It might happen that the target clutter actor that we return on call
of st_texture_cache_load_sliced_image might be destroyed while the
loading task is still running. To protect from this, let's connect
to "destroy" signal and when this happens we use a cancellable to
stop the task.
This allows to safely reuse the return value of this function to
cancel the execution and avoiding that load_callback is called
even for a request that is not anymore under our control.
https://bugzilla.gnome.org/show_bug.cgi?id=765011
When loading an actor for a sliced image actor, we can now use the
REQUEST_CONTENT_SIZE request-mode for the actor since we the content image
has now a predictable size and thus we can be sure that the size will be applied
taking care of the resource scale.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/5
Create StImageContent as a simple ClutterImage with preferred width/height
properties in order to be able to use explicit sizing when creating clutter
contents that will be applied to actors whose size depends on the content itself.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/5
Instead of just passing a scale when getting a cached icon, pass both a
'paint_scale', the scale of which the icon will be painted on the
stage, and a 'resource_scale', the scale of the resource used for
painting.
In effect, the texture size will use the scale 'paint_scale * resource_scale'
in a ceiled value while the size of the actor will use 'paint_scale' when
determining the size.
this would load a bigger texture, but the downscaling would keep the visual
quality.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/5https://bugzilla.gnome.org/show_bug.cgi?id=765011
Create the surfaces for background shadows at scaled sizes and then draw on them
using logical coordinates, by setting the surface device scale accordingly.
Use the said surface scale when generating the actual shadow cairo pattern
but in such case, to reduce the number of code changes, is better to work in
absolute coordinates, and to do so:
1) Create a temporary shadow-spec copy with scaled values to absolute sizes
2) Invert the scaling on the shadow matrix
3) Do the actual painting in absolute coordinates
4) Set the shadow matrix scaling back to the logical coordinates.
Finally scale down the created shadow pattern surface size when painting it,
applying again a reverse scale to the matrix.
https://bugzilla.gnome.org/show_bug.cgi?id=765011https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/5
Pass resource-scale to drawing phase, and use it to create texture
surfaces scaled with the widget current scaling.
Also redraw by default widgets when the resource scale changes.
https://bugzilla.gnome.org/show_bug.cgi?id=765011
A fractional resource scale would mean we never use the fast path for
creating the shadow, because we'd cast the int to a float before
comparing, which would never match.
Instead compare the expected texture size with the source texture, to
actually potentially trigger the fast path.
https://bugzilla.gnome.org/show_bug.cgi?id=765011
The fade shader will draw the fade effect up until the border pixel. If
we set the bottom right coordinate to the outer edge of the pixel we
might end up not drawing the fade effect on all of the pixels. This
could for example happen if one logical pixel (clutter stage pixel)
consists of more than one physical pixel.
https://bugzilla.gnome.org/show_bug.cgi?id=765011
This commit makes StWidget manage the scale of which its associated
resources should be multiplied with. The resource scale is calculated
by clutter, and is retrieved by clutter_actor_get_resource_scale(). Due
to the resource scale not always being available, the getter may fail,
and the actual widget that draws the content will have to deal with
this situation.
As the resource scale depends on where on the stage the widget is drawn,
the resource scale will in general be available once the widget is
mapped.
https://bugzilla.gnome.org/show_bug.cgi?id=765011
instead of explicit screen arguments and `gdk_screen_get_number()`, as nowadays there
is always only one screen. This silences some deprecation warnings and removes
deprecated API.
Bonus: some code style cleanups
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/422
Since commit deec0bf255, the texture cache is based on ClutterImage
rather than ClutterTexture. As ClutterImage (like all ClutterContent)
is only concerned with painting, it doesn't influence the size of the
actor it is added to at all, and the returned actor will now stay at
size 0x0 after the image has been loaded.
Set up the actor to follow the content's size instead, to get closer
to the previous behavior.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/402
As the app menu is being phased out, it is no longer a good indicator
for GtkApplications. Instead, base the check directly on the appropriate
D-Bus properties.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/624
In order to replace GTK+'s GtkDirectionType. It's bit-compatible with it,
too. All callers have been updated to use it.
This is a purely accessory change in terms of X11 Display usage cleanup,
but helps see better what is left.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/317
In order to replace GTK+'s GtkPolicyType. It's bit-compatible with it, too.
All callers have been updated to use it.
This is a purely accessory change in terms of X11 Display usage cleanup,
but helps see better what is left.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/317
It apparently uses no ClutterTexture specifics, and instead
only tries to match what StTextureCache used to return.
Since StTextureCache now returns ClutterActor, also use a
ClutterActor on ShellApp.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/373
_st_create_shadow_pipeline_from_actor creates shadow pipelines
from actors. This function special-cases ClutterTexture as a
small performance improvement, since we can have access to the
CoglTexture easily with it. However, recent commits removed all
usage of ClutterTexture from GNOME Shell, rendering this optimization
useless. Instead, actors now may have a ClutterImage set as
their content, that can be used instead.
Replace the check for ClutterTexture with a check for ClutterImage,
and use the texture of the image when it is available.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/371
The Magnifier class uses a small subtree of actors to track the
current cursor's position and sprite. Specifically, it uses the
deprecated ClutterTexture to paint the cursor sprites.
Add a new, very simple ClutterContent implementation to track the
cursor sprite, and replace the ClutterTexture by a ClutterActor.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/371
This drops usage of Gtk/X11, replacing it with code that is dependent
of the Clutter backend in use. Another positive side effect is that
the keymap state will now be correct on wayland, since there were no
guarantees that X11 key state would reflect the current reality.
https://bugzilla.gnome.org/show_bug.cgi?id=762881
This is an expensive operation that is best avoided in the main loop. Given
the call doesn't care much about returning error or status, it can just
be made async within.
Every operation on a given file will be destructive wrt previous
operations on the same file, so we just cancel any pending operation on
it before batching the current one.
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/815
The last patch in the series, this one adapts StShadowHelper
to received a CoglFramebuffer. This is where we first touch
JavaScript with Cogl types, and as such, it depends on the
latest Mutter. Earlier versions of Mutter didn't have its
Mutter-Clutter GIR to generate types for various Cogl types.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/283
Same case of the previous patch; _st_paint_shadow_with_opacity()
uses cogl_get_draw_framebuffer(), and this patch makes it receive
a CoglFramebuffer as a parameter instead.
The cautious reader might notice that this commit apparently goes
against the long-term goal, for it introduces more instances of
cogl_get_draw_framebuffer(). This is not wrong, but these introduced
ones will be removed later on, when ClutterActor.paint() receives
a CoglFramebuffer as a parameter instead.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/283
This is in preparation for a future where only explicit frambuffer
APIs are available, i.e., cogl_get_draw_framebuffer() does not
exist.
There is absolutely no functional changes in this patch (nor the
following ones in this series), only rearrangements so that various
functions receive a CoglFramebuffer instead of using the draw
framebuffer.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/283
This was called here just to end up emitting ::installed-changed,
which would trigger other g_app_info_get_all() calls. Cache it here
so it may be reused later on.
Handling those events is neccessary if a touch event that pressed down a
button turns out to be a gesture. In this case the button should be
released without emitting the clicked signal.
After loading the GdkPixbuf, StTextureCache unconditionally
creates a ClutterImage and, if it's not in the cache, add
it to the cache. That's a waste of resources when the image
is already committed to the texture cache.
Fix that by reusing the ClutterImage of the cache if it is
already there; otherwise, create a new ClutterImage as we
were previously doing.
ClutterTexture is a deprecated class that is simultaneously
an actor, and the content of the actor. Clutter's new model
is to separate painting (via ClutterContent) from actors.
Currently, StTextureCache relies on ClutterTexture to store
the loaded textures. This not only does not match the latest
practices of Clutter, but also generates various compile-time
warnings.
Port StTextureCache to store ClutterImages instead of storing
ClutterTextures. ClutterImage exposes the internal CoglTexture,
so no helpers are needed to match the current StTextureCache
API. Aspect ratio was dropped, but from my testing, it doesn't
change anything.
All callers have been updated to use MetaSoundPlayer. This drops direct
usage of libcanberra-gtk, and the X11 connection indirectly. One thing
worth noting is that we pass less metadata (eg. event x/y that might be
used for surrounding effects). This was all largely unused, so the
MetaSoundPlayer was made simpler.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/327
When computing the effective border color, we operate on colors with
premultiplied alpha to simplify the calculations, then unpremultiply
the result. However we miss a bounds check in the last check, so any
color component can overflow the allowed maximum of 0xff and shift the
result in unexpected ways.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/305
If the actor is not on the stage yet (i.e. does not have a theme
node), but has a paint state cached, we currently fail to invalidate
it, which will lead to the actor painting with old contents once it
gets onto the stage.
This commit fixes the issue by changing our invalidation strategy;
previously we were looking at the widget's own theme node to determine
if it should be invalidated or not.
Now we look at the theme nodes of our cached paint states. When the
widget is mapped on stage, those are the same as the widget's own
theme node, but when the widget is not on the stage, we'll still be
able to invalidate them.
