v0.9.0

This release makes the prebuilt packages actually installable on ordinary Linux distributions, and fixes two problems reported from the field.

Prebuilt binaries could not start on any non-Nix system

Release artifacts are built with Nix, and every binary carried a /nix/store path as its ELF interpreter. On a normal distro that path does not exist, so the kernel refused to start the process and the shell reported:

/usr/bin/ht32paneld: cannot execute: required file not found

This affected the deb, the rpm, the AppImage and the tarball — every prebuilt artifact since packaging was introduced. The interpreter is now the standard /lib64/ld-linux-x86-64.so.2, and the build fails outright if a store path ever slips back in.

The packages also now declare the libraries they need — libusb-1.0 for the daemon and CLI, libdbus-1 for the tray applet — so a missing library is caught at install time instead of first start.

Requires glibc 2.39 or newer (Ubuntu 24.04+, Debian 13+, Fedora 40+). Build from source on older systems.

Fixes #1.

heartbeat and refresh_interval had no effect

The shipped config.toml.example listed both keys underneath the [web] header. TOML scopes a bare key to the section above it, so both were read as web settings and then discarded — editing heartbeat in the example file, the natural thing to do, changed a value the daemon never looked at, and the panel silently kept the 1000 ms default.

Both keys have moved above the first section header. The daemon now logs a warning naming any key it had to ignore, so a misplaced setting says so instead of failing quietly:

WARN Config: `heartbeat` is set under [web] and is being IGNORED: it is a
     top-level setting, so it must appear ABOVE the first [section] header

Existing config files keep working — this warns, it does not reject.

If you previously edited heartbeat or refresh_interval and saw no change, move them above [web] and the value will now take effect.

A failing panel no longer floods the journal

A panel dropping in and out logged a warning every few seconds despite the throttle being set to a minute, because a successful write reset the throttle and made the next failure look like the first. The throttle is now strictly time-based: one line per interval, whatever the device does in between.

The message itself was also unhelpful — hidapi error: hid_error is not implemented yet is a placeholder from hidapi's libusb backend, not a diagnosis. It now describes what actually went wrong.

LCD self-healing

The daemon recovers on its own when the panel stops accepting writes — a USB drop, a re-enumeration, or the device being claimed elsewhere. After a run of consecutive failures it drops the handle and reopens the device in-process; if the panel stays dark long enough it exits for systemd to relaunch it. Tunable via lcd_failure_threshold, lcd_reconnect_interval_ms, lcd_exit_after_ms and lcd_error_log_interval_ms — see the configuration guide.

Previously the panel reverted to its built-in default screen after any write failure and stayed there until the service was restarted by hand.

Headless installs work out of the box

The deb and rpm now ship the D-Bus system policy the daemon needs to own org.ht32panel.Daemon on the system bus. Without it a fresh headless install could not register its bus name — and ht32panelctl and the web UI could not reach it — until an admin wrote the policy by hand.

Thanks to @tekgnosis-net for the LCD self-healing and D-Bus packaging work, and to everyone who reported these.