If a source is associated with an app, ignore the app name provided
by libnotify, as that is often garbage.
This fixes the "XChat-GNOME OSD" problem.
https://bugzilla.gnome.org/show_bug.cgi?id=681143
Reduce padding around persistent sources, and ensure that spacing
around resident notifications is only applied once.
Also, add some padding to the clock.
https://bugzilla.gnome.org/show_bug.cgi?id=681143
As description of the setting says, color-saturation ranges from
0.0 (grayscale) to 1.0 (full color), but the real outcome was the
opposite. The reason is that clutter provides a desaturation effect,
and color-saturation was passed directly to that effect. This patch
renames the effect and compute the desaturation value.
GetUserVerifier can only be called from the greeter session,
and fails with AccessDenied in all other cases. Also, calling it
hides the real error from OpenReauthenticationChannel, which
instead should be logged.
https://bugzilla.gnome.org/show_bug.cgi?id=680750
gnome-session and gnome-settings-daemon rely on the screensaver
interface to know the locked state. Since gnome-screensaver is no
longer running, it's up to gnome-shell to provide it.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
When the screen shield is activated from the user menu, animate
it instead of showing it abruptly. Also, ensure that the animation
had time to finish before calling UPower to suspend, to avoid
showing it when resuming.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Track screen lock status in the message tray, and filter banner
notifications. The message tray is completely hidden when the screen is
locked, but exceptions can be made for individual transient notifications,
such as shell messages and the on screen keyboard.
Non transient sources are shown in the middle of the lock screen. Resident
notifications (such as those from Rhythmbox) are shown in full, while
persistent ones are displayed as icon and message count.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Track locked status and use it to provide a reduced version of
the panel in the locked screen. Accessibility, input sources and
volume menus are preserved, without the link to the control center.
Network, battery and user menu are reduced to pure indicators,
with no menu.
This is similar to the design but not exactly, because designers
in IRC said that network needs more analysis before exposing, and
because the design didn't account for a11y and IM (so the one menu
metaphor is not really appropriate).
https://bugzilla.gnome.org/show_bug.cgi?id=619955
The design calls for the curtain to appear in the gdm greeter too.
Implement this by having the screenshield manage the login dialog
(delegating its creation to SessionMode).
https://bugzilla.gnome.org/show_bug.cgi?id=619955
This separates the screen shield into two main screens. One is
the lock screen, and it is shown when coming back from idle status
and when failing authentication. The other is the actual unlock
dialog.
Moving from the first to the second is possible by pressing Escape
or by dragging an arrow on the bottom on the screen.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
This ensures that the screen shield is created at the right
stacking level, so the message tray is visible in the lock screen
(showing PAM messages, critical notifications and the on screen
keyboard)
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Ensure that the lightbox is above everything (including the screenlock
itself) when fading in - this allows for fading while showing the
unlock dialog. Also, don't pushModal again when already locked, or
we won't get out of it.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
When the screenshield is deactivated, instead of going back to the
session immediately, prompt the user for authentication.
This essentially reinstates what used to be provided by gnome-screensaver.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
We are replacing the gnome-screensaver module with with a screen shield
that is part of gnome-shell.
This patch fades out the screen on idle and shows a shield with a background
image when there is activity again. The shield can be removed with a key or
button press.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Add 'default' parameter to setButtons, that controls the binding
of Return (unless overridden) and applies the 'default' pseudo-class.
Currently it has no effect, but it will start having after the
login dialog redesign.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
To allow more than one summary icon actor for a source we split
the model of the source icon (which is iconName, if the default
implementation is used, or a GIcon otherwise) and replace
createNotificationIcon() with a generic createIcon(size). Also,
the actual source actor is split into a separate class, that handles
the notification counter automatically.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
The :reactive property is used on StButton to like the :sensitive
property on GtkWidgets, that is, to indicate that the user is not
(yet) expected to click the button, and therefore should affect
styling too.
This allows to remove some code at the JS layer.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Previous code would access the array element before checking that
the index was within bounds, and therefore cause a TypeError.
It wasn't noticed earlier because at least one visible children
is in each panel box in all session modes.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Destroyed modal actors should be completely removed from the modal
stack automatically, including leaving modality if needed.