As part of this, we move the invalidation API to StThemeNodePaintState,
which is a more natural place for our use case.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/314
In X11, pointer emulated touch events are replicated with normal PRESS, RELEASE
pair events which are generated by the server. Thus for a single tap we get:
- TOUCH_BEGIN -> TOUCH_END, PRESS -> RELEASE
This will cause st-button to send two "clicked" signals, instead of just one,
breaking extensions (like dash-to-dock) that show buttons in the main stage
which will be checked two times or that will receive the same signal two times.
Back in the day, there was a proposed system of tracking apps in a
specific context.
The inspiration was that you may have used apps in multiple modes:
Firefox may have been used in both "Programmer Reference" and
"Kitten Videos" contexts. Early user response to the feedback wasn't
too positive - context switching is something that humans have trouble
doing implicitly, let alone explicitly. The old codebase still has a
few remnants of this around; let's finally put them to rest.
Note that we still write out a dummy context tag to the XML file - old
versions of the shell will flat out crash if you don't have one of those
in there, so just leave it in for compatibility sake.
https://bugzilla.gnome.org/show_bug.cgi?id=673767
cogl_texture_new() is used in a few places in GNOME Shell, but
it's a deprecated Cogl function. The replacement is the less
verbose cogl_texture_2d_new_with_size(), that is very much a
straightforward replacement.
Remove the few places where this function is used, replacing
it by the CoglTexture2d counterpart.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/287
StTextureCache installs file monitors that invalidate caches when
contents of the underlying file change.
At the moment, the cache uses the Gio.FileMonitorEvent.CHANGED event
type to make that determination.
However, that is suboptimal for at least two reasons:
- while a file is being written to disk, many CHANGED events will be
emitted in sequence. That will cause needless cache invalidations,
and we will risk loading the file before it's fully loaded.
- if an existing file is replaced, e.g. with g_file_replace(), we may
not get a CHANGED event but a CREATED one instead, so the cache ends
up never getting invalidated.
The good news is that in both of those cases GFileMonitor will send a
CHANGES_DONE_HINT event after changes have settled, or after the file
is replaced.
This commit fixes both cases by switching from the CHANGED event to
CHANGES_DONE_HINT to determine that a file has in fact changed.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/286
If the initialization fails for some reason, for example by
running 'gnome-shell --replace', we should not crash because
of an attempt of unregistering an unregistered agent handle.
Fix that by checking if the handle is not NULL before calling
the unregistering routines.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/66
According to Clutter documentation, "[…] actors implementing the
ClutterContainer interface should override the default implementation
of the class handler of this signal and call clutter_actor_destroy()
on their children."
StBin was doing that in GObject:dispose() instead. Move the child
destruction to a new ClutterActor:destroy() vfunc override.
StBoxLayout implements StScrollable, which, semantically, means that
the StBoxLayout size may not match the minimum size reported by the
layout manager. In this specific case, the layout manager by is a
ClutterBoxLayout by default. For example:
+--------------+
| Viewport |
+------+--------------+-----------------+
| | | |
| | | Content |
| | | |
+------+--------------+-----------------+
| |
+--------------+
So, assuming that:
- ContentSize = the minimum size of the content;
- ViewportSize = the allocated size of the viewport;
When allocating StBoxLayout, it must assume ViewportSize, but must
pass ContentSize to the layout manager. That way, the children of
StBoxLayout are correctly placed within it, even if it's bigger than
ViewportSize.
And here's the problem: right now, StBoxLayout assumes ViewportSize
AND also passes it to layout manager. Commit 77c4c6b6d specifically
exposed this bug by relying entirely on StBoxLayout to arrange the
app and window icons.
Fix that by using ViewportSize to allocate StBoxLayout itself, but
passing ContentSize to the layout manager.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/153
gjs now relies entirely on introspection data to determine parent
types and implemented interfaces, so in order to have all methods
and properties resolve correctly, we must include the corresponding
GIRs of all types used.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/574
There's no relation between a window being hidden from overview/taskbars
and a window not being closable - currently we effectively disable the
fallback quit action for any application with open transients, which
simply doesn't make sense.
Instead, only exclude windows for which the close action has been
explicitly disabled.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/217
Using a single resource file for all JS sources saves a couple of
build system instructions, but has some serious downsides:
- bundling the entire shell code with the tools blows
up their size unnecessarily
- the tools are rebuilt unnecessarily for any shell
code change
Autotools was painful enough to let this slip, but with meson we
don't have any excuses - using the actual dependencies speeds up
the build a tiny bit and reduces the tools' sizes from over 2M
to about 50k.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/192
Simply reusing the same dependencies as gnome-shell itself not only
means that we link tons of stuff unnecessarily, but also that we
have to do the whole mutter rpath dance for nothing. Just use the
dependencies those executables actually need for a nice cleanup.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/192
We need this in the main gnome-shell executable in order to locate
the private Shell and St typelibs, but those aren't useful or even
usable in the extension-prefs/portal helper tools.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/192
GSettings now recognizes per-desktop overrides that can be used
to change schemas' default values for a particular desktop. This
is not entirely unlike our existing custom override mechanism in
mutter, except that it is not limited to keys in org.gnome.mutter,
and it doesn't require a separate schema - the latter means that
we (and gnome-teak-tool) no longer have to figure out the correct
schema for the current login session and just use the original one.
https://bugzilla.gnome.org/show_bug.cgi?id=786496
Add a debug command (to be executed manually via Alt+F2) to check
that all of gnome-shell's file descriptors have the CLOEXEC flag set.
This is important so that internal file descriptors do not get passed
to apps when they are launched.
It prints a warning message for every fd that does not have the flag set.
fdwalk() is used from the standard library if available (it is not
available in glibc), otherwise we use the same implementation as glib
has internally.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/132
When the amount of free memory on the system is somewhat low, gnome-shell
will sometimes fail to launch apps, reporting the error:
fork(): Cannot allocate memory
fork() is failing here because while cloning the process virtual address
space, Linux worries that the thread being forked may end up COWing the
entire address space of the parent process (gnome-shell, which is
memory-hungry), and there is not enough free memory to permit that to
happen. This check is somewhat irrelevant because we are only forking
to immediately exec(), which will discard the whole virtual address
space anyway.
This issue can be avoided by using a new optimized gspawn codepath in
the latest glib development version, which uses posix_spawn() internally.
For the optimized codepath to be used, we must not pass a child_setup
function, so the the file descriptor management is reimplemented here
using new glib API to pass fds to the child process. The old API will
continue to be used on older glib versions.
We must also change the spawn flags for this code path to be hit.
I checked that gnome-shell's open file descriptors are all CLOEXEC
so using G_SPAWN_LEAVE_DESCRIPTORS_OPEN should be safe.
This will result in more resilient app launching when memory is low,
since the optimized spawn path avoids cloning the virtual address
space of the parent process (gnome-shell) and avoids the irrelevant
memory overcommit check.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/132
While can_open_new_window() uses some elaborate heuristics to predict
whether an application can open multiple windows, open_new_window()
will always simply relaunch the application. This is often the best
we can do, but when an application provides a "new-window" action in
its .desktop file or on the bus, it is much more likely to work as
expected than blindly activating the app and hoping for a particular
behavior.
https://bugzilla.gnome.org/show_bug.cgi?id=756844
Graphical applications like GIMP or GIMP allow picking colors from
any location on-screen. In order to keep supporting this feature
on wayland and in sandboxed apps, we will expose an appropriate
method in the Screenshot interface, so first add a corresponding
method to ShellScreenshot.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/286
A custom callback type is more convenient, but only as long as no
other callback type is required. We are about to add functionality
that does not return the filename to a screenshot saved on disk, so
prepare for that by moving to GIO's generic async callback pattern.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/286
So that we can be started by systemd --user, instead of gnome-session.
There are three units:
- gnome-shell.service: Start gnome-shell itself.
- gnome-shell-x11.target, gnome-shell-wayland.target: Sync points for
units that need to care if x11 or wayland is in use.
gnome-settings-daemon will use these, for example.
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/138
All adjustment setter functions take good care of avoiding emission of
notify:: when it's not needed. The set_property() implementation already
calls into the setter functions, so mark the properties as EXPLICITY_NOTIFY
in order to optimize notify:: emission away through g_object_set (rather
common from JS code).
The actor allocation doesn't change per-se, but apply_transform()
will practically transform it. In order to have the paint volume
update accordingly, queue a relayout.
The default get_paint_volume() implementation will do the union
of children, and the child ClutterText paint volume may expand
beyond StEntry size when text overflows.
We actually want all content to be clipped to the StEntry, so
implement get_paint_volume() and tell it so.
And constrain it in StScrollView instead (instead of falling back to an
infinite paint volume, as the actor as paint/pick impls, but no
corresponding get_paint_volume one).
Fixes artifacts with the AppView (and possibly other places) when paint
volumes are aggressively cached.
Remove any usage of MetaScreen, as it has been removed from libmutter
in the API version 3. The corresponding functionality has been moved
into three different places: MetaDisplay, MetaX11Display (for X11
specific functionality) and MetaWorkspaceManager.
https://bugzilla.gnome.org/show_bug.cgi?id=759538
The default keyed_surface is meant to handle CoglTextures thus we can't
add cairo surfaces to it, as the DestroyNotify function won't handle them.
Then the quicker way is to just add another Hash table for handling
such types of textures, with proper destroy function.
This might cause a crash when cleaning up the cache as the hash table has
cogl_object_unref as DestroyNotify function but that assumes that
the passed object is a valid CoglObject.
