Use StWidget:track-hover rather than doing it ourselves. Don't assume
that hover is always TRUE after an enter_event or FALSE after a
leave_event, since we have a pointer grab and will be getting other
actors' events.
Don't ungrab the pointer when it leaves the button, since that
destroys the whole point of getting a grab in the first place.
Only consider the button to have been clicked when it has both grab
(meaning the mouse was pressed over the button) and hover (meaning the
mouse was released over the button).
Also remove the virtual pressed/released methods, which weren't being
used anyway.
https://bugzilla.gnome.org/show_bug.cgi?id=633853
Like all X API, XFixesGetCursorImage returns arrays of 32-bit
quantities as arrays of long; on 64-bit systems we need to
convert to an array of 32-bit words before creating a texture
from the result.
https://bugzilla.gnome.org/show_bug.cgi?id=633591
Add StWidget:can-focus, st_widget_navigate_focus(), and
st_container_get_focus_chain(), and implement as needed to allow
keyboard navigation of widgets.
https://bugzilla.gnome.org/show_bug.cgi?id=621671
In d66e7dd49 I got confused between border_texture and
background_texture. The background_texture was being created as normal
but in the one place that it gets drawn I accidentally made it use the
border_material instead. This patch makes it create a
background_material similar to the border_material and uses it to
paint.
A few places in st-theme-node-drawing create one-shot material, paint
with it and then free it. This is suboptimal with current Cogl because
it will end up compiling an ARBfp program just for that single paint
and then it will throw it away when the material is destroyed.
There is a new function in st-private.c called
_st_create_texture_material. This creates a simple material for a
texture based on a common parent material that points to a dummy
texture. Any materials created with this function are likely to be
able to share the same program unless the material is further modified
to contain a different number of layers. It would be possible to use
cogl_set_source_texture for this instead except that it's not possible
to modify the material's color in that case so we couldn't render the
texture with opacity.
The corner textures are now stored as a handle to a material that
references the texture rather than storing the texure directly. There
is also a separate border_material member which always points to
border_texture as the only layer.
https://bugzilla.gnome.org/show_bug.cgi?id=633340
Plugging the GtkSocket icon can fail, in which case NaTrayManager emits
tray-icon-removed but the icon has never emitted plug-added, thus the
ShellTrayIcon has not been created. This results in various criticals
and exceptions in the tray-icon-removed handler.
https://bugzilla.gnome.org/show_bug.cgi?id=633028
MutterWindow and MutterPlugin have been renamed to MetaWindowActor
and MetaPlugin, mutter_plugin_list_windows() to
meta_plugin_list_window_actors(). Adapt to those changes.
https://bugzilla.gnome.org/show_bug.cgi?id=632500
Although within St itself there are situations where the semantics of
these functions (return TRUE or FALSE and return the actual value in
an out parameter) is useful, it's mostly just annoying at the
application level, where you generally know that the CSS property is
going to specified, and there is no especially sane fallback if it's
not.
So rename the current methods to lookup_color, lookup_double, and
lookup_length, and add new get_color, get_double, and get_length
methods that don't take an "inherit" parameter, and return their
values directly. (Well, except for get_color, due to the lack of (out
caller-allocates) in gjs.)
And update the code to use either the old or new methods as appropriate.
https://bugzilla.gnome.org/show_bug.cgi?id=632590
Gjs changed to separate the "basic" API from "embedder" API. Unfortunately
due to our use of the importer API and a few other bits, we need to use
the "module" one. In the future I plan to extend the core API to
support this use case.
Add volume control indicator which uses API from gnome-volume-control
to interact with PulseAudio and shows both input and output volumes.
Also adds a small wrapper around libcanberra in ShellGlobal, used by the
volume indicator to provide auditive feedback.
https://bugzilla.gnome.org/show_bug.cgi?id=629455
The library is introspected, and should not require using
Pulseaudio directly.