This allows for destroying modal dialogs without calling close().
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Change visibleInFullscreen to be trackFullscreen. If true, visibility
is fully bound to fullscreen status, if false, no change is made.
This allows to avoid set_skip_paint(), while not messing with
visibility of actors that are sometimes hidden for other reasons.
The flag was reversed because only the panel uses it, so false is
a more useful default.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
Make it possible to control the visibility of "Show/hide text" item
at runtime, to reuse the same entry for both password and non-password
prompts.
https://bugzilla.gnome.org/show_bug.cgi?id=619955
When GDM was moved over to GDBus it dropped the libgdmgreeter
library and introduced a new libgdm library with a somewhat
different API.
The main differences in the API are:
1) open_connection is now implicit and automatic
2) conversations don't need to be started explicitly, they're
started just-in-time when verification is requested
3) The functions are split up between the client, and new
helper objects that correspond to the dbus interfaces
they were generated from (one for user verification,
one for greeter specific operations, and a couple more
that aren't used by gnome-shell).
4) libgdm supports reauthenticating in an already running
session, so user switching should now affect the users
session more like screen unlocking does.
This commit moves the shell over to the new library.
Based on work by Giovanni Campagna <gcampagna@src.gnome.org>
https://bugzilla.gnome.org/show_bug.cgi?id=676401
As part of wanting to reuse some of the looking glass components,
don't use Main.lookingGlass, but instead pass the parent around.
Don't adjust the evaluator just yet, though. We'll split it into
a separate class soon.
https://bugzilla.gnome.org/show_bug.cgi?id=679944
This makes sense, since we are covering input methods and keyboard
layouts in the same keyboard menu. The effect of this change is that
the ibus status icon no longer appears in the system status area.
Changing the number of workspaces while the popup was visible (which
happens when moving windows on the last non empty workspace) resulted
in a wrong layout. Fix that, by listening to workspace-added and
workspace-removed signals, and by always requesting an updated size
from the actor.
https://bugzilla.gnome.org/show_bug.cgi?id=679005
We connect to the IBus daemon asyncronously and use it to query info
about input sources of the type 'ibus'. In case the daemon is or
becomes unreachable we just skip showing input sources of this type.
https://bugzilla.gnome.org/show_bug.cgi?id=641531
The calendar grid is build by giving each element right and bottom
borders, all top-most elements a top border, and all left-most
elements a left border. However in RTL locales, we currently add
the left border to the *right-most* elements, resulting in the grid
appearing clipped on the left side.
https://bugzilla.gnome.org/show_bug.cgi?id=679879
While modal dialogs were attached to the parent's titlebar, it
made sense to leave the top of the parent window at full color.
With the new position of modal dialogs, it makes more sense to dim
the entire parent window, so we can use a combination of Clutter's
BrightnessContrast- and DesaturateEffect instead of our own custom
shader.
https://bugzilla.gnome.org/show_bug.cgi?id=674499
With modal dialogs no longer being attached to their parents'
titlebar, the current animation no longer works too well. Use
a simple fade animation instead.
https://bugzilla.gnome.org/show_bug.cgi?id=674499
Show a notification when we receive a show-unmount-progress signal on
the mount operation we use for unmounting.
The notification will either turn fade out automatically with a
completion message when the unmount successfully completes, or will
disappear in case the operation is aborted underway (for example because
the device has been unplugged in the meantime).
https://bugzilla.gnome.org/show_bug.cgi?id=676125
When activating the app menu while displaying a startup notification
animation, the application shown in the menu does not match the
application providing the menu. To avoid this case, make the menu
button unreactive while playing the animation.
https://bugzilla.gnome.org/show_bug.cgi?id=672322
This is a bare-bones copy/replace. It does not implement ChangeLog
support. If we cannot get System Updates integration, I will implement
notification support.
https://bugzilla.gnome.org/show_bug.cgi?id=679099
When PackageKit signals that it prepared an update, offer an option
to reboot and apply it, using a helper that will setup the next
reboot and then calling to gnome-session.
https://bugzilla.gnome.org/show_bug.cgi?id=677394
This commit adds a grayscale effect to the magnifier, similar to
the lightness, brightness and contrast effects that are already there.