Fixes: #210
The instance is owned by the actor (being its child), and thus when the
disposal happens for the parent the text is disposed too, thus it's just
safer to nullify its reference so that we won't try to access to invalid
objects later, and this might be the case since the JS objects could be kept
around until they aren't finalized.
https://bugzilla.gnome.org/show_bug.cgi?id=788931
The gnome-shell-calendar-server calls to refresh queries even when it
has no time range set, which results in:
a) waste of resources (for example after login),
b) many runtime warnings in the journalctl log, related to
incorrect time range being used.
https://gitlab.gnome.org/GNOME/gnome-shell/issues/244
While it is theoretically fine to run --perf=hwtest without passing --hwtest,
the test will hang in this case because it fails to start gedit. Always append
Gedit to the default value for --extra-filter when running the "hwtest" test
to make it work fine without --hwtest.
Call _st_set_text_from_style() when updating the entry's style, so
that CSS style properties such as text-decoration or letter-spacing
are applied over the internal ClutterText instance.
If an actor is pending a relayout when get_allocation_box() is called,
the method forces an allocation update. In case of StWidget, this might
then result in a style update and a consecutive invalidation of the
shadow spec.
A helper method that invalidates one of its parameters as a side effect
(and by extension its return value as well) is most unexpected, so cur-
rently _st_create_shadow_pipeline_from_actor() poses an easy trap to
callers to run into.
Remove that trap by calling get_size()/get_position() instead, which
don't have the unintended side effect - it is still a good idea to fix
callers who were running into this to not waste resources on creating
shadows that are invalidated before the next paint, but throwing un-
defined behavior at them is harsh ...
https://bugzilla.gnome.org/show_bug.cgi?id=788908
The imagedir and userdatadir variables are not fetched from
constants/environment but created with g_strdup*/g_build*. Acknowledge that
and also free the strings in finalize.
Whenever a new cursor image is loaded also save it's hot x/y
coordinates. The drawing code already respects these values, however
they were never set. This change will place the cursor image at the
correct location as seen on screen. shell-screenshot.c served as
reference.
https://bugzilla.gnome.org/show_bug.cgi?id=792860
The st_button_release() call wouldn't happen because StButton does not
set priv->button_mask on touch events. And if we make it called, we can't
try to unset the device grab at the end of the function, as device/sequence
are unset earlier on.
After we receive one of the tracked signals, in case we get stuck inside
the gjs_dumpstack () call, we use an alarm to raise the previously emitted
signal, however without using SA_NODEFER, the raise inside the alarm handler
will be ignored.
To avoid to handle new signals caused by the handler calls, once we get the
first signal, we just ignore them all as they could only lead to dirty traces.
Also, cleaning up a bit the code, and disabling the shell log handler
in dump_gjs_stack_alarm_sigaction since this might lead to a new
gjs_dumpstack () request.
https://bugzilla.gnome.org/show_bug.cgi?id=789237
In order to debug issues triggered by JS code we might need to
see the stack of it, this is not normally visible in static stack traces,
thus we need to call gjs_dumpstack () before dying the process.
Intercepting signals SIGABRT, SIGTRAP (needed for catching fatal glib
errors) SIGFPE and SIGIOT by default, while introducing a new
'backtrace-segfaults' flag for the SHELL_DEBUG environment variable to
do the same on SIGSEGV and SIGBUS (this is a precaution to avoid that we
corrupt the stack for automatic errors trackers).
In any case after dumping the stack we raise the signal again, in order
to make the system aware of it.
https://bugzilla.gnome.org/show_bug.cgi?id=789237
Similar to what it's done in _shell_app_add_window(), rely on calling
shell_app_sync_running_state() when removing windows too, to avoid
transitioning to STOPPED while it's still in the STARTING state.
This makes the logic compliant with the startup notification spec and
prevents stopping an application too early if it has a splash screen
that has been closed before the application's main window is shown.
https://bugzilla.gnome.org/show_bug.cgi?id=787905
Meson now warns when specifying the rpath in linker flags instead
of using the newly added build_rpath option - now that we already
depend on the latest release, there's no reason anyway to not use
the nicer syntax.
https://bugzilla.gnome.org/show_bug.cgi?id=786343
When creating a shadow for a ClutterTexture, we currently use the
underlying CoglTexture directly instead of rendering the actor to
an offscreen buffer. This assumes that the CoglTexture is directly
suitable as shadow source, which isn't necessarily the case - it
may have a very different size than what is shown and scaled up or
down by the hardware. In that case we end up with a scaled shadow
texture as well, which messes up the desired blur effect - the
result will be too light when scaling up, or too sharp when scaling
down. To fix this, only take the shortcut when a ClutterTexture's
underlying texture has the correct size and fall back to offscreen
rendering otherwise.
https://bugzilla.gnome.org/show_bug.cgi?id=788039
If a tray icon gets a mapped and unmapped and the mapped again
in quick succession, we can end up with multiple handlers
listening for window creation events.
This commit tries to guard against that by only listening for
window-created events when we don't know the actor associated
with the icon.
https://bugzilla.gnome.org/show_bug.cgi?id=787361
Unlike pango_font_description_from_string(),
pango_font_description_set_family() requires a already properly
formatted font family string. The proper format is a comma seperated
list of font families, but we generated a "comma space" separated list.
Passing a incorrectly formatted font family string to pango seems to
cause wierd issues, where the wrong font is sometimes selected.
For example, this fixes a font selection issue on zh_TW.UTF-8 locale for
chinese characters, where previously the "Droid Sans" font was selected
instead of "Source Han Sans TW" even though fontconfig had placed
"Source Han Sans TW" before "Droid Sans".
https://bugzilla.gnome.org/show_bug.cgi?id=786868
The theme widget passed to shell_tray_manager_manager_screen() is simply
a means to receive style information. There is nothing which ties the two
object's life cycles together, so it is entirely possible for the tray to
be finalized with the widget still around - we shouldn't try to update the
stale object on widget style changes, so make sure we properly disconnect
the signal handler.
https://bugzilla.gnome.org/show_bug.cgi?id=786526
Compiling the generated source for each consumer of the dependency
means we end up trying to register the enum types multiple times,
resulting in a fatal failure on startup. Luckily code outside libst
itself only depends on the header, which doesn't cause those issues.
st_built_sources contains the source and header generated by mkenums,
not any other generated sources. Clarify that in the name, as we are
about to use source and header separately.
We cannot rely on any build order, except the one we specify ourselves.
St depends on various generated files; other targets depend on those
files existing, so they can be included. There is no direct relationship
between targets and files, unless we declare a dependency, using the
Meson declare_dependency() constructor — which allows us to replace the
various `link_with` directives with the more appropriate `dependencies`
one, and also allows us to specify sources that must exist by the time
we build those targets.
In parallel builds we may end up with st-enum-types.c being built inside
separate targets outside of src/st which may not have the ST_COMPILATION
pre-processor symbol defined. For this reason, we need to define it
ourselves in the source file, before including other headers, to avoid
the single-include guard.
This is not strictly necessary when linking the shell with DT_RPATH as
the runtime paths will be extracted from the introspected libraries and
used from gobject introspection to find them when loading them based on
the typelib files, but when linking with DT_RUNPATH, as it's the case
for some linkers (e.g. ld.gold, or ld.bfd in some distros like Debian).
By explicitly prepending this directory as a library path for GIR, we're
making sure that the wrapped shared objects installed in Mutter's and
GNOME Shell's private directories can be found in all cases.
https://bugzilla.gnome.org/show_bug.cgi?id=777519
Meson is on track to replace autotools as the build system of choice,
so support it in addition to autotools. If all goes well, we'll
eventually be able to drop the latter ...
https://bugzilla.gnome.org/show_bug.cgi?id=783229
When using the meson build system, gvc will be built as a subproject.
As meson restricts submodules to a subprojects/ directory in the
source tree root, move the submodule there.
https://bugzilla.gnome.org/show_bug.cgi?id=783229
libmutter-clutter fixed stage capturing when views are scaled, and for
that to work properly when compositing the screenshot, we need to deal
with the cairo device scale here too.
https://bugzilla.gnome.org/show_bug.cgi?id=765011
Even though the API documentation doesn't say so, the underlying
Cogl texture of a ClutterTexture may be unset, so check for that
case to avoid a runtime warning.
https://bugzilla.gnome.org/show_bug.cgi?id=784353
Otherwise UTC is used which results in a different interpretation of
floating time events. This then could lead to a mismatch with events
generated e_cal_client_generate_instances_sync() which always uses the
default timezone. Such a mismatch would then cause constant invalidation
and reloading.
https://bugzilla.gnome.org/show_bug.cgi?id=781950
This allows a full ClutterActor to be used as hint in the entry, instead
of a simple string.
The string case has been now re-implemented on top of the hint actor.
https://bugzilla.gnome.org/show_bug.cgi?id=783484
This is the same as the previous commit, but for StEntry.
We don't have any need to explicitly destroy this actor in our dispose
implementation, and doing so breaks the assumption that we can access
the clutter_text from within destroy.
https://bugzilla.gnome.org/show_bug.cgi?id=783483
There's no need to explicitly destroy the ClutterText actor inside the
label; doing so is actually harmful, as it will break the normal
reference cycle between container and children.
