### System details

Omarchy 4.0.3-1 - See additional details in Description

### What's wrong?

# Hyprland/Aquamarine crash-loop on boot, plus bar position corrupted to "left"

## Summary

After a routine package install + reboot, Hyprland repeatedly crashed on startup
(3 coredumps in ~10 minutes, identical stack trace each time), leaving the
screen blank/unusable through 5 login attempts. Only a hard power-down
recovered it. On the subsequent clean boot, the top bar (`omarchy-shell`)
came up anchored to the **left** edge instead of **top**, because it started
before any display outputs were available and wrote that bad state to the
user's live config (`~/.config/omarchy/shell.json`), not just a runtime
placeholder.

## Environment

- Omarchy 4.0.3-1
- Hyprland 0.56.2-2 (`efb50993780079460b0cbed1363e2166a2de1d9f`, tag v0.56.2)
- aquamarine 0.15.0-2
- quickshell 0.3.1-1
- Linux 7.2.3-arch1-3
- Hardware: MSI Vector 17 HX A14VHG, hybrid graphics (Intel Raptor Lake iGPU +
  NVIDIA RTX 4080 Max-Q). External monitor connected via a Plugable
  USB4-HUB3A dock (DisplayPort Alt-Mode tunneled over Thunderbolt to the
  Intel iGPU, appears as `DP-4`). Internal panel is `eDP-2`.

## What triggered the reboot

Installing `xpadneo-dkms` 0.10.4-1.1 (unrelated Xbox/Bluetooth controller
driver) pulled in `linux-headers`, rebuilt the DKMS module and the UKI, and
the machine was rebooted afterward as prompted. The controller driver itself
is almost certainly incidental — the failure that followed is entirely in
the display/compositor stack — but noting it since it's what prompted the
reboot that exposed the bug.

## The crash (3x, identical stack trace)

```
Process NNNN (Hyprland) of user 962 dumped core.

Stack trace:
#0 Aquamarine::CDRMBackend::flushAsyncCommitEvents()
#1 Aquamarine::CDRMBackend::cancelAsyncOutput(Aquamarine::CDRMOutput*, bool)
#2 Aquamarine::SDRMConnector::disconnect()
#3 Aquamarine::CDRMBackend::~CDRMBackend()
#4 Hyprutils::Memory::CSharedPointer<Aquamarine::CDRMBackend>::_delete(void*)
#5 Aquamarine::CBackend::~CBackend()
#6 Hyprutils::Memory::CSharedPointer<Aquamarine::CBackend>::_delete(void*)
```

Occurred at 16:52:47, 16:55:15, and 16:58:10 (boot id
`44c5f4327f214ca4a2e5b5e23ef1b2a8`), each time on a fresh Hyprland instance
launched by SDDM after a login attempt. The crash is inside the DRM backend's
own teardown path (`~CDRMBackend()` disconnecting a connector as part of
normal cleanup) — not an unusual physical hotplug event. Between/around the
crashes, other attempts instead failed differently:

```
wayland-session-waitenv.service: start operation timed out. Terminating.
[email protected]: start operation timed out. Terminating.
sddm-helper-start-wayland: Wayland socket watcher timed out
```

Login (PAM auth) succeeded on nearly every attempt — only one incidental
mistyped-password failure in the middle of this. The repeated "had to enter
the password several times" experience was actually the session dying right
after each successful login, not auth failing.

No kernel-level DRM/connector hotplug messages were logged around any of the
three crash timestamps, suggesting the crash trigger is internal to
Hyprland's own shutdown/backend-teardown logic rather than a real electrical
disconnect on `DP-4`.

Recovery required a hard power-down — no clean shutdown path was reachable
with the screen blank. Last log line before the boot ends is
`systemd-logind: Lid opened.`

## The bar-position corruption (separate but related)

On the next (clean) boot, `omarchy-shell` logged, right at startup:

```
qt.qpa.wayland: There are no outputs - creating placeholder screen
Layershell screen does not correspond to a real screen. Letting the compositor pick.
```

i.e. it started before Hyprland had any real outputs ready. As a result, it
wrote `"position": "left"` into `~/.config/omarchy/shell.json`, overwriting
the intended `"top"` (still the default in
`/usr/share/omarchy/config/omarchy/shell.json`). This isn't just a one-time
render glitch — it persisted on disk and stayed wrong across the still-running
`quickshell` process until manually reverted.

## Suggested areas to look at

1. Whatever `CDRMBackend`/`SDRMConnector::disconnect()` is doing during
   backend teardown in aquamarine 0.15.0 that can null-deref/use-after-free
   in `flushAsyncCommitEvents()` — may be worth cross-filing with
   hyprwm/aquamarine or hyprwm/Hyprland directly if this repo doesn't own
   that code.
2. `omarchy-shell`'s handling of the "no outputs yet" startup race — it
   should not persist a bar-position value derived from the placeholder
   screen assignment back into the user's real config file.

## Reproduction

Not reliably reproducible on demand (root cause of the *first* startup
failure that led into the crash loop isn't fully isolated) - occurred once
so far, triggered by a normal reboot after a DKMS package install. Happy to
gather more logs if it recurs; full `journalctl -b <id>` for the affected
boot is available on request while it's still in the journal.