Logging

This document defines the logging standards for HelixScreen. All new code should follow these patterns; existing code will be migrated incrementally.

Log Levels

Level	CLI Flag	Purpose	Examples
ERROR	(always)	Unrecoverable failures	“Failed to connect”, “NULL pointer”
WARN	(always)	Recoverable issues, guards	“Double init detected”, “Fallback used”
INFO	(production default), `-v`	User-visible milestones	“Connected to Moonraker”, “Setup complete”
DEBUG	`-vv`	Troubleshooting info	Component init summaries, state changes
TRACE	`-vvv`	Wire-level details	Per-item loops, JSON-RPC protocol, observer plumbing

When to Use Each Level

ERROR

Unrecoverable failures that require user action
NULL pointer dereferences (guarded)
Missing critical configuration
Failed operations that cannot proceed

WARN

Recoverable issues with automatic fallbacks
Guard checks (double initialization, missing optional components)
Deprecated API usage
Situations that might indicate a problem but don’t prevent operation

INFO

Use INFO sparingly for milestones the user cares about:

System startup completion: “HelixScreen UI Prototype”
Connection events: “✓ Connected to Moonraker”
Discovery summaries: “Capabilities: QGL, bed_mesh, chamber_sensor”
Panel setup completion: “[Home Panel] Setup complete!”
Major operations: “File list updated: 126 G-code files”

NOT INFO (use DEBUG or TRACE instead):

Navigation events: “Switched to panel 2” → DEBUG
Mock backend operations: “Returning mock file list” → DEBUG
Per-item updates: “Updated slot 0 info” → DEBUG
Internal wiring: “Queuing switch to panel” → DEBUG

DEBUG

Use DEBUG for troubleshooting information:

Component initialization summaries: “Subjects initialized”
Configuration details: “Target: 800x480, DPI: 160”
State change summaries: “Printer connection state changed: Connected”
Batch operation summaries: “Auto-registered 21 theme-aware color pairs”
Discovery details: “Detected probe: probe”

NOT DEBUG (use TRACE instead):

Per-widget XML apply: “Applied size preset: 64x32” → TRACE
Per-file metadata: “Using cached thumbnail: X.png” → TRACE
RPC method calls: “Mock send_jsonrpc: method” → TRACE
File listing results: “Found 11 files”, “Directory has X items” → TRACE
UI state toggles: “Overlay backdrop visibility set to: true” → TRACE
Spoolman lookups: “get_spoolman_spool(1) -> Polymaker PLA” → TRACE
Theme file parsing: “Parsing X.json in legacy format” → TRACE

TRACE

Use TRACE for deep debugging only:

Per-item processing in loops: “Registering color graph_bg: selected=#2D2D2D”
Wire protocol: “send_jsonrpc: {…}”, “Registered request 5 for method X”
Observer/callback registration: “Registering observer on X at 0x…”
Subject value changes: “Subject value now: 2”
Per-widget creation: “Created widget for slot 3”

Message Format

Prefix Standard: `[ComponentName]`

Correct:

spdlog::debug("[Theme] Auto-registered 21 color pairs");
spdlog::info("[Home Panel] Setup complete!");
spdlog::trace("[Moonraker Client] send_jsonrpc: {}");

Incorrect:

spdlog::debug("Theme: Auto-registered 21 color pairs");  // NO colon
spdlog::info("[Home Panel]: Setup complete!");           // NO double colon
spdlog::debug("Home Panel] Setup complete!");            // Missing bracket

Prefix Naming Rules

Pattern	Use For	Example
`[ClassName]`	Classes	`[MoonrakerClient]`, `[PrinterState]`
`[Feature Name]`	Multi-word features	`[Home Panel]`, `[Print Select Panel]`
`[Subsystem]`	Subsystems	`[Theme]`, `[AMS AFC]`

Message Content

Be specific: Include relevant data (counts, states, identifiers)
Be consistent: Use the same wording for similar events
Be actionable: For errors/warnings, include what went wrong

Good:

spdlog::info("[PrinterState] Initialized 6 fans (version 1)");
spdlog::debug("[Controls Panel] Populated 3 secondary fans");
spdlog::trace("[Theme] Registering spacing space_lg: selected=16");

Bad:

spdlog::info("fans initialized");  // No context
spdlog::debug("Done");             // Not actionable

