The use of the core idle monitor means that focus change events
are also delayed by keyboard interaction. Since the magnifier is
already in the business of pointer tracking, it's easy enough to
fire the pointer rest timeout from here, so focus changes are
accumulated and delayed.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/984
We may get several a11y events setting the caret on the same
coordinates it previously was. Make focus tracking ignore those,
as we're jumping to the same coordinates again during eg. mouse
operation.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/984
If the crosshair is clipped so it doesn't cover the pointer cursor,
the clip rectangle is skewed towards the bottom/right. This was
made so to accomodate the default pointer, but the unevenness stays
on other pointer cursors, and it makes the crosshair look odd on
short crosshair length.
Make all lines clip to an even distance from the center instead.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/984
It is somewhat unexpected that crosshair color and pointer cursor colors
remain the same across changes in color inversion settings, and may lead
to contrast issues. Apply the effect on the common container, so it
applies to these all.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/984
If the icon proper has opacity of zero then that's probably because a
clone of it is animating. So avoid animating the source actor too.
And if there's any other reason for the opacity being zero, still don't
animate it because we can't see it :)
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/2167
So as to guarantee the unmapped state sticks and doesn't get toggled
back to mapped before we return.
Being in a mapped state when `FolderIcon.vfunc_unmap()` returned was
causing an assertion failure in `clutter_actor_set_mapped` and crashed
the shell.
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/2170
The portal helper is rather sensitive because potentially-hostile Wi-Fi
networks can decide to launch it whenever they want (by blocking the
user's connection to the nmcheck domain) and load whatever web content
they want into it. So having this unsandboxed is really extraordinarily
risky. Previously it was a risk we had to accept, because WebKit did not
have a web process sandbox, but now it does. So let's bubblewrap all the
things!
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/983
Commit 147a743d8d moved the suspend and power-off actions into
the submenu that contains the log-out and switch-user actions,
but did not update the submenu visibility logic to account for
the additional actions.
As a result, the submenu is hidden when log-out and switch-user
are unavailable (like on the login screen), even if suspend and
power-off are enabled.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/2169
While gnome-shell will now check for extension updates, the checks
are performed infrequently. Opening the Extensions app implies that
the user's current focus is on extensions, so it is an appropriate
time to schedule another updates check.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
Now that we can download, apply and display extension updates, it is time
to actually check for updates. Schedule an update check right on startup,
then every 24 hours.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
Now that we have support for extension updates in the shell, we
need some place to display the updates to the user.
As we are establishing the Extensions app as the primary way for
managing extensions, it's a natural place for that functionality.
Show which extensions have updates available, and offer a log out
button (so gnome-shell can apply the updates when logging back in).
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
Until now, it didn't matter whether an extension was installed in the
user's home or system-wide. However with support for uninstallation,
there is now a significant different, as that action is only available
for user extensions.
Account for that by separating extensions by type, so that users don't
have to second-guess which extensions can be fully-managed and which
appear as part of the system.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
The newly added expander gives us a place where we can display
more details without cluttering the interface.
Take advantage of that by including the extension website, version
and author.
(Author is in the mockups, but will not actually be shown until
the extensions website is changed to include it in its metadata;
however best to have UI and string in place for the freezes)
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
The description can be useful information, but also increases the
visual complexity of the extensions list. Move it into a hidden
details area that can be expanded, which unclutters the interface
while keeping the information readily available.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
The current fixed two-line label requires a custom widget, which
make moving to a widget template harder.
As the description will soon move elsewhere anyway, just go back
to a single line with a standard label for now.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
Extensions can have a major impact on stability and performance. Now that
the tool will become the main way for users to manage their extensions, it
is an appropriate place to warn the user of that risk.
Add a small info popover to the headerbar to display that warning, together
with the previously removed hint of where to go for finding new extensions.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
Currently when the extensions list is scrolled, it is possible to
keynav out of view, as the scrolling doesn't follow the key focus.
Hook up the adjustment to fix that.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
Currently the main window is a plain Gtk.ApplicationWindow that is
built and managed from within the application.