As StLabel doesn't hold any extra reference to the ClutterText actor and
just uses clutter_actor_add_actor() to add it to itself, let the normal
container dispose cycle run to dispose of the reference.
https://bugzilla.gnome.org/show_bug.cgi?id=783483
The static library was split out to avoid duplicate compilation
of mutter-independent libgnome-shell sources when we had a
wayland/non-wayland split. With the separate libmutter-wayland
library long gone, there's no reason to make the build more
complicated than it needs to be, so fold libgnome-shell-base
back into the regular libgnome-shell.
https://bugzilla.gnome.org/show_bug.cgi?id=783206
Most of the libsecret API was declared stable in 0.18 and the stable API
now provides everything we need, so switch from libsecret-unstable to
libsecret-1.
https://bugzilla.gnome.org/show_bug.cgi?id=782637
Commit ffe4eaf00d broke the handler by fetching the instance private
from the wrong actor - as we don't use the ::primary-icon-clicked signal,
and the ::secondary-icon-clicked signal still works by accident, nobody
noticed until now ...
https://bugzilla.gnome.org/show_bug.cgi?id=782190
When extracting the sliced image, the GTask grants data ownership on
g_task_propagate_*, so the pixbuf list must be properly freed. On async
load, we just left a dangling reference when returning on the async
task.
https://bugzilla.gnome.org/show_bug.cgi?id=642652
st_theme_node_get_border_image() may return %NULL, leading to a
segfault in st_border_image_get_file() when glib is compiled with
G_DISABLE_CHECKS.
https://bugzilla.gnome.org/show_bug.cgi?id=780381
The built-in telepathy integration is far less prominent than it
used to be, and doesn't really justify a link time dependency to
send debug message over D-Bus.
https://bugzilla.gnome.org/show_bug.cgi?id=771721
We are still using DBUS_NAME_FLAG_REPLACE_EXISTING from dbus-glib
in some places (whose headers are drawn in via telepathy-glib).
As both GIO and dbus-glib use the values that will end up being
sent on the bus, the define and enum value are interchangeable,
but it's clearly better to consistently use the library that is
explicitly included.
https://bugzilla.gnome.org/show_bug.cgi?id=771721
Rename the signals which have been used to handle XDnd events to more inclusive
ones. So that these signals can be used to handle the DnD events in Wayland.
https://bugzilla.gnome.org/show_bug.cgi?id=765003
Mutter now provides versioned libraries and pkg-config files, meaning
an application using libmutter and friends need to depend on a specific
version of the API.
https://bugzilla.gnome.org/show_bug.cgi?id=777317
Solaris 11.3.5 and later have a Linux-compatible implementation of
/proc/self/cmdline, so the code to use it can be enabled in gnome-shell
on Solaris - if used on an older OS, it simply fails to open the file
and returns without doing anything, just as the code did before enabling
this.
https://bugzilla.gnome.org/show_bug.cgi?id=776199
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
It isn't useful to move the keyboard focus to a hidden actor, so
only include visible actors in the focus chain - this is in fact
the documented behavior of st_widget_get_focus_chain(), so having
the default implementation return all children has always been
unexpected.
https://bugzilla.gnome.org/show_bug.cgi?id=778158
This fixes the shell not picking updates to the contents of the underlying
.desktop files for application launchers, or when a .desktop file is
overriden by a user-installed one under ~/.local/share/applications
https://bugzilla.gnome.org/show_bug.cgi?id=773636
After having embedded a Gtk window, we start to listen for the
corresponding MetaWindow to know when it has been mapped by the
compositor. If the embedder does not realize the window immediately,
and some other window is created and mapped before the embedder
realizes the window, we would prior to this patch try to dereference
the not created GdkWindow. Fix this by NULL checking the GdkWindow
before dereferencing. We will never miss the needed MetaWindow creation
since it can not have been mapped have it not yet been realized.
https://bugzilla.gnome.org/show_bug.cgi?id=776130
This will be required in the upcoming version of GJS.
The reference count on ShellGlobal is 2 at this point, because JS holds a
reference due to the "window.global = Shell.Global.get()" line in
ui/environment.js. Therefore, destroy the GjsContext first, then unref
the ShellGlobal object.
Cleaning up ShellGlobal was previously only enabled behind a debug
environment variable, but it should be required now.
https://bugzilla.gnome.org/show_bug.cgi?id=775374
Sliced images are loaded into a group actor with one child actor
per slice. In case loading the image fails, we currently quietly
return the empty group actor, which makes diagnosing problems
unnecessarily hard - just be a bit more verbose on failure.
https://bugzilla.gnome.org/show_bug.cgi?id=774805
This might be a good fit for extension developers: With
this option one doesn't need to restart the whole Shell
in order to see their changes in effect.
https://bugzilla.gnome.org/show_bug.cgi?id=772593
And adapt existing callers to the new API. This will allow us to
implement a way to launch applications on the discrete GPU for systems
where an "Optimus" system exists.
https://bugzilla.gnome.org/show_bug.cgi?id=773117
You can define a new importer object by importing a subdirectory in GJS.
This is undocumented, but it is likely to at least hold until the whole
thing moves to ES6 modules, after which we'll be able to do this purely
in JS with Reflect.Loader.
Since this was the only thing the ShellJS library did, we can remove it
altogether.
This allows us to discontinue use of the gjs-internals-1.0 embedder API.
https://bugzilla.gnome.org/show_bug.cgi?id=772386
Our window matching currently fails frequently with Flatpak
applications, as one of the primary hints used to link windows
with .desktop files - the WM_CLASS - no longer matches when
flatpak renames the exported .desktop file. Worse, as Flatpak
applications are run in their own PID namespace, different
apps frequently share a common _NET_WM_PID, resulting in
unrelated apps being grouped together by one of the fallback
paths. To match Flatpak applications reliably, take the newly
exported Flatpak ID into account.
https://bugzilla.gnome.org/show_bug.cgi?id=772615
Other windows like the mutter Xwayland selection bridges might deal with
the clipboard, which would result in events visible on st-clipboard event
filters.
In order to avoid unintended results, ignore events that are not meant for
the clipboard window.
https://bugzilla.gnome.org/show_bug.cgi?id=760745
Mutter may capture the screen in multiple capture regions. If this is
done, take these images and composite them onto one large image, which
is then passed into the gstreamer source.
When clutter gives us multiple captures (multiple cairo_surface_t's),
composite them into one large image and use that as final screenshot
result. This makes screenshooting work when mutter uses multiple views.
https://bugzilla.gnome.org/show_bug.cgi?id=770128
intltool is unmaintained nowadays while upstream gettext gained
support for formats like GSettings schemas and .desktop files,
and offers a mechanism to teach it about other XML formats not
yet supported out of the box which we can use for the rest.
So there's nothing stopping us, just make the switch ...
https://bugzilla.gnome.org/show_bug.cgi?id=769074
Negative values don't make sense to timestamps, and an unsigned
timestamp as returned by shell_global_get_current_time() or
Gdk/Clutter events may exceed the range of signed integers.
https://bugzilla.gnome.org/show_bug.cgi?id=769216
In commit 7e0e224e0, when moving from e_cal_recur_generate_instances()
to e_cal_client_generate_instances(), the return value of the
ECalRecurInstanceFn callback was accidentally removed; add it
back.
https://bugzilla.gnome.org/show_bug.cgi?id=769156
There is no longer any guarantee that there'll be one single
framebuffer to read pixels from. In order to still read pixels from the
stage, use the new clutter_stage_capture API.
https://bugzilla.gnome.org/show_bug.cgi?id=768979
Instead of querying the client for a list of objects and using
e_cal_recur_generate_instances() to get occurrences for each of
them, we can use e_cal_client_generate_instances_sync() which
combines the functionality of both functions. This doesn't only
save us some lines of code (yay!), but also gives us access to
the real recurrence ID of an event, so we can get rid of the hack
of faking one.
https://bugzilla.gnome.org/show_bug.cgi?id=748226
We use the triplet of source ID, UID and recurrence ID to create
an ID to unambiguously identify an event, which we use to implement
hiding dismissed events from the calendar. However we currently
try to fetch the recurrence ID from the objects returned by
e_cal_client_get_object_list_sync(), which are always the primary
events with no recurrence ID. Instead, we need a recurrence ID
associated with each occurrence.
https://bugzilla.gnome.org/show_bug.cgi?id=748226
When running as a wayland compositor, MetaCursorTracker might
legitimately not have a displayed cursor because e.g. a client unsets
the cursor surface.
Note that, under an X session, the assumption that the tracker always
has a cursor texture is true because XFixesGetCursorImage() always
returns data even if the cursor isn't visible.
https://bugzilla.gnome.org/show_bug.cgi?id=767001
We're using an unitialized box resulting in an undefined shadow box
size.