With help from Giovanni Campagna <scampa.giovanni@gmail.com>
(introspection annotations, build fixes)
https://bugzilla.gnome.org/show_bug.cgi?id=629455
st-theme stores some loaded stylesheets in a custom
stylesheets list. When removing items from this list
it unrefs them, but when adding items to the list it
neglects to ref them. This means that under certain
circumstances the list will contain items that have
already been freed.
https://bugzilla.gnome.org/show_bug.cgi?id=632477
StDrawingArea uses the member variable needs_repaint to keep track
of whether it needs repainting. The variable is set to TRUE correctly,
e.g. on allocation or style changes - alas, it is never set to FALSE,
resulting in the area being repainted continuously.
https://bugzilla.gnome.org/show_bug.cgi?id=632197
Since we're just using a cut-and-paste of gdm code there's a
slight impedance mismatch between how that code and the rest of
the shell manage debug messages. In GDM, they're out of sight by
default, but in gnome-shell they're visible and quite verbose.
This muddies up and masks the useful messages that other parts
of the code generate.
This commit just mutes the messages unconditionally by default,
since they aren't that useful anyway.
The GDM code upstream talks to the account service now,
has better introspection annotations, and is more
asynchronous.
This commit updates the shell's copy to the latest
upstream.
Note, the API changed somewhat and so the callers will
need to be fixed up subsequently.
https://bugzilla.gnome.org/show_bug.cgi?id=631888
1. move logic to shell-app.c
2. change state to RUNNING only after startup sequence complete
3. correct handle state for applications with several .desktop files
https://bugzilla.gnome.org/show_bug.cgi?id=623688
Non-uniform border-radii are already supported when using a gradient
background, this patch adds support for solid colors as well.
The currently applied technique of using corner textures and filling
the remaining area with rectangles is extended, so that each corner is
padded with rectangles to the size of the largest corner.
Add border-radius.js to test cases, to test non-uniform border-radii
with both solid color and gradient backgrounds.
https://bugzilla.gnome.org/show_bug.cgi?id=631091
Icons can be loaded as St.Icon.SYMBOLIC, FULLCOLOR, APPLICATION or
DOCUMENT. The first will look for a symbolic equivalent, the second
looks for a full-color version (and does fallback, eg, from
"drive-harddisk-usb" to "drive-harddisk"). APPLICATION and DOCUMENT do
full-color icons without fallback (as specified by the icon spec).
And update various callers to use the right flags.
Based on a patch from Matt Novenstern.
https://bugzilla.gnome.org/show_bug.cgi?id=621311
Switch to using the layer combine constant rather than the material
primary color for representing the opacity of the material; this
avoids triggering a Cogl bug where changing the primary color corrupts
the layer state.
https://bugzilla.gnome.org/show_bug.cgi?id=629616
The actor emitted by ShellTrayManager is now ShellTrayIcon, a subclass
of ShellGtkEmbed which has several properties on it which are (or will
soon be) useful to the shell.
Part of the rearranging to use ShellTrayIcon means that we now show
the ShellEmbeddedWindow before creating its ShellGtkEmbed, which
requires a few modifications to ShellEmbeddedWindow (notably, telling
it at construct time what stage it will be drawn on, since it needs to
know that before it has a ShellGtkEmbed now).
https://bugzilla.gnome.org/show_bug.cgi?id=608869
Previously the code to keep the X window aligned with its actor made
various assumptions that were true in the panel status tray area but
that are not true in the message tray. Fix it up by repositioning the
X window at clutter_actor_paint() time, which will definitely be
called every time the actor moves, whether it's because of its own or
its parent/ancestor's changes.
https://bugzilla.gnome.org/show_bug.cgi?id=608869
As of 1.4, ClutterGLXTexturePixmap is identical to
ClutterX11TexturePixmap (with the differences having been moved into
the cogl layer). So make ShellGtkEmbed use the (introspected) X11
version, which then allows us to make the instance and class structs
public as well.
https://bugzilla.gnome.org/show_bug.cgi?id=608869
Override the new mutter preference /apps/mutter/general/attach_modal_dialogs
to attach modal dialogs to their parent window. Animate the modal dialogs
expanding from the top of the parent window. Slowly dim the parent window
after the dialog comes up.
https://bugzilla.gnome.org/show_bug.cgi?id=612726
* Use --warn-all, --warn-error
* Fix various broken gtk-doc
* Drop unused shell_get_event_related
* For header defines, we currently require them to end in _H to be skipped
* Drop the no-longer-necessary fix-meta-rectangle.py hack
* Move to the convention of using -private.h for headers that are,
well, private.
* Add shell-wm-private.h
Move shadow helper functions from st-theme-node-drawing to st-private
to make them available to widgets which want to add shadows to internal
actors.