As the application becomes more complex, it makes sense to decouple
the two and handle the window from a separate ExtensionsWindow class.
Not least this is a prerequisite of using a widget template for the
window.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
GNOME Software will remove its extension support, so we should stop
referencing it in addition to extensions.gnome.org.
In fact, the placeholder is not the best place to hint at where new
extensions can be found, as the user will never see it in case the
distribution includes pre-installed extensions.
So remove the hint altogether, we will add it back in a more prominent
place later.
With the whole placeholder now being much lighter, we can stop dimming
the remaining elements.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1968
If all password entries in dialogs are hidden, there is either an entry
that has visible characters or no entry at all. That means we don't have
to show the caps lock warning at all, so hide it.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/942
Since the headlines of the dialogs now use a much larger font, the
strings need to be shorter so they won't be ellipsized. So use a shorter
strings for those titles and also adjust the title-strings of the
notifications sent by the NetworkAgent to be consistent.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/942
Since the wiggle effect will be used by the redesigned prompt-dialogs
and we always want to use the same parameters, move those as defaults
for the wiggle function to the util.js file.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/942
Since the caps-lock warning adds a lot of spacing to dialogs and the
lock screen, hide it by default and only show it when necessary. To make
the transition smooth instead of just showing the label, animate it in
using the height and opacity.
Also add some bottom padding to the label so we can show or hide that
padding, too.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/952
The caps-lock warning is more related to entries than dialogs and is
also used in gdm, which is not realated to dialogs at all. Rename the
css class to caps-lock-warning-label and move it to the entry
stylesheet.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/952
The Animation class inherits from St.Bin and manages the scale factor
in the image loading, but the widget size doesn't change and doesn't
depend on the scale factor so when the scale factor is different
from 1 the widget size doesn't match the image size.
This patch resizes the Animation widget using the scale factor so the
widget will match the animation images sizes.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/1746
Now that we have a way to check for updates and download them, we
should actually apply them as well. Do this on startup before any
extensions are initialized, to make sure we don't run into any
conflicts with a previously loaded version.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/945
While it is possible that an extension has a newer version available
than the previously downloaded update, it's more likely that we end up
downloading the same archive again. That would be a bit silly despite
the usually small size, so we can either use the metadata from the
update, or exclude the extension from the check.
The latter is much easier, so let's go with that for now.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/945
Currently the method installs updates instead of merely checking for
them (or it would do, if it actually worked).
This is not just surprising considering the method name, the whole idea
of live updates is problematic and will not work properly more often
than not:
- imports are cached, so any local modules will stay at their
original version until a shell restart
- GTypes cannot be unregistered
So change the method to only download available updates, and set the
extensions' hasUpdate state accordingly.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/945
The current support for extension updates is half-baked at best.
We are about to change that, and implement offline updates similar
to gnome-software.
As a first step, add a hasUpdate property to the extension state
which will communicate available updates.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/945
Since a11f417cd0, both drag and scroll
gestures are added to Main.layoutManager.overviewGroup actor, while
previously drag gesture was added to Main.overview._backgroundGroup
instead. Since we cannot use 2 different actors for dragging and scrolling
anymore. just disable the swipe tracker while dragging a window.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/2151
Since the orientation lock menu entry is a proper menu entry instead of
a icon-only button now, we also show a description-text for that entry,
so update this text depending on whether orientation is locked or not to
better reflect what clicking the menu entry will do.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/964
Since we don't really know what the buttons we're adding to the dialogs
are about, we can't configure a button to "be clicked" when the escape
key is pressed. So add a separate escape key handler to fix that, return
-1 and abort the request.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/961
When updating the buttons of the mount dialogs, compare the new buttons
to the old ones and only create new buttons in case something changed.
This makes sure key focus isn't reset or lost unnecessarily while a
dialog is opened.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/961
Since there is a generic layout for dialogs like that now, use it. Also
remove the functionality of focussing a window when clicking a list
item, it's not discoverable at all and pretty unexpected.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/961
On Wayland, the display server is the Wayland compositor, i.e.
`gnome-shell` itself.
