Proof Portal

Project overview

Omarchy

ProbeLabs65 findings · 95 requirements

A proof layer — requirements, tests and verified fixes — for two of Omarchy's subsystems: the application menu (launcher scripts, QML model, JSONC config, search and selection) and the lock screen (lock scripts, QML, PAM authentication). Scope is deliberately limited to those components of omacom/omarchy; the rest of the distribution is not covered.

Inbox
ReportR-4561GitHub issuePublicReceived

Shell tests dump core instead of skipping when WAYLAND_DISPLAY is set but the compositor is unreachable

Reported by dhh · filed Aug 12, 2026Closed upstream on 2026-08-12 by the merge of PR #6749 (closed by dhh). GitHub close reason: completed.

Received and recorded verbatim. Nothing has been checked yet.

At a glance

1 · What is claimed?
Only the report text below — no validator has restated it as checkable claims yet.
2 · What does it affect?
Not yet mapped to any requirement or component.
3 · Is it reproducible?
No validation has been run yet.
4 · What remains unclear?
  • Not validated yet: no claims, no reproducer.
5 · What happens next?
Our staff decide whether the report is valid. Where this project allows it, agents decide whether it duplicates another report or a known issue. No decision has been recorded yet.

Original report

Shell tests dump core instead of skipping when WAYLAND_DISPLAY is set but the compositor is unreachablerevision 1 · captured 2026-10-04 11:56 UTC via poll

Summary

The 12 test/shell.d/* tests that launch Quickshell gate on:

if [[ -z ${WAYLAND_DISPLAY:-} ]]; then
  pass "no Wayland compositor; skipping ..."
  exit 0
fi

That checks whether the variable is set, not whether the compositor is reachable. In any environment that inherits WAYLAND_DISPLAY but cannot connect to the socket, the guard passes, Quickshell aborts inside the QGuiApplication constructor, and each run leaves core dumps behind instead of skipping cleanly.

Reproducing

Any agent sandbox that passes the environment through but blocks /run/user/1000. Found via codex exec review (CODEX_PERMISSION_PROFILE=:workspace, CODEX_SANDBOX_NETWORK_DISABLED=1), which ran:

bash ./test/shell.d/bar-test.sh && bash ./test/shell.d/runtime-smoke-test.sh

What happens

ERROR quickshell.paths: Could not create instance runtime directory at
      "/run/user/1000/quickshell/by-id/53gpknjt"
 WARN : Failed to create wl_display (Operation not permitted)
        Failed to connect to socket /run/user/1000/bus: Operation not permitted
 INFO qt.qpa.plugin: Could not load the Qt platform plugin "wayland" in ""
      even though it was found.
 INFO qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in ""
      even though it was found.
FATAL: This application failed to start because no Qt platform plugin could
       be initialized.

Symbolized stack — it never reaches QML, so no shell code is involved:

init_platform                                  qguiapplication.cpp:1368
QGuiApplicationPrivate::createPlatformIntegration   qguiapplication.cpp:1616
QGuiApplicationPrivate::createEventDispatcher       qguiapplication.cpp:1634
QCoreApplicationPrivate::init                  qcoreapplication.cpp:870
QGuiApplication::QGuiApplication
qs::launch::launch                             launch.cpp:290

qFatal → abort() → SIGABRT, two cores per launch (the process plus its forked crash reporter).

Test-visible result: not ok - test shell exited before plugins were listed.

Not only a sandbox problem

The same signature shows up with a different errno when the compositor dies mid-run — Failed to create wl_display (Connection refused), one second after a Hyprland SIGABRT. The weak guard collapses several unrelated "no usable compositor" conditions into identical core dumps, which makes them hard to tell apart after the fact.

On one machine this has produced 141 quickshell core dumps since Jul 27, all carrying the mktemp -d test-harness command-line shape (quickshell -p /tmp/tmp.XXXXXXXX/... --no-color).

Affected files

All launch Quickshell behind the same guard:

  • test/shell.d/bar-icon-geometry-test.sh
  • test/shell.d/bar-widget-contract-test.sh
  • test/shell.d/button-border-stability-test.sh
  • test/shell.d/indicator-contract-test.sh
  • test/shell.d/lock-fingerprint-indicator-test.sh
  • test/shell.d/lock-password-overflow-test.sh
  • test/shell.d/manifest-entrypoints-test.sh
  • test/shell.d/plugin-registry-contract-test.sh
  • test/shell.d/pointer-move-gate-test.sh
  • test/shell.d/runtime-smoke-test.sh
  • test/shell.d/screenshot-sanity-test.sh
  • test/shell.d/tray-menu-test.sh

Three more use the same guard without launching Quickshell, so they are cheaper but equally wrong: shell-ipc-display-test.sh, sleep-lock-environment-migration-test.sh, systemd-test.sh.

Suggested fix

Probe the compositor instead of the variable. bin/omarchy-launch-shell already carries the right helper:

compositor_alive() {
  local attempt
  for attempt in 1 2 3; do
    hyprctl -j monitors >/dev/null 2>&1 && return 0
    (( attempt < 3 )) && sleep 0.5
  done
  return 1
}

Hoisting that into the shared test preamble and replacing the -z ${WAYLAND_DISPLAY:-} check covers both the sandbox case and the dead-compositor case. Skipping when CODEX_SANDBOX_NETWORK_DISABLED is set would be a cheap belt-and-braces addition, since Codex exposes that variable specifically so builds can detect the sandbox.

