The time computation isn't just unnecessary in that case, it's likely
wrong as well: If we don't have a valid allocation, we may well end
up with a negative value, NaN or Infinity.
Now that redisplaying is a lightweight operation that only
adds and removes what changed, we can not be concerned about
redisplaying on folder changes.
Redisplaying will be necessary when custom order in the app
grid is implemented, in order to update not only which icons
are hidden, but also their position.
FolderView and AllView currently check if the item is
present in the BaseAppView._items map, in order to avoid
adding the same icon multiple times.
Now that BaseAppView._loadApps() has a different role --
it returns a list with all app icons, and BaseAppView
diffs with the current list of app icons -- checking the
BaseAppView._items map is wrong.
Make sure there are no duplicated items in the temporary
array returned by all _loadApps() implementations. Remove
the now unused BaseAppView.hasItem() method.
BaseAppView currently removes all icons, and readds them, every
time the list of app icons needs to be redisplayed. In order to
allow animating app icon positions in the future, however, we
cannot destroy the actors of the app icons.
Previous commits paved the way for us to do differential loading,
i.e. add only the icons that were added, and remove only what was
removed.
Make the BaseAppView effectively implement differential loading.
The BaseAppView.removeAll() method is removed, since we do not
remove all icons anymore. BaseAppView._loadApps() now returns an
array with the new apps, instead of putting them directly at the
BaseAppView lists.
Now that the three views follow the exact same loading routine
(remove all + load apps + load grid), we don't need each view
call loadGrid() directly anymore.
This is an important step in order to animate adding and removing
icons, since now we can diff old and new app icons properly.
Move all calls to BaseAppView.loadGrid() to a single one after
BaseAppView._loadApps(). Also add the underscore prefix, since
this is now considered a protected function.
frequentView: Use BaseAppView.addItem() and loadGrid()
FrequentView is another view that is slightly not unified with how
BaseAppView expects subclasses to load app icons. Instead of using
BaseAppView.addItem() and then calling BaseAppview.loadGrid(), it
adds the app icons directly to the icon grid.
Make FrequentView add icons using BaseAppview.addItem(), and load
the icons using BaseAppView.loadGrid().
Future patches will diff the old and new icons of views, in order to
animate them when necessary (e.g. adding an app icon to a folder, or
from a folder back to the app grid). In order to do that, all views
must be streamlined in how they load app icons.
Currently, FrequentView and AllView are already following the behavior
expected by BaseAppView, but FolderView isn't. Its icons are loaded by
FolderIcon, and FolderView doesn't implement BaseView._loadApps(),
which makes it impossible to diff old and new apps.
Move the app icon loading routine from FolderIcon to FolderView, by
implementing the _loadApps() method.
Florian Müllner [Thu, 1 Aug 2019 23:13:10 +0000 (01:13 +0200)]
cleanup: Use milliseconds for animation times
The different units - seconds for Tweener and milliseconds for
timeouts - are not a big issue currently, as there is little
overlap. However this will change when we start using Clutter's
own animation framework (which uses milliseconds as well), in
particular where constants are shared between modules.
In order to prepare for the transition, define all animation times
as milliseconds and adjust them when passing them to Tweener.
Benjamin Berg [Fri, 19 Apr 2019 22:21:55 +0000 (00:21 +0200)]
data: Update systemd units
The existing units were never used as the corresponding support was
never merged into gnome-session.
This commits updates units to be usable with the newer gnome-session
unit definitions. Also added is appropriate failure/restart logic
including the ability to disable extensions. Note that extensions will
only be disabled if a failure happens in the first two minutes after
login.
Benjamin Berg [Mon, 5 Aug 2019 19:46:05 +0000 (21:46 +0200)]
data: Flag in desktop file that gnome-shell can be started using systemd
As gnome-shell is a required component for GNOME sessions, gnome-session
will currently always try to autostart it. However as we are moving towards
using systemd's user instance for session startup, gnome-session should only
be responsible for launching the shell when either not running under systemd
or when we were built without systemd support.
gnome-session can detect the former but not the latter, so communicate this
via the newly added X-GNOME-HiddenUnderSystemd key in the .desktop file.
Our current Weather integration depends on poking around the app's
settings, which we cannot do when the app is sandboxed (as its
filesystem is "hidden away" in a container in that case).
So instead, use our own GSettings schema for the settings, and sync
it with GNOME Weather via a custom D-Bus interface.