The effect is configured with the
org.gnome.desktop.a11y.magnifier.color-saturation setting, which
can take values from 0.0 (grayscale) to 1.0 (full color).
Based on a patch by Matthias Clasen <mclasen@redhat.com>
https://bugzilla.gnome.org/show_bug.cgi?id=676782
When connecting to virtual machines with usb-device redirection, such as Spice
enabled vms, automount may get in the way. Specifically if auto-usbredir is
enabled in the vm-viewer, then the usbredir code and the automount code race
for who gets to the device first.
If the automount code wins the race this is a problem, since usbredir causes a
device-disconnect (iow the usb mass storage driver sees an unplug), so in the
end usbredir always wins, and we end up with a non clean potentially corrupt
filesystem. Also see:
https://bugzilla.redhat.com/show_bug.cgi?id=812972
There for the need exists to be able to inhibit gnome-shell's automounting,
since all other inhibits run through gnome-session, I've chosen to do the same
for the automount-inhibiting. I've also submitted a patch to gnome-session to
reserve flag value 16 for this, see bug 678595.
This patch adds support to gnome-shell to honor this new inhibit flag.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
https://bugzilla.gnome.org/show_bug.cgi?id=678597
The log messages about presence changes unnecessarily cluttered the
notification.
Instead, we now present the presence states (online, offline, away, busy)
with an icon placed right next to the avatar. We also no longer show
notifications on presence changes.
https://bugzilla.gnome.org/show_bug.cgi?id=669508
When selecting "Open Calendar" in the date menu, the configured
application is launched via command line, so we don't get any
startup notification. If Evolution is used as calendar application,
launch it via the .desktop file added by the last commit instead in
order to fix the issue.
https://bugzilla.gnome.org/show_bug.cgi?id=677907
By disconnecting the 'notify::connection-status' signal as soon as the account
is disabled, we were missing the signal telling us when the status was moving
from CONNECTING/CONNECTED to DISCONNECTED and so the status icon was never
updated.
What we really want is to disconnect the signal when the account is removed
from the account manager as we don't care about it any more.
https://bugzilla.gnome.org/show_bug.cgi?id=669112
Use our native JS error system in the "extension system" API, only
using the signal/log-based error reporting at the last mile. Additionally,
delete the directory if loading the extension failed, and report the error
back over DBus.
https://bugzilla.gnome.org/show_bug.cgi?id=679099
Instead of using the 'extension-state-changed' signal to relay errors,
use DBus's native error mechanism to inform the method caller that the
call has failed. This requires making the method actually asynchronous
so that we don't block the browser, which is stuck waiting for a reply
from the browser plugin. To ensure this, we need to modify the browser
plugin API to ensure its extesion installation method is asynchronous.
Additionally, this lets us remove the awful, broken hacks that we used
when a user clicked the "Cancel" button, replacing it by a DBus return
value.
https://bugzilla.gnome.org/show_bug.cgi?id=679099
When the extension downloader was originally designed, the information
downloading part was inserted at the last minute, along with the modal
dialog as a security feature to make sure an extension didn't silently
get installed on the user's machines either due to a security issue in
the browser-plugin, or an XSS issue on the extensions website. Correct
the mistake I made when writing the code; instead of dropping an error
on the floor, log it correctly. This "bug" has already bitten a number
of users who forgot to configure proxy settings in the control center.
https://bugzilla.gnome.org/show_bug.cgi?id=679099
Pam seems to give us different strings, sometimes 'Password:',
sometimes 'Password: '. Look for both of these when replacing
them with a translated prompt.
https://bugzilla.gnome.org/show_bug.cgi?id=675300
The current position below "System Settings" is problematic - the
items are unrelated, and misclicks will result in a scary system
modal dialog that has to be cancelled.
