v0.99.49 — AD5X / ZMOD crash-loop hotfix
A targeted hotfix release for AD5X (ZMOD firmware) and other platforms with aggressive supervisor lifecycles. Real-time triage of bundle EE8X6GSK plus a live SSH session on the affected printer revealed a crash-loop chain combining three independent bugs: ZMOD's display-handoff Klipper macro periodically killalls helix-screen, our SIGTERM handler ran the full Application::shutdown() teardown, the teardown crashed with SIGBUS in late-stage cleanup (the crash handler was uninstalled in the first line of shutdown(), so no diagnostic landed on disk), and the watchdog burned through restart credits leaving a blank screen. All three sides are now hardened.
Fixed:
- HelixScreen no longer crash-loops on platforms whose firmware aggressively respawns the UI process (AD5X / ZMOD). SIGTERM now triggers immediate _exit(0) without running Application::shutdown(). External supervisors only care about a clean exit code, not whether teardown ran. Skipping the fragile teardown path (LVGL deinit, observer cleanup, static destructors) avoids the L081-family SIGBUS that ZMOD's killall-and-respawn cycle was hitting on AD5X every ~20s after Klipper restart. SIGINT (Ctrl+C from terminal) keeps the graceful shutdown path. Persisted state (settings, telemetry queue, crash history) is written on each change so nothing is lost.
- Crash handler stays installed through the entire shutdown sequence. crash_handler::uninstall() was the first line of Application::shutdown(); now the last. Any SIGBUS/SIGSEGV during widget deletion, observer cleanup, or lv_deinit is now captured to crash.txt instead of falling through to the kernel default with no diagnostic. Critical for surfacing the L081-family teardown crashes seen post-v0.99.46.
- Crash diagnostics survive misconfigured config dirs (chroot, overlay, PrivateTmp). When the signal handler can't open() the primary crash path, it now falls back to writing the dump to stderr (with === HELIX_CRASH_DUMP === marker prefix). Stderr is always open and gets captured by both journald (systemd) and the watchdog's stderr pipe, so users on platforms where $HELIX_CONFIG_DIR resolves to a non-writable layer no longer get silent diagnostic blackouts.
- Debug bundles now include crash data on platforms that set $HELIX_CONFIG_DIR (ZMOD, RatOS, etc.). Collector previously used an ad-hoc $HOME/helixscreen/config probe that missed overlay installs entirely. Now uses the canonical helix::get_user_config_dir() resolver, matching where the crash handler writes. New crash_txt bundle key includes the raw signal-handler dump even when the next-boot reporter hasn't run yet.
- Webcam discovery skipped on platforms without a camera widget. Moonraker server.webcams.list and the local-endpoint probes (127.0.0.1:8080/8081/4408) now respect the HELIX_HAS_CAMERA compile-time gate. Skipped on AD5M/AD5X/CC1/K1/K2/MIPS/SnapmakerU1. On AD5X specifically, the firmware's kernel H.264 codec driver crashes on every v4l2_open (dma_coherent_mem_available NULL+12 deref) — we never opened /dev/video* directly so this isn't the trigger for the recent crash family, but probing for a webcam we can't render is wasted RPC.
- Installer disk-check uses POSIX 'head -n 1' so BusyBox doesn't reject 'head -1' (AD5M/AD5X/K1/K2).
- Updater download-directory candidate list now includes the install root — previously skipped install-root candidates, breaking updates on platforms where the install root is the only writable location.