_st_paint_shadow_with_opacity() already computes the shadow's bounding
box from the source actor's box so we just need to pass that along.
https://bugzilla.gnome.org/show_bug.cgi?id=767954
on_custom_stylesheet_changed() would set properties_computed to FALSE
without freeing the old properties, then the properties pointer would
be overwritten in ensure_properties().
https://bugzilla.gnome.org/show_bug.cgi?id=710230
Checking offscreen for COGL_INVALID_HANDLE is not sufficient,
as cogl_offscreen_new_with_texture doesn't initialize framebuffer
objects but lets Cogl solve this the lazy way.
cogl_offscreen_new_with_texture will never return COGL_INVALID_HANDLE
anyways.
https://bugzilla.gnome.org/show_bug.cgi?id=764898
For shortcuts that involve a letter (like <ctrl>c), we currently only
accept the lower-case variant. This makes shortcuts awkward to use when
caps-lock is active, and is inconsistent with GTK+, so accept upper-case
variants as well.
https://bugzilla.gnome.org/show_bug.cgi?id=766325
We need to point executables to our private cogl/clutter forks.
Commit 093fd54e2 did this for the main executable, but forgot
the extension-prefs tool and other helpers.
We used to take window visibility into account when comparing apps
until commit 1dfc38d078, following changes in the window switcher
due to auto-minimization. However auto-minimization was abolished
and the window switcher changes reverted, so it makes sense again
to sort apps without non-minimized windows last again.
https://bugzilla.gnome.org/show_bug.cgi?id=766238
We now link to cogl/clutter forks in a private location, so make
sure we set the rpath of executables to point the runtime linker
to the correct location.
While CoglError is a define to GError, it doesn't follow the convention
of ignoring errors when NULL is passed, but rather treats the error as
fatal :-(
That's clearly unwanted for a compositor, so make sure to always pass
an error parameter where a runtime error is possible
https://bugzilla.gnome.org/show_bug.cgi?id=765061
If cogl_framebuffer_allocate fails in _st_create_shadow_pipeline_from_actor, the
CoglOffscreen* that was allocated earlier in the function is leaked.
https://bugzilla.gnome.org/show_bug.cgi?id=735705
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
The non-interactive requests for 'vpn' settings are forwarded to the UI because
it is able to talk to the auth helpers. However, the VPN requests are identified
by the connection type instead of setting type. That is incorrect and the UI
is not prepared to handle such requests; tries to construct a dialog and fails
miserably:
Gjs-Message: JS LOG: Invalid connection type: vpn
(gnome-shell:13133): Gjs-WARNING **: JS ERROR: Error: No property 'text' in property list (or its value was undefined)
NetworkSecretDialog<._init@resource:///org/gnome/shell/ui/components/networkAgent.js:60
wrapper@resource:///org/gnome/gjs/modules/lang.js:169
_Base.prototype._construct@resource:///org/gnome/gjs/modules/lang.js:110
Class.prototype._construct/newClass@resource:///org/gnome/gjs/modules/lang.js:204
NetworkAgent<._handleRequest@resource:///org/gnome/shell/ui/components/networkAgent.js:724
wrapper@resource:///org/gnome/gjs/modules/lang.js:169
NetworkAgent<._newRequest@resource:///org/gnome/shell/ui/components/networkAgent.js:715
wrapper@resource:///org/gnome/gjs/modules/lang.js:169
https://bugzilla.gnome.org/show_bug.cgi?id=760999
This never worked since the code landed but apparently no-one noticed
until now.
The intent here is to return the accessible's default role if none has
been explicitly set on the StWidget instance.
https://bugzilla.gnome.org/show_bug.cgi?id=760945
Commit ffe4eaf00d changed this code to
call st_widget_get_accessible_role() instead of using the value
directly which would be an infinite recursion if that function didn't
have a bug. As it is, this just resulted in
CRITICAL **: atk_object_get_role: assertion 'ATK_IS_OBJECT
(accessible)' failed
https://bugzilla.gnome.org/show_bug.cgi?id=760945
In get_secrets_keyring_cb, we own a ref on the 'attributes' hash table
from secret_item_get_attributes), and a ref on the 'secret' object (from
secret_item_get_secret(), but in the SHELL_KEYRING_SK_TAG case, we unref
these once before breaking out of the loop, and the second time after
breaking out of the loop.
https://bugzilla.gnome.org/show_bug.cgi?id=759708
Since commit 48a54e8ac4, paint() has an explicit framebuffer parameter,
however a couple of submethods are still using the draw framebuffer,
which breaks when rendering to an offscreen buffer.
The polkit agent does not take ownership of the request's cancellable,
so it may no longer be valid after the request finished and we need
to disconnect it before that. Before the GTask port in 8b4249ef2,
it was safe to do that when freeing the request data because requests
completed in an idle, but now we need to do it before returning the
task result.
https://bugzilla.gnome.org/show_bug.cgi?id=757668
If we are trying to render a shadow at a size that is very large in one
direction, but small in the other direction (so that we don't 9-slice
the texture), then allocating the backing texture for the offscreen
buffer may fail due to texture-size limits. Don't crash in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=757150
There are quite a few crashes in retrace.fedoraproject.org that are a result of
of cairo_pattern_get_surface() failing, then a subsequent call to
cairo_image_surface_get_width() crashing because no surface was returned to the
out parameter. Knowing what causes these is hard - my best guess is widgets getting
allocated at ridiculous sizes - but avoiding the crash makes sense in any case.
See https://bugzilla.redhat.com/show_bug.cgi?id=1206754https://bugzilla.gnome.org/show_bug.cgi?id=756983
The text-scaling-factor GSetting was not being properly propagated
to clutter and the Pango font map; under X this is done by Clutter,
which listens to XSETTINGS directly.
https://bugzilla.gnome.org/show_bug.cgi?id=756447
gdk_window_set_composited() is deprecated, and we don't need it as
we end up setting it to the default anyway.
We do need gtk_widget_set_double_buffered() though, which has been
deprecated as well because it is only meaningful on X11 - but so
is XEmbed, so just shut up that warning.
Composite is a hard requirement for mutter's compositor, so if we get
to the point of managing the tray, we already know that the extension
is supported. So let's get rid of yet another deprecation warning.
It has been deprecated for a while in favor of GdkRgba, though in our
case translating ClutterColors to an intermediate color type before
setting _NET_SYSTEM_TRAY_COLORS is a bit silly, so just move to
ClutterColor as the rest of the code base.
Cut down on boilerplate by using the (no longer that) new helper
macros. We don't care about breaking ABI in private libraries, so
use G_DECLARE_FINAL_TYPE even where the class struct used to be
exposed in the header, except for types we inherit from ourselves
(obviously) or where the class exposes any vfuncs (where changes
could affect inheritance in extensions).
Both ShellAppSystem and ShellTrayManager are used as singletons, which
significantly reduces the usefulness of inheritance - it's unlikely for
extensions to inherit from them anyway (AND use any of the vfuncs), so
drop them to allow defining the types as final in an upcoming commit.
Cut down on boilerplate by using the (no longer that) new helper
macros. We don't care about breaking ABI in private libraries, so
use G_DECLARE_FINAL_TYPE even where the class struct used to be
exposed in the header, except for types we inherit from ourselves
(obviously) or where the class exposes any vfuncs (where changes
could affect inheritance in extensions).
They are unused, as we don't use them ourselves and the class is not
exposed to introspection. Drop them to allow defining the type as final
in an upcoming commit.
There is nothing preventing callers from replacing the internal
layout manager, and as long as the replacement is a (or derives
from) ClutterBoxLayout, everything should work fine except for
losing a bit of automatic property mapping - and the latter is
easily fixable by moving the setup out of the constructor.
https://bugzilla.gnome.org/show_bug.cgi?id=708472
We use the newly introduced feature from Mutter to hook up our own
fullscreen and unfullscreen animations.
To give the illusion of a transition as smooth as possible, we create a
snapshot of the current contents of the actor before its state is
changed, and crossfade between the two states while the size changes.
https://bugzilla.gnome.org/show_bug.cgi?id=707248
The JS code could still be holding on to a reference to a window-backed app
after all windows have vanished. (For example, the dash queues an idle to
refetch apps and display them.) Avoid dying with an error message if we
attempt to activate or otherwise manipulate such a window.
https://bugzilla.gnome.org/show_bug.cgi?id=674799
The default pipeline color is opaque white and blending is turned
off. If we only draw with that color (e.g. because animations are
disabled and we're always drawn with opacity == 255), blending is kept
disabled since cogl_pipeline_set_color() returns early if the color
doesn't change from what was there before.
In our case we always want blending to be enabled which we can achieve
by setting the blending string ourselves.
https://bugzilla.gnome.org/show_bug.cgi?id=755827
Since commit 4f1f226828 we only consider buttons clicked when the
release event had a corresponding press event. However as we use
the hover state to check whether a release event actually occurred
on the button, we dismiss any clicks in cases where we missed the
enter event - most likely due to some other actor holding a grab.
Instead, check whether the button contains the event's source, which
should be less error-prone.
https://bugzilla.gnome.org/show_bug.cgi?id=748919
GObject-Introspection started warning for wrong annotations, and
StGenericAccessible::set-current-value has a return value annotation
even if it returns nothing. This generates the warning:
src/st/st-generic-accessible.c:146: Warning: St:
StGenericAccessible::set-current-value: invalid return annotation
Which, coupled with fatal warnings, breaks the Shell build.
It may be 2015, but users still stumble upon the occasional .desktop
file that uses a filename encoding other than UTF-8. We currently
fail quite spectacularly in that case by not displaying any apps at
all - handle this case more gracefully, by only filtering out the
offending apps.
https://bugzilla.gnome.org/show_bug.cgi?id=651503
When the last interesting window of an app-backed window is removed,
we'll transition it back to STOPPED, but we transition the state and
send out the signal before we clear the running state.