Move items around a bit to avoid this problem.
https://bugzilla.gnome.org/show_bug.cgi?id=678887
Right now the "move to workspace" keyboard shortcut transition isn't smooth. It
shows the window disappearing, hopping to the next workspace, and then sliding
into view. "Pin" the window to the stage while the animation is in progress,
then release it afterwards.
https://bugzilla.gnome.org/show_bug.cgi?id=660839
Install a custom handler for move-to-workspace-* keybindings that
shows the workspace switcher, which gives the user a sense of
direction when navigating with the keyboard.
https://bugzilla.gnome.org/show_bug.cgi?id=674104
Most of code implementing workspace switches was repeated with
minor differences on each direction. Instead, consolidate it
and use the new meta_workspace_get_neighbor.
https://bugzilla.gnome.org/show_bug.cgi?id=674104
When using dbus-glib, single return values were special-cased to
be returned verbatim rather than as array with a single element.
This is no longer true since switching to GDBus, so fix the places
where the change was overlooked.
https://bugzilla.gnome.org/show_bug.cgi?id=678852
We strongly expect applications to use the same values for the
'Title'/'Icon' fields in their search provider .ini file as the
'Name'/'Icon' fields in their .desktop file. Rather than requiring
applications to duplicate those fields, allow them to specify a
'DesktopId' field instead to point to the corresponding .desktop
file, which makes it possible to ship search provider files without
translatable strings (which is nice given that merging translations
into search provider files lacks a standard rule).
https://bugzilla.gnome.org/show_bug.cgi?id=678816
Currently we pass an icon name for the 'icon' parameter of the
RemoteSearchProvider constructor. In hindsight, using a GIcon
instead will give us a bit more flexibility, so change it.
https://bugzilla.gnome.org/show_bug.cgi?id=678816
Use the ShellMountOperation dialogs we have to implement a DBus API
allowing other processes to display them.
Since GtkMountOperation now tries to call into our DBus implementation,
every application that uses a GtkMountOperation will gain integration
with our shell dialogs (but will still handle the actual communication
with GVfs).
https://bugzilla.gnome.org/show_bug.cgi?id=678516
Wait until the completion of the mount operation before dismissing the
passphrase dialog, so in case it fails, we can re-use the same dialog
with an error message (like e.g. PolicyKit auth dialogs) instead of
showing a brand new one.
https://bugzilla.gnome.org/show_bug.cgi?id=674962
Wifi and mobile broadband have signal indicators and are thus
more useful than vpn icons in the panel. Therefore, in the case
we have both wifi/3g and VPN we prefer the former as the "primary
icon" and add a lock next to it.
Behavior when VPN is added to wired or other connections is still
preserved: the wired icon is replaced by vpn.
https://bugzilla.gnome.org/show_bug.cgi?id=672591
Sorting by strength is what the other OSes do by default, and it
provides a better UX (by offering your hotspot and router before
the one from your neighbor).
https://bugzilla.gnome.org/show_bug.cgi?id=658946
Only apply the allowAutorun flag for transient notifications, not for
mounts that end up in the resident notification well.
Also, stop looking at volume.can_automount() here, since we already
checked that previously in the mounter, and allowAutorun is enough.
https://bugzilla.gnome.org/show_bug.cgi?id=660595
Previously, a volume was being ignored from autorun if one of these two
conditions were met:
- its mount root file had a native scheme and was mounted in a
non-hidden location
- it had a volume that could have been automounted, and had a flag set
by the shell to allow autorun
In order to effectively ignore volumes that we don't mount ourselves
from our notification system, we have to meet both conditions at the
same time instead.
https://bugzilla.gnome.org/show_bug.cgi?id=660595
This is a fallout from some changes in MessageTray.Source, which now
requires either defining the iconName/iconType properties on it, or
implementing createNotificationIcon, and we're not doing any of those.
Fix it by storing the gicon of the source object and using a helper
method to create the icon actor on demand, to avoid any case when the
same actor might be added twice to different containers.
https://bugzilla.gnome.org/show_bug.cgi?id=678428
The dependency chain spirals out from folks->zeitgeist->xapian...and
I'm really not interested in pulling in all of that into the core
shell.
There is work on splitting out contact search into gnome-contacts; I'd
add a bug link but Bugzilla is down.
When the shell takes control of the screen (for example to show
a modal dialog or to lock the screen), it must reestablish itself
on top of the stack, and in particular restore any unredirected
window so that it is composited below the Shell UI.