Also add a new helper function for creating the shadow material from a
ClutterActor.
https://bugzilla.gnome.org/show_bug.cgi?id=624384
Reorganize the existing code which parses the -st-shadow property
to allow parsing different shadow properties and add support for
the text-shadow property.
https://bugzilla.gnome.org/show_bug.cgi?id=624384
If a shadow property is inherited from a parent, multiple StThemeNodes
share a common StShadow. It would be possible to use st_shadow_copy()
for this purpose, but proper reference counting is nicer.
https://bugzilla.gnome.org/show_bug.cgi?id=624384
Add basic support for background-position, which only supports absolute
values.
Also don't require an unit to be specified for 0 (because the unit does not
really matter here 0 is 0 regardless of the unit).
https://bugzilla.gnome.org/show_bug.cgi?id=624375
Add st_theme_node_paint_equal() and use that to do two things:
1) Avoid animating transitions where nothing changes.
2) Copy cached painting state from the old theme node to the new
theme node.
https://bugzilla.gnome.org/show_bug.cgi?id=627083
Creating an FBO may be expensive, so we should avoid the operation
if possible. When transitioning between theme nodes, the widget's
opacity is used to paint to the offscreen textures which are blended
together - this means that the textures have to be recreated each time
the widget's opacity changes. It is much more effective to paint the
textures at full opacity and respect the widget's paint opacity when
blending the textures together.
https://bugzilla.gnome.org/show_bug.cgi?id=627085
The current search system uses the OR operator to concatenate search
terms. While results which are matched multiple times sort before
other matches, it is almost guaranteed that adding an additional term
to the search increments the number of results, which is rather
surprising.
https://bugzilla.gnome.org/show_bug.cgi?id=610955
gdk_screen_get_rgb_colormap() has been removed from GTK+; there's
no longer a special visual that is used for drawing true-color
images distinct from the system visual. So, we don't need to
check for it; if the visual isn't the system visual, we just
fall through to the case where we create a new GdkColormap.
The function has been upstreamed as clutter_actor_contains() - with
the switch to clutter-1.4 it is now available to the Shell, so it
is no longer necessary to keep a copy in-tree.
https://bugzilla.gnome.org/show_bug.cgi?id=626512
Add support for the CSS :first-child and :last-child properties to
StContainer, and thus to all containers in St and ShellGenericContainer.
The internal ordering of the container's children is used to determine
the child to which to attach the pseudo class, not the children's
positions. This means that containers where positions can differ from
the ordering (ShellGenericContainer / StGroup) may behave unexpectedly,
so some caution is required.
https://bugzilla.gnome.org/show_bug.cgi?id=625316
When we are manually starting dconf in the case where it can't
be activated via D-Bus, we need to set LD_LIBRARY_PATH in its
environment, since dconf is no longer linked with a rpath.
(Unlike the rest of GNOME, it doesn't use libtool.)
There's an assertion in calculate_gaussian_kernel() to avoid a division
by zero - due to an unnecessary cast from float to int this assertion
is triggered incorrectly for small (but non-zero) values, e.g. a blur
radius of 1px.
Recently, g-ir-scanner got a little stricter with regard to
GSList, marking functions not introspectable if the element
type is not annotated. This fixes a visible JS Error by g-ir
and gdmuser/gdm-user-manager.c
Add SHELL_STAGE_INPUT_MODE_FOCUSED, to move the keyboard focus to the
shell without grabbing the keyboard or mouse, and make
stage_input_mode into a GObject property so that (among other things),
callers can tell when MODE_FOCUSED reverts back to MODE_NORMAL.
https://bugzilla.gnome.org/show_bug.cgi?id=623429
g-ir-scanner is currently buggy and confuses the Gdk.Rectangle alias
with MetaRectangle. Since this is moderately hard to fix in
gobject-introspection and the fix would conflict with in-progress
changes, work around by doing a 'sed job' on the generated Meta.gir.
https://bugzilla.gnome.org/show_bug.cgi?id=623640
The conversion of GdkRectangle to a typedef for cairo_rectangle_int_t
in GTK+-3 makes it no longer a proper boxed type (it's still registered
boxed, but gobject-introspection doesn't know that.) So, switch to using
MetaRectangle, which is now registered as a boxed type by Mutter.
https://bugzilla.gnome.org/show_bug.cgi?id=621845
"text-align" allows setting the alignment of text, with respect to
other lines and allocated space, without requiring a reference to
the ClutterText (which is private for most widgets).