Loop Logging Pattern

When logging inside loops, use TRACE for per-item and DEBUG for summaries:

int registered = 0;
for (const auto& item : items) {
    spdlog::trace("[Theme] Registering {}: value={}", item.name, item.value);
    // ... registration logic ...
    registered++;
}
spdlog::debug("[Theme] Auto-registered {} items", registered);

Common Patterns

Initialization Guards

if (initialized_) {
    spdlog::warn("[MyComponent] init_subjects() called twice - ignoring");
    return;
}
// ... initialization ...
spdlog::debug("[MyComponent] Subjects initialized");

Error Handling

if (!ptr) {
    spdlog::error("[MyComponent] Cannot process: NULL pointer");
    return;
}

Milestone Completion

// After significant setup work
spdlog::info("[MyComponent] Setup complete!");

Wire Protocol

spdlog::trace("[Moonraker Client] send_jsonrpc: {}", rpc.dump());
spdlog::trace("[Moonraker Client] Registered request {} for method {}", id, method);

Implementation Notes

Use spdlog exclusively (not printf, std::cout, or LV_LOG_*)
Every log line carries the emitting thread id and (on console/file) an ms timestamp — see “Per-Sink Log Patterns” below
For errors that should notify the user, use NOTIFY_ERROR() macro

Log Destinations & Retrieval

There are two independent log streams on every deployment:

Structured app log — every spdlog::* call from helix-screen, helix-watchdog, helix-splash. Routed through spdlog sinks.
Launcher subshell capture — helix-launcher.sh echoes, libc/glibc abort output, anything written to stdout/stderr before init_logging() runs. Captured by the SysV init script’s >> "$LOGFILE" 2>&1 redirect. Never sees per-line spdlog output (see “Console sink” below).

These two streams do not overlap. spdlog never writes to the launcher file (in daemon mode); the shell redirect never sees structured spdlog calls. They are complementary diagnostics, not redundant.

spdlog Sinks & Auto-Detection

Logging is brought up in two phases, both implemented in src/system/logging_init.cpp:

helix::logging::init_early(), called at the top of Application::run() before any startup phase, so nothing can log into a void.
helix::logging::init(config), called from Application::init_logging() in Phase 3, once CLI args (Phase 1) and settings.json (Phase 2) have been read. This builds the full sink set for the resolved target.

Everything below the “Target Resolution” heading describes phase 2. Phase 1 and the handoff between them are in “Ring-Buffer Sink Lifecycle”.

Target Resolution

The runtime target is set by precedence (highest → lowest):

--log-dest=<dest> CLI flag (passed through by helix-launcher.sh)
HELIX_LOG_DEST env var — read twice, independently: helix-launcher.sh translates it into --log-dest=, and Application::init_logging() also reads it directly via helix::logging::log_env_override(). The direct read is what makes the variable work under a systemd unit’s Environment=, a hand-run binary, or a third-party init script that execs helix-screen without our launcher (#1249). An unrecognized value warns and is ignored — never fatal, because a typo in helixscreen.env must not crash-loop an appliance.
/log_dest in settings.json
LogTarget::Auto (default) → resolved by detect_best_target():
- HELIX_PLATFORM_ANDROID → Android (__android_log_print)
- Linux with HELIX_HAS_SYSTEMD AND /run/systemd/journal/socket exists → Journal (systemd_sink_mt)
- Other Linux → Syslog (syslog_sink_mt via libc syslog())
- macOS/other → Console (stdout_color_sink_mt)

Valid LogTarget values: auto, journal, syslog, file, console, android.

Per-Sink Log Patterns

Each sink gets its own spdlog pattern (applied via sink->set_pattern() right after construction — not a global spdlog::set_pattern(), since the formats differ). The pattern strings come from the pure helper helix::logging::pattern_for_sink(SinkKind) in logging_init.h, so the format decision is unit-testable without constructing real sinks (tests/unit/test_log_pattern.cpp, tag [logging][pattern]).

