v2.4.4

v2.4.4 — per-preset source override via UA_<key>_SOURCE

Port of the x86 sibling's feature. Agreed with Bertrand Clech (renesenses)
after v2.4.3: Tune OS has no /home to point a preset at even in principle
— no user accounts exist there at all. His plan is a dedicated persistent
mount instead.

Adds a silent per-preset override, UA_<key>_SOURCE, a plain env-var
lookup, not a new interactive prompt.

v2.4.3

v2.4.3 — verify a genuinely separate mount, not a /home/* guess

Port of the x86 sibling's fix. Bug found by Bertrand Clech (renesenses)
against Tune OS's actual disk layout (single grown root partition, no
separate /home) — the source-path check was a /home/* string prefix, not
a mount-boundary test, so a pair would silently give a false sense of
protection on any host without a separate /home mount.

_ram_persist_add_pair now resolves the actual backing filesystem of both
the source and / (findmnt --target) and refuses, not just warns, when they
match.

v2.4.2

v2.4.2 — Tune unit detection + E/P chaining for unattended mode

Port of the x86 sibling's fix. Two bugs found by Bertrand Clech
(renesenses) while wiring v2.4.0's Persistent paths into Tune OS:

1. The Tune Server preset only checked for tune-server.service, never
   confirmed against a real unit file — Tune OS ships tune.service, so the
   preset silently never matched. Fixed: comma-separated candidate list
   (tune.service,tune-server.service).

2. RAM_MODE_ACTION=e alone never reached the persistent-paths questions in
   a single unattended pass. Fixed: enabling RAM mode now automatically
   walks into Persistent paths right after.

v2.4.1

v2.4.1 — newbie walkthrough covers the P (persistent paths) prompt

The EN + FR newbie walkthroughs' step-by-step prompt table and the
post-install verification section didn't mention v2.4.0's new P
(Persistent paths) choice for module 14-ram-mode. Added. Ported from the
x86 sibling.

v2.4.0

v2.4.0 — RAM-mode persistent paths

RAM-mode gains a Persistent paths option (module 14, new P menu choice):
bind-mount an app's mutable state from /home (untouched by either RAM-mode
strategy) onto the path it actually expects, so it survives reboots even
though /var — or all of / under overlay — is otherwise wiped every time.
Auto-detects installed apps and offers two verified presets (Lyrion Music
Server, Tune Server's self-updating /opt/tune), migrates existing data on
first enable, and a symmetric removal restores it. Independent of the
Enable/Disable choice.

Ported from the x86 sibling, inspired by a HiFi-forum member's manual LMS
bind-mount setup; validated end-to-end on real hardware (add, reboot-
survival, remove/restore) — one stdin-redirection bug in the removal
prompt found and fixed along the way.

v2.3.3

v2.3.3 — dracut tmpdir fix

Module 04's optional /var/tmp tmpfs (256M) starved dracut's own initramfs
build directory on every subsequent kernel update, failing with the same
cryptic "No space left on device" as v2.3.2 but from a different cause
(plenty of room on /boot, none in dracut's scratch space). Now redirects
dracut's tmpdir to /tmp (not resized by this repo) alongside the /var/tmp
tmpfs entry. Ported from the x86 sibling after a live diagnosis on
Dominique's TuneOS box.

v2.3.2

v2.3.2 — RAM-mode /boot space preflight check

dracut -f writes the rebuilt initramfs to a .tmp file next to the
existing one for that kernel before renaming over it, so /boot briefly
needs room for both at once, not just the new image. Ported from the
x86 sibling fix after a user (hd3291) hit a raw "No space left on
device" from zstd there (same module, same bug).

Module 14 now checks free /boot space against the current initramfs
size before every dracut -f call (overlay enable and disable paths)
and fails with an actionable message — free up a kernel, or use
Strategy V (volatile), which needs no dracut rebuild at all.

Ported from the x86 sibling after a report by hd3291.

v2.3.1

v2.3.1 — RAM-mode disable bug fix (id-based BLS targeting)

Fixes a production bug reported by Auke: disable sometimes left
audiophile.overlay=1 in the boot cmdline because grubby --update-kernel=DEFAULT
resolved ambiguously when the recovery entry shares the same kernel path.

The module now reads saved_entry from grubenv (the one authoritative source),
edits the BLS .conf file directly, and verifies the change after every
add/remove. A reboot prompt (default Y) is also offered immediately after
disable — the current session is still on the RAM layer until an actual reboot.

Reported and fixed by Auke.

v2.3.0

v2.3.0 — Unattended mode

setup.sh --unattended [--answers file.env] lets the entire wizard run
without a TTY. Every prompt is driven by UA_<KEY> environment variables
set in the answers file or the shell environment. Enables scripted
appliance builds, kickstart %post, and CI provisioning.

extras/answers-example.env lists all keys with their defaults and
typical values for a headless audio host. RPi-specific keys added:
PI_TWEAKS, PI_TWEAKS_WIFI_BT, PI_TWEAKS_HDMI (module 13-pi-tweaks);
RAM mode is module 14 here.

Ported from the x86 sibling (Bertrand Clech / renesenses, PR #3).

v2.2.0

v2.2.0 — FFmpeg 8.0.1 on Pi 5, RAM-mode warning at wizard launch

- 10-install-drup: FFmpeg 8.0.1 minimal confirmed working on Pi 5
- FFmpeg menu references updated for DRUP v2.5.7
- setup.sh now warns before any action when the running session discards
  writes on reboot (overlayfs: everything is lost; systemd.volatile: /var
  only), so a full install — or a git pull — can no longer evaporate
  silently at the next boot. Detection reads /proc/cmdline (the kernel
  actually running), not grubby (which reports the next boot).

RAM-mode warning reported by Auke.