If not specified, all text is left-aligned.
https://bugzilla.gnome.org/show_bug.cgi?id=622447
* Add dconf and dependencies to the moduleset and add
libxml2 dependency to gnome-shell-build-setup.sh
* if we can't ping the dconf D-Bus service, try to
activate it manually.
* Stop forcing GConf via GSETTINGS_BACKEND
https://bugzilla.gnome.org/show_bug.cgi?id=622308
It has probably crossed the line to evil by a mile or so, but here
it is: a Tweener.slowDownFactor replacement used by all animations
without exception.
While at it, update Tweener to use the new setTimeScale() upstream
function instead of adjusting the timeline directly.
https://bugzilla.gnome.org/show_bug.cgi?id=622249
Windows are only added to an application if they are considered
"interesting". If we keep it that way, we cannot unconditionally
call _shell_app_remove_window() - applications without interesting
windows are not considered running, so the call crashes the shell.
https://bugzilla.gnome.org/show_bug.cgi?id=622236
Use GSettings for all Shell configuration. GConf is kept to read
configuration from external programs (Metacity, Nautilus and Magnifier),
but ShellGConf is removed because it's mostly useless for the few calls
we still have. Also get rid of unused GConf code in ShellAppSystem.
A basic GConf schema is still used to override Metacity defaults and
configure Magnifier in a system-wide fashion. GConf is also used as
GSettings backend via the GSETTINGS_BACKEND environment variable.
All of this will be removed when these programs have been ported
to GSettings and able to use dconf.
GLib 2.25.9 is required. Schemas are converted to the new XML format,
and compiled at build time in data/ so that the Shell can be run from
the source tree. This also requires setting the GSETTINGS_SCHEMA_DIR
environment variable both when running installed or from source tree,
in src/gnome-shell.in and src/gnome-shell-clock-preferences.in.
https://bugzilla.gnome.org/show_bug.cgi?id=617917
This is a small memory usage optimization, and cleans up the code.
In particular, this will help for later patches which perform
more substantial operations on running apps.
https://bugzilla.gnome.org/show_bug.cgi?id=621203
Nothing uses ShellOverflowList any more, so remove it. (Even if we
want that sort of functionality again later, we'd want it to be
StWidget-based, and with a name less likely to be confused with
StOverflowBox.)
https://bugzilla.gnome.org/show_bug.cgi?id=621582
This function reads a single image file which contains multiple
images internally.
The image file will be divided using @grid_width and @grid_height;
note that the dimensions of the image loaded from @path
should be a multiple of the specified grid dimensions.
https://bugzilla.gnome.org/show_bug.cgi?id=598349
The framebuffer setup code can fail, e.g. when the texture used as
color buffer has a height/width of 0. In that case, the call to
cogl_pop_framebuffer() will crash the shell trigger an assert. Add
error checking to fail gracefully.
Add _st_actor_contains() in st-private for use within St, and
monkey-patch in a Clutter.Actor.contains() for use by javascript, and
then replace all the duplicate implementations with one or the other
of those.
https://bugzilla.gnome.org/show_bug.cgi?id=621197
Our behavior of reversing the animation when the widget's style
changes back to the previous one is sound. On the other hand, when
there's a change to a new style while a transition is active, we
simply cancel the ongoing transition. Updating the transition
correctly so that the new one starts from an intermediate state
is hard.
Nevertheless, if the style changes before any time of the transition
has elapsed, we should do better than the current behavior.
https://bugzilla.gnome.org/show_bug.cgi?id=621140
First, we were passing an incorrect timestamp to
meta_display_focus_the_no_focus_window - fix that.
The invocation of set_focus_app to the started app there couldn't
really work, because (if the above call had worked) we'd get the
X reply *after* the started app.
What we need to untangle here is the distinction that's now made in
ShellApp between _STATE_STARTING and _STATE_RUNNING. A nice way to
start doing this is to rebase ShellWindowTracker to only be concerned
with app states. Concretely, the current "has windows implies
running" logic now lives just inside shell-app.c.