Sink (`SinkKind`)	Pattern	Notes
Console (`stdout_color_sink`)	`[%H:%M:%S.%e] [%^%l%$] [%t] %v`	ms timestamp, colored level, thread id
File (`rotating_file_sink`)	`[%H:%M:%S.%e] [%^%l%$] [%t] %v`	same string — `%^…%$` are no-ops on the non-color file sink
journald (`systemd_sink`)	`[%l] [%t] %v`	no time token — journald stamps its own time
syslog (`syslog_sink`)	`[%l] [%t] %v`	no time token — syslog stamps its own; `%l` kept for grep-ability of `/var/log/messages`
Android (`android_sink`)	`[%t] %v`	logcat adds its own timestamp/level/tag metadata
Crash breadcrumb (`CrashErrorLogSink`)	`[%H:%M:%S.%e] [%l] [%t] %v`	feeds crash context; the ring actually stores `msg.payload`, so this pattern is for any other consumer of the stream

Why the thread id (%t) is on every sink: the worst crash family in this codebase (L081 / async-delete) is about main-thread-vs-background-thread (WebSocket / HTTP worker) confusion. Knowing which thread emitted a line is the single highest-value field for diagnosing it. The [logging][pattern] test fails if %t is dropped from any sink or if a time token is added to the system sinks (which would double-stamp the journal/syslog clock).

A console/file line now looks like:

[14:32:07.918] [debug] [140351827234560] [PrinterState] Initialized 6 fans (version 1)

Ring-Buffer Sink Lifecycle

The in-memory ring (MonotonicRingSink, process-global g_ring_sink) is what the debug bundle’s log_tail and helix-screen ctl log read. It is installed by init_early(), and init() adopts that same instance rather than constructing a new one.

That ordering is the whole point. Application::run() does Phase 2 init_config() before Phase 3 init_logging(), so the entire config diagnostic trail - corrupt settings.json, restore-from-backup, parse failures, migration output - is emitted while only the early logger exists. With the ring created in init(), every one of those lines was already gone by the time a ring existed, and no bundle could ever show them. Bundle XGVDYEB5 is the case: a user-visible “settings were corrupted, restored from backup” toast, and not one line of the config trail anywhere in a 20,000-line log_tail.

	`init_early()`	`init(config)`
Console sink level	Pinned at WARN (`EARLY_CONSOLE_LEVEL`)	`config.level`
Ring sink level	`debug`, or WARN when `HELIX_BUNDLE_LOG_DEBUG=0`	`debug`, or `config.level` when `HELIX_BUNDLE_LOG_DEBUG=0`
Logger floor	The more verbose of the two above	Same rule
Ring instance	Created	Adopted if one exists, else created

Three things fall out of this that are easy to get wrong:

The early console sink must pin its level explicitly. The logger floor now drops to debug so the ring can capture it, and spdlog gates at the logger before any sink sees a message - a sink’s own default level is trace. Without the explicit pin the early console would start echoing debug to stdout, which a daemonized launch redirects straight into the journal or the log file. Stdout volume is unchanged from before the ring moved.
HELIX_BUNDLE_LOG_DEBUG=0 still works, with one wrinkle. Before init() there is no configured level to fall back to, so the early ring matches the console at WARN. That leaves the early phase behaving exactly as it did when no early ring existed at all.
Adoption is one-shot. The flag is cleared on adoption, so a later init() rebuilds the ring exactly as it always did. Production calls init() once (Application::init_logging, helix_watchdog), but tests re-initialize the logger constantly and rely on each init() handing them a clean buffer. The watchdog build has no early phase at all, so init() creates the ring there.

Adopting rather than replacing also keeps set_runtime_level()’s identity check against g_ring_sink valid, keeps tail_ring_buffer()’s shared_ptr copy pointing at live data across the handoff, and keeps the sink’s clock-step detector’s memory continuous.

Capacity (HELIX_LOG_RING_LINES, else scaled from total RAM) is resolved in init_early(). That is safe this early: resolve_ring_capacity() reads only the env var and /proc/meminfo, neither of which needs Config. The cost is one extra PlatformCapabilities::detect() at startup, not extra memory - there is still exactly one ring.

Console Sink (Stdout) — When It’s Attached

The console sink is opt-in by detection. The whole decision is the pure function helix::logging::should_add_console(target, enable_console, force_console, test_mode, stdout_kind) (logging_init.cpp), which init() calls with classify_stdout() — the only part that touches the real fd. classify_stdout() distinguishes Tty / Pipe / File / Socket / Other, which is more than isatty() can tell you and is what the gate below turns on.

