We will start using both URLHighlighter and the _fixMarkup() helper
method the same way it's used in MessageTray. Usually we should
make fixMarkup() public and call the existing methods, but we are
planning for them to go away soon, so just keep two copies until
the original one is removed.
https://bugzilla.gnome.org/show_bug.cgi?id=744817
We will soon replace regular menu items in the calendar drop-down
with more complex elements.
However there will still be items that should close the drop down
when activated - rather than making the menu available throughout
the hierarchy (and eventually from outside as well when we add the
notifications list), have a public method on a global object just
like the ubiquitous Main.overview.hide().
https://bugzilla.gnome.org/show_bug.cgi?id=744817
Menu items are on their way out of the calendar drop down, so let's
start with the easy one. This one is removed without replacement,
but then the Date & Time panel should be a one-time stop for most
users anyway, so not having a direct shortcut should not be much of
a problem. It is also the last remaining Settings item outside the
system menu ...
https://bugzilla.gnome.org/show_bug.cgi?id=744817
The new design has the events list on the left and the calendar on the
right, so swap them around and remove the vertical separator between
them in favor of some additional whitespace as in the mockups.
https://bugzilla.gnome.org/show_bug.cgi?id=744817
This used to be the style-class for status icons (i.e. icon-only
top bar items). It got unused a while ago when the class used
by status icons stopped using it - except for the keyboard indicator,
which set the class manually to appear as status icon despite not
being a real icon.
Now that the button highlight is provided by the .panel-button class
on a parent, the obsolete class results in a double border on the
keyboard indicator when active - just drop it from there as well
to fix.
So we can style it differently than :hover.
We already have a active state for the menu items which includes
more than hover. For example, when the keyboard focus moves to a item
or we select programatically a item.
For this reason we need a style class named active for the meaning we
give to it in menu items, and a pseudo class active with the meaning
CSS has.
https://bugzilla.gnome.org/show_bug.cgi?id=744680
Don't assume struts are on the primary monitor while computing
the strut side. Instead, find the first monitor that overlaps the
strut and compute the strut side using it.
https://bugzilla.gnome.org/show_bug.cgi?id=744183
This capability was only really useful for media players, and with
music notifications no longer being special, we can simplify a bit
by removing support for it.
https://bugzilla.gnome.org/show_bug.cgi?id=744815
Music is no longer a special type of notification according
to the design. If we want to resurrect the functionality, we
can reimplement it with a dedicated API like MPRIS rather
than piggy-bagging on the notification system.
https://bugzilla.gnome.org/show_bug.cgi?id=744815
According to the design, hotplug notifications should no longer offer
an eject action and use regular notification buttons (but using icon
and text), the default action when clicking the notification itself
is to launch the file browser.
Also as the corresponding resident notification is gone, it no
longer makes sense to make the notification transient.
https://bugzilla.gnome.org/show_bug.cgi?id=744815
The new notification system will no longer give access to actions
from the notifications list, so a notification that is never
displayed as banner does not make sense here.
https://bugzilla.gnome.org/show_bug.cgi?id=744815
Tray icons make for a terrible UI on their own, but trying to
shoehorn them into the notification system has only made them
worse. At least for the time being this removal is temporary
and support for tray icons will be back, but no longer as part
of the notification system.
https://bugzilla.gnome.org/show_bug.cgi?id=744815
Users currently can't switch VTs while at the login screen.
This commit fixes that, by adding the relevant keybindings
to the login screen's allowed keybindings list.
https://bugzilla.gnome.org/show_bug.cgi?id=744800
drawing cursor is on by default, so the code:
if (options['draw-cursor'])
recorder.set_draw_cursor(options['draw-cursor']);
never lets you unset it.
Fix is to use 'draw-cursor' in options instead.
https://bugzilla.gnome.org/show_bug.cgi?id=744599
When opening a panel menu, we set it's max-height to the available
work-area height to keep menus with scrollable content from growing
outside the monitor. However a menu that extends all the way down
to the bottom edge does not look great either, so also take margins
into account here.
https://bugzilla.gnome.org/show_bug.cgi?id=744498
Ouch, classy - we are telling translators to use a '24h' time
format for the '12h' time format string.
Luckily, only a handful of translations actually followed the
comment (de,hu,id,is,kk,nb,nl), and most of the corresponding
locales do not support 12-hour format anyway (only is_IS, at
least on Fedora).
