openclaw - 💡(How to fix) Fix [Bug]: Subcommand --help segfaults + status/health/doctor silently exit in rootless Podman (Fedora 44)

Bug type

Regression / container compatibility

Beta release blocker

No

Summary

In a rootless Podman container on Fedora 44, OpenClaw 2026.5.12 has two distinct failure modes:

1. Several subcommands silently exit with no visible output (status, health, doctor, onboard). Their output appears to be drawn into the terminal's alternate-screen buffer and then discarded on exit, leaving the user with no diagnostic information.
2. Several other subcommands segfault outright, including subcommand --help invocations. The most damning example: openclaw gateway --help crashes with Segmentation fault (core dumped) — just asking for help on the subcommand is enough to crash.

Top-level openclaw --version and openclaw --help work normally, which suggests the crash happens during subcommand module load, before the help text is rendered.

Likely related to #60670 (same "subcommand-help segfaults" pattern, but on macOS).

Steps to reproduce

1. On Fedora 44 host with rootless Podman 5.8.1
2. Launch a rootless container from registry.fedoraproject.org/fedora:44 with --userns=keep-id, non-root user uid 1000, no extra capabilities
3. Install Node.js 22.22.2, npm, git inside
4. Install OpenClaw via the official installer: curl -fsSL https://openclaw.ai/install.sh | bash
5. Add ~/.npm-global/bin to PATH
6. Verify: openclaw --version → OpenClaw 2026.5.12 (f066dd2) ✓
7. Try any subcommand listed below

Expected behavior

• Subcommand --help should always print help text, not segfault.
• status, health, etc. should produce some visible output even when the gateway isn't running.
• TUI commands should fail gracefully (e.g. "no TTY detected, run interactively") rather than crash.

Actual behavior

openclaw doctor 2>&1 | tee /tmp/log captures the banner + ┌ OpenClaw doctor header. Anything after that is drawn into the alt-screen buffer and not captured.

OpenClaw version

2026.5.12 (f066dd2)

Operating system

Linux, Fedora 44, kernel 6.19.10-300.fc44.x86_64, x86_64

Install method

curl -fsSL https://openclaw.ai/install.sh | bash (which uses npm install -g openclaw@latest and switches the npm prefix to ~/.npm-global)

Logs, screenshots, and evidence

Environment

• Host: Fedora 44, kernel 6.19.10-300.fc44.x86_64
• Container runtime: rootless Podman 5.8.1
• Base image: registry.fedoraproject.org/fedora:44
• Container args of note: --rm --userns=keep-id -v <host>:/workspace:z -v <home>:/home/work:z
• Container user: work (uid 1000), matching host uid via keep-id
• No --privileged, no --cap-add
• Node.js v22.22.2 (meets engines.node >= 22.16.0); npm 10.9.7

Native bindings present and all lib deps satisfied per ldd:

• @mariozechner/clipboard-linux-x64-gnu/clipboard.linux-x64-gnu.node — all libs resolved
• @napi-rs/canvas-linux-x64-gnu/skia.linux-x64-gnu.node — all libs resolved
• @lydell/node-pty-linux-x64/prebuilds/linux-x64/pty.node — all libs resolved
• tree-sitter-bash/prebuilds/linux-x64/tree-sitter-bash.node — all libs resolved
• koffi/build/koffi/linux_x64/koffi.node — all libs resolved

(Package ships prebuilds for many platforms. The "not found" libs in ldd for musl/openbsd builds are expected and irrelevant on glibc Linux.)

Suspected root cause

A native module imported during subcommand module load is crashing in this environment. Candidates in rough order of suspicion:

1. @lydell/node-pty-linux-x64 — node-pty interacts with /dev/ptmx and /dev/pts/, which behave differently in rootless Podman's user namespace.
2. koffi — FFI library doing dynamic linking; sometimes incompatible with user namespaces.
3. @napi-rs/canvas-linux-x64-gnu — Skia.

openclaw gateway --help crashing is the strongest signal that this is in module-import code, not in actual subcommand logic.

Things that did NOT help:

• CI=1 openclaw doctor — still silent
• TERM=dumb openclaw doctor — still silent
• NO_COLOR=1 openclaw doctor — still silent
• openclaw doctor --fix — same silent behavior

Things I haven't yet tried (in case maintainers want to suggest):

• Downgrade to 2026.5.7 (a known-working version from a different host) to verify if 5.8→5.12 is a regression
• --cap-add ALL on the podman run
• Running as root inside the container

Impact and severity

Blocks all interactive use of OpenClaw in a rootless-podman sandbox, which is a natural deployment target for a tool that runs untrusted agent code. The sandbox isolation is precisely the reason one would want to containerize OpenClaw.

Additional information

• Related: #60670 — segfault on openclaw models -h on macOS. Same "subcommand-help segfault" pattern, different platform. May share root cause.
• Related: #48404 (closed) — startup segfault on ImmortalWrt, different OS but same family of issue.
• Happy to attach a core dump if useful; tell me where to send it.