Resolved target	Console sink?
`Console`	Always — it’s the only sink
`Android`	Never — stdout is invisible to logcat, and `test_mode` does not override this
`Syslog` / `Journal` / `File`	`test_mode` (`--test`) → any stdout kind. Otherwise: a Tty always; a Pipe only with `force_console`; a regular File, Socket, or Other → never

enable_console = false vetoes every row.

force_console is set by an explicit -v/--log-level, or by HELIX_LOG_LEVEL (which the launcher also translates into --log-level=, so it has always reached this flag on a launcher-started device).

What that yields in practice:

./build/bin/helix-screen --test -vv from a shell → console sink on, colored output to terminal.
ssh -t pi 'helix-screen ...' → tty allocated → console sink on.
helix-screen -vv | tee run.log → pipe + force_console → console sink on. A human is watching through tee; discarding the output there was #1105.
helix-screen --test > out.log → regular file, but test_mode → console sink on.
helix-screen -vv > out.log without --test → regular file, no test mode → console sink off. A plain redirect is indistinguishable from the daemon redirect, so this is an accepted tradeoff; use | tee instead.
SysV daemon launch where stdout is redirected to a file → console sink off, even though the launcher passed --log-level.
systemd daemon launch where stdout is a journald socket → console sink off (journald already has the structured stream via the journal sink).

The file/socket exclusions prevent the “double-log” mode that caused the Snapmaker U1 print failure where spdlog at trace wrote ~35 lines/sec to stdout, the init script captured stdout to a tmpfs file, and 498 MB filled /tmp. They are also why the ZMOD AD5X launcher redirect (>> /opt/config/mod_data/log/helixscreen.log 2>&1) contains no app log: stdout there is a regular file.

Reading `--test` Logs When stdout Isn’t a TTY

--test sets LogConfig::test_mode, and should_add_console() attaches the console sink for any stdout kind under test mode — pipe, regular file, socket, background run, non-interactive agent shell. So ./build/bin/helix-screen --test -vv > run.log 2>&1 & captures the logs in run.log, and | tee / | grep work too. That is safe because --test never runs in production: no systemd unit, init script, procd shim, or launcher passes it, so test mode cannot reach a daemonized double-log path.

Without --test the narrower gate applies (see the table above): a pipe needs -v/--log-level to force the sink, and a plain > file redirect gets nothing. Read the resolved system sink instead:

HELIX_MOCK_PRINTER=ad5m ./build/bin/helix-screen -v &   # note: no --test
sleep 6 && kill %1
journalctl --since "1 min ago" | grep helix          # all lines
journalctl --since "1 min ago" | grep '\[PrinterDetector\]'   # one subsystem

--log-dest=file is a workable dev escape hatch too. With no --log-file, resolve_log_file_path() probes /var/log for writability and falls back to $XDG_DATA_HOME/helix-screen/helix.log when it is not writable, so a non-root run lands in ~/.local/share/helix-screen/helix.log rather than failing. An explicit --log-file at an unopenable path is not fatal either: the sink construction is caught, and the platform’s normal system sink takes over with a warning.

Verbosity still applies (-v=info, -vv=debug, -vvv=trace). Detection lines such as [PrinterState] Printer type set to: '…' are info-level, so -v is enough to see them.

Per-Platform Routing Summary

Platform	spdlog target (default)	Where structured logs land	How to read
Raspberry Pi (systemd)	Journal	systemd journal	`journalctl -u helixscreen -f`
x86/x86_64 (systemd)	Journal	systemd journal	`journalctl -u helixscreen -f`
Snapmaker U1 (Debian Trixie, SysV)	Syslog	`/var/log/messages` (rsyslogd persists on overlay)	`grep helix-screen /var/log/messages` or `journalctl -t helix-screen` if journald-only
AD5M Forge-X/KMod (BusyBox SysV)	File (hook)	`/data/.helixscreen/logs/helix.log`	`tail -f /data/.helixscreen/logs/helix.log`
AD5M ZMOD / AD5X (ZMOD MIPS, BusyBox)	File (hook)	`/opt/config/mod_data/log/helix.log` — under `/opt/config` so ZMOD’s `TAR_CONFIG` archives it (#1249)	`tail -f /opt/config/mod_data/log/helix.log`
Creality K1/K1C (BusyBox, in-memory syslog)	File (hook)	`/usr/data/helixscreen-state/logs/helix.log`	`tail -f /usr/data/helixscreen-state/logs/helix.log`
Creality K2 (BusyBox procd)	File (hook)	`/mnt/UDISK/helixscreen-state/logs/helix.log`	`tail -f /mnt/UDISK/helixscreen-state/logs/helix.log`
Elegoo CC1 / COSMOS (BusyBox)	File (hook)	`/user-resource/helixscreen-state/logs/helix.log`	`tail -f /user-resource/helixscreen-state/logs/helix.log`
SonicPad (Debian)	Syslog	`/var/log/syslog`	`grep helix-screen /var/log/syslog`
Android	Android	logcat	`adb logcat -s HelixScreen`
Dev workstation (macOS / interactive Linux)	Console	stdout in terminal	visible directly