As a result, we cannot spawn `gnome-shell-perf-helper` before
`gnome-shell` is started, as `gnome-shell-perf-helper` needs to connect
to the display server.
So, instead of spawning `gnome-shell-perf-helper` from the perf tool,
start it from `gnome-shell` itself.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/941
When do-not-disturb is enabled, non-critical notifications will not
be shown as banners. It therefore makes sense to indicate that state
to the user, so they don't accidentally miss notifications.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/239
Currently the indicator pad requests a size of 0x0 if the corresponding
indicator is hidden. Right now this is enough to balance out the indicator,
but it won't be when we add spacing to the parent container.
Properly hide the pad with the indicator to avoid that issue.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/239
We've had the ability to temporarily disable notification banners
all the way back to 3.0, but we stopped exposing it in the UI with
the 3.16 notification redesign. With the message list being more
concise nowadays and the "Clear" button reduced to a single icon,
we now have space for a "Do Not Disturb" switch again.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/239
We try to make the OSK 1/3rd as big as the monitor. On landscape
layouts we usually get away with it as there's plenty of horizontal
space to enlarge the OSK while keeping the OSK aspect ratio.
In portrait layout, the horizontal space is a lot more scarce so
it means we'll still have plenty of space after making the OSK as
wide as it can possibly be, which will look as odd blank space in
the OSK panel.
In order to fix this, let the OSK panel height be less than 1/3rd
the monitor size if we are dealing with portrait layouts.
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/2132
It will happen before next map anyway, and most notably at times
actor sizes produce correct results.
Fixes oddities in emoji pager visibility after showing the emoji
panel, moving to another page, and hiding the OSK with the downward
arrow button. The next time the emoji panel would be shown, panels
had a chance to remain invisible.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/943
It is already scheduled to be set on first map. Doing so here triggers
size and theme node checks the actor tree is not ready for, as the
toplevel actor is not yet attached to the stage.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/943
This functionality always suffered from discoveribility
problems, and now that the folder dialog can rename the
app folders, there's just no reason to keep it.
Remove the rename popup.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/896
Now that the folder dialog handles the adaptToSize workaround,
there is no need to propagate the call throughout the class
hierarchy.
Remove the propagating calls to adaptToSize, and all the satellite
code.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/896
Now the the folder popup behaves like a dialog, it must be
above the app grid, and not be affected by the scroll view
translation.
Add the folder popup to the AllView itself, instead of the
internal Shell.Stack that is inside the scroll view.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/896
Make the AppFolderPopup behave much more like a dialog than a
popup itself. To do that, remove the BoxPointer and replace it
by a StBoxLayout. The dialog is is also bind-constrained to the
view selector.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/896
Since the overview search results now have a dark background layer, we
can brighten the background image a bit again. As a factor we use 0.5,
since that ensures the texts in the IconGrid are still readable and the
background image is visible, too.
Fixes https://gitlab.gnome.org/GNOME/gnome-shell/issues/2133
If a new SwitcherPopup is created and there are no windows or apps to
switch through found, instead of returning from _init(), still
initialize the SwitcherPopup and let the check in SwitcherPopup.show()
return false to terminate the popup.
In both cases, with or without the return statements,
WindowManager._startSwitcher() will call SwicherPopup.destroy(), which
will try to disconnect signal handlers, destroy actors etc. Now if the
constructor can't finish creating the popup, some of the functions
called from _onDestroy() will fail and throw errors.
One of those cases is when window-switcher is limited to the current
workspace, and a WindowCyclerPopup is initiated on an empty workspace.
Because this._highlight hasn't been created, _onDestroy() will fail when
trying to destroy the actor of this._highlight.
Also, the actor of this._switcherList will not get destroyed in case
show() returns because this._items is empty. For example, this will
happen when a new AppSwitcherPopup is initialized with at least 1
running app, but 0 windows on the active workspace.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/168
When pressing the menu key to show the popupMenu inside a ShellEntry,
the menu is currently aligned with the end of the entered text, this
causes a bug in case the text is overflowing the width of the entry: The
menu will be shown outside of the entry field, because it's aligned with
the (invisible) end of the text.
