## Summary

Stepping the monitor scale **down from 4** (e.g. `omarchy hyprland monitor scaling` 4 → 1.25) on a 6K monitor makes the Quickshell shell request a scene-graph layer texture of **18432 × 7680**, which exceeds the GPU's `GL_MAX_TEXTURE_SIZE` of 16384. Qt logs `QSGRhiLayer: Unsupported size requested` and refuses the texture, but the render thread proceeds into `endFrame`/`swapBuffers` anyway and segfaults inside Mesa (`dri2_query_image`). `omarchy-shell` catches the crash and restarts, so the desktop recovers in ~1 s, but the bar/background flash out.

18432 × 7680 is exactly **3×** the monitor's native 6144 × 2560, and **4 ÷ 1.33333 = 3.0** — the shell's full-screen background layer is being sized with the pre-change scale factor against the post-change logical size during the live output reconfigure.

## Environment

- Omarchy: `omarchy 4.0.2-1` (`/usr/share/omarchy/version`: `4.0.0.alpha`)
- quickshell `0.3.1-1`, qt6-base `6.11.2-2`, qt6-declarative `6.11.2-1`, qt6-wayland `6.11.2-1`
- mesa `1:26.2.1-1`, kernel `7.1.9-arch1-2`, Hyprland `0.56.2`
- GPU: Intel UHD Graphics 770 (Raptor Lake-S, `8086:a780`), i915, Mesa OpenGL 4.6 — `GL_MAX_TEXTURE_SIZE` = 16384
- Monitor: single **Dell U5226KW**, native **6144 × 2560** ("6K"), Hyprland scale 1.33333

## Steps to reproduce

1. Single 6K monitor (6144 × 2560) at a normal scale (e.g. 1.33).
2. `omarchy-hyprland-monitor-scaling 4` (or the display panel / keybinding) — set scale to 4.
3. A few seconds later, step the scale back down: `omarchy-hyprland-monitor-scaling 1.25` (cleans to 1.33333 on this mode).
4. The Quickshell render thread segfaults during the reconfigure; `omarchy-shell` logs the crash and restarts the shell.

Observed on this machine from `~/.local/state/omarchy/monitor-scaling.log`:

```
at=2026-09-08T22:34:51+02:00  requested=4     current=1.3333334  new=4        monitor=DP-2
at=2026-09-08T22:35:07+02:00  requested=1.25  current=4          new=1.33333  monitor=DP-2   <- crash at 22:35:07
```

## Evidence

Shell log, immediately before the crash:

```
22:35:07  WARN: QSGRhiLayer: Unsupported size requested: [18432, 7680]. Maximum texture size: 16384
22:35:07  ERROR: Quickshell has crashed under pid 125305
22:35:08  omarchy-shell: hyprland ipc event "configreloaded"
```

This `QSGRhiLayer` warning appears exactly **once** in the entire journal history — at the crash second. `~/.config/hypr/monitors.lua` was rewritten at `22:35:07.084` (matches the `sed -i` in `omarchy-hyprland-monitor-scaling`).

Symbolized backtrace of the crashing thread (`coredumpctl` + Arch debuginfod; Mesa frames unsymbolized — no debug package):

```
#0  __pthread_kill_implementation
#3  <quickshell crash handler>
#4  <signal handler called>
#5  dri2_query_image                                          libgallium-26.2.1-arch1.1.so   <-- SIGSEGV
#6  ??                                                        libEGL_mesa.so.0
#7  ??                                                        libEGL_mesa.so.0
#8  ??                                                        libEGL_mesa.so.0
#9  QtWaylandClient::QWaylandGLContext::swapBuffers           qwaylandglcontext.cpp:497
#10 ??                                                        libQt6Gui.so.6
#11 QRhi::endFrame(QRhiSwapChain*, QFlags<QRhi::EndFrameFlag>)
#12 QSGRenderThread::syncAndRender                            qsgthreadedrenderloop.cpp:773
#13 QSGRenderThread::run                                      qsgthreadedrenderloop.cpp:977
```

Not OOM (77 GiB RAM free at the time). First and only quickshell crash recorded on this machine.

## Analysis

`plugins/background/Background.qml` keeps full-screen scene-graph layers on the background window:

- `revealMask` — an `Item` anchored to the full screen with `layer.enabled: true` **unconditionally**.
- `incomingLayer` — full-screen `Item` with `layer.enabled` + `layer.smooth: true` + `layer.effect: MultiEffect` during wallpaper cross-fades.

A layer FBO is therefore (re)allocated at the background window's pixel size on every output reconfigure. During a live scale change from 4 to ~1.33 the layer is momentarily sized as `native_px × (old_scale ÷ new_scale)` = `6144×2560 × 3` = `18432×7680`, over the 16384 limit. `QSGRhiLayer` rejects it (logs the warning) but the frame still goes through `endFrame` → `swapBuffers`, and Mesa dereferences a buffer for the texture that was never created.

## Suggested fixes

- In the shell: clamp any screen-sized `layer` / render-target to `GL_MAX_TEXTURE_SIZE`, and/or skip the frame (don't call `swapBuffers`) when `QSGRhiLayer` reports the size is unsupported.
- In `omarchy-hyprland-monitor-scaling`: on a large panel, don't offer scale presets (3, 4) that drive a screen-sized shell layer past the GPU texture limit during a step-down, or apply scale changes in a way that doesn't leave the shell layer transiently over-sized.
- The hard segfault itself is an upstream robustness bug (Qt continues rendering with a rejected layer target; Mesa faults instead of erroring) — worth an upstream report too, but the trigger here is entirely within Omarchy's scaling flow + shipped shell.

---

Filed by Claude Sonnet 5 via Claude Code, at the machine owner's request.