“File (hook)” means platform_pre_start in that platform’s assets/config/platform/hooks-*.sh exports HELIX_LOG_DEST=file + HELIX_LOG_FILE (plus a 1–2 MiB × 3 rotation cap). Without the hook these targets would all fall back to detect_best_target() → Syslog, which on the BusyBox boxes is an in-memory ring that dies with the reboot you are trying to diagnose. logread | grep helix-screen still shows anything logged before the hook takes effect, and remains the right command on Snapmaker U1 / SonicPad.

Debugging the C / libhv Layer On-Device

spdlog is C++-only. When you need temporary instrumentation inside a C dependency (libhv, the DNS resolver in lib/libhv/base/), do not rely on fprintf(stderr) — the app manages its own stdout/stderr and a raw stderr write may not be captured. Use syslog(3) (#include <syslog.h>, syslog(LOG_WARNING, "[TAG] ...")); it lands in the platform’s syslog (/var/log/messages, logread, etc.) right alongside spdlog’s syslog sink, with no plumbing. spdlog’s own warn/error (fd1/syslog) is reliably captured; --log-dest=console -vv forces the console sink for redirected/non-tty runs.

Trap: before trusting absence of instrumentation output, confirm the code is actually in the deployed binary: strings <binary> | grep <MARKER>. A patched file compiled into a static .a (e.g. hsocket.o in libhv.a) can be a stale cached object that never picked up your edit — “no log output” then means “not compiled in,” not “not reached.” See BUILD_SYSTEM.md § Patch Gotchas.

Launcher Subshell Capture (`launcher.log`)

config/helixscreen.init runs the launcher in a backgrounded subshell with ( ... exec "$LAUNCHER" ) >> "$LOGFILE" 2>&1 &. Everything written to stdout/stderr inside that subshell ends up in $LOGFILE:

[helix-launcher] ... shell echoes (binary selection, splash detection, log level, exit code)
glibc abort output: *** glibc detected ***, MALLOC_CHECK_=3 diagnostics, std::terminate without active exception — happens after spdlog is dead and can be the only diagnostic on a crash
Early-init spdlog console output before Application::init_logging() swaps to the real logger (a handful of lines, warn-level)
LVGL warnings emitted before the LVGL log handler is wired up

It is NOT the structured app log. For that, use the syslog/journal commands above.

Path resolution

LOGFILE is chosen at script-start by config/helixscreen.init:

Probe /var/log — writable AND its backing filesystem (from /proc/mounts) isn’t tmpfs or ramfs → LOGFILE="/var/log/helixscreen/launcher.log" (the FHS-conforming location).
Otherwise (tmpfs/ramfs /var/log, or /var/log not writable) → LOGFILE="${DAEMON_DIR}/logs/launcher.log" where DAEMON_DIR is the install directory (always persistent ext4 on supported targets).

Platform	Resolved path
Raspberry Pi	`/var/log/helixscreen/launcher.log` (systemd platforms don’t use this script — uses `ExecStart` via systemd journal instead)
Snapmaker U1	`/var/log/helixscreen/launcher.log` (overlay’s upperdir is persistent ext4)
AD5M Forge-X/KMod	`/opt/helixscreen/logs/launcher.log` (`/var/log` is tmpfs on BusyBox)
K1 / K1C	`/usr/data/helixscreen/logs/launcher.log`
K2	`/mnt/UDISK/helixscreen/logs/launcher.log`
AD5X (ZMOD)	`/opt/config/mod_data/log/helixscreen.log` — ghzserg ships their own fork of the init script with `LOGFILE` hardcoded, so the `/var/log` probe above never runs. `INSTALL_DIR` is `/srv/helixscreen`, not `/usr/data/helixscreen`. The app log is a separate file, `/opt/config/mod_data/log/helix.log` (see below)
CC1 / COSMOS	`/user-resource/helixscreen/logs/launcher.log`

Size is capped at 5 MB at every start (the init script truncates if larger). This is belt-and-suspenders; the file shouldn’t grow that big now that the console sink is gated off in daemon mode.