We assume that applications that export a 'new-window' action can open
a new window, so we add an appropriate entry to the context menu.
However this duplicates functionality if the application already
exposes the action via the desktop file - don't add our own entry
in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=744446
This DBus API is intended to be used by gnome-control-center's
displays panel to show monitor labels.
Each output (i.e. hardware monitor) identified by its
org.gnome.Mutter.DisplayConfig API ID has at most one label. On
mirrored setups, all the labels for outputs corresponding to the same
logical monitor (i.e. showing the same contents in the same mode) are
shown together.
At most, only one DBus client at a time is allowed to show labels.
https://bugzilla.gnome.org/show_bug.cgi?id=743744
Because there's nothing (in single-monitor setups) that could
take the drop in this case.
* js/ui/appDisplay.js:
AllView._loadApps(), FrequentView._loadApps(): Pass
an isDraggable parameter when creating the AppIcons,
depending on whether the favorite-apps key is locked.
AppIcon._init(): Check for isDraggable in the params and
do not create _draggable if it was specified, to prevent a
drag from starting.
AppIcon.popupMenu(): Check _draggable before trying to call
fakeRelease on it.
* js/ui/dash.js: Dash._createAppItem(): Check AppIcon._draggable
before trying to connect to its signals.
https://bugzilla.gnome.org/show_bug.cgi?id=741325
In a lockdown scenario, where the favorite-apps GSettings key is not
writable, hide the menu items for adding and removing favorites from the
dash menu. Additionally, reject drops to the dash for DND.
https://bugzilla.gnome.org/show_bug.cgi?id=741325
After the login banner is shown and hidden, the first user
in the user list becomes non-reactive. This is because the
banner is given an opacity of 0, but still allocated.
This commit fixes that by hiding the banner explicitly.
https://bugzilla.gnome.org/show_bug.cgi?id=743370
Normally when a user uses the login screen to log in, the
login screen gets killed and the user session takes over
the display.
This doesn't happen for wayland sessions, though. Instead,
the login screen gets reset, and the wayland session is started
on another VT.
The greeter proxy object needs to be recreated after this reset,
since it's associated with state no longer coupled to the login
screen after the reset.
This commit moves greeter proxy creation to happen at reset time.
https://bugzilla.gnome.org/show_bug.cgi?id=743371
The code described by the comment was moved away in commit eda27d51,
so it is not misleading at best. It wasn't too useful to begin with,
so kill it off rather than moving it to the correct place ...
When using dynamic workspaces, a new workspace will be appended
when moving a window down to the last (empty) workspace. It makes
sense to extend the behavior in the opposite direction, and prepend
a new workspace when moving a window up from the first workspace.
https://bugzilla.gnome.org/show_bug.cgi?id=665764
New workspaces are inserted by shifting all windows on workspaces
below the insertion position down. As a result, when the new
workspace is inserted before the active one, we end up with
the illusion of a workspace switch. Instead, activate the workspace
on which the windows from the active one ended up.
https://bugzilla.gnome.org/show_bug.cgi?id=665764
We are not supposed to mess around with OR windows, so don't try
to shift them to a different workspace. This fixes a warning with
newer versions of mutter.
https://bugzilla.gnome.org/show_bug.cgi?id=665764
We will soon allow to insert a new workspace by other means than
DND in between workspace thumbnails, so move the relevant code
to a new windowManager method.
https://bugzilla.gnome.org/show_bug.cgi?id=665764
Input method preedit text needs to be disabled on password entries
for security and usability reasons.
IBus 1.5.7 provides the signal set-content-type so that panel UIs can
handle these special purpose input entries:
https://github.com/ibus/ibus/commit/6ca5ddb302c9
Unfortunately IBus versions older than 1.5.10 have a bug which causes
spurious set-content-type emissions when switching input focus that
temporarily lose purpose and hints defeating its intended semantics
and confusing users. We thus don't use it in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=730628
We check for (metasNeeded.length == 0) at the beginning of the function,
which is only ever called when when a non-zero number of results is
received back from the provider. Effectively, this means that
(metas.length != metasNeeded.length) will also catch (metas.length == 0)
and print a nicer message to the log.
