Writting the screenshot to a file can take a relativly long time
in which we block the compositor, so do that part in a separate
thread to avoid the hang.
https://bugzilla.gnome.org/show_bug.cgi?id=652952
DashItem labels have initial delay before showing up, but once the
first label in the dash is visible (meaning the user is very likely
exploring things) and the pointer is moved along the dash, the label
will follow immediately.
https://bugzilla.gnome.org/show_bug.cgi?id=666170
Signed-off-by: Seif Lotfy <seif.lotfy@collabora.co.uk>
- We should only call workspaceRemoved() for workspaces that are
are actually being removed.
- When we have multiple monitors, a window on a secondary monitor is
on all workspaces, so it ends up in all workspaces _allWindows
lists, so we can't use previous presence in that list to determine
whether we need to go ahead and add the actor; allWindows is simply
the list of windows where we are listening to notify::minimized.
https://bugzilla.gnome.org/show_bug.cgi?id=667652
Since the application proxy is created asynchronously, at the time
the GActionGroup (GActionMuxer) is created, there is no GDBusMenu yet.
Defer creating the menu in that case.
Also, clear out signal handlers if have no target application.
https://bugzilla.gnome.org/show_bug.cgi?id=633028
We add a drag monitor to check whether the pointer is inside
the workspace selector, and update the visibility of the drop
placeholder consequently.
https://bugzilla.gnome.org/show_bug.cgi?id=664201
According to the GIO docs, sections can have labels too. We support
them by inserting a non reactive menu item at the beginning of the
section. This item is specially flagged to be ignored while processing
changed signals from the model (since it does not correspond to any
model item)
https://bugzilla.gnome.org/show_bug.cgi?id=666681
If there's a single small window (e.g. empathy chat) in the overview, it
looks usable, because it's as big as outside of the overview, but when
you start to type, overview search is launched, which is confusing.
Fix that by setting maximum scale for window clones to 0.7
https://bugzilla.gnome.org/show_bug.cgi?id=646704
In overview when closing a window and afterwards dragging a window it can
happen that you pick a wrong window or no window if windows' positions is
updated while initiating the drag.
Fix that by delaying window rearrangement when cursor is over a window.
https://bugzilla.gnome.org/show_bug.cgi?id=645325
DND code assumes it can query the size of the actor before parenting,
while StWidget asserts that get_preferred_size() is only called
after the actor is on stage. This fixes a crash while dragging
"Connect to..."
https://bugzilla.gnome.org/show_bug.cgi?id=633028
With GJS' GDBus implementation, we get the invocation paramters as an
array if we declare a method as async.
This is bad and not consistent with what GJS does for synchronous
methods, but it's the way it is, and other classes in gnome-shell
implement this correctly by exploding the array into its components in
the method implementation, but not the screenshot methods.
Also, we're supposed to return a value using the provided invocation
object, not with a callback now, so do that.
https://bugzilla.gnome.org/show_bug.cgi?id=667662
Commit 25948f214e replaced the old hardcoded scaling behavior of
background-images with the CSS-compliant option to control that
behavior with the background-size property. Fix some fallout from
the changed default scaling behavior.
That way different system notifications, such as the ones about battery power
and the ones about software updates, are shown with separate message tray
sources.
https://bugzilla.gnome.org/show_bug.cgi?id=664138
After an item is destroyed, all its signals were disconnected,
except for 'destroy' itself. This could lead to exceptions, if
destroy was called more than once on the item.
https://bugzilla.gnome.org/show_bug.cgi?id=665680
Instead of using an St.Tooltip to show the app's name under the icon,
manually position a new St.Label ourselves. Make sure to keep the label
hidden when right-clicking so it doesn't get in the way of the popup menu.
Only one tooltip/label will be displayed at a time.
https://bugzilla.gnome.org/show_bug.cgi?id=666166
GTK+ also exports window-specific actions, by putting the object path
for the exported action group in the _DBUS_OBJECT_PATH X property.
We add this action group to the app's muxer with a 'win' prefix,
since that is what the exported menu expects. Whenever the focus
window changes, we update the window-specific actions of its
application, and emit notify::action-group to cause the app
menu to be updated.
GDBusActionGroup api has changed again, adapt to that.
Also, use a GActionMuxer to add the 'app.' prefix to actions,
instead of manually stripping it out of the action names.