Fix that by simply aligning the popup menu with the cursor of the entry,
which is a behavior that makes sense when pressing the menu-key anyway.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/934
`MessageDialogContent.messageBox` is not really needed and was only
needed to show icons, which is now no longer supported. The styling can
also be done using other CSS classes and this makes it a bit more
straightforward to add actors to the MessageDialogContent.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/886
Since MessageDialogContent.messageBox is going to be removed in a
subsequent commit, move the parts where it's used out of messageBox and
into the contentLayout instead. This will introduce wrong spacings in
some dialogs, which we're going to fix when implementing the redesigned
individual dialogs.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/886
Since the last commit we set the gicon property of StIcon to NULL if an
empty string was given when setting the icon name, this means
`st_icon_get_icon_name()` will return NULL instead of an empty string.
When `getIndicatorIcon()` returns an empty string, the icon_name
property will now be set to NULL, so compare it to NULL here.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/888
Animating the icon spring using the `translation-x/y` properties instead
of the `x/y` properties avoids relayouts. There are still other non-icon
actors moving, but it's a big improvement.
Before: 595 relayouts per spring
After: 94 relayouts per spring
Reducing relayouts reduces reallocation, which reduces CPU-intensive
JavaScript execution.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/926
String.prototype.substr() doesn't support a negative length value
(to subtract from the full length), so we end up with a filename
of '' when hitting that code paths (a relative filename with '.png'
suffix).
Fix this by switching to String.prototype.replace() instead.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/2091
Replace existing panning and touchpad scrolling by SwipeTracker.
Since SwipeTracker only references one actor, redirect scroll events
from page indicators to the main scroll view.
Change programmatic scroll animation to use easeOutCubic interpolator
to match the gesture.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/826
Add a unified swipe tracker supporting dragging, four-finger swipe on both
touchscreen and touchpad, and touchpad scrolling.
The shared logic is largely same as the one in WebKit and libhandy.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/826
Commit 24e631ffe2 changed the shield animation to use translation
instead of position.
However once the shield is raised, only an animation will lower it
again, which means the shield is missing when it's supposed to be
shown without animation (for example after an idle blank).
Fix this by resetting the translation-y property in that case.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/927
Together with the forecast icon, the temperature label is the most
important information in the weather section. To emphasize it more,
reduce its space requirement by removing the temperature unit, then
make the text bold.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1143
Since commit 90a08ba0b6, we only open a network secret dialog immediately
in response to user action, and show a notification otherwise.
While for the actual request VPNs are handled separately from other connections,
this isn't true when we show the notification - we need to handle 'vpn' together
with the other types there, or we fall through to the default 'invalid type'
exception.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/2008
Now that both ThumbnailsBox and WorkspacesDisplay use single adjustments for
controlling indicator and scrolling, create the adjustment in OverviewControls
and pass it to both objects, effectively syncing indicator to scrolling.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/821
Instead of having a scroll adjustment in each WorkspacesView, and using the
one from primary screen in WorkspacesDisplay, have just one adjustment in
WorkspacesDisplay, and sync the changes between WorkspacesView.
This will allow to share the adjustment between WorkspacesDisplay and
ThumbnailsBox in the next commits.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/821
Instead of using the 'y', which queues a full relayout and
thus forces effects to be reapplied, use the 'translation_y'
property, that doesn't force relayouts and allows a future
blur effect to actually use the cached framebuffers a lot more.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/864
In an earlier iteration of the ease patch set, animatable properties
and easing parameters were different arguments.
This wasn't the case in the final version, so remove the left-overs.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/917
We don't want the icon to fill extra space, so set the alignment
accordingly. Otherwise we get an unexpected result when adding
a background just to the icon part (as far as I can tell: just
system-action-icon).
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/909
The screenshot code has a fair bit of nested callbacks, which means
that it is a good use case for async functions and Promises.