This means that any listeners to the state-changed signal might
encounter a window-backed app that has a running state, but no
windows. If they call, e.g. shell_app_get_name, while in this state,
they'll encounter an assertion fail.
Apps that are starting might have uninteresting windows like splash
screens pop up and then go away (like LibreOffice), even when
startup-notification hasn't completed yet. In those cases, we don't
want to transition the app back to stopped -- it should remain in
the running state.
Our StartUpWMClass heuristics use a StartupWMClass -> .desktop ID
mapping built from the list of all installed applications. In case
of multiple .desktop files setting the same StartupWMClass, we
currently simply pick the last one returned by g_app_info_get_all (),
which can be a bit surprising:
A window with WM_CLASS 'emacs', launched through a .desktop file
named 'emacs.desktop' with a StartupWMClass of 'emacs' maps to ...
'emacsclient.desktop'!
Make this case a bit less random by preferring the app info whose
ID matches the StartupWMClass.
https://bugzilla.gnome.org/show_bug.cgi?id=751541
Use our own locking and queue instead of async_queue.
Implement unlock and unlock_stop to make the create function return
FLUSHING. This is important to be able to pause the pipeline after some
error occured in the pipeline.
Implement start/stop to clear the queue and its state.
Currently GNOME Shell doesn't support mnemonics and prompters may
send labels with it.
Remove the mnemonics indicator for now.
Signed-off-by: Stef Walter <stefw@redhat.com>
* Fixed style issue
https://bugzilla.gnome.org/show_bug.cgi?id=750465
libcaribou was designed to generate X events which works under wayland
sessions for X clients but obviously doesn't work for wayland clients
and for shell chrome.
This patch adds a simple caribou display adapter which inherits from
its X display adapter and allows us to continue to work for X clients
and at the same time makes the OSK work on shell text entries by
sending key events directly to the focused text actor.
Making the OSK work for wayland clients requires much bigger changes
at various levels in the stack and either not using libcaribou or
re-working it substantially so that's left for future work.
https://bugzilla.gnome.org/show_bug.cgi?id=747274
We currently don't have any code either in gnome-shell or
gnome-shell-extensions setting margins directly with the Clutter API.
On the other hand, the current behavior doesn't allow us to remove a
style class with margins and have that be reflected, so removing this
special casing seems like the right thing to do at this point.
https://bugzilla.gnome.org/show_bug.cgi?id=746902
NVIDIA drivers don't preserve FBO contents across suspend / resume
cycles which results in broken backgrounds. We can work around that by
forcing a refresh when coming out of suspend.
https://bugzilla.gnome.org/show_bug.cgi?id=739178
The current code assumes that an embedded window will correspond to
a single MetaWindow over the entire lifetime of the icon, which is
incorrect - the embedded window is unmanaged on Unmap and a new
MetaWindow will be created the next time the embedded window is
mapped. As we currently ignore the new MetaWindow completely, it
will be shown normally in the scene instead of the embedded clone
as intended.
Fix this by setting up clone and window actor each time the embedded
window is mapped rather than once in shell_gtk_embed_set_window().
https://bugzilla.gnome.org/show_bug.cgi?id=745824
Each event returned by GetEvents includes the (currently unused)
UID, which is not always enough to unambiguously identify an event
(different calendar sources, recurring events, ...).
As we will start using the property to record events that have been
dismissed and should be persistently hidden from the calendar, change
it to a truly unique ID.
https://bugzilla.gnome.org/show_bug.cgi?id=744927
If users click outside the search entry while it's empty we reset and
thus give up key focus. This means that when using an input method
with candidate popups, interacting with the popup with a mouse click
cancels the current input method context if there's no other text in
the entry besides the preedit string.
To avoid this we can check if the entry has preedit in addition to
checking if it has normal text.
https://bugzilla.gnome.org/show_bug.cgi?id=745167
Commit 5a47ad837b changed the definition of NC_() to include the
message context, so that gettext can correctly disambiguate strings
like "S". However this does only work when gettext actually returns
a translation from the catalogue; if the string is not found, we
need to strip the context again to not have it show up in the UI.
https://bugzilla.gnome.org/show_bug.cgi?id=745521
The mode is never set after the removal of the bottom tray, so it
no longer makes sense to pass it to allowKeybinding(). We can also
safely remove it from the ActionModes flags altogether without
requiring a synchronized update with gnome-settings-daemon, as
the latter never used any flag value above LOGIN_SCREEN.
Commit 1c1f63a7d7 changed the shadow
pipeline to use cogl_framebuffer_ortographic() instead of cogl_ortho(),
but the two functions take their arguments in a different order.
Fixes graphical corruption for text shadows in the login screen.
https://bugzilla.gnome.org/show_bug.cgi?id=745061
When it comes to keybindings or gestures, there's not really a good
reason why popups associated with the top bar should behave differently
from any other shell menus. Just set the action mode generically for
all menus, so actions like screenshots or media-keys start working
with menus like the background- or app launcher context menus.
https://bugzilla.gnome.org/show_bug.cgi?id=745039
This is wasteful, since GResource does not support file monitoring.
Further, doing so will trigger a fallback code path in GLib that polls
every second in a thread, which is doubly wasteful.
https://bugzilla.gnome.org/show_bug.cgi?id=744013
Currently we have been using the vp8 codec because it was the best unencumbered codec at that time. With vp9 we now have a successor that leads to smaller
files at at the same video quality and has been supported by current browsers
for a while.
With the raise of hidpi and 4K displays we need a better codec that handles
those resolutions better, so switch to vp9.
https://bugzilla.gnome.org/show_bug.cgi?id=742744
The lookup table used by get_app_from_window_pid contained only pids of
apps launched by gnome-shell itself, but not pids of apps running before
gnome-shell was (re-)started. Also the pids in that table might not even
be the pid of the process that is actually showing the window if wrapper
scripts are used.
Instead use shell_window_tracker_get_app_from_pid which uses the pids
from the windows themselves.
This removes the only use of launched_pid_to_app which will be removed
in the following commit.
This fixes: https://bugzilla.gnome.org/show_bug.cgi?id=736527
g_dbus_proxy_new() (and library calls that wrap it) has an hardcoded
timeout of 25 seconds, which is insufficient for starting up e-s-r
in certain setups. Avoid a timeout error by starting the service
manually with a longer timeout before hand.
Also demote the error to a warning + exit failure instead of
a crash, to avoid triggering abrt reports.
https://bugzilla.gnome.org/show_bug.cgi?id=735308
Themes - namely the HighContrast one - may now request symbolic
icons rather than fullcolor ones. In order to have recoloring
work as expected in that case, we will need a theme node to pick
up colors from - using an StIcon instead of manually loading a
texture from the cache gives us that for free, so do that.
https://bugzilla.gnome.org/show_bug.cgi?id=740447
GTK+ added support for a -gtk-icon-style property in themes to
enforce a particular icon style. Do the same for shell themes
with an -st-icon-style property, with the same set of possible
values as the GTK+ variant:
'requested' - use symbolic or fullcolor icon depending on the
icon name (default)
'regular' - enforce fullcolor icons
'symbolic' - enforce symbolic icons
https://bugzilla.gnome.org/show_bug.cgi?id=740447
The split between st_texture_cache_load_gicon() and load_gicon_with_colors()
no longer makes any sense, so just move the code into the public method.
https://bugzilla.gnome.org/show_bug.cgi?id=740447
Using a separate property to show when the application is busy rather
than cramming it into the state property makes the code clearer. In most
places we only care if an app is running or not, not whether it is
actually busy.
https://bugzilla.gnome.org/show_bug.cgi?id=736492
This simplifies the code and fixes a race where an application could
call g_application_mark_busy() before the shell subscribed to change
notifications on the application's busy state.
https://bugzilla.gnome.org/show_bug.cgi?id=736492
The window actor shouldn't have been picked since it has an empty
input shape, but bugs in mutter weren't properly interpreting that.
Since the Wayland restructure, the MetaWindowActor isn't actually
reactive anymore, it's the MetaSurfaceActorX11, so this code isn't
doing anything.
It's implemented similar to the padding property, but instead of taking
into account the margin values at drawing time in node-drawing, we set
the clutter actor margins in StWidget when the style is computed.
In the case that a CSS margin is not specified, we don't to set a value
of 0 to the clutter actor margin. In this manner it allows to use
Clutter margin values set in the code. However, the margins that are set
both in the code and in the CSS on the same side, the result is
unpredictable.