In the future, the muxer will also contain per-window actions
with a 'win.' prefix.
By the time the window is first mapped and the app menu button is
synced, we may not have finished reading the menu. In that case,
connect to notify::menu and update accordingly.
Use the new GApplication support in ShellApp to create the application
menu. Supports plain (no state), boolean and double actions.
Includes a test application (as no other application uses GApplication
for actions)
https://bugzilla.gnome.org/show_bug.cgi?id=621203
The loop can exit with an interval of length one or one of
length zero. In the first case it is correct to check which side
of the interval to return, in the second case no comparison should
be made (since there is only one possible value).
In practice, this usually results in one comparison more than needed,
but in some cases (when the position was past the end of the array),
would call the comparator with undefined.
https://bugzilla.gnome.org/show_bug.cgi?id=666614
When transitioning from gnome-panel to gnome-shell in 3.0 we
lost the ability to summon the wisdom of the mythical fish.
This patch restores this, for the few adepts that are aware of
the magical incantation.
(Not as configurable as the original one, but it's an easter egg
after all...)
https://bugzilla.gnome.org/show_bug.cgi?id=666606
Forcing the icon size will distort it unnecessarily, and will
in any case not work if showing an animation (which is a ClutterGroup).
Instead, set the size on the bin, and make it align its child
if needed.
https://bugzilla.gnome.org/show_bug.cgi?id=666606
Previously the code in _accessPointAdded was iterating over the
the network list to find a good place, and at that time, added both
the network to the list and the item to the menu. When I refactored
to call queueCreateSection, I forgot to add code to insert the
network in the list.
Add it now, using the new Util.insertSorted function.
https://bugzilla.gnome.org/show_bug.cgi?id=666429
Adds two new functions, Util.lowerBound and Util.insertSorted,
that take an array, a value and a comparator, and find the
first position at which the value can be inserted without
violating the order, in optimal time.
https://bugzilla.gnome.org/show_bug.cgi?id=666429
By using Main.queueDeferredWork, we can ensure that most of the
menu contents (in particular, the heaviest parts like the list of
wifi networks) are not updated immediately as we receive signals
from NetworkManager. Instead, the menu is rebuilt some time later,
or as soon as the user opens the menu.
This means that it is no longer needed to optimize for the
access-point-added case, replacing a lot of buggy code with a safer
call to _queueCreateSection, which in turn should ensure that the
more menu, if existing, is always at the end and that at most 5 networks
are visible outside it.
https://bugzilla.gnome.org/show_bug.cgi?id=664124
We need to notify when the channel dispatcher calls HandleChannels on
us with a channel we already handle. However, we don't want to notify
if we claim a new incoming channel which doesn't actually have
anything interesting in it yet.
For example, a new channel pops up just to give a delivery
notification. We want (or, need) to handle it but don't want to notify
for it.
Fixes: https://bugzilla.gnome.org/show_bug.cgi?id=666243
Signed-off-by: Jonny Lamb <jonnylamb@gnome.org>
The window overlays may be shown erroneously if hideOverlays is
called while the corresponding clone has an uncompleted tween which
calls showOverlays in its onComplete handler, for instance when
quickly leaving the overview before the initial overview animation
has finished. To fix, remove all existing tweens when hiding the
overlays.
https://bugzilla.gnome.org/show_bug.cgi?id=666020
When in overview, window labels flicker or are temporarily hidden on a
number of occasions - when simply clicking around the area the windows
are displayed in, dragging a window, sliding in the workspace list,
adding new workspaces etc. This patch makes the label for any window
visible at any given moment when in overview and the said window is
not being dragged around.
https://bugzilla.gnome.org/show_bug.cgi?id=644861
When wifi or wwan are blocked by hardware killswitch, we should not
allow changing the switch (it won't work anyway), and show
"hardware disabled" instead, similar to what we already do in the
bluetooth menu.
https://bugzilla.gnome.org/show_bug.cgi?id=665194
When placing networks in _createSection, we were taking in
consideration that _activeNetwork is always first, by adding 1,
but then kept this offset also for networks following it (normally,
all of them, since _activeNetwork is also the most recently used),
that instead should not be affected by the movement.