The updateSearch() function is called in SearchResults every time new
search hits are available from a search provider; SearchResults will
wait for updateSearch() to complete in a callaback, to update the
overall progress of the search operation.
updateSearch() will call _ensureResultActors(), which will in turn call
getResultMetas() on the search provider, which is an operation that can
fail arbitrarily or return inconsistent data, as it's entirely in the
hands of the search provider.
In case _ensureResultActors() returns a failure, updateSearch() is
currently failing to notify the passed-in callback, which might leave
SearchResults in an inconsistent state: make sure the asynchronous flow
always ends up with a notification to the updateSearch() callback.
So that we'll recreate it the next time we want to show it. Otherwise,
we'll try to call things on a half-destroyed ResizePopup and end up
causing errors instead of showing the user their resize popup.
This will allow g-s-d to handle actions differently based on the
current mode - namely, allow the power button when locked, but
make sure to never show any dialogs in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=711682
Adding new parameters to the signal currently will break keybindings
until gnome-settings-daemon is updated to the new API as well.
Put additional parameters into a dictionary instead to make future
extensions easier.
https://bugzilla.gnome.org/show_bug.cgi?id=711682
When animating workspace switches, windows on the old and new workspaces
are temporarily reparented. If windows are restacked, those windows will
thus be ignored by mutter until meta_switch_workspace_completed() resyncs
the stacking at the end of the animation.
As a result, activating a window on another workspace that is not on top
of the stack is very noticeably a two-step operation of switching workspace
and raising the window. There is a technical reason for that order[0], but
we can avoid the visible disruption by manually syncing the stack during
the switch operation.
[0] https://git.gnome.org/browse/mutter/tree/src/core/workspace.c#n590https://bugzilla.gnome.org/show_bug.cgi?id=741680
Just like keybindings and the message tray pointer barrier, gestures
don't always make sense - for instance, swiping up the screen shield
should not trigger the message tray just as the SelectArea action around
the left edge should not open the overview.
To avoid this, restrict gestures based on the current keybinding mode.
https://bugzilla.gnome.org/show_bug.cgi?id=740237
Frequently banner messages are longer than can reasonable
fit in a one column view, which leads to a smooshed layout.
This commit changes the layout to a two column view, with the
banner on the left and the prompt on the right, if the banner
message is long enough that it can't fit well above the prompt.
If there isn't enough space for two columns then we keep the
one column layout but add scrollbars.
https://bugzilla.gnome.org/show_bug.cgi?id=703972
The login screen supports showing a banner message which admins
can use to mention login rules or disclaimers.
This message only shows up currently if the user list is enabled.
Most people who want to show a banner message also want to disable
the user list.
This commit moves the banner message to display when the user is
prompted for login credentials instead of when showing the user
list. It also adds a scrollbar if the message is too long.
https://bugzilla.gnome.org/show_bug.cgi?id=703972
The login screen is pretty custom full screen container and the standard
layout managers aren't really a good fit for the kind of layout that's
happening. This will be even more problematic with upcoming changes
to login banners, so we need to switch techniques.
This commit moves login dialog over to using a custom allocate handler
that has specific domain knowledge of the parts of the login screen
and where they go.
https://bugzilla.gnome.org/show_bug.cgi?id=703972
In certain cases the timeout for starting the calendar helper can
be reached but the calendar helper still loads fine. If so, just
ignore the timeout and wait until we get a notification from
dbus of the successful start.
https://bugzilla.gnome.org/show_bug.cgi?id=735308
While the default Shell style is fairly decent with regard to
accessibility requirements, having the ability to tweak certain
aspects where the regular style works less well is still useful.
For this purpose, try to load a -high-contrast theme variant of
the default stylesheet when a high-contrast theme is requested
(as determined by the GTK+ theme name).
https://bugzilla.gnome.org/show_bug.cgi?id=740447
_hideDone checks _shown to determine if anything has shown the overview
while we hid it, and if so, shows the overview forward just in case.
In a local patch that called _hideDone immediately inside _hide for
testing, this broke. While we don't actually depend on this anywhere,
it doesn't hurt so that the next person to hack this up (perhaps me!)
doesn't get stuck debugging it for 20 minutes.
Since moving to a GFile based API in commit 642bf2b778,
setThemeStylesheet() no longer accepts %null to revert to
the default theme. We should have some way to revert to the
default and the least intrusive option is to return to the
old behavior, so do that.
