the initial value
before animating (0 when animating in, and 255 when animating out),
and to the final value after the spring animation is done (vice-versa).
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1518>
To run the spring animation, IconGrid uses the transformed position and size
of the icons. This works okay when IconGridLayout doesn't need to update the
icon sizes, but looks bad when IconGridLayout selects a different icon size.
Wait for the icon sizes to be recalculated, and the icons beallocated, before
running the spring animation. If no icon size update is pending, run the spring
animation directly.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1518>
In the future, both AppDisplay and FolderView will be horizontal, and thus
using the orientation to determine whether to use animated indicators won't
be enough.
Use a different flag to control that, and make FolderView not use animated
page indicators.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1518>
SwipeTracker connects to signals of the stage, but doesn't disconnect on
destroy, leaving them hanging and potentially running callbacks for
destroyed objects.
This is visible when removing a folder by dragging all icons out, and
running the swipe gestures, which will produce a bunch of warnings.
Explicitly destroy, remove, and disconnect the swipe tracker.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1518>
Make it subclass ClutterActor, since we don't need any of StWidget's
features. Pass the source actor of the bind constraint in the
constructor, and remove the extra method to set the source actor.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1518>
ControlsLayout is a tiny layout manager whose only purpose
is emit "allocation-changed" after allocation. This signal
was listened to update the workspaces actual geometry.
However, since d66cd0d206, ControlsManager doesn't listen
to this signal anymore, rendering the class useless.
Remove ControlsLayout.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1518>
How and if notifications are shown is controlled by NotificationPolicy
objects. But ever since 098bd45, only notification daemon sources or
notifications associated with an app are hooked up to GSettings.
The hardcoded default policy for built-in notifications (including
those provided by extensions) arguably made sense back then, but
now that the main setting has been rebranded as "Do Not Disturb"
and is exposed prominently in the calendar drop-down, following
GSettings is a better default.
https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3291
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1511>
When selecting the best icon size for the available area, we
iterate from the biggest icon size to the smallest one, and
stop when finding a size that fits the available area.
However, the 'bestSize' variable is only updated when the
available area is positive. This is problematic in super bad
cases like when none of the icon sizes actually fit the availabe
area, which was hit with a previous iteration of this branch.
Make sure to update the best size while iterating, so that the
smallest size is selected even in such bad cases.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1510>
Use the Clutter.ActorAlign.FILL alignment by default, which
expands the grid until max-row|column-spacing is hit. This
was the behavior we originally wanted for the icon grid, and
it's finally being realized.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1510>
This is a new property to control the padding around each page,
as opposed to the padding around the entire container.
Following the original design of IconGridLayout [1], changing
the page-padding property doesn't trigger relayouts; the container
is responsible for queueing a relayout appropriately.
[1] 3555550d5e
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1510>
Plain classes are private to their file, so accessing them from
another module results in the following warning:
That property was defined with 'let' or 'const' inside the module.
This was previously supported, but is not correct according to the
ES6 standard.
Fix by assigning the class to a public variable instead.
(Eventually switching to ES6 modules with proper imports/exports will
fix this as well)
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1512>
The grid icons in the search results page doesn't expand to
acommodate the multiline label, resulting in the multiline
label to overflow behind the list search results. However,
after 548d3b62d, it was possible to trigger this behavior
with keyboard focus.
Don't update multiline labels for search results.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1509>
Adding children to the WorkspaceLayout without calling addPreview() is
not supported, so let's log an error in case that happened.
We also have to allocate that child an empty ClutterActorBox, otherwise
Clutter will complain loudly.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1481>
Clutter expects actors overriding the allocate vfunc to allocate all
mapped children of the actor, otherwise bad things happen.
So make sure we actually allocate all our visible children in our custom
allocation functions, and since we don't want to give them a real
allocation, just pass them an empty ClutterActorBox.
It would be nice if we had a way to hide children during the allocation
process where no relayout is queued like gtk allows with
gtk_widget_set_child_visible(), then we could avoid those weird empty
ClutterActorBoxes.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3098
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1481>
When logging out or terminating gnome-shell, mutter will unmanage all
open windows, triggering the window-close animation in gnome-shell and
very quickly after that emitting "kill-window-effects". That means we'll
call _destroyWindowDone() to cleanup our animation data, but at this
point the MetaWindow of the window is already gone, so we get an error
that get_meta_window() returns NULL.
Fix that by checking whether get_meta_window() returned NULL and if it
did, don't access the window.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/2018
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/1483>