This resulted in the menu showing 4 networks + More... instead of
5.
https://bugzilla.gnome.org/show_bug.cgi?id=664124
The three boxes for the ripple animation are visible when created. This
means that the drag and drop code that searches for an actor to handle
the drag can find the ripple boxes instead of the Activities button or
hot corner. The latter can handle drag and drop while the ripple boxes
can't.
This is only a problem if drag and drop is attempted before the ripple
animation has been played: the boxes are made invisible at the end of
the animation. The fix is to just create the boxes invisible.
Instead of leaving the tray covering the whole last pixel row when it's
hidden, hide it completely. This avoids mouse events not being delivered to
application windows on the last pixel row.
To summon the tray we use a single reactive pixel on the corner.
https://bugzilla.gnome.org/show_bug.cgi?id=663366
The dash handles 'window-drag-cancelled', to be able to do the
animations for drag snap-back and size changes in parallel. As
the signal is not emitted for previews in the workspace switcher,
it does not work in that case.
If workspaces-only-on-primary is false, swipe scrolling is now
broken with multiple monitors. To fix, let workspacesDisplay
handle swipe scrolling for all views.
https://bugzilla.gnome.org/show_bug.cgi?id=652580
If workspaces-only-on-primary is false, workspaces should be shown
on each monitor; rather than letting the existing workspaces span
the entire screen, manage one workspacesView per monitor (similar
to the extra workspaces in WorkspacesView when the setting is true).
https://bugzilla.gnome.org/show_bug.cgi?id=652580
Extra workspaces are special, in that they collect windows from
all workspaces for a particular monitor. This matches the default
behavior, but we need more than a single workspace per monitor if
workspaces-only-on-primary is false, so don't create the extra
workspaces in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=652580
WorkspacesDisplay was introduced to manage the workspace objects
and views; however, the overview still accesses the view held
by the workspacesDisplay directly, which is a bit odd.
Add some additional methods needed by the overview, and make the
view a private property.
https://bugzilla.gnome.org/show_bug.cgi?id=652580
When installing an extension at runtime, we accidentally swapped the 'type'
and 'enabled' parameters. While this doesn't directly affect anything right
now, as everything works coincidentally, future patches that look at the
'type' parameter to decide what to do would get the wrong answer.
The last patch in the sequence. Every place that was previously
setting prototype has been ported to Lang.Class, to make code more
concise and allow for better toString().
https://bugzilla.gnome.org/show_bug.cgi?id=664436
All classes that have at least one other derived class (and thus
benefit from the framework) have been now ported. These includes
NMDevice, SearchProvider, AltTab.SwitcherList, and some other
stuff around.
https://bugzilla.gnome.org/show_bug.cgi?id=664436
Third step in the class framework port, now it's the turn of
MessageTray.Source and MessageTray.Notification, as well as
the various implementations around the shell.
https://bugzilla.gnome.org/show_bug.cgi?id=664436
The Lang module in gjs has recently gained a small yet powerful
Class framework, that should help improve the readability of code
when using complex inheritance.
This commit starts porting shell code, by rewriting all classes in
popupMenu.js (and all derived classes) to Lang.Class.
https://bugzilla.gnome.org/show_bug.cgi?id=664436
No idea why connecting a key-press-event to a non-reactive actor
used to work, but some Clutter update broke it. Obvious fix is
to make the actor reactive.
https://bugzilla.gnome.org/show_bug.cgi?id=664582
We prefer to ask the user for his own password. If PolicyKit
is not configured to accept that, try the root password. If
PolicyKit does not accept that either, ask for password of
the first user that PolicyKit _will_ accept. The last case
is a bit broken, but should rarely occur in real-life
configurations.
https://bugzilla.gnome.org/show_bug.cgi?id=651547
If no telepathy accounts have been set up or enabled, the IM status
chooser won't have any effect. To avoid confusing behavior, make
the status selector insensitive in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=662800
ComboBoxMenuItems use ClutterClones to reconstruct the active item
in the associated ComboMenu, so pseudo class changes due to state
changes of the ComboBoxMenuItem don't have the intended effect
(since the actual style information is taken from the associated
ComboBoxMenu item).