Rename the app-running-changed signal to be app-state-changed. This
will ultimately be useful so that inside the panel, we can track
the last started app.
https://bugzilla.gnome.org/show_bug.cgi?id=620899
The current code only sets up / updates transitions when the new
node's transition-duration is non-zero. It should cancel an
existing transition if the duration is 0.
It is sometimes desirable to fade smoothly between two styles
instead of changing it abruptly.
Add transitions controlled by the transition-duration CSS property.
https://bugzilla.gnome.org/show_bug.cgi?id=619025
StThemeNodes may have properties - namely shadows - which paint
outside an actor's allocation. This is not a problem unless drawing
is redirected to an offscreen buffer, in which case the actually
painted size is needed in advance when setting up the buffer.
Add a convenience method to calculate an allocation large enough to
paint the node.
https://bugzilla.gnome.org/show_bug.cgi?id=619025
Add st_theme_node_equal() - two nodes are considered equal iff they
refer to identical elements, so e.g. .example and .example:hover are
not equal, even if no .example:hover rule exists in the CSS.
https://bugzilla.gnome.org/show_bug.cgi?id=619025
Previously we were trying to match up remote windows with local
.desktop files, which is definitely wrong. This patch simply
falls back to the app-from-window case for this; better handling
would need design.
https://bugzilla.gnome.org/show_bug.cgi?id=620855
Separate out the main app view into different sections based on the categories
in the desktop file. The configuration is done via gmenu and the desktop menu
specification, we set XDG_MENU_PREFIX="gs-" on startup, so that gmenu reads
gs-applications.menu, which we install.
There is no support for "submenus" - only the menus directly under
Applications will be displayed as categories.
https://bugzilla.gnome.org/show_bug.cgi?id=614131
Currently we only relayout when the screen size changes, this gets
the cases where a monitor gets added/removed but not when the primary
monitor changes.
We need to relayout on all monitor layout changes.
Remove ShellGlobal::screen-size-changed signal as it is no longer used, Gdk is
used to track changes now.
A ShellGlobal::gdk-screen property is added for this purpose.
https://bugzilla.gnome.org/show_bug.cgi?id=620377
When hovering over the entry's ClutterText, the entry itself
currently receives a leave event and updates the hover state.
Apply the same logic as StWidget itself to treat children as
part of the entry.
https://bugzilla.gnome.org/show_bug.cgi?id=620381
Add an event when we receive an event on buffer swap completion; we'll
only get this if Clutter is using the INTEL_swap_event GLX extension,
but it's useful to see the actual timing of video frames.
The recorded event includes the actual timestamp of the swap, since
we are given that in the GLX event - on my system it tends to be
consistently 80-100us before we log the event, but if something was
going wrong in event handling (too much synchronous work), then that
could could show up as a longer delay.
https://bugzilla.gnome.org/show_bug.cgi?id=619516
Currently shadows disregard the overall opacity, so e.g. setting
an ancestor's opacity does not effect the shadow. Fix this by
deferring the setting of the shadow's color until it is painted.
Also move duplicated drawing code from st_theme_node_paint() into
its own function.
https://bugzilla.gnome.org/show_bug.cgi?id=619083
The preference dialog for the panel clock does not set its gettext
domain, so it's not translated even if though translations are
already available.
Fix by setting the domain in the GtkBuilder file and binding the
domain to the locale directory.
https://bugzilla.gnome.org/show_bug.cgi?id=618873
Currently the size of an StGroup depends exclusively on the group's
children - it should be possible to override this behaviour with
fixed values in the CSS.
https://bugzilla.gnome.org/show_bug.cgi?id=613907
Update the GNOME Shell version to 2.31.2, and make corresponding
updates to dependencies:
clutter => 1.2.8
mutter => 2.31.2
gobject-introspection => 0.6.11
gjs => 0.7
Typo: 10000000 instead of 1000000 for a million. Also, use
G_INT64_CONSTANT instead of LL for a 64-bit constant, to avoid
extraneous 128-bit arithmetic on 64-bit machines.
Add a command line option to upload the generated performance report
to a web service. The options for the upload (url, system name, secret
key) are read from ~/.config/gnome-shell/perf.ini.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Switch from having separate METRICS and METRIC_DESCRIPTIONS objects
in a perf module to a single METRICS array. This is done so the
perf module can define the units for each metric.
