Introspecting the dynamic headers is low-level and full of implementation
details.
It seems that gnome-shell was assuming the address of the string table is
given by taking the DT_STRTAB dynamic symbol table's value (d_un.d_val)
and casting it to an absolute pointer. However, in some environments
(apparently including Debian mips64el) the DT_STRTAB dynamic symbol
table's d_un.d_val can be merely an offset relative to the base address
of the executable in memory. The base address would have to be parsed from
a different ELF header, which as far as I know is not part of _DYNAMIC.
Rather than trying to look up the Shell executable's base address using
dlinfo() or similar, during unit testing it seems easier to pass in the
uninstalled library search path in the environment, along with other
uninstalled paths that we are going to need anyway.
The reason given for not using GI_TYPELIB_PATH in commit 9bc89b82
"main: Prepend RPATH or RUNPATH paths to gir search paths" was that
g_irepository_prepend_library_path() takes precedence, but we can easily
define our own environment variable that takes even higher precedence.
If it's set, we might as well also bypass the PKGLIBDIR, since while
running uninstalled we *only* want to use the uninstalled libraries.
When not running unit tests, it's desirable to make
`./build/src/gnome-shell` something you can run directly, so continue
to parse the ELF headers - but only on x86_64, where we're reasonably
confident that it works and any regressions with a newer glibc will
be noticed promptly.
Resolves: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/6877
Bug-Debian: https://bugs.debian.org/1042980
Signed-off-by: Simon McVittie <smcv@debian.org>
We have always defaults to an empty ornament, so that menu items
are always aligned, even when radio items are used.
However radio items are fairly rare, so most of the time we end
up with an extra margin with no purpose. The design team now
prefers radio items to only align with each other, so that regular
items get the expected margin.
Change the defaults accordingly.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2843>
Soon only radio items should use a visible ornament, to avoid
unnecessary extra margins in regular items.
Network items can act as both radio- and regular items, so
update the ornament accordingly.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2843>
Layout items use the ornament to indicate the active layout, so
their ornament should always be NONE or DOT.
The default is about to change to HIDDEN, so explicitly initialize
the ornament to NONE to keep the current radio item appearance.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2843>
We currently consider idle nodes as in use to avoid too fast state
changes and thus flickering of the indicator icon. However, the current
idle timeout is rather long and arguably confusing for users.
Thus switch to only consider running nodes as in use, but delay
disabling of the indicator by 500ms.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2854>
Settings no longer exposes a slider for the keyboard brightness,
leaving keyboard shortcuts as the only way of adjusting it.
This is good enough in most cases, because devices with keyboard
backlight usually include corresponding keys on their keyboard.
However for devices without those keys, it would be good for the
settings to be exposed somewhere again. Quick settings seems like
a more appropriate place than "proper" Settings, since it gives
quick access that doesn't require a major focus change.
https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/6765
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2820>
Extensions must now export a class with a fillPreferencesWindow()
method in their prefs. That is less convenient for extensions
with simple preferences than the old buildPrefsWidget() hook, as
they must wrap their widget in page/group widgets.
Address this by adding a default fillPreferencesWindow() implementation
that calls a getPreferencesWidget() method and wraps it as necessary.
This is flexible enough to support different cases fairly conveniently,
from simple single-widget prefs over tweaking the window to complex
multi-page prefs:
```js
class SimplePreferences extends ExtensionPreferences {
getPreferencesWidget() {
return new SimplePrefsWidget();
}
}
class TinkerPreferences extends ExtensionPreferences {
getPreferencesWidget() {
return new SimplePrefsWidget();
}
fillPreferencesWindow(window) {
super.fillPreferencesWindow(window);
window.set_default_size(123, 456);
}
}
class FullPreferences extends ExtensionPreferences {
fillPreferencesWindow(window) {
const page1 = new GeneralPage();
window.add(page1);
const page2 = new FoobarPage();
window.add(page2);
}
}
```
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
Use the new defineTranslationFunctions() method from the previous
commit to create gettext functions for the module, instead of
re-exporting from the shared module.
It is now up to extension developers to use the more effective
```js
import {Extension} from 'etensions/extension.js';
const {gettext: _} =
Extension.defineTranslationFunctions(import.meta.url);
```
or the more convenient
```js
import {Extension, gettext} from 'extensions/extension.js';
const _ = gettext;
```
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
The method can be used to define a set of gettext functions that
call the corresponding method of an extension.
Those functions are very similar to the gettext functions we are
exporting, except that:
- looking up the extension is delegated to the
Extension/Preferences class
- it is possible to avoid examining the stack
when called with `import.meta.url`
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
With convenience API like getSettings() now being provided by
the ExtensionObject subclass, extensions will need to access
their entry point more often.
Having to pass a pointer through the hierarchy can be annoying,
so add a static method that allows them to look it up:
```js
const ext = Extension.lookupByURL(import.meta.url);
this._settings = ext.getSettings();
```
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
The Extension/Preferences classes are where we will focus for
future extension convenience API, so developers should be
encouraged to use them as entry points.
Adjust the existing templates to do that.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
Extensions now must export a class that conforms to a particular
interface both for the actual extension as well as for prefs:
enable()/disable() methods for the former, fillPreferencesWindow()
for the latter.
This is quite similar to the previous method-based entry points,
but it also gives us a more structured way of providing convenience
API in form of base classes.
Do that in form of Extension and ExtensionPreferences classes on
top of a common ExtensionBase base class.
getSettings(), initTranslations() and the gettext wrappers are
now methods of the common base, while openPreferences() moves
to the Extension class.
Based on an original suggestion from Evan Welsh.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
Extensions often need to set up additional resources from their
directory, like settings, translations or image assets.
So far extensions have used getCurrentExtension() to access the
shell's internal extension object which contains path and dir
properties. That's far from ideal, first because it requires
generating a stack to figure out the current extension, and
second because the internal object also contains state that
extensions shouldn't meddle with.
Just include those properties in the metadata we pass to the
extension constructor.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2838>
Pipewire allows us to easily track whether any cameras are in used by
checking the state of camera nodes. Add a simple camera monitor to the
shell, allowing us to show e.g. a status indicator.
Naturally the monitor is limited to apps using Pipewire for camera
access and thus subject to the same chicken-egg problem like the camera
portal - it could confuse users that apps may use the camera without
being noticed by the monitor. The hope and assumption here is that a
better shell integration might speed up adoption of the new camera APIs
Pipewire 0.3.49 is required for refcounted `pw_init()`/`pw_deinit()`.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2840>
Use the new privacy indicator class for the input one and move it next
to the other privacy indicators.
While on it move all privacy indicators to the front, following the
system-status-indicators mockup.
Part-of: <https://gitlab.gnome.org/GNOME/gnome-shell/-/merge_requests/2840>