As a fix, propagate relevant pseudo class changes to the active
ComboBoxMenu item.
https://bugzilla.gnome.org/show_bug.cgi?id=662799
ComboBoxMenuItems use ClutterClones to reconstruct the active item
in the associated ComboMenu to not impose a particular MenuItem type
in the menu. However, this results in style changes (for instance
those triggered by icon-theme or text-scaling-factor changes) of
the ComboBoxMenuItem not having a visual effect until the ComboBoxMenu
is shown.
As a fix, force a style update on the ComboBoxMenu when the item's
style changes.
https://bugzilla.gnome.org/show_bug.cgi?id=662799
The option was merge with 'visual-bell-type' with the GSettings
port, but the change turned out too disruptive for the universal
access menu / settings panel, so gsettings-desktop-schemas commit
a5819b2a4e9 re-added the separate option.
- replace some left-over references to GnomeSession.Presence.setStatus()
- the correct replacement for GnomeSession.Presence.getStatus()
is *not* GnomeSession.Presence.connectSignal('StatusChanged')
This continues the series of patches for GDBus porting, affecting
all code that accesses remote DBus objects. This includes modemManager,
automount, autorun (for the hotplug sniffer), calendar, network (for
nm-applet only), power, scripting (for perf monitor interface)
https://bugzilla.gnome.org/show_bug.cgi?id=648651
Port org.gnome.ScreenSaver and org.gnome.SessionManager glue code
to use GDBus, and move /org/gnome/Shell/EndSessionDialog to the
GDBus connection, so it is backed by the org.gnome.Shell name.
https://bugzilla.gnome.org/show_bug.cgi?id=648651
Rewrite code acquiring dbus names so that it uses GDBus, and rewrite
ShellDBus so that it is exposed on the GDBus connection. Ports of
the other objects will follow.
https://bugzilla.gnome.org/show_bug.cgi?id=648651
When changing _findNetwork with _findExistingNetwork, I changed
the return value to avoid searching twice for the access point,
and changed some names. I forgot to update all points where those
names were used.
https://bugzilla.gnome.org/show_bug.cgi?id=663278
Previously, we connected to notify::strength only if there was
already a signal connected, and the AP changed (thus, by induction,
we never connected). As a result, the icon became stale and different
from that shown inside the menu (which is correctly updated).
https://bugzilla.gnome.org/show_bug.cgi?id=650007
We must look for the actor under the pointer in the whole message tray and not
just in the notification. This will avoid us to capture focus when a
notification comes up with the pointer on the whole tray area.
https://bugzilla.gnome.org/show_bug.cgi?id=661358
Add "Mute"/"Unmute" option to the right click menu for chats to allow muting conversations
without blocking the sender or disabling all non-urgent notifications. Muting a conversation
prevents the pop up of notifications on new messages from the muted source, while these
messages are still available from the summary notification in the message tray.
https://bugzilla.gnome.org/show_bug.cgi?id=659962
SubMenuMenuItems close automatically with their parent, however
closing fails when the parent item is a MenuSection, as those
currently ignore any open()/close() requests.
At some minimal handling by emitting the 'open-state-changed' signal,
so children like SubMenuMenuItems work as expected.
https://bugzilla.gnome.org/show_bug.cgi?id=661029
PopupMenu.firstMenuItem returns a PopupMenuItem, not an apObj. We
need to retrive the latter using the _apObj property.
Also, somehow the property from the number of elements in a menu
was changed from .length to .numMenuItems, and this broke the
destruction of the menu upon emptying it.
https://bugzilla.gnome.org/show_bug.cgi?id=659277
Calling nm_access_point_get_ssid() in the handler of the
access-point-removed signal can result in DBus request, which will
then fail because the object was already removed at the server side.
Instead, use a difference function to retrieve the access point
object (the network), that compares directly by object identity.
https://bugzilla.gnome.org/show_bug.cgi?id=651378
Allow push_modal to optionally only work with a keyboard only grab and
use that in altTab as a fallback to allow switching windows while a pointer grab
is in effect (like during DND operations).
https://bugzilla.gnome.org/show_bug.cgi?id=660457
When two extensions monkey-patch the same area, enable() and disable() may
behave badly and completely wreck things. To solve this, when disabling
an extension, "rebase" the extension list so that monkey patches should be
added and removed in order.
https://bugzilla.gnome.org/show_bug.cgi?id=661815
Rather than loading and enabling all extensions at Shell init time, save some
time and gain some basic security by not loading extensions if they're
not enabled.
https://bugzilla.gnome.org/show_bug.cgi?id=661815
The GFileMonitor on ~/.gtk-bookmarks was block-scoped in the
PlacesManager._init() function, which caused it to be destroyed as soon
as the constructor was done. This caused changes in bookmarks to never
be notified to possible watchers (such as the places-menu extension).
