## What happened

`omarchy restart shell` killed the running Omarchy shell over Quickshell IPC. During teardown the process SIGSEGV'd. Quickshell's crash handler restarted a replacement; `omarchy-restart-shell`'s `quickshell kill` loop then killed that replacement too, and it died on the same stack. A third launch a few seconds later stayed up. The bar flickered and came back. No user data was lost.

This is a shutdown-path crash, not an idle crash.

## What was expected

`quickshell kill` / `omarchy restart shell` should exit the shell cleanly, then start a new instance.

## Steps to reproduce

1. Omarchy shell running (`quickshell -n -p /usr/share/omarchy/shell`).
2. Run `omarchy restart shell` (which loops `quickshell kill -p "$CONFIG_DIR" --any-display` until no instance remains, then `hyprctl dispatch` of `omarchy-launch-shell`).
3. Journal shows `Exiting due to IPC request`, then SIGSEGV.

Reproduced twice in one restart (PIDs 1690679 and 1692614, four seconds apart, matching stacks). Not retested after removing a third-party plugin that was loaded at the time (see below).

## Evidence

Journal, 2026-09-15 16:41 MDT:

```
16:41:21  omarchy-shell: INFO: Exiting due to IPC request.
16:41:21  omarchy-shell: WARN qml: data-goblin.fileblade: drive watcher failed (1): Backend stopped
16:41:22  systemd-coredump: Process 1690679 (quickshell) ... signal 11/SEGV
16:41:22  omarchy-shell: ERROR: Quickshell has crashed under pid 1690679
16:41:22  omarchy-shell: ERROR: Quickshell has been restarted.
16:41:26  second SIGSEGV, PID 1692614, same stack
16:41:29  third IPC exit; replacement PID 1694054 survived
```

`coredumpctl info` for both dumps: signal 11, `si_code: SI_TKILL`. That is Quickshell's crash handler catching the original fault and re-raising. The original work is the Qt frames below.

### Stack (Quickshell crash report, PID 1690679)

Config: `/usr/share/omarchy/shell/shell.qml`. Report: `~/.cache/quickshell/crashes/0ph4ch2flt/report.txt`.

```
#4  QQuickWindowPrivate::polishItems()     libQt6Quick.so.6
#6  QQuickWindow::event(QEvent*)
#8  QPlatformWindow::deliverUpdateRequest()
#17 libQt6QmlWorkerScript.so.6             (nested event-loop spin)
#19 QQmlEngine::~QQmlEngine()
#23 QObject::~QObject()
#25 QObject::event(QEvent*)                (deferred delete)
#33 QEventLoop::exec
#34 QCoreApplication::exec
```

PID 1692614 (`~/.cache/quickshell/crashes/0ph4ycgflt/report.txt`) is the same path.

gdb on the systemd core confirmed `<signal handler called>` between Quickshell's re-raise and the `libQt6Quick` polish frames. Arch debuginfod was enabled but did not finish symbolizing Qt internals in time; unnamed frames are left unnamed. They sit in `libQt6Quick` above `polishItems` and in `quickshell` itself.

## Mechanism (proved vs inferred)

**Proved:** IPC kill starts teardown; `QQmlEngine` is destroyed; a nested event loop still delivers a pending window update; `polishItems` then SIGSEGVs.

**Inferred:** destroy-engine-while-a-`QQuickWindow` still has a polish/update queued. `omarchy-restart-shell`'s kill loop makes that two crashes instead of one, because the crash handler respawns an instance the loop then kills.

## Third-party code in the address space

`data-goblin.fileblade` was loaded in the dying process and logged `Backend stopped; retry in 360 ms` during exit. The crashing frames are Qt/Quickshell C++, not plugin QML. Fileblade was removed after this dump; this restart has **not** been reproduced on a stock plugin set. Do not treat Fileblade as the root cause from this report alone.

Not a duplicate of:

- #10098 — `IpcHandler::updateRegistration` on a replacement that is still starting
- #8547 — `QQuickRepeater` teardown on Hyprland workspace moves, not an IPC kill

## System

| | |
|---|---|
| Omarchy | `omarchy-dev 4.0.0.r2147.g2fbac0c-1` (edge) |
| Quickshell | 0.3.1-1 |
| Qt | 6.11.2 (`qt6-base` 6.11.2-3, `qt6-declarative` 6.11.2-2) |
| Hyprland | 0.56.2-3 |
| Kernel | 7.2.5-3-omarchy |
| GPU | Intel TigerLake-LP GT2 / i915 (`8086:9a49`) |
| Machine | Dell Vostro 14 5410, 16 GiB |

`omarchy-debug --no-sudo --print` log (24h): https://logs.omarchy.org/11w9pR

Cores remain in systemd-coredump for PIDs 1690679 and 1692614 on this machine.

Filed by Grok 4.6 via Grok Build.