Pre-v0.99.62 (legacy) path

Older installs wrote to /tmp/helixscreen.log on every SysV target. Debug bundles still scan this path for backward compatibility (src/system/log_collector.cpp:default_file_paths()).

Controlling Log Verbosity

CLI

Flag	Effect
`-v`	INFO and above
`-vv`	DEBUG and above
`-vvv`	TRACE and above
`--log-level=trace\|debug\|info\|warn\|error\|critical\|off`	Explicit level (overrides `-v`)
`--log-dest=auto\|journal\|syslog\|file\|console\|android`	Override destination
`--log-file=<path>`	Override path when `--log-dest=file`

Env / config

helixscreen.env (in <install_dir>/config/, but on Klipper platforms it’s symlinked to ~/printer_data/config/helixscreen/helixscreen.env):

HELIX_LOG_LEVEL=info   # trace, debug, info, warn, error, critical, off
HELIX_LOG_DEST=auto    # auto, journal, syslog, file, console
HELIX_LOG_FILE=        # path; only used when HELIX_LOG_DEST=file

Precedence: --log-* CLI flag > HELIX_LOG_* env > /log_* in settings.json > defaults (production: info; test mode: debug). libhv’s own logger stays at warn or quieter whatever the app level (libhv_level_for()), because it fsyncs every line into its own file in the working directory.

The env tier is enforced in C++ (Application::init_logging() → helix::logging::log_env_override() / resolve_log_setting()), not just by the launcher’s flag translation, so it holds however the binary was started. Validation is shared with the CLI parser (is_valid_log_target() / is_valid_log_level() in logging_init.h) so the accepted sets cannot drift, but the two disagree on what to do with a bad value on purpose: a bad flag is fatal (the user is at a prompt and can retry), a bad env value logs a warning and falls through to the next tier (a typo in helixscreen.env must not turn every boot of an appliance into a crash-loop).

Note the platform hooks are the real source of HELIX_LOG_DEST / HELIX_LOG_FILE on most embedded targets — six of the seven assets/config/platform/hooks-*.sh export them from platform_pre_start. helix-launcher.sh therefore resolves LOG_DEST/LOG_FILE/LOG_LEVEL/DEBUG_MODE after sourcing the hooks; resolving before them read unset variables (#1249).

Runtime (in-app)

The Settings → System → Log Level dropdown calls helix::logging::set_runtime_level(), which is spdlog::set_level() globally + persists to settings.json. This survives restart via the precedence chain (config-file level applies on next launch if no CLI/env override is set).

Klipper-style helixscreen.env convention

On Klipper-based platforms (Pi, AD5M, K1, K2, Snapmaker U1, etc.), setup_config_symlink() in scripts/lib/installer/platform.sh creates the canonical env file at ~/printer_data/config/helixscreen/helixscreen.env (visible/editable from Mainsail/Fluidd) and symlinks <install_dir>/config/helixscreen.env → that path. Edit either location; both point to the same file. The launcher only reads the install-dir copy, so a broken symlink silently makes user edits invisible — make deploy-<platform> now verifies and repairs the symlink during deploy.