To fix this, the 'monitor' variable has been promoted to an object
instance member.
https://bugzilla.gnome.org/show_bug.cgi?id=661921
When requesting a presence change, the actual presence set by
mission control does not necessarily match the requested presence
(if an active account does not support the requested presence),
which may result in the wrong presence being restored.
As a fix, be more cautious about saving status by assuming that
users do not request presence changes between an automatic presence
change request and the actual change.
https://bugzilla.gnome.org/show_bug.cgi?id=661485
Historically, when applications set "image-data" they expect it to show up
as an icon. So we display it as such if an icon is not specified with an
"app_icon" argument to Notify(). We also use "image-path" for an icon if
an icon is not specified.
We only display a large image specified with "image-data" or "image-path"
if an icon is also specified.
https://bugzilla.gnome.org/show_bug.cgi?id=659158
Users don't expect the bullets they just typed into an entry
field to disappear as soon as they hit enter.
Instead, they want the dialog to become insensitive during the
authentication process, so that it's clear that what they typed
in is being processed.
https://bugzilla.gnome.org/show_bug.cgi?id=657894
The message tray focus grabbing code sets the stage input
mode to Shell.StageInputMode.FOCUSED when the overview is
not visible. This ensures the stage window's input
region gets reshaped to include the notification chrome,
and so that input events get delivered appropriately to
the notification that grabbed focus.
The message tray code never tries to restore the stage input
mode later. Instead, the code relies on the stage input
mode (and input region) getting reset to
shell.StageInputMode.NORMAL automatically when focus moves
back from the shell chrome to a window in the user's session.
It's not really correct to set the stage input mode based
on the overview's visibility, though. At the login screen,
even though no overview is visible, the stage input mode is
Shell.StageInputMode.FULLSCREEN which is sufficient
for the notification's needs, Furthermore,
Shell.StageInputMode.FOCUSED is insufficient for the login
dialog's needs since the login dialog isn't considered
part of the shell's chrome and won't get included in the
stage input region.
This commit changes the message tray code to only set the
stage input mode if the current stage input mode isn't good enough,
rather than assuming the input mode isn't good enough just because
the overview is hidden.
https://bugzilla.gnome.org/show_bug.cgi?id=660919
Because we were sorting the Alt+Tab list by user_time rather than
stacking order / MRU, it was possible for the currently-focused window
to sometimes not be the first app in the list. Fix this by using
meta_display_get_tab_list() to get the proper MRU ordering of windows
on the current workspace, and then convert that to an ordered list of
apps.
https://bugzilla.gnome.org/show_bug.cgi?id=645026
In case _adjustIconSize() is called while the the dash icons are
animating, some extra work is required to yield the expected result.
Skip those extra steps when the icons are not actually animating.
https://bugzilla.gnome.org/show_bug.cgi?id=649248
The current code uses the dash's height and current icon size to
calculate the new icon size. However, the height does not correctly
relate to the icon size while the icons are animating, in which
case the resulting icon size may be wrong.
Rework the function to be independent from the actual icon sizes,
so that a correct size is calculated even when called during an
animation.
https://bugzilla.gnome.org/show_bug.cgi?id=649248
Rather than relying on the caller to hide the remove target and
removed items before calling _adjustIconSize(), move that logic
into _adjustIconSize() itself.
https://bugzilla.gnome.org/show_bug.cgi?id=649248
In case _adjustIconSize() is called while the the dash icons are
animating, some extra work is required to yield the expected result.
Skip those extra steps when the icons are not actually animating.
https://bugzilla.gnome.org/show_bug.cgi?id=649248
Rather than relying on the caller to hide the remove target and
removed items before calling _adjustIconSize(), move that logic
into _adjustIconSize() itself.
https://bugzilla.gnome.org/show_bug.cgi?id=649248
Use ShellEntry.addContextMenu() to add context menus to most
existing entries, with the exception of:
- the login dialog - it may act be used to enter either the
username (e.g. no password entry) or the
password, and copy/paste does not make sense
(nowhere to copy from, nowhere to paste to)
- notifications - while adding a context menu is useful here as
well, it will require changes to the tray's
focus grab handling, so leave those entries
out for now
https://bugzilla.gnome.org/show_bug.cgi?id=659275
Add addContextMenu() to support context menus on right-click/long-press.
