### System details

Intel Core i9-12900KF, NVIDIA GeForce RTX 3080 LHR (nvidia-open-dkms 610.57.04), Omarchy 4.0.2-1 (stable), Hyprland 0.56.2 (`efb50993780079460b0cbed1363e2166a2de1d9f`), kernel 7.1.9-arch1-2. Dual monitors: BenQ 2560x1440 (DP-3) + Samsung QCQ90 3840x2160@30 (HDMI-A-1). Desktop, Gigabyte Z690 UD AX DDR4.

`omarchy-debug --no-sudo --print` captured locally 2026-09-05 if more of it is useful.

### What's wrong?

**Super+T (Omarchy default “Toggle window floating/tiling”) SIGSEGVs Hyprland and kills the session** when the focused window is grouped and its layout target still holds a `CSpace` whose parent workspace has already been destroyed.

This is a Hyprland 0.56.2 compositor crash. Omarchy’s default bindings are the trigger path:

- `SUPER + T` → `hl.dsp.window.float({ action = "toggle" })` in `/usr/share/omarchy/default/hypr/bindings/tiling.lua`
- `SUPER + G` → `hl.dsp.group.toggle()` in the same file
- `SUPER + SHIFT + <n>` → move window to workspace

Hyprland `main` still has the unguarded call (checked 2026-09-05). Hyprland no longer accepts GitHub issues (see hyprwm/Hyprland#16121, same `toggleTargetFloating` → `updateWindows` path, different trigger, bot-closed). Filing here because the default Omarchy tiling/grouping workflow can take down the whole session.

#### What happened

2026-09-05 ~11:52 CDT, Excalibur:

1. Hyprland PID 1415 received SIGSEGV (signal 11, fault address `0x10`).
2. Wayland clients died with it (`hyprsunset` SIGABRT, Synology Drive `cloud-drive-ui` SIGABRT).
3. `uwsm` respawned Hyprland; existing windows were gone. I rebooted from tty5 (`sudo shutdown -r now`).

Not OOM: no oom-killer in the journal.

#### Crash stack (from `~/.cache/hyprland/hyprlandCrashReport1415.txt`)

```
Hyprland received signal 11(SEGV)
Version: efb50993780079460b0cbed1363e2166a2de1d9f
Tag: v0.56.2

#4  CWorkspace::updateWindows()
#5  Layout::CSpace::toggleTargetFloating(...)
#6  Layout::CLayoutManager::changeFloatingMode(...)
#7  Config::Actions::floatWindow(...)
#8  Config::Lua::Bindings::Internal::registerDispatcherBindings
    … lua_pcallk …
#16 CKeybindManager::handleKeybinds(...)
```

#### Forensics from the core (PID 1415)

- Focused window: `foot` (`m_class = "foot"`), mapped on **workspace 5**.
- It was in a single-window `Desktop::View::CGroup`; layout target was `Layout::CWindowGroupTarget`.
- That group target’s `CSpace` (`0x55b37b2edfb0`) was created for **workspace 2**.
- Workspace 2 had since been destroyed (empty non-persistent workspace). `CSpace::m_parent` was expired (`impl_->_data = nullptr`).
- The `CSpace` object itself stayed alive because the group target still held `SP<CSpace> m_space`.
- Super+T then called `CSpace::toggleTargetFloating()` → `m_parent->updateWindows()` with a null `this` (`%rdi = 0`). `CWorkspace::updateWindows()` did `mov 0x10(%rdi), %rdi` and SIGSEGV’d at address `0x10`.

#### The unguarded call

`src/layout/space/Space.cpp` already treats a missing parent as valid in `remove()`:

```cpp
if (m_parent) // can be null if the workspace is gone
    m_parent->updateWindows();
```

`toggleTargetFloating()` (v0.56.2 and current `main`) does not:

```cpp
void CSpace::toggleTargetFloating(SP<ITarget> t) {
    t->setWasTiling(true);
    m_algorithm->setFloating(t, !t->floating());
    t->setWasTiling(false);

    m_parent->updateWindows(); // crash when the workspace is gone

    recalculate();
}
```

The deeper layout bug is that a grouped window moved onto another workspace did not get `assignToSpace()` onto that workspace’s `CSpace`, so it kept the zombie space from the destroyed one. The missing null check is what turns that into a session-killing SIGSEGV.

#### Likely reproduction (inferred from the core; not a second controlled run yet)

1. Open `foot` on workspace 2. Group it (`Super+G`) — even as a single-window group.
2. Move it to workspace 5 (`Super+Shift+5`). Leave workspace 2 empty so Hyprland destroys it.
3. On workspace 5, press `Super+T`.

Workaround that should avoid the crash: ungroup (`Super+Alt+G`) before toggling float or moving across workspaces.

#### Related, not the same bug

- omacom/omarchy#4883 — Super+T crash on a *misplaced special-workspace* window; stack is `bad_variant_access` in `CDwindleAlgorithm::removeTarget` on Hyprland 0.54.
- omacom/omarchy#5914 — Super+O on a pinned window stranded by a cross-monitor workspace move.
- omacom/omarchy#4854 — Super+O after a workspace switch (Hyprland 0.54, closed).
- hyprwm/Hyprland#16121 — same `toggleTargetFloating` → `updateWindows` frames, but SIGABRT on *scrolling-layout drag-end*, closed because Hyprland issues are discussions-only now.

Happy to attach the crash report / a `coredumpctl dump` of PID 1415 if that’s useful. I can also try a controlled re-repro.

— Derek Scherer ([@Dayruke](https://github.com/Dayruke)), Omarchy 4.0.2 on Excalibur

Filed by Grok 4.6 via Grok Build.