`g_object_notify()` actually takes a global lock to look up the property
by its name, which means there is a performance hit (albeit tiny) every
time this function is called. For this reason, always try to use
`g_object_notify_by_pspec()` instead.
We currently assume that user icons are square, which is the case when
set by the users settings panel, but not enforced by AccountsService.
Handle that case by moving the pixel size back to the actor and using
an appropriate background-size style property of 'cover' (which means
the smallest dimension of the image is scaled to fit the desired size).
A generic, introspectable Shader effect is not only more flexible
than a shader actor, it will also make it much easier to turn
Lightbox into an actor subclass and replace Tweener with Clutter's
own animation support.
Jonas Dreßler [Sat, 27 Jul 2019 17:29:36 +0000 (19:29 +0200)]
shellDBus: Use floats instead of percentages for OSD levels
Due to typecasting being done when converting floats to integers in
gnome-settings-daemon, a volume of 0.9% in g-s-d will end up as 0% in
gnome-shell. This can lead to a mismatch of icons between the volume OSD
(the icon to use is determined by g-s-d itself) and the shells own
volume indicator (the icon to use is determined by the shell using the
volume received from g-s-d).
To fix this, simply get rid of the conversion from float to percentage
in g-s-d and back to floats in the shell and just send a float/double
value on DBus.
verdre [Sat, 2 Feb 2019 16:58:47 +0000 (17:58 +0100)]
osdWindow: Remove LevelBar and use BarLevel directly
LevelBar is not really used, all the checks are implemented inside
BarLevel as well and the accessible name is wrong because the osdWindow
doesn't only show the volume, but also the brightness and other things.
The workaround for updating the bars width is also no longer needed now
that we have BarLevel.
verdre [Sat, 2 Feb 2019 17:01:56 +0000 (18:01 +0100)]
barLevel: Return when trying to set value to already used value
There are some cases (for example when tweening value changes), where
the level will be set to the same value it already is at. Avoid those
unnecessary repaints by checking whether the value is already used and
returning if it is.
The check is supposed to run for all merge requests that touch javascript
files, but for some reason I see it's skipped quite often. Better have
the test run unnecessarily some times than let bugs through, so remove
the limitation.
AppIcon makes itself draggable, and handles the various DnD
routines such as 'drag-begin' and 'drag-end' by making the
Overview emit the appropriate signals.
However, when destroyed, the AppIcon does not try to finish
any drag operations that started. That causes the event
blocker in AllView not to be updated correctly when dragging
icons to outside folders.
Make AppIcon emit 'item-drag-end' when a drag operation
started and it's destroyed.
Florian Müllner [Sun, 30 Jun 2019 11:05:04 +0000 (13:05 +0200)]
ci: Add eslint job to review stage
We have now reduced the number of eslint errors enough to add it to
the CI pipeline. There are still plenty of errors left though, so we
cannot simply run eslint and fail on any errors. So instead, run it
through a fancy script that:
- generates an eslint report using the "regular" configuration
- generates an eslint report using the "legacy" configuration
- creates a combined report with errors common to both configurations
When the pipeline is running for a branch or tag, the final report is
printed out and the job succeeds (we know there are errors left);
when the pipelne is running for a merge request, we fail if any errors
are reported for the lines modified/added by the MR.
GNOME apps use 'review' for the CI stage that generates and exports a
flatpak bundle after a successful build, so they need some other name
for anything that is checked before building; that's how we ended up
with the somewhat awkward 'source_check' stage (inherited from Polari).
But since the commit log check added a 'review' stage that runs pre-build,
use that for all checks that are performed before the build stage.
Florian Müllner [Sat, 29 Jun 2019 13:14:37 +0000 (15:14 +0200)]
perf: Shut up another eslint error
The functions here are asynchronous to handle control back to the
mainloop while waiting for an action to complete, not to run operations
in parallel. That is, the race condition the rule is protecting against
isn't an issue here, so disable the error.
Florian Müllner [Fri, 1 Feb 2019 00:16:20 +0000 (01:16 +0100)]
style: Disable camelcase rule for GObject properties
For GObject properties, we follow the convention of all-lowercase,
dash-separated names. Those translate to underscores in getters/setters,
so exempt them from the newly added "camelcase" rule.
