Currently, modules polling hardware nodes continuously wake up kernel drivers even when the monitor is asleep (e.g., DPMS off), flooding dmesg.
- Adds 'disable-on-sleep' config flag
- Hooks GTK window map/unmap signals to track DPMS state
- Propagates suspend/resume calls to AModule worker threads
Fixes a bug where removing a hot-plugged device mid-loop causes a filesystem error that turns into a fatal runtime error that crashes waybar entirely, changed to warning instead of runtime error.
### Problem
Waybar segfaults on startup when `niri/workspaces` is the only configured
module (default options). Right after the first `WorkspacesChanged` event:
Thread 1 "waybar" received signal SIGSEGV
#0 gtk_label_set_markup ()
#1 waybar::modules::niri::Workspaces::doUpdate () at src/modules/niri/workspaces.cpp:106
### Root cause
[doUpdate()](cci:1://file:///home/lj/Downloads/Waybar/src/modules/niri/window.cpp:27:0-90:1) set the label markup via:
static_cast<Gtk::Label*>(button.get_children()[0])->set_markup(name);
The button's child *is* a valid `GtkLabel`, but gtkmm's `get_children()`
returns it wrapped as a generic `Gtk::Widget` (confirmed: `dynamic_cast`
to `Gtk::Label*` yields `nullptr`). The unchecked `static_cast` then
performs an invalid downcast, producing a corrupt pointer whose `gobj()`
is garbage (`0x1`), so `gtk_label_set_markup()` dereferences it and crashes.
This is reliably triggered when no other module has instantiated a
`Gtk::Label` yet (so the `Gtk::Label` wrapper isn't registered), which is
exactly the case for a `niri/workspaces`-only bar. The same idiom exists in
`sway/workspaces` and `wayfire/workspaces`; it's masked there because typical
configs include other label-using modules.
### Testing
- Reproduced the crash on 0.15.0 / current `master` with a minimal
`"modules-left": ["niri/workspaces"]` config under niri.
- After the fix: no crash; three workspace buttons render with correct
names/labels (`niri-workspace-1/2/3`).
Tray::update() already had logic to hide the module when all child items were passive, but it was only re-run on item add/remove — never on a status change. When the last visible item transitioned to Passive, the item hid itself but the now-empty module remained visible.
Connect Tray to each item's Gtk::EventBox signal_show/signal_hide so update() runs on every visibility transition, and simplify update() to check child->get_visible() directly instead of inspecting the `passive` CSS class. The Tray-level `show-passive-items` read becomes redundant since Item already honours it when deciding its own visibility; remove it along with the now-unused Tray::show_passive_ member.
Fixes: #3721
Move buildLuaDispatch and isLuaProtocol from private to
protected/public so IPCTestHelper can access them.
Add 7 tests covering all buildLuaDispatch branches, dispatch
error path, and isLuaProtocol cache behavior.
Hyprland 0.54 replaced the text-based dispatch socket protocol with a
Lua-based one. Commands like "dispatch workspace 1" are now interpreted
as invalid Lua (return hl.dispatch(workspace 1)), breaking workspace
clicks and scroll navigation.
Add IPC::dispatch() that probes the running Hyprland on first call and
routes commands through the new hl.dsp Lua API when the Lua protocol is
detected, falling back to the old text format otherwise.
Alexis Rouillard