All the mode or visibility changes require `wl_surface_commit` to be
applied. gtk-layer-shell will attempt to force GTK to commit, but may
fail if the surface has stopped receiving frame callbacks[^1].
Thus, we could get stuck in a state where the bar is hidden and unable
to regain visibility.
To address this, a new API has been added to gtk-layer-shell,
`gtk_layer_try_force_commit`, which does `wl_surface_commit` with the
necessary safety checks to avoid corrupting GTK internal state.
Note: this change bumps gtk-layer-shell requirement to 0.9.0.
[^1]: https://github.com/wmww/gtk-layer-shell/issues/185
This fixes a major inconsistency with the swaybar implementation of
these modes[^1]. `overlay` layer no longer has security implications due
to a wide adoption of `ext-session-lock`, so it's safe to use.
Following config will restore the previous behavior:
```json
"modes": {
"hide": { "layer": "top" },
"overlay": { "layer": "top" }
},
```
[^1]: https://github.com/swaywm/sway/commit/2f7247e08a16610228067c9ec34d2b6d897e15fa
Hyprland hasn't been using TCP sockets for IPC since the first release,
so this getaddrinfo call and its result was never needed.
Additionally, it leaks the `aiRes`, causing test failure under ASan.
The bus error when the daemon is not reachable prevents the initial
update and keeps the module visible, as an empty section on the bar.
Do the update explicitly before connecting to set initial visibility.
While we at it, remove a couple of redundant `update()` calls.
```
../include/util/date.hpp:34:26: warning: literal operator suffixes not preceded by ‘_’ are reserved for future standardization [-Wliteral-suffix]
34 | constexpr decltype(auto) operator""d(unsigned long long d) noexcept {
```
While we at it, eliminate use of non-portable GCC conditional expression
syntax. There are no significant side-effects that would justify use of
the language extension.
BREAKING CHANGE: gtk-layer-shell is now required and unconditionally
used. The corresponding config option is removed.
As a part of preparation for future versions of GTK, remove an ability
to use wlr-layer-shell directly. The APIs it required were dropped in
GTK4, and with the menus/tooltips positioning issue being practically
unsolvable it doesn't make sense to keep maintaining the code.
We already use it without checking (`<gio/gdesktopappinfo.h>` in
wlr/taskbar), it's a transitive dependency of GTK and it's always
available on Unix platforms.
The `<experimental/filesystem>` and `-lc++experimental` aren't needed
since LLVM 9.0. And since we now require C++20, checking for the
`<filesystem>` support shouldn't be necessary either.
Previously, the only way to select all the module labels was with the
following kind of selector:
```css
.modules-left > widget > label,
.modules-center > widget > label,
.modules-right > widget > label {
/* ... */
}
```
(and a matching block for the `box` containers).
Now, this can be expressed as
```css
label.module, box.module {
/* ... */
}
```
Use chrono Calendars and Time Zones (P0355R7, P1466R3) when available
instead of the `date` library.
Verified with a patched build of a recent GCC 13 snapshot.
- Add tests for global locale.
- Warn about missing locales.
- Downgrade REQUIRE to CHECK.
- Skip tests if localized formatting does not work as expected.
There were two main issues with fmtlib and C++20 mode:
- `fmt::format` defaults to compile-time argument checking and requires
using `fmt::runtime(format_string)` to bypass that.
- `std::format` implementation introduces conflicting declarations and
we have to specify the namespace for all `format`/`format_to` calls.
The structure was used to pass the locale instance to the date
formatter. All the supported versions of `fmt` are passing the locale
parameter via `FormatContext.locale()` so we can remove the struct and
simplify the code.
While we at it, drop `date::make_zoned` in favor of CTAD on a
`date::zoned_time` constructor.
Linux power_supply sysfs interface allows checking if the battery powers
the whole system or a specific device/tree of devices with `scope`
attribute[1]. We can use it to skip the non-system power supplies in the
battery module and avoid adding HIDs or other peripheral devices to the
total.
The logic is based on UPower, where it is assumed that "Unknown" devices
or devices without a `scope` are system power supplies.
[1]: https://lore.kernel.org/lkml/alpine.LNX.2.00.1201031556460.24984@pobox.suse.cz/T/
As we always use the enum to compare or initialize uint32_t values, it
would be better to declare it with the right type. This way we could
avoid `-Wnarrowing` warnings or unnecessary type casts.
Stop using private implementation details of the `formatter<std::tm>`.
We never needed anything from the class besides the format specifier,
which is easily obtainable with public API.
Allow changing existing modes and adding new ones via `modes`
configuration key.
`modes` accepts a JSON object roughly described by the following type
```typescript
type BarMode = {
layer: 'bottom' | 'top' | 'overlay';
exclusive: bool;
passthrough: bool;
visible: bool;
};
type BarModeList = {
[name: string]: BarMode;
};
```
and will be merged with the default modes defined in `bar.cpp`.
Note that with absence of other ways to set mode, only those defined in
the `sway-bar(5)`[1] documentation could be used right now.
[1]: https://github.com/swaywm/sway/blob/master/sway/sway-bar.5.scd
Intermittent CI failures without any useful diagnostics could be caused
by the OOM killer. 1024MB is not really enough to run 3 parallel jobs
with a modern C++ compiler.
`show_all` call from `Tray::update` attempts to walk the widget tree and
make every widget visible. Since we control individual tray item
visibility based on `Status` SNI property, we don't want that to happen.
Modify `Tray::update` to control the visibility of a whole tray module
only and ensure that the children of `Item` are still visible when
necessary.
On the `Passive` value of `Status` tray items would be hidden unless
`show-passive-items` is set to true.
On the `NeedsAttention` value of `Status` tray items will have a
`.needs-attention` CSS class.
The changes in GCC 11.x made `std::condition_variable` implementation
internals `noexcept`. `noexcept` is known to interact particularly bad
with `pthread_cancel`, i.e. `__cxxabiv1::__force_unwind` passing through
the `noexcept` call stack frame causes a `std::terminate` call and
immediate termination of the program
Digging through the GCC ML archives[1] lead me to the idea of patching
this with a few pthread_setcancelstate's. As bad as the solution is, it
seems to be the best we can do within C++17 limits and without major
rework.
[1]: https://gcc.gnu.org/legacy-ml/gcc/2017-08/msg00156.html
Moving rfkill to the main event loop had unexpected side-effects.
Notably, the network module mutex can block all the main thread events
for several seconds while the network worker thread is sleeping.
Instead of waiting for the mutex let's hope that the worker thread
succeeds and schedule timer thread wakeup just in case.
Aleksei Bavshin