Correctly computing the ISO week number is tricky and we already
have code in the platform to do it, so just refer its computation
to GDateTime rather than doing it ourselves.
https://bugzilla.gnome.org/show_bug.cgi?id=736722
Using a separate property to show when the application is busy rather
than cramming it into the state property makes the code clearer. In most
places we only care if an app is running or not, not whether it is
actually busy.
https://bugzilla.gnome.org/show_bug.cgi?id=736492
If the user list is disabled and the user clicks cancel quickly enough
after typing their username, they can get in a state where the
auth prompt gets stuck in the insensitive state.
This is because the login dialog code makes the prompt insensitive
while while pam is processing the provided username, but the prompt
only makes itself sensitive again when it is hidden.
This commit makes it sensitive right before asking for a username again.
https://bugzilla.gnome.org/show_bug.cgi?id=740141
Once verification has succeeded, the train's already
left the building and we shouldn't allow canceling.
This commit renders the cancel button non-reactive
and makes the cancel function be a noop after
verification succeeds.
https://bugzilla.gnome.org/show_bug.cgi?id=740141
Due to a typo we were always removing the first (index 0) connection
from the global list of connections instead of the correct one.
This resulted in some connections remaining in the shell's connection
list long after they were removed. In particular, this resulted in
multiple copies of a bluetooth connection appearing after suspend/resume
(when the device was readded and the cached connection list was
rescanned).
https://bugzilla.gnome.org/show_bug.cgi?id=740227
I was going to add another DBus property to signal when the shell was
done loading and was idle, and while implementing that I noticed we
aren't emitting PropertyChanged for, well, any property. Let's fix
OverviewActive.
It's unfortunate it's so tedious to correctly implement a DBus
property =/
https://bugzilla.gnome.org/show_bug.cgi?id=704163
Currently, shellDBus only uses the passed in monitor index if it's
strictly > 0. A zero-index monitor is a valid one though, so don't
restrict this to strictly positive indices.
https://bugzilla.gnome.org/show_bug.cgi?id=740074
Commit 1291bcd0c8 implemented it for dateMenu, but the function
is already used in screenShield as well. Just add it globally as
we do for other standard gettext "macros".
Normally users switch xkb input sources and ibus input sources.
But currently the first input source only is running. It's also good
to preload all ibus engines in the logging session so that users switch
input sources quickly without the launching time of input sources.
The following is the ibus change:
https://github.com/ibus/ibus/commit/cff35929a9https://bugzilla.gnome.org/show_bug.cgi?id=695428
WorkspacesDisplay relies on being hidden to disable workspace switches
by scrolling or panning. Usually viewSelector will hide the previous
page on page switch, but we currently miss the case when opening the
overview at the app picker, where the workspaces page is still shown
for the transition, but never hidden.
Fix this by calling hide() in addition to setting the opacity to 0 at
the end of the overview animation.
https://bugzilla.gnome.org/show_bug.cgi?id=737534
Calling g_dbus_proxy_new without any flag means that the caribou
daemon will be launched through D-Bus activation, when creating
a proxy. It smoked out some corner cases in caribou and at-spi2-core,
but generally it would be good to avoid creating unused process.
This patch delays the invocation until the "Run" method is called.
https://bugzilla.gnome.org/show_bug.cgi?id=739712
Leading zeros are common in the 24h format, and indeed used in the
wallclock in the top bar. Convention and consistency within the
same clock format trumps inconsistency between different time formats,
so reverting commit 316f825b2a.
https://bugzilla.gnome.org/show_bug.cgi?id=658675
Splits instantiation of the event source into a separate method,
allowing extensions to subclass the DateMenuButton and provide its
own calendar source.
https://bugzilla.gnome.org/show_bug.cgi?id=672500
There is currently no simple way to inject into AppIcon's state change,
so an extension that wants to do this has to destroy/remove/update all
icons in the Shell (i.e. in the Dash, AllView, FrequentView) on enable()
and disable() after updating AppIcon.prototype._onStateChange, or the
extension must require a restart of the Shell.
To solve this issue, we rename _onStateChanged to _updateRunningStyle,
and connect the notify::state signal with an anonymous function that
calls _updateRunningStyle.