Reviewed-By: drago01 in IRC.
There is a race if a channel is invalidated during its preparation: the
'invalidated' signal is already emitted so the Shell will never notice.
We fix this by simply checking if the channel is already invalidated when
receiving it from telepathy-glib.
In the approving case, we reject the full ChannelDispatchOperation as we only
support approving one channel at the time.
https://bugzilla.gnome.org/show_bug.cgi?id=677457
Initially, extensions were loaded after they shell had fully created
the session and all objects, but this didn't allow extensions easy
ways to monkey patch prototypes, as most functions had already been
bound. Remove the historical vestigal function, and just merge the
two together.
https://bugzilla.gnome.org/show_bug.cgi?id=677586
A large amount of extensions have something like this in them:
function init() {}
Since we have encouraged extension authors to try and not make any
changes in init, it feels weird and strange to have to create an
initialization function that does nothing. From now on, don't require
it.
https://bugzilla.gnome.org/show_bug.cgi?id=677586
This allows us to move to a file-monitor based approach in the future.
Since we need signals, we convert the current set of functions to an
object we attach signals too, leading to the new ExtensionFinder object.
https://bugzilla.gnome.org/show_bug.cgi?id=677586
We explicitly include NoDisplay applications in the ShellAppSystem because
we want app tracking for them, but we explicitly filter NoDisplay applications
out when showing them to the user because we don't want to show them to the
user. We also based our "All" apps view on a flattened list of apps. While
we did check for NoDisplay on the app item itself, we didn't check against
its parents. Refactor the app display view to not use a separate flat list
of applications, but instead a concatenation of all the applications in all
the loaded categories.
https://bugzilla.gnome.org/show_bug.cgi?id=658176
clutter_actor_get_children requires making a temporary GSList from
a linked list structure, and then creating a JS Array from that GSList.
For simple cases like the number of children, use clutter_actor_get_n_children.
https://bugzilla.gnome.org/show_bug.cgi?id=677426
Show the dialog on the monitor containing the pointer, rather than
the monitor with active focused window. This brings it inline with
the behaviour seen when launching applications.
Remove the focusMonitor/focusIndex from LayoutManager. These
properties were only used by the modal dialogs. Remove them since
they are not being used elsewhere.
https://bugzilla.gnome.org/show_bug.cgi?id=642591
Don't show non urgent notifications when the primary monitor
is in fullscreen (user watching a movie, having a presentation,
playing a game ...).
Once the user leaves fullscreen show the messagetray so that the user don't
miss any notification (same as in "back from idle").
https://bugzilla.gnome.org/show_bug.cgi?id=677590
Do not overflow currentItems array. If the Menu section is filled
via model items-changed callback the position in the model passed to
the RemoteMenu _modelChanged can be a to be added asynchronously
action-added. Thus the item does not yet exists in the currentItems.
https://bugzilla.gnome.org/show_bug.cgi?id=676447
_startTimer adds a tweener to implement the description countdown, which
updates the entire content on each iteration, including the icon. This
causes a significant impact on performance, especially when accessibility
is enabled, as it causes a flood of AtkObject:state-change:showing events.
As the countdown only affects the description, factor out _updateDescription
and use it in _startTimer, and only do a full update of all contents when
necessary.
https://bugzilla.gnome.org/show_bug.cgi?id=674210
When displaying children, we need to make sure that we use the rowLimit
property that we pass to iconGrid, rather than assuming that it's
always MAX_SEARCH_RESULTS_ROWS in case some subclass (in an extension maybe)
wants to do something different.
https://bugzilla.gnome.org/show_bug.cgi?id=675527
Depending on the number of accounts, the type or simply the network,
there may be a noticable lag between setting the status and the actual
status change.
Use the new user-status-pending icon to indicate progress.
https://bugzilla.gnome.org/show_bug.cgi?id=659067
Refactor NMDeviceVPN to be more like the other NMDevices, including
having a valid getSectionTitle() and emitting signals when the
underlying connection changes state.
Use the existing notification infrastructure to hook these signals
to actual notifications (including some code consolidation).
https://bugzilla.gnome.org/show_bug.cgi?id=676330
Currently we only connect to the 'notify::n-workspaces' signal the
first time the overview is shown, which means we will miss any
changes to the workspace layout in the meanwhile.
