Fono

Fixed

• Robust headless / systemd startup. Five regressions surfaced
  when running fono daemon on a headless inference box (Debian 13,
  no DISPLAY, systemd) all collapsed into one pass:
  • Vulkan probe crash on shutdown. ash::Entry::load() at
    daemon start enumerates every Vulkan ICD on the host (incl.
    Mesa lvp / LLVMpipe), which spawns CPU worker threads still
    parked in futexes when glibc dl_fini unmaps libvulkan on
    exit — segfault. The probe now runs in a disposable subprocess
    (re-exec self with FONO_INTERNAL_VULKAN_PROBE=1) and the
    parent reads a single tab-delimited result line off stdout
    cached in a OnceLock. Any spawn / timeout / parse failure
    collapses to Outcome::NotAvailable so the daemon never crashes
    on a broken Vulkan stack.
  • global-hotkey null-display crash. global-hotkey 0.6.4's
    X11 events_processor calls XOpenDisplay(NULL) and then
    dereferences the result via XDefaultRootWindow without a
    NULL check, segfaulting on hosts without DISPLAY /
    WAYLAND_DISPLAY. fono_hotkey::spawn_listener is now gated
    on is_graphical_session(), the same runtime check the tray
    already uses.
  • Systemd crash-loop on first run. The implicit first-run
    wizard ran whenever ~/.config/fono/config.toml was missing;
    under systemd dialoguer aborts with IO error: not a terminal
    and the unit restart-loops. The implicit wizard is now gated on
    stdin().is_terminal(); with no TTY, Fono writes
    Config::default() and continues. Explicit fono setup is
    unchanged.
  • Redundant daemon --no-tray in the systemd unit. daemon
    is the implicit default and the tray is already runtime-gated
    on is_graphical_session(), so the flag was dead weight.
    ExecStart is now plain /usr/local/bin/fono.
  • Silent install failures. systemctl enable --now returns
    success the moment ExecStart is spawned, so a unit that
    crashes a second later (Restart=on-failure loop) was invisible
    at install time. sudo fono install now waits 2 s, runs
    systemctl is-active, and on failure dumps the last 20 journal
    lines plus the recommended follow-up command.

Changed

• --no-tray flag removed; system IPC socket tried first. The
  tray is already runtime-gated on is_graphical_session(), making
  --no-tray redundant. CLI clients (fono toggle, fono history,
  fono use …, …) now try the system-wide IPC socket
  (/run/fono/fono.sock) before the per-user one, so a fono
  process running under the system fono.service unit can be
  driven from any user account on the box without per-user setup.
  Documentation in docs/wayland.md and docs/troubleshooting.md
  updated to match.

• general.sound_feedback config field, tray "Start/stop chimes"
  toggle, and the chime playback action removed. The chime path
  was a vestige from before the audio-visualisation overlay landed
  in v0.6.0; the overlay's bottom-centre panel + right-side VU bar
  now serves the same "did the recording start?" feedback role
  without spawning a separate audio process. Existing configs that
  set the field are silently ignored — no migration needed.

• [overlay].waveform now defaults to true. The standalone
  batch-mode overlay was off by default; new users had to discover
  the setting and edit ~/.config/fono/config.toml to turn it on.
  The push-to-talk feedback panel (volume bars / oscilloscope /
  FFT / heatmap) is the kind of UX that's better-on-than-off.
  Existing configs are unaffected: the field is #[serde(default)]
  on the Overlay struct, so a config that omits the line picks
  up the new default; configs with an explicit waveform = false
  stay opted-out as the user wrote them.

Full Changelog: https://github.com/bogdanr/fono/compare/v0.6.0...v0.6.1