This extra function call should allow extensions to just extend the
updateRunningStyle function in the prototype.
https://bugzilla.gnome.org/show_bug.cgi?id=739497
This patch inlines the function _ellipsizeEventTime into its only caller
_addEvent. This also removes the need for the global const
EventEllipses and is thus removed by this commit as well.
https://bugzilla.gnome.org/show_bug.cgi?id=727302
With commit dc6a60dde, the calendar displays the ending day and time of
a continuing multi-day event on its ending day. This results in the list
not appearing to be sorted. This patch sorts the list according to the
displayed day/time.
With the two appointments
Thursday 0800-1000 Foo, and Wednesday 0900-Friday 1200 Bar and today
being Monday, the rest of the week list currently displays:
F ...1200 Bar
T 0800 Foo
With this patch, the displaying order is switched because Friday comes
after Thursday.
https://bugzilla.gnome.org/show_bug.cgi?id=727302
Currently, multi-day events are shown as individual appointments on each
day. This patch ellipsizes multi-day events to indicate continuation on
the prior or following day (or other time-period.)
The time label spot is now replaced by a box layout that contains the
prefix ellipsis label, the time label and the postfix ellipsis label.
In order to keep the alignment, ellipses are merely invisible (zero
opacity) when hidden.
The ellipses are styled using the events-day-time-ellipses class which,
by default, take the color of the event text.
When RTL is used, the box contents are adjusted accordingly (clutter
does that for us).
An event spanning three days now displays "...All Day..." in the
calendar on the second day.
https://bugzilla.gnome.org/show_bug.cgi?id=727302
It is really annoying for the user to acknowledge multiple notifications
when they queue up. So, to prevent a notification flood that has to be
handled by the user one-by-one, a summarized-notification feature is
added which leaves a single summarized-notification for the user,
replacing multiple notifications if the number exceeds 1, which they may
or may not acknowledge. When this summarized-notification is acknowledged,
the message-tray is opened where they can view the notifications that were
summarized. This helps the user concentrate on his primary task
simultaneously informing them about the new notifications.
https://bugzilla.gnome.org/show_bug.cgi?id=702460
We commonly mark strftime format strings for translation to account
for date/time representations without an existing strftime shortcut
("Yesterday %H%p"). As those translations are looked up according to
the locale defined by LC_MESSAGES, while the conversion characters
themselves are resolved according to LC_TIME, the result can be
rather odd when mixing locales ("Den 27. January"). The correct
solution would be to install translations for format strings in
the LC_TIME catalogue and look them up with dcgettext(), but we
don't have the infrastructure to do that easily. Work around this
by adding a helper method that looks up a string in LC_MESSAGES
using the locale defined by LC_TIME and use that to translate
format strings, which has the same result.
https://bugzilla.gnome.org/show_bug.cgi?id=738640
Since the background rework, SystemBackground is no longer a transparent
actor that you have to stack on top of a solid background, it is an
opaque actor. Fix the color of the background actor, and remove places
where we were setting the background color underneath the system background
and expecting blending - in particular, we can always set no_clear_hint
on the stage.
https://bugzilla.gnome.org/show_bug.cgi?id=738652
When monitors change, the previous index might not mean the same
physical monitor anymore, in fact, it might become invalid. In the
latter case, we'll actually get a JS error when accessing
this.keyboardMonitor in _updateKeyboardBox() . To avoid this, let's
just always reset the OSK to the primary monitor.
https://bugzilla.gnome.org/show_bug.cgi?id=738536
Incorrect braces meant that if the ShellUserVerifier was destroyed before
the call to fprintManager.GetDefaultDeviceRemote(), the reply would result in
an error.
https://bugzilla.gnome.org/show_bug.cgi?id=738256
Most of the code handles the sources setting being empty and
InputSourceManager.currentSource being null because previously the
"model" (i.e. the sources list) was kept in gnome-settings-daemon.
But this is fragile and since we're now the canonical place where the
list lives we can force it to never be empty even if the gsetting is
empty or contains only invalid entries. Adding the default keyboard
layout in that case is the safest thing to do.
https://bugzilla.gnome.org/show_bug.cgi?id=738303
Commit a4475465f1 fixed the wrong alignment for the fully visible
control, but regressed the partially slid-out one; take the slideX
factor into account to get the right offset for both cases.
Controls that slide left are located on the left, so the offset to
align them with the corresponding edge is always 0. It's controls
on the right that need a different offset when the available width
exceeds the child's width.
https://bugzilla.gnome.org/show_bug.cgi?id=728899
Currently we use the same amount of total delay divided by all items,
but that makes the items animate slow if the amount of items is small.