Florian Müllner [Thu, 31 Jan 2019 14:07:06 +0000 (15:07 +0100)]
cleanup: Mark globals used from other modules as exported
eslint cannot figure out that those symbols are used from other modules
via imports, so they trigger unused-variable errors. To fix, explicitly
mark those symbols as exported.
Florian Müllner [Thu, 31 Jan 2019 14:08:00 +0000 (15:08 +0100)]
cleanup: Mark unused (but useful) variables as ignored
While we aren't using those destructured variables, they are still useful
to document the meaning of those elements. We don't want eslint to keep
warning about them though, so mark them accordingly.
Florian Müllner [Mon, 4 Feb 2019 11:30:53 +0000 (12:30 +0100)]
cleanup: Remove some unhelpful unused arguments
Those unused arguments aren't bugs - unbeknownst to eslint, they all
correspond to valid signal parameters - but they don't contribute
anything to clarity, so just remove them anyway.
Florian Müllner [Mon, 28 Jan 2019 00:37:12 +0000 (01:37 +0100)]
lint: Allow marking variables/arguments as unused
Unused variables or arguments can indicate bugs, but they can also
help document the code, in particular in case of signal handlers
and destructuring.
Account for this by keeping the error, but set up patterns that allow
us to opt out of if for individual variables/arguments. For arguments
we pick a '_' prefix, while for variables we go with a suffix instead,
to not accidentally exempt private module-scope variables.
This was left-over in commit 2743f18af, and probably is the real reason
why the busy spinner wasn't using the shared AnimatedIcon.Spinner class:
The animation there was much slower.
Still, let's keep the code as-is for now, if we really need a different
animation time, we can add an optional constructor parameter to the
Spinner class.
Ray Strode [Thu, 18 Jul 2019 18:49:30 +0000 (14:49 -0400)]
appDisplay: Stop watching FolderIcon parent view when destroyed
When a FolderIcon is opened, it asks the parent view to allocate
space for it, which takes time. Eventually, the space-ready
signal is emitted on the view and the icon can make use of the new
space with its popup. If the icon gets destroyed in the
interim, though, space-ready signal handler still fires.
This commit disconnects the signal handler so it doesn't get called
on a destroyed icon.
Ray Strode [Mon, 22 Jul 2019 14:57:57 +0000 (10:57 -0400)]
appDisplay: Add destroy handler for FolderIcon
It is important that the FolderView of a FolderIcon always
gets destroyed before the AppFolderPopup, since the view
may or may not be in the popup, and the view should
get cleaned up exactly once in either case.
This commit adds a destroy handler on FolderIcon to ensure
things get taken down in the right order, and to make sure
the view isn't leaked if it's not yet part of the popup.
Ray Strode [Mon, 22 Jul 2019 15:02:10 +0000 (11:02 -0400)]
appDisplay: Add AppFolderPopup destroy handler
At the moment AppFolderPopup calls popdown on destruction,
which leads to open-state-changed getting emitted after
the actor associated with the popup is destroyed.
This commit handles ungrabbing and closing from an
actor destroy handler to side-step the open-state-changed
signal.
Ray Strode [Mon, 15 Jul 2019 17:52:58 +0000 (13:52 -0400)]
appDisplay: Don't leak duplicate items in AppView
If an icon already exists in an app view with the same id, the
duplicate is not added on a call to addItem. Unfortunately,
since it's not added, the icon actor gets orphaned and leaked.
This commit address the problem by introducing a new hasItem
method and disallowing callers to call addItem with a duplicate
in the first place.
js: Add missing return values to later_add() handlers
meta_later_add() is modelled after g_idle_add() and friends, and
the handler's boolean return value determines whether it should
be scheduled again or removed. There are some places where we omit
the return value, add them (although the implicit return value of
"undefined" already gives us the intended result).
The startup/busy indication in the app menu was left out of commit 22e21ad7d1 because it doesn't use a hard-coded image, but as the
image in the CSS is actually the same used by the spinner class,
drop the "custom" styling and use the regular spinner.
Florian Müllner [Tue, 12 Feb 2019 10:07:55 +0000 (11:07 +0100)]
lint: Enforce consistent return
While all javascript functions have a return value - either an explicit
one from a return statement, or an implicit "undefined" - mixing both in
the same function is almost certainly an oversight, and more often than
not a bug.
Enable the corresponding eslint rule to catch those errors.
Trying to disable an extension that is enabled by the session mode
currently has no effect, which is clearly confusing. We could update
the various extension UIs to reflect that via sensitivity, but being
unable to configure extensions based on which session the user picked
at login isn't obvious either.