Most code uses GIO's async pattern, which means it can be easily turned
into promises with Gio._promisify(); first handle the couple of cases
that need custom code though, starting with SelectArea.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/903
Some GrabHelper uses are in the form:
doPreGrabStuff();
this._grabHelper.grab({
onUngrab: () => {
undoPreGrabStuff();
},
});
A promise-based variant allows to write this more cleanly as:
doPreGrabStuff();
await this._grabHelper.grabAsync();
undoPreGrabStuff();
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/903
We currently handle the case where the indicator itself is disabled
(read: hidden), but not when the entire top bar is invisible (for
instance when the primary monitor is in fullscreen state).
It is odd to pop up a top bar menu without the top bar, so check for
the indicator's mapped- instead of visible state.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/2002
Use the new StPasswordEntry for password entry fields
and remove all direct handling of clutter text of the entry
via clutter_text_set_password_char to show/hide the password
text. StPasswordEntry will provides a peek-password-icon which
will allow to show/hide the password present in the field to
the user in subsequent commits.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/619
shellEntry should not need to query the clutter-text directly
in order to know if the entry is for passport-input purpose.
shellEntry can easily determine a password-entry by querying
if it is an instance of StPasswordEntry and use it's API
whereever relevant.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/619
This is necessary to make DnD operations work from tablet devices on
wayland, as it's not the same onscreen pointer sprite than mice. Fixes
window DnD in the overview on tablet devices, no longer having them stick
to the wrong pointer.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/897
Since commit 87e60ed97843, geoclue no longer pretends that authorization
is useful for system-installed apps (as they can easily lie about their
ID). Unfortunately this broke our auto-location support in case Weather
is installed non-sandboxed, as we are waiting for an authorization that
will never happen.
Unbreak it by only requiring authorization when installed as Flatpak.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1823
Just like switcher popups, popup menus don't play well together with
system modals, and generally have a lower priority. So just like
switcher popups, close popup menus when a system modal dialog pops
up.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1536
As system modal dialogs may open without user interaction (for instance
polkit or network agent requests), it is possible for them to pop up
while the app/window switcher is up.
The current result of having both up simultaneously is clearly broken,
so we can either dismiss the popup or prevent the modal dialog from
opening. Assume that the dialog indicates a more important action and
should therefore take precedence, so go with the former.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1536
Since both the `_items` object and the `_allItems` array include the
same items, the difference between those variables seems unclear. The
real difference between them (except the different data type) is that
`_allItems` is ordered in the same order as the visible grid, so rename
`_allItems` to `_orderedItems` which makes that more obvious.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/799
We have three interactions with an extension's prefs module:
- we import the module
- we call its init() hook
- we call its buildPrefsWidget() hook
The first two are one-time actions where we expect most getCurrentExtension()
calls (local imports, initTranslations() etc.).
However it's still possible that the extension will use the utility function
in buildPrefsWidget() as well, either directly or via other functions like
getSettings(): Make sure getCurrentExtension() returns the correct extension
in that case, not the last one whose preferences were initialized.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/873
The icon grid currently sorts icons by their names. When creating new
folders, the folder may end up being in a different page, and that's
confusing since we don't actually move to where the new folder is.
Move the icon grid to the newly created folder.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/883
We want extended keys to have the same size as their parent key,
but this is currently broken and the keys end up with their
parent key's preferred size (which is smaller than its allocated
size).
This is due to the way ClutterActor's width/height properties work,
which only return the "real" (i.e. allocated) size when the allocation
is valid, and fall back to the preferred size otherwise.
As changing an StWidget's hover state involves adding or removing
the `:hover` pseudo class, this is currently always the case.
Creating the extended keys first means the keyButton's allocation
is probably valid, so do that.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/1976
Mutter and Clutter was changed to pass around the current target
framebuffer via the paint context instead of via the deprecated Cogl
framebuffer stack.
The framebuffer stack has also been removed from Cogl so change to use
the one in the paint context instead.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/827
While still leaving them unused, pass around ClutterPaintContext and
ClutterPickContext when painting and picking.
The reason for splitting this change up in two is to make it possible to
bisect easier in between the API change and the change to using the
framebuffer passed around with the temporary contexts.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/827