To avoid that, use a smaller total amount delay for an amount of items
smaller than four.
https://bugzilla.gnome.org/show_bug.cgi?id=737017
This commit ensures the login screen gets focused after
the screen shield is raised.
The code affects the unlock screen as well, but it's
less important since the unlock screen gets destroyed
and recreated each time the curtain moves, so it
has an opportunity to take focus on its own.
https://bugzilla.gnome.org/show_bug.cgi?id=708105
We currently allow infinite number of screenshot requests to be active at
the same time, which can "dos" the system and cause OOM.
So fail subsequent requests for the same sender when a screenshot operation
is already running.
https://bugzilla.gnome.org/show_bug.cgi?id=737456
We currently just clear out the map of recorders (which results
in the top bar indicator to be hidden), but don't stop the actual
recordings.
Spotted by Adel Gadllah on IRC.
Instead of waking up the JS every second to set the clock and update a
date label the user will rarely see, simply use property binding to
bypass JS string handling, and update the date in the menu when the menu
is opened.
Checks for a duplicate search before setting the current search
to true and before cancelling the current search. This ensures that
if a duplicate search occurs while the previous search is still active,
the duplicate search will not incorrectly cancel or change the state
of the previous search.
Indicate to NetworkManager that the Shell's agent supports VPN
hints, and pass those hints to VPN auth dialogs that also indicate
that they support hints.
VPN plugins can request new secrets, for example if the previous
ones are incorrect (eg, user mis-typed the password) or some other
reason (next token code required to re-sync a hardware token).
The specific secret that the VPN wants, and a VPN-specific message,
are passed in hints from the plugin, to NetworkManager, to the
agent (GNOME Shell) and then to the auth dialog.
https://bugzilla.gnome.org/show_bug.cgi?id=737592
If we have a modal, the stage's input region doesn't really matter --
all events go to us anyway. To avoid doing extra work doing animations
when we have a modal, like menus, the overview, and the message tray,
just fizzle out all updates.
To make sure we catch updates, update the input region whenever we end a
modal.
https://bugzilla.gnome.org/show_bug.cgi?id=737001
When the overview is created, search is populated with one actor for
each provider. As they're not hidden though, they will contribute to the
overall size request of the search page, which will shift upwards the
overview grid.
Reviewed-By: Jasper St. Pierre <jstpierre@mecheye.net>
Since (mutter) commit 49400657caae27 we disable the "Always on Top"
action for maximized windows, however when commit e7af257814
moved the window menu into the shell, this behavior was accidentally
extended to partially maximized (including tiled) windows.
As it can be desirable in this case to keep the window visible while
interacting with a different one, restore the previous behavior.
https://bugzilla.gnome.org/show_bug.cgi?id=737382
MetaBackgroundCache doesn't itself use file watches, so when a image
that we are monitoring changes we need to purge it from the cache,
so that when we load it again we get the new image.
https://bugzilla.gnome.org/show_bug.cgi?id=710756
When the keyboard is destroyed, we destroy the keyboard actor, but the
keyboard's menu isn't part of the key itself, so it's never tracked.
The menus are actually tracked actors, so they slow down the layout
manager's code to rebuild regions and other things. Keeping this list
small is a good idea.
To prevent leaking menus, destroy the menu when the key is destroyed.
https://bugzilla.gnome.org/show_bug.cgi?id=736999
The spring animation has to use clones to escape any clip set on the
parent, as it mainly occurs outside the parent. The same does not apply
to the pulse animation, which is very much in place - in fact, if the
parent is clipped (for instance a scrolled app folder), not having the
clip applied to the animated icons is indeed wrong.
Just animate the original grid actors instead, which gives the expected
result.
https://bugzilla.gnome.org/show_bug.cgi?id=736885
Trying to open an empty folder currently leaves the parent view in a
rather confused state. While we should look into fixing this in the
future, empty folders are not useful at all to begin with, so hide them.
https://bugzilla.gnome.org/show_bug.cgi?id=736910
The summary container will trap the focus if any sources are present,
making the message tray menu unreachable by keynav. Apply the same
hack as in searchDisplay and add a focusTrap to move the focus manually
as necessary.
https://bugzilla.gnome.org/show_bug.cgi?id=707799
