### System details

MacBook Air 9,1 (Intel Iris Plus, i915), Omarchy 4.0.4, Hyprland 0.56, Quickshell 0.3.1 (Arch package), Qt 6.11.2

### What's wrong?

`quickshell` aborts with SIGSEGV on a Wayland event dispatch thread inside `libgdk-3` → `libjemalloc` (heap corruption
or invalid free) while Hyprland headless outputs are created, resized and removed. The same crash class happened on
2026-09-29 13:44 and again on 2026-10-04 01:57. Quickshell restarts itself, but the bar and all panels disappear for a few
seconds and the shell loses its state.

## Environment

- Omarchy 4.0.4 (Arch), Hyprland 0.56, Quickshell 0.3.1 (Arch package, RelWithDebInfo, jemalloc ON), Qt 6.11.2
- MacBook Air 9,1 (Intel Iris Plus, i915), one physical output
- Session environment (from the crash report): `QT_QPA_PLATFORMTHEME=gtk3`, `QT_QPA_PLATFORM=wayland;xcb`, `QS_NO_RELOAD_POPUP=1`

## What happens before the crash

A remote-desktop service on this machine (Omarchy Remote, a WebRTC sender for phone/tablet clients) uses Hyprland
headless outputs (`hyprctl output create headless`, named `OmarchyRemote-<n>`, a new name each time) as the screen for a
session and switches the physical output off for the duration:

```
01:56:54  session ends, physical output back on
01:57:03  headless output removed
01:57:06  next session starts
01:57:08  new headless output created (OmarchyRemote-269838), resized twice (1200x2360 -> 2240x1200 -> 1200x2360)
01:57:23  quickshell: SIGSEGV
```

Quickshell logs `qt.qpa.wayland: There are no outputs - creating placeholder screen` each time the physical output is
off and the headless one is not yet there (01:56:50 .. 01:57:09). Changes to `~/.config/omarchy/shell.json` during the
session (reloads of all plugins) made the window of risk longer; they were reduced afterwards, the crash class is the same
as on 2026-09-29, before that change existed.

## Stack trace (coredump, thread 851232)

```
#0  libc.so.6 + 0x9a17c
#1  raise
#2  quickshell + 0x124389          (crash handler)
#3  libc.so.6 + 0x3e6f0
#4  libjemalloc.so.2 + 0xd3226
#5  libjemalloc.so.2 + 0x1e450
#6  libgdk-3.so.0 + 0x6467e
#7  libffi.so.8 + 0x8db6
#8  libffi.so.8 + 0x4a6d
#9  ffi_call
#10 libwayland-client.so.0 + 0x47cd
#11 libwayland-client.so.0 + 0x56b9
#12 wl_display_dispatch_queue_pending
#13 libgdk-3.so.0 + 0x5a431
#14 gdk_display_get_event
#15 libgdk-3.so.0 + 0x62dc0
#16 libglib-2.0.so.0 + 0x61c76
#17 libglib-2.0.so.0 + 0x63ed7
```

The faulting code is GDK 3's own Wayland connection (it is loaded by Qt's `gtk3` platform theme), handling an
output event (`wl_output` removed/added) in a libffi-dispatched callback, and then jemalloc detects the corruption.
It is not in Quickshell's own Wayland code.

## Reproduction (probable)

1. Hyprland session with Quickshell 0.3.1 and `QT_QPA_PLATFORMTHEME=gtk3`.
2. `hyprctl output create headless <name>`, enable it, disable the physical output, resize the headless output
   (`hl.monitor{...}` with a different mode) a few times, remove it, create another one a few seconds later.
3. After several cycles `quickshell` crashes (not on every cycle).

## Ideas

- Does the GTK3 platform theme need its own Wayland connection in Quickshell at all? Switching the platform theme
  away from `gtk3` for the shell would take GDK out of the output-hotplug path (not tried yet).
- Quickshell side: keeping the placeholder screen alive instead of creating/destroying it when the last output goes away.

Crash report directory on the machine: `~/.cache/quickshell/crashes/cznjqcmt`, coredump kept in
`/var/lib/systemd/coredump/` (`coredumpctl info 851232`). I can attach the report or test a patch.