We avoid to set the clutter actor margin values to 0 if there's no CSS
margin values defined, so we still allow clutter actors to use margin
set in the code.
https://bugzilla.gnome.org/show_bug.cgi?id=728437
GTK+ added a new PolicyType which currently triggers compiler warnings
about unhandled values in switch statements. We also have a use case for
it already, so add support for the new policy type.
https://bugzilla.gnome.org/show_bug.cgi?id=739379
We commonly mark strftime format strings for translation to account
for date/time representations without an existing strftime shortcut
("Yesterday %H%p"). As those translations are looked up according to
the locale defined by LC_MESSAGES, while the conversion characters
themselves are resolved according to LC_TIME, the result can be
rather odd when mixing locales ("Den 27. January"). The correct
solution would be to install translations for format strings in
the LC_TIME catalogue and look them up with dcgettext(), but we
don't have the infrastructure to do that easily. Work around this
by adding a helper method that looks up a string in LC_MESSAGES
using the locale defined by LC_TIME and use that to translate
format strings, which has the same result.
https://bugzilla.gnome.org/show_bug.cgi?id=738640
The pointer to ->accessible was cleared too early in dispose, which
resulted in another accessible object being created when the actor
was removed from its parent in clutter_actor_dispose(). Use a
weak reference instead to clear the ->accessible pointer.
https://bugzilla.gnome.org/show_bug.cgi?id=738147
Wayland applications don't tend to have very useful WM Class properties,
as GTK+ isn't very good at picking an appropriate application ID. While
we should likely fix GTK+ to pick a better app ID, we do have the
existing gtk_shell for more accurate information. The only problem is
that the gtk_surface is set after the MetaWindow is constructed, and
we're not listening for changes on the GTK+ application ID.
Listen to changes on the GTK+ application ID to fix app tracking for
most GTK+ applications under Wayland.
We currently allow infinite number of screenshot requests to be active at
the same time, which can "dos" the system and cause OOM.
So fail subsequent requests for the same sender when a screenshot operation
is already running.
https://bugzilla.gnome.org/show_bug.cgi?id=737456
These keybindings are well-established on the CLI (e.g. "kill-line"
and "unix-line-discard" in readline(3)), and adding support for them
is cheap ...
https://bugzilla.gnome.org/show_bug.cgi?id=737346
All current code assumes that the list of window actors corresponds to the
list of windows; however as the list returned by meta_get_window_actors()
now includes actors during the destroy animation, that assumption breaks.
Eventually we should make everyone move to a more appropriate API, but
for now make it work again by returning a filtered list of "good"
window actors.
https://bugzilla.gnome.org/show_bug.cgi?id=735927
Performance testing was producing inconsistent values at different
times in the day since the GNOME default background is animated and
sometimes has a single layer, and sometimes two blended layers.
So we have consistent numbers, install a simple animated background
with GNOME Shell that has 40-year long transition ending in 2030,a
and set an environment variable in gnome-shell-perf-tool so that the
background is override with that background. (The background depends
on files installed by gnome-backgrounds; we assume that the person
running performance tests is doing so within the scope of a full
GNOME install.)
https://bugzilla.gnome.org/show_bug.cgi?id=734610
If the application reports itself as single window (through
an explicit indication in the desktop file or some heuristics),
not show a "New window" item that doesn't actually open a new window.
https://bugzilla.gnome.org/show_bug.cgi?id=722554
Add a --hwtest option to gnome-shell-perf-helper which runs the
tests in perf/hwtest.js with the appropriate environment, and then
logs the results using the 'gnome-hwtest-log' utility which is
available in the hardware testing environent.
(For development of hwtest.js in a normal environment, run the tests
as: gnome-shell-perf-tool --perf=hwtest --extra-filter=Gedit)
https://bugzilla.gnome.org/show_bug.cgi?id=732350
It's useful to know how long frames are taking to render on the GPU.
This is impossible to measure in the normal case because frames may
parallelize with previous frames, but by calling glFinish() at the
end of the frame, we can create a (somewhat artificial, but useful)
environment where we have a meaningful timestamp for the frame
finishing drawing.
https://bugzilla.gnome.org/show_bug.cgi?id=732350
Instead of always logging frame timestamps for every frame - which
was using >26 bytes of memory per frame, or 5MB per hour of continuous
redrawing - make frame timestamps something that defaults off and is
turned turned on using a new ShellGlobal::frame-timestamps property by
the perf scripts.
https://bugzilla.gnome.org/show_bug.cgi?id=732350
Add an option for windows created with Scripting.createTestWindow()
to continually redraw themselves; this is for testing performance
of application updates.
https://bugzilla.gnome.org/show_bug.cgi?id=732350
On X11 we don't need to scale up fonts because font scaling is already handled
by clutter based xft-dpi. On wayland we need to set the resolution by ourselves
so do that when the scale factor changes.
https://bugzilla.gnome.org/show_bug.cgi?id=732537
Add a small DBus-activated GtkApplication that embeds a WebKitWebView
and implements some minimal logic to see if the login succeeds.
It will try to connect to a custom NM-provided url (the portal login
page), if one exists, or to www.gnome.org in the normal case of
a portal doing redirect.
https://bugzilla.gnome.org/show_bug.cgi?id=704416
This reverts commit e23c2ffecc.
The patch was intended as a cleanup but accidently removed the setting of the
event base, breaking the swap event handling.
With the event base setting removing the other code isn't much of a cleanup so
just revert it.
Send EOS to the complete pipeline instead of only to our own source.
When there are multiple sources in the pipeline (for example when we also
record audio), the pipeline will send the EOS to all sources in order to
shut down the complete pipeline.
Commit 6c2f3d1d17 moved pref overrides into JS to implement
session mode specific overrides in a clean and generic way.
However that approach comes with a cost - doing the overrides only
after having handled over control to JS means that the core will
be initialized with the non-overridden settings before changing
to the correct values. In the best case this is unnecessary work,
but it can in fact have a worse effect: when initializing workspaces,
we will restore the previous number of workspaces when using
dynamic-workspaces and reset to the configured number otherwise.
As the non-overridden default for dynamic-workspaces is FALSE, we
can easily end up moving the user's windows to the "wrong" workspace.
Now GSettings is expected to grow support for session specific defaults,
which will render our entire override system obsolete (yay!). Given
that, it seems acceptable to use a less generic (and uglier) approach
in the meanwhile, in order to fix aforementioned problems. So move
overrides back before core initialization and just hardcode the
session-mode => override-schema relation.
https://bugzilla.gnome.org/show_bug.cgi?id=695487
Having the full geometry of the menu's source button (if any) will
allow us to address several misbehaviors of window menus, so use
that instead of show_menu().
https://bugzilla.gnome.org/show_bug.cgi?id=731058
Rework the way we re-exec the shell on OpenBSD so that it does not only
work the first time it is re-exec'd.
Plug a small leak in the __linux__ case while here.
https://bugzilla.gnome.org/show_bug.cgi?id=727763
We now allow "appmenu" in the button layout to make synchronizing it
with GTK+'s client-side decorations easier, but as some people tweak
their settings to get in-window app menus even when using the shell,
actually pop up the app menu when the button is activated.
https://bugzilla.gnome.org/show_bug.cgi?id=730752
The switch to GNOME_COMPILE_WARNINGS() caused -Wall and other
warnings to not actually be used since GNOME_COMPILE_WARNINGS()
just sets WARN_CFLAGS. Add WARN_CFLAGS to AM_CFLAGS so that
it takes effect.
Add -Wno-error=deprecated-declarations so that when -Werror is
enabled, we don't fail on all the deprecated cogl and clutter
symbols.
https://bugzilla.gnome.org/show_bug.cgi?id=730408
It can be useful to augment the shell's search path by doing
GNOME_SHELL_JS=resource:///org/gnome/shell:<mypath>
But this doesn't work because resource: is split off. Special
case path elements that are just 'resource' and recombine
them with the next element.
https://bugzilla.gnome.org/show_bug.cgi?id=730409
The hover state of a widget can become persistent if
the widget becomes reactive while a pointer grab.
To avoid that, remove hover state if the reactive property
is disabled.
https://bugzilla.gnome.org/show_bug.cgi?id=728343
We need to use a GdkPixbufLoader instead of the straightforward
gdk_pixbuf_new_from_file(), since we want to load the image already
scaled if possible - e.g. if it's an SVG file.
https://bugzilla.gnome.org/show_bug.cgi?id=726907
Since rebasing our AppSystem on GLib's facilities, we only ever
append to the id-to-app cache. So if an application is uninstalled,
shell_app_system_lookup_app() will still happily return it if it
was cached previously. For instance if a favorite app is uninstalled,
it keeps lurking in the dash until a restart.
To fix, filter out removed apps from the cache when handling
GAppInfoMonitor::installed-changed.
https://bugzilla.gnome.org/show_bug.cgi?id=726414
Currently we update the scale factor on startup and when we get a
monitors-changed signal, which is not the only cases where the setting changes. We cannot listen for gdk-window-scaling-factor changes because it is not
exported to gdk.
So use gtk-xft-dpi which also indicates a scale factor change.
When someone changes gtk-xft-dpi directly without changing the scale factor
we will just re-read the gdk-window-scaling-factor so no harm is done.
https://bugzilla.gnome.org/show_bug.cgi?id=726238
It's possible that FBO creation fails due to hw limits or the
driver not exposing the EXT_framebuffer_object extension.
In that case, just give up on creating square icons.
https://bugzilla.gnome.org/show_bug.cgi?id=724977
Instead of poking through IDLETIME, which confuses the state tracking
and can prevent automatic suspend, send a special signal to GSD
when the screen is to be waken up for a notification.
