We weren't freeing the whole pixbuf, nor were we the GList of
subpixbufs. Plug both of these leaks in the error handling path
and in the default case.
All of the unreffing/cleanup should happen in the GDestroyNotify
for the result, not some in the handler.
https://bugzilla.gnome.org/show_bug.cgi?id=636489
According to the designers, system notifications should be transient;
so now that transient notifications are supported properly, make use
of it instead of using a timeout to remove the associated source.
Transient notifications are removed after being shown. If the summary
is being shown while they appear, they are represented in it by a new
source icon.
We always create a new source for new transient notifications to
ensure that they don't replace the latest persistent notification
associated with the source. Because we generally don't want any
new or resident notifications to be replaced by others, associating
multiple notifications with a source is the next thing we will
implement.
https://bugzilla.gnome.org/show_bug.cgi?id=633412
Resident notifications don't get removed when they are clicked or
one of their actions is invoked, and are only removed when the app
that created them requests them to be removed or sends another
notification.
Remove the source when a notification associated with it is removed.
Except if the source is a tray icon.
Make sure that we pop down the tray when a notification is clicked
or one of the actions of a non-resident notification is selected.
Based on the initial patch by Jonathan Matthew.
https://bugzilla.gnome.org/show_bug.cgi?id=633412
Given that the grid view is gone there is no point in animating the
window previews on all workspaces anymore so just do it for the current
one avoid taking a slow down caused by animating windows on other workspaces.
https://bugzilla.gnome.org/show_bug.cgi?id=637353
In both tests the scrolled actor's width was fixed to the stage
width, so that the scroll bars ended up outside the visible area.
Fix by adding an outer container with a fixed width and expanding
the scroll view to fill the available area.
Mixing submenu menuitems and toggle menuitems results in poor layout.
The fix is to right-align the submenu arrows. Since we already need to
right-align the battery percentages as well, add alignment support to
PopupBaseMenuItem.addActor(), and update stuff for that.
Also remove the "column" param from addActor() since it hadn't
actually been implemented correctly before.
https://bugzilla.gnome.org/show_bug.cgi?id=633476
While non-uniform border widths were parsed correctly, an arbitrary
side's width was picked when painting, so that each border ended up
with the same width and the widths specified in CSS were ignored.
At least for sides between non-rounded corners, using a different
border width can be reasonable, for instance at screen edges.
Different border widths around rounded corners are kind of crack,
but then it would be lame not to support it ...
https://bugzilla.gnome.org/show_bug.cgi?id=607500
For gradient backgrounds, borders were implemented by filling the
background shape with the border color first, and then scaling down
the path to draw the background.
The result is not correct[0], which is especially visible if the border
width is greater than the border radius - so instead of scaling down
the original path, use a separate path for the background.
The result is consistent with the borders we draw for non-gradient
backgrounds, and much closer to the correct standard behavior.
[0] http://www.w3.org/TR/css3-background/#the-border-radiushttps://bugzilla.gnome.org/show_bug.cgi?id=607500
non-absolute paths specified as url()'s in
stylesheets are resolved to be relative to the location
of the stylesheets they are in.
Inline styles don't have physical styleshseets sitting on disk,
which leads to a crash in the url resolving code.
This commit ensures that we don't try to use the stylesheet associated
with a url, if there isn't one to use.
This commit doesn't try to handle relative paths in inline styles.
It only prevents crashes when absolute paths are used.
https://bugzilla.gnome.org/show_bug.cgi?id=636975
This prevents a race at start up that could prevent the user
from getting loaded properly.
In the near future we need to drop the gdm user code entirely
and switch to using the accountsservice library.
Since we are keeping a current pointer position anyways, we
don't have to continually call global.get_pointer() which is
a round trip to the magnifier; make ZoomRegion simply fetch
a current position stored in the Magnifier object.
https://bugzilla.gnome.org/show_bug.cgi?id=633553
Change the proportional algorithm so stop moving the zoom region
when cursor is in a "padding region" at the edge of the screen.
(The padding region is a 10th of the screen at 2x zoom, and smaller
for higher zooms.)
Based on earlier versions from Jon McCann and Florian Muellner.
https://bugzilla.gnome.org/show_bug.cgi?id=629950
This basic point of this change is to avoid always creating a
hidden Clutter.Clone actor for the default present-but-not-active
zoom region. The position of the viewport and region of interest
are now stored in member variables, and the actors are only created
and updated when the region is active.
Other significant changes:
* Unused public functions are removed or made private
* The mouse tracking position is immediately updated when options
like the zoom are changed, not just on the next mouse motion.
* ZoomRegion.setROI() now updates the zoom, not just the position;
a FIXME is added to the D-Bus interface for a place where the
D-Bus interface contains duplicate possibly conflicting information
* Lens-mode is now only effectively off when the magnifier is
fullscreen, instead of actually modifying the member variable;
this makes things work properly when changing out of full-screen
mode.
* When the clamping to screen edges is turned on, we now immediately
clamp.
* The handling of setting the position to fullscreen as compared
to just setting the viewport to fullscreen is untangled.
https://bugzilla.gnome.org/show_bug.cgi?id=633582
libcanberra is an (optional) dependency of Mutter; since we have
it in the moduleset we need to mark the dependency so things get
built in the right order.