In addition to improving the output in the web interface, the purpose
of having units is to give some clue about how to pick from multiple
values from different runs. In particular, with the assumption that
"noise" on the system will increase run times, for time values we want
to pick the smallest values, while for "rate" values, we want to pick
the largest value.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Add --perf-output=<filename> option to gnome-shell that combines
the reports written for each run by the C/Javascript code into
a complete report.
If this option is not specified, a brief human-readable summary
is printed to stdout instead.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
When SHELL_PERF_OUTPUT is set, instead of just dumping out the metrics, dump
a more complete report with:
- Event descriptions
- Metric descriptions and value
- Event log
Helper functions shell_perf_log_dump_events() and shell_perf_log_dump_log()
are added to ShellPerfLog to support this. The gnome-shell wrapper is adapted
to deal with the changed report format.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Add a helper function to write a string as UTF-8 to a GOutputStream.
The signature of GOutputStream:
gboolean g_output_stream_write_all (GOutputStream *stream,
const void *buffer,
gsize count,
gsize *bytes_written,
GCancellable *cancellable,
GError **error);
Can't currently be handled by GJS.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Add gnome-shell options:
--perf-iters=ITERS"
Numbers of iterations of performance module to run
--perf-warmup
Run a dry run before performance tests
Make a successful run of a performance test return 0 not non-zero,
and handle the difference between that and a 0-exit in normal
usage (meaning replaced) in the wrapper.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Move the code that starts gnome-shell and waits for it to exit
to a function in preparation for running it multiple times when
doing iterations of a performance test.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Add some basic statistics for allocated memory based on mallinfo(),
and use that to define two metrics:
usedAfterOverview: bytes used after the overview is shown once
leakedAfterOverview: additional bytes used when the overview is
shown a second time.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
Add a facility for defining and recording numeric statistics
as performance events
shell_perf_log_define_statistic()
Define a statistic and the corresponding event type
shell_perf_log_update_statistic_[ix]() -
Update the stored statistic value
shell_perf_log_add_statistics_callback()
Add a function called before collecting statistics
shell_perf_collect_statistics()
Update statistics and record events for them
In addition to the explicit collection, statistics are
recorded at a periodic interval (currently 5s)
https://bugzilla.gnome.org/show_bug.cgi?id=618189
We want to be able to summarize the behavior of the shell's
performance in a series of "metrics", like the latency between
clicking on the Activities button and seeing a response.
This patch adds the ability to create a script under perf/
in a special format that automates a series of actions in the
shell, writing events to the performance log, then collects
statistics as the log as replayed and turns them into a set
of metrics.
The script is then executed by running as gnome-shell
--perf=<script>.
The 'core' script which is added here will be the primary
performance measurement script that we use for shell performance
regression testing; right now it has a couple of placeholder
metrics.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
ShellPerfLog provides a way for different parts of the code to
record information for subsequent analysis and interactive
exploration. Events exist of a timestamp, an event ID, and
arguments to the event.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
To support scheduling performance-measurement scripts that want to run
a number of actions in series, add shell_global_run_at_leisure() to run
a callback when all work is finished.
The initial implementation of this is not that accurate: we track
business in Tweener.js via new shell_global_begin_work(),
shell_global_end_work() functions, and we also handle the case
where the main loop is continually busy.
https://bugzilla.gnome.org/show_bug.cgi?id=618189
The API docs for ShellApp claimed it sorted by the last time the
user interacted with the app, but if one closed a window, then
we would fall back to comparing against a possibly much older
timestamp from another window. Fix this by just keeping a
user time per app.
Also clean up the comparison function to explicitly check the state
instead of deferring to the window list.
https://bugzilla.gnome.org/show_bug.cgi?id=618378
The ShellGlobal initialization performs several actions like connecting
to the X server, ensuring directories exist, etc., that are problematic
because we were creating the object even when running the binary for
introspection scanning. During compilation we may not even have X11
available in e.g. autobuilder type environments, and it's just a
bad idea to connect even if we do.
Avoid this by deferring creation of the ShellGlobal object
until the plugin is actually started.
Now that we're initializing things later, remove the connection to
screen changes, and initialize cached ShellGlobal state at the point
when the plugin is set. The root pixmap actor is now sized initially
on creation too. Instead of relying on screen-size-changed being
emitted on startup, explicitly invoke _relayout().
https://bugzilla.gnome.org/show_bug.cgi?id=618371