Someday we'll bring over all the state tracking and avoid this
ping-pong between gnome-shell and gnome-settings-daemon, but
that day's not today.
https://bugzilla.gnome.org/show_bug.cgi?id=712706
While the named commit was correct for VPN connections, it didn't
work correctly for the initial secrets requests like when connecting
to a new access point. In that case, secrets *should* be requested
when none are found, but only if interaction is enabled. The
bits of 17726abb which removed checking secrets against the hints
*were* correct, but 17726abb removed too much.
Also, to ensure passwords don't get inadvertently cleared when
simply reading them from the keyring, don't save passwords
unless something might have changed.
https://bugzilla.gnome.org/show_bug.cgi?id=724779
Currently running the perf tool results into no wm running
afterwards making it hard for the user to get the results from a terminal
and generally does not make it easy for users to run it to gather numbers.
So restore the shell after the test has completed.
https://bugzilla.gnome.org/show_bug.cgi?id=724870
clutter_device_manager_get_core_device calls XIGetClientPointer, which
requires a round-trip to the server. Since we do this on StWidget
creation, this means a full round-trip for every created StWidget.
Replace this with get_device with the ID of the VCP/VCK, since mutter
doesn't support MPX, and we know this is what the device is.
If the sprite is NULL, like if a Wayland app wanted to hide the cursor,
then we need to hide the ClutterTexture on our side, as ClutterTexture
has no easy way to tell it to paint nothing.
If gdk_screen_get_setting fails, like if it's running without XSettings,
then the GValue will have a value of 0. A lot of code tries to divide by
the scale factor. This produces NaN, and combined with the fact that NaN
is "leaky", we very quickly end up spinning out of control.
jsapi.h has some bad warnings with gcc. gjs-module.h already includes
jsapi.h and uses a complicated set of GCC pragmas to mask them out, so
just kill our include.
Since we started tracking non-interesting windows, we can no longer
assume that if we manage to find an app associated with the focus window,
it should appear focused - we now can find apps for docks, the DESKTOP
window etc.
To restore the old behavior, make sure that the focus window or one of
its parents is "interesting".
https://bugzilla.gnome.org/show_bug.cgi?id=722928
The log handler can be invoked at bad times, and in particular
it can be invoked from gsignal with the signal lock taken.
At that time, calling into arbitrary high-level APIs can
cause a dead-lock.
Instead, only send to telepathy the tp-glib debug messages.
Everything else is in the journal anyway.
https://bugzilla.gnome.org/show_bug.cgi?id=724256
So far we have assumed that whether or not a window is interesting
is static. In general this is the case, but as it is legal for the
underlying properties to change at any time, there are of course
offenders that actually do this (flash I'm looking at ya).
While we used the property to determine whether a window should be
tracked or not, the worst case was showing windows that should be
hidden or missing windows that should be shown.
However as we nowadays base an app's running state on the number of
interesting windows, we need to be more careful in order to avoid
ending up with running apps with no windows.
https://bugzilla.gnome.org/show_bug.cgi?id=723308
The code from shell_window_tracker_is_window_interesting() is equivalent
of MetaWindow's skip-taskbar property, so use it to avoid code duplication.
https://bugzilla.gnome.org/show_bug.cgi?id=723308
With the lastest ShellApp changes, an app is considered stopped
when the last "interesting" window is closed. However the app
may still track non-interesting windows, so if we unref the
running state on the state transition, we hit an assertion later-on
when trying to remove the non-interesting window.
Fix this by keeping the running state around until the last window
is closed.
https://bugzilla.gnome.org/show_bug.cgi?id=722840
An app should be considered running if it has at least one "interesting"
window, however the code considers an app running if it has at least
one tracked window. This was fine while we were only tracking interesting
windows, but since commit d21aa0d85f this is no longer the case.
So keep track of the number of interesting windows as well and use that
to determine the running state.
https://bugzilla.gnome.org/show_bug.cgi?id=722690
It is possible to associate an application's window with a different
application using the transient_for hint. However we currently only
consider the hint in get_window_app() and not when making the original
association, which opens the door to some confusing inconsistencies;
for instance, get_window_app() will not necessarily return the same
value for all windows retrieved via shell_app_get_windows().
Fix this by looking at the transient_for hint when making the original
association, not just in get_window_app().
https://bugzilla.gnome.org/show_bug.cgi?id=722434
Using the new list_actions() API in Gio, add entries for static
actions specified in .desktop files in the right-click app menus,
in the dash, app well and search.
https://bugzilla.gnome.org/show_bug.cgi?id=669603
Use a new ShellGLSLQuad actor class to build a RadialEffect that can be
enabled on Lightboxes to achieve a radial effect similar to the overview
one. Then enable it for modal dialogs.
https://bugzilla.gnome.org/show_bug.cgi?id=669798
StWidget::popup-menu is emitted when Menu/<Shift>F10 is pressed,
not released (for consistency with Gtk+), so we need to forward
that. Note that for key press we don't emit the matching key
release, because the app will take a grab and get the event directly
from X when the key is physicall released.
https://bugzilla.gnome.org/show_bug.cgi?id=721267
Since commit 1ebb162a00 moved JS sources into resources,
the extension-prefs tool was broken. To fix it, we would either
need to generate an external GResource in addition to the generated
C code and teach gjs-console about loading it before evaluating
the script, or turn gnome-shell-extension-prefs into a binary with
the JS resources compiled in.
https://bugzilla.gnome.org/show_bug.cgi?id=722334
Specifying the session mode on the command-line doesn't play
well with session management (since the saved session desktop
file well either drop the specified session mode, or force it
always, even if the user picked a different mode at the login
screen)
This commit adds support for specifying the session mode via an
enviroment variable. For now, keep the old command line interface
for backward compatibility
https://bugzilla.gnome.org/show_bug.cgi?id=720894
All get_app_from_*() helpers are transfer full, but
get_app_from_gapplication_id() was directly returning the result
of lookup_app(), which is transfer none.
https://bugzilla.gnome.org/show_bug.cgi?id=721439
The hash table must keep a copy of the IDs, because the GAppInfos
are unreferenced (and thus freed) at the end of the function.
This was possibly not a problem if the GAppInfos were referencing
the memory-mapped cache, but it becomes one for regularly parsed
desktop files in ~/.local.
https://bugzilla.gnome.org/show_bug.cgi?id=721039
get_secrets_keyring_cb() contained an optimization (copied over from
nm-applet) that avoided a D-Bus round-trip when NetworkManager sent
secrets hints that were not satisified by the user. This code did
not properly handle empty hints though, and proceeded to always
request new secrets whenever empty hints were sent. Remove this
code entirely since the complexity is not worth it (per Jasper).
Second, get_secrets_keyring_cb() was mishandling VPN secrets which
were marked as "always ask". Because the VPN secrets are not GObject
properties because they cannot be pre-defined, they are passed in
a hash table that is a GObject property marked 'secret'. Unfortunately,
that means that the shell agent cannot determine their secret flags.
But since the VPN plugin auth dialogs have much better information
about what's required than the shell agent does, always ask the VPN
auth dialogs to handle the secrets requests after grabbing any that
already exist from the keyring. This is also what nm-applet does.
https://bugzilla.gnome.org/show_bug.cgi?id=719815
Filtering out "non-interesting" windows beforehand as we currently do
means that we may get properties that should be based on all windows,
like the last time the application was used, wrong.
Just track all windows and filter out non-interesting windows manually
in the one place we actually care about the difference.
https://bugzilla.gnome.org/show_bug.cgi?id=719824
Sorting actors by the distance in the axis of movement first and against
the axis otherwise means that if we have a situation like:
A F
B
where "F" is the focused actor, and it slightly overlaps with B vertically,
then we'll choose "B" to go left, rather than "A", which is most likely
what the user intended.
This is especially apparent in the overview where slight window size
differences mean we might not get an exact grid shape.
https://bugzilla.gnome.org/show_bug.cgi?id=644306
Rather than scanning all apps for searching, use Ryan's new desktop
file index and the glib support APIs for app searching instead of our
own system.
https://bugzilla.gnome.org/show_bug.cgi?id=711631
While unfortunate that we still have to scan all apps with get_all(),
support for this feature will be short-lived, so hopefully we can drop
it in the future as new apps adapt to the desktop file / app ID
recommendations.
For now, simply scan all desktop IDs.
https://bugzilla.gnome.org/show_bug.cgi?id=711631
Systemd-for-the-user-session would also do this, but that's a deeply
invasive change that I may not actually get to this cycle. This
change is tiny and non-invasive, but provides an important benefit:
You can actually reliably tell *which* applications are logging which
messages (assuming they're launched by the shell).
This actually complements a recent change in DBus:
See https://bugs.freedesktop.org/show_bug.cgi?id=68559
which does a similar thing for bus activated apps.
https://bugzilla.gnome.org/show_bug.cgi?id=711626
Rather than create all ShellApps up-front, create them lazily. We really
had no reason to do this before as we were scanning GMenu to get all the
apps, but doing this can remove a need for get_all, which is slow and
memory-hungry.
We want to transition to a system in the future where we have a desktop
file cache. As we no longer differentiate categories or similar, it no
longer makes sense to have app visibility based on categories. Thus,
we no longer need to use gnome-menus to list all apps. The potential
issue here is reloading all desktop files when new files are created,
but this can be dealt with individually.
The "All Applications" view still uses gnome-menus.
https://bugzilla.gnome.org/show_bug.cgi?id=698486