Worth deciding separately: whether codex exec review should be running the GUI suite at all — it cannot pass in that sandbox by construction.

View raw text
## Summary

The 12 `test/shell.d/*` tests that launch Quickshell gate on:

```bash
if [[ -z ${WAYLAND_DISPLAY:-} ]]; then
  pass "no Wayland compositor; skipping ..."
  exit 0
fi
```

That checks whether the variable is *set*, not whether the compositor is *reachable*. In any environment that inherits `WAYLAND_DISPLAY` but cannot connect to the socket, the guard passes, Quickshell aborts inside the `QGuiApplication` constructor, and each run leaves core dumps behind instead of skipping cleanly.

## Reproducing

Any agent sandbox that passes the environment through but blocks `/run/user/1000`. Found via `codex exec review` (`CODEX_PERMISSION_PROFILE=:workspace`, `CODEX_SANDBOX_NETWORK_DISABLED=1`), which ran:

```
bash ./test/shell.d/bar-test.sh && bash ./test/shell.d/runtime-smoke-test.sh
```

## What happens

```
ERROR quickshell.paths: Could not create instance runtime directory at
      "/run/user/1000/quickshell/by-id/53gpknjt"
 WARN : Failed to create wl_display (Operation not permitted)
        Failed to connect to socket /run/user/1000/bus: Operation not permitted
 INFO qt.qpa.plugin: Could not load the Qt platform plugin "wayland" in ""
      even though it was found.
 INFO qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in ""
      even though it was found.
FATAL: This application failed to start because no Qt platform plugin could
       be initialized.
```

Symbolized stack — it never reaches QML, so no shell code is involved:

```
init_platform                                  qguiapplication.cpp:1368
QGuiApplicationPrivate::createPlatformIntegration   qguiapplication.cpp:1616
QGuiApplicationPrivate::createEventDispatcher       qguiapplication.cpp:1634
QCoreApplicationPrivate::init                  qcoreapplication.cpp:870
QGuiApplication::QGuiApplication
qs::launch::launch                             launch.cpp:290
```

`qFatal` → `abort()` → SIGABRT, two cores per launch (the process plus its forked crash reporter).

Test-visible result: `not ok - test shell exited before plugins were listed`.

## Not only a sandbox problem

The same signature shows up with a different errno when the compositor dies mid-run — `Failed to create wl_display (Connection refused)`, one second after a Hyprland SIGABRT. The weak guard collapses several unrelated "no usable compositor" conditions into identical core dumps, which makes them hard to tell apart after the fact.

On one machine this has produced 141 `quickshell` core dumps since Jul 27, all carrying the `mktemp -d` test-harness command-line shape (`quickshell -p /tmp/tmp.XXXXXXXX/... --no-color`).

## Affected files

All launch Quickshell behind the same guard:

- `test/shell.d/bar-icon-geometry-test.sh`
- `test/shell.d/bar-widget-contract-test.sh`
- `test/shell.d/button-border-stability-test.sh`
- `test/shell.d/indicator-contract-test.sh`
- `test/shell.d/lock-fingerprint-indicator-test.sh`
- `test/shell.d/lock-password-overflow-test.sh`
- `test/shell.d/manifest-entrypoints-test.sh`
- `test/shell.d/plugin-registry-contract-test.sh`
- `test/shell.d/pointer-move-gate-test.sh`
- `test/shell.d/runtime-smoke-test.sh`
- `test/shell.d/screenshot-sanity-test.sh`
- `test/shell.d/tray-menu-test.sh`

Three more use the same guard without launching Quickshell, so they are cheaper but equally wrong: `shell-ipc-display-test.sh`, `sleep-lock-environment-migration-test.sh`, `systemd-test.sh`.

## Suggested fix

Probe the compositor instead of the variable. `bin/omarchy-launch-shell` already carries the right helper:

```bash
compositor_alive() {
  local attempt
  for attempt in 1 2 3; do
    hyprctl -j monitors >/dev/null 2>&1 && return 0
    (( attempt < 3 )) && sleep 0.5
  done
  return 1
}
```

Hoisting that into the shared test preamble and replacing the `-z ${WAYLAND_DISPLAY:-}` check covers both the sandbox case and the dead-compositor case. Skipping when `CODEX_SANDBOX_NETWORK_DISABLED` is set would be a cheap belt-and-braces addition, since Codex exposes that variable specifically so builds can detect the sandbox.

Worth deciding separately: whether `codex exec review` should be running the GUI suite at all — it cannot pass in that sandbox by construction.

Validation

Not checked yet. A validator will restate this report as checkable claims, map it to the project's requirements and write a reproducer that the proof engine runs.

Linked

Not linked to any known issue, other report or change yet.

Decision

Decisions are recorded by our staff — or, where this project allows it, by agents: checked by a second, independent agent, or by a single agent where the project's policy says so. Any decision reached so far is shown below.

No decision has been recorded yet.