So instead, add a 'disabled-extensions' gsettings key to list extensions
that should not be enabled which takes precedence over 'enabled-extensions'
and can be used to disable session mode extensions.
extensionSystem: Move extension loading into ExtensionManager
Now that extension loading and the extensions map are no longer shared
between the gnome-shell and gnome-shell-extension-prefs processes, we
can move both into the ExtensionManager which makes much more sense
conceptually.
Didier Roche [Thu, 1 Nov 2018 12:55:17 +0000 (13:55 +0100)]
extensionPrefs: Switch to D-Bus API to get extension live state
By direclty using the underlying GSetting, whether or not an extension
appears as enabled or disabled currently depends only on whether it is
included in the 'enabled-extensions' list or not.
However this doesn't necessarily reflect the real extension state, as an
extension may be in error state, or enabled via the session mode.
Switch to the extensions D-Bus API to ensure that the list of extensions
and each extension's state correctly reflects the state in gnome-shell.
extensionPrefs: Override getCurrentExtension() for extensions
Extensions are used to calling the getCurrentExtension() utility function,
both from the extension itself and from its preferences. For the latter,
that relies on the extensions map in ExtensionUtils being populated from
the separated extension-prefs process just like from gnome-shell.
This won't be the case anymore when we switch to the extensions D-Bus API,
but as we know which extension we are showing the prefs dialog for, we
can patch in a simple replacement that gives extensions the expected API.
Didier Roche [Thu, 1 Nov 2018 12:50:30 +0000 (13:50 +0100)]
extensionPrefs: Attach extension object to each row
Each row represents an extension, so it makes sense to associate the
rows with the actual extensions instead of linking rows and extensions
by looking up the UUID in the external extensions map in ExtensionUtils.
This will also make it much easier to stop using the shared extension
loading / map in favor of the extension D-Bus API.
Didier Roche [Tue, 4 Dec 2018 08:31:27 +0000 (09:31 +0100)]
extensionSystem: Add canChange property to extensions
Whether or not an extension can be enabled/disabled depends on various
factors: Whether the extension is in error state, whether user extensions
are disabled and whether the underlying GSettings keys are writable.
This is complex enough to share the logic, so add it to the extension
properties that are exposed over D-Bus.
The existing 'ExtensionStatusChanged' signal has a fixed set of parameters,
which means we cannot add additional state without an API break. Deprecate
it in favor of a new 'ExtensionStateChanged' signal which addresses this
issue by taking the full serialized extension as parameter.
Didier Roche [Thu, 1 Nov 2018 12:55:17 +0000 (13:55 +0100)]
extensionUtils: Add functions to (de)serialize extensions
Serializing an extension for sending over D-Bus is currently done by the
appropriate D-Bus method implementations. Split out the code as utility
function and add a corresponding deserialization function, which we will
soon use when consuming the D-Bus extension API from the extension-prefs
tool.
Didier Roche [Wed, 17 Jan 2018 12:43:11 +0000 (13:43 +0100)]
extensionSystem: Add methods to enable/disable extensions
Extensions are currently enabled or disabled by directly changing the
list in the 'enabled-extensions' GSettings key. As we will soon add
an overriding 'disabled-extensions' key as well, it makes sense to
offer explicit API for enabling/disabling to avoid duplicating the
logic.
For the corresponding D-Bus API, the methods were even mentioned in
the GSettings schema, albeit unimplemented until now.
extensionSystem: Make methods to call extension functions private
While public methods to enable/disable extensions make sense for an
extension manager, the existing ones are only used internally. Make
them private and rename them, so that we can re-use the current
names for more useful public methods.
Florian Müllner [Thu, 7 Mar 2019 00:45:45 +0000 (01:45 +0100)]
extensionSystem: Turn into a class
The extension system started out as a set of simple functions, but
gained more state later, and even some hacks to emit signals without
having an object to emit them on.
There is no good reason for that weirdness, so rather than imitating an
object, wrap the existing system into a real ExtensionManager object.
extensionUtils: Move ExtensionState definition here
It makes sense to keep extension-related enums in the same module instead
of spreading them between ExtensionSystem and ExtensionUtils.
More importantly, this will make the type available to the extensions-prefs
tool (which runs in a different process and therefore only has access to
a limited set of modules).
projects / gnome-shell.git / log