In particular, the decision of whether the workspace switcher should be
shown is taken before the dynamic workspace handling takes over, and is
thus based entirely on the value of the num-workspaces user preference
rather than the actual number of workspaces.
Just connect the signal in _init() (with the nice side-effect to make it
explicit that the signal handler won't ever be disconnected).
https://bugzilla.gnome.org/show_bug.cgi?id=673198
role/label_actor should be used for the actor that will receive
the focus. In some cases it was also (wrongly) set on the
container, so using an AT like Orca, it exposed both.
https://bugzilla.gnome.org/show_bug.cgi?id=672242
It's used right away to discard some Empathy notifications.
This regression has been introduced during the 3.4 cycle when 'hints' has been
turned to a GVariant.
https://bugzilla.gnome.org/show_bug.cgi?id=675370
Ensure that the UI is updated when a connection changes name or id,
even if it was already known by a device.
Also, use less private properties on NMConnection objects, as they
can become stale and cause problems.
https://bugzilla.gnome.org/show_bug.cgi?id=677097
We now drop all status icons that are not explicitly enabled, which
breaks the ibus icon. Add it to the list of allowed icons until we
merge it with the keyboard one.
https://bugzilla.gnome.org/show_bug.cgi?id=677058
These methods were initially introduced when I was planning on having
an explicit DisableExtension/EnableExtension, instead of hooking up
a gsettings notify. This behavior was changed at the last minute, but
the methods were kept to avoid having to change the browser-plugin.
Consumers of this API should just set the GSettings key directly
instead now.
https://bugzilla.gnome.org/show_bug.cgi?id=676837
In Shell.SessionType.USER mode, two separate setup functions were
used during startup. With the new feature-based checks, the second
one is now almost empty, so move its remaining code into the first
function and remove it.
https://bugzilla.gnome.org/show_bug.cgi?id=676156
Add a sessionMode.allowKeybindingsWhenModal property, which determines
whether keybindings should still be handled while a modal dialog is
up or not.
https://bugzilla.gnome.org/show_bug.cgi?id=676156
Instead of falling back to a set of default values or crashing the
window manager when an invalid mode is specified, check the value
of the ShellGlobal:session-mode property before taking over as WM
and make a clean exit if it cannot be resolved to an existent mode.
https://bugzilla.gnome.org/show_bug.cgi?id=676156
Rather than accessing global.session_type / global.session_mode
all over the place, delegate mode information to a dedicated
sessionMode object. While not very useful for now, we will replace
checks for a particular mode with checks for particular properties
that sessionMode defines based on global.session_mode.
https://bugzilla.gnome.org/show_bug.cgi?id=676156
Since commit 518282e169, we hide both "Switch User" and "Log out"
on single-user systems. However, if there is more than one session
available, users may still want to get back to the login manager
to change sessions.
Add both "Log out" and "Switch Session" items in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=675802
Bluetooth PINs are required to have 6 digits, so enforce that
condition by making the PIN request notification's confirm
button insensitive unless the entered PIN has the correct length.
https://bugzilla.gnome.org/show_bug.cgi?id=651251
The availability of a notification action may depend on conditions,
so add a method to control the sensitivity of buttons which have
been added with addButton().
https://bugzilla.gnome.org/show_bug.cgi?id=651251
When the appMenu is not available, for instance when no windows are
open (on the current workspace), we make its actor unreactive to
"hide" it from keynav. However the menu can still be triggered
erroneously when using the corresponding keyboard shortcut, so
add a check for the actor's reactivity there as well.
https://bugzilla.gnome.org/show_bug.cgi?id=676316
For most subclasses, this is a direct swap -- a lot of the time, the
constructor was a blank class that override createNotificationIcon,
and called _setSummaryIcon in _init.
https://bugzilla.gnome.org/show_bug.cgi?id=661236
Rather than ask most users of Source to subclass it to simply set their icon,
just allow them to create a new instance and add it without any complex magic.
https://bugzilla.gnome.org/show_bug.cgi?id=661236