Depending on the parameters passed, the context menu only contains
"Copy"/"Paste" actions or an additional "Show/Hide Text" toggle action
for password entries.
https://bugzilla.gnome.org/show_bug.cgi?id=659275
While the ability to show the password can be useful at times,
the existing implementation is problematic:
1) The use of a switch is wrong (as even noted in a code
comment).
2) It is inconsistent with any other password dialog (login screen,
polkit).
In lack of a properly designed solution (for all password dialogs),
the designers agreed to remove the switch for now.
https://bugzilla.gnome.org/show_bug.cgi?id=658948
The keyboard hides prematurely when the user is typing into
an empty search box because the click is a captured event that
triggers a loss of entry focus. By adding a keyboard check to
this event, the problem is solved.
https://bugzilla.gnome.org/show_bug.cgi?id=661340
Currently BoxPointer/Menus always point to the center of the
associated source actor. This is generally what we want, but
add some API to adjust that behavior for the cases where it
isn't.
https://bugzilla.gnome.org/show_bug.cgi?id=659274
Right now we show "(or swipe finger)" at the user login prompt
any time we detect a fingerprint reader.
Checking for the presense of a fingerprint reader isn't really
sufficient for knowing if it is appropriate or not to show the
message, though. Often, a user's fingerprint won't be enrolled
in the system even if the machine has a fingerprint reader.
In this scenario, we end up in a situation where the code will
fade out the message right after fading it in, or worse, fade
out the message while fading it in.
The former case looks flickery and bad, and the latter case
causes the login dialog to lock up since it never completes its
"show prompt" animation and we don't procede with the login
process until after that animation.
If a user is enrolled in the system, the fingerprint pam module
tries to tell the user to swipe their finger. We never show the
user that message because it's redundant with our own "(or swipe
finger)" message and because it uses techy words like "UPEK" and
"TouchStrip".
This commit changes the code to defer showing "(or swipe
finger)" until the fingerprint pam module forwards us its own
message. This makes it less likely we'll show the message when
fingerprint login won't work, and also removes the fingerprint
animation from the critical path "show prompt" animation.
https://bugzilla.gnome.org/show_bug.cgi?id=660492
Nvidia's twin view option does not align monitors properly, but with
a one pixel overlap. It looks safe to ignore an overlap this small
to make this case work.
https://bugzilla.gnome.org/show_bug.cgi?id=661387
js2-mode is no longer developed and we recommend js-mode these days,
so switch the modelines to specify that, and make them consistent
across all files.
https://bugzilla.gnome.org/show_bug.cgi?id=660358
Right-click menus in the dash can be dismissed by clicking anywhere
outside the menu. However, if a window clone is located beneath the
pointer when doing so, the window is activated and the overview
closed.
The cause of this unexpected behavior is that window previews are
activated on button-release, which is delivered to the preview after
the menu releases its grab on button-press. Use a ClutterClickAction
instead and let Clutter do the right thing, i.e. only trigger a
'clicked' signal when a button-release event is matched by a
corresponding button-press event.
https://bugzilla.gnome.org/show_bug.cgi?id=661151
tp-glib can sometimes emit a notify::alias signal when the alias doesn't
actually change. Bail out early instead of pushing an alias change message.
https://bugzilla.gnome.org/show_bug.cgi?id=660774
We now match individuals on other properties than alias, so take
this into account when representing a contact in search results
to avoid having them show up as "Unknown".
https://bugzilla.gnome.org/show_bug.cgi?id=660580
When workspace "previews" in the overview were just tiny gray
rectangles, it made sense to provide a way to move windows
directly between workspaces (by switching workspaces when dragging
a window to the corresponding screen edge). As the overview has
evolved however, the workspace switcher provides a good and
intuitive drop target already, so the alternative provided by the
screen edges is no longer necessary. As it also conflicts with
moving windows between monitors when using a vertical layout,
just remove it.
https://bugzilla.gnome.org/show_bug.cgi?id=660838