Debug Bundles

DebugBundleCollector::collect() (src/system/debug_bundle_collector.cpp) assembles a debug bundle on user request (Settings → Help & About → Upload Debug Bundle, or the helix_debug Moonraker shell command), reading the on-disk log cascade through src/system/log_collector.cpp. It captures:

log_tail - the structured app log, read from the in-memory ring rather than from disk (src/system/debug_bundle_collector.cpp), so it is always the live process and always fresh. Because the ring is installed in init_early() it reaches back past Phase 2 config load; see “Ring-Buffer Sink Lifecycle”. The companion log_meta field records the active sink target, the level the persistent sinks were configured at (which the ring may have been more verbose than), and whether the tail came from the live ring or the on-disk fallback
Last N lines from each candidate launcher-log path (/var/log/helixscreen/launcher.log, ${install_dir}/logs/launcher.log, legacy /tmp/helixscreen.log, etc.)
Last N lines of syslog (/var/log/messages, /var/log/syslog)
systemd journal entries when available (journalctl -u helixscreen)
Crash report (if recent)
settings.json (sanitized)
printer_config — Klipper’s printer.cfg and every config it [include]s, fetched from /server/files/config/<path> and sanitized per line (see below)
filament_system.gcode_macros — bare gcode_macro names from /printer/objects/list

Why `printer_config` is its own field

Klipper re-dumps printer.cfg into klippy.log on every start, so the log tail used to carry it for free. It no longer does: every line of a config is a unique shape, so it survived condense_klipper_log()’s shape-collapse intact and spent the whole line budget on config — 84/63/58% of klipper_log on AD5X bundles 4QA7SZAM / LYGVE39Y / XSNN7PX5. ce4f21914 added strip_klipper_config_dumps() to elide it so the incident window survives.

Fetching the files into a separate field gives the content back without putting it back in competition with the incident, and beats the log copy anyway — that one arrives head-truncated whenever the byte window slices through the dump.

Budget: MAX_CONFIG_BYTES (512 KB) and MAX_CONFIG_FILES (40). A full ZMOD AD5X config is ~250 KB across its includes, so it fits whole. Hitting either cap records printer_config.truncated with the reason rather than silently shortening.

Sanitizing config bodies

Config bodies go through sanitize_text_block(), which applies sanitize_value() per line. This is not interchangeable with calling sanitize_value() on the file: that function replaces any single string over 4 KB with [REDACTED_LONG_VALUE], so a whole-file call would redact every config in the bundle. Line granularity still catches what actually turns up in a printer.cfg — notification macros carrying Pushover/Telegram tokens, camera and Spoolman URLs with embedded credentials, emails, MACs.

IP addresses: private kept, routable tokenised

sanitize_value() finishes by running helix::redact::ips_in_text() (include/log_redact.h), the same helper the WiFi log call sites use, so the log path and the bundle path share one implementation. It keeps RFC1918, loopback, link-local, CGNAT and IPv6 ULA addresses verbatim — 192.168.1.50 is identical in millions of houses and is the whole diagnostic value of a “cannot reach Moonraker” report — and replaces globally routable IPv4 and IPv6 with a session-stable ip#4f2a91 token. A routable IPv6 is an ISP-allocated prefix plus a stable interface ID, which identifies a household far more directly than the MAC on the same object. The scanner is boundary-checked so version strings (0.99.115, v1.2.3.4, 1.2.3.4.5) and log timestamps are not mistaken for addresses.

Hardware serials are handled by key instead: is_sensitive_key() matches serial_number, which covers cpu_info.serial_number (the board serial) and sd_info.serial_number under /machine/system_info. It deliberately does not match a bare serial, so Klipper’s serial: /dev/serial/by-id/… MCU path — which names the chip — is not swept up.

What it does not redact: filesystem paths, so an [include /home/<user>/printer_data/…] ships the username. That is the same exposure update.install_root has always had, not a new class, but it is worth knowing before pasting a config excerpt into a public issue. The existing rule applies unchanged — scrub bundle excerpts before they go anywhere public.

Bundles are uploaded to crash.helixscreen.org with an 8-char uppercase alphanumeric share code (e.g. UK9QCFY3, CGR6C7PA). The code is minted server-side; nothing in this repo validates its length, so treat the server as the authority if that ever changes. Retrieval: ./scripts/debug-bundle.sh <CODE> --save. Note that --save writes debug-bundle-<code>.json to the current working directory — run it from /tmp (cd /tmp && …) so bundles never land in the repo, and don’t commit them.

DEVELOPMENT.md#contributing - Code standards
DEVELOPMENT.md - Build and debug workflow
CLAUDE.md - AI assistant rules (includes verbosity flags)
ENVIRONMENT_VARIABLES.md - All HELIX_* env vars including HELIX_LOG_LEVEL
CRASH_REPORTER.md - How crashes are captured and shipped
../user/TROUBLESHOOTING.md - User-facing guide to reading logs