v1.2.1

Patch release. Three small bugs found by exercising v1.2.0 against a real OpenWrt 25.12.4 router, plus one polish item (an honest error code when the daemon you're configuring isn't installed).

Fixed

• packages/installed always returned [] on apk-tools 3.x. list_installed() called apk info --installed, a flag that exists on apk-tools 2.x but not 3.x (which OpenWrt 25 ships). Plain apk info is the right command; it prints one installed package name per line. The new implementation also filters output to the package-name regex so diagnostic lines never end up surfacing as fake packages.
• network/interfaces ipaddr surfaced as an array on modern uci. OpenWrt 25 uses list ipaddr for static-proto multi-address interfaces (loopback ships as list ipaddr '127.0.0.1/8'); uapi was returning the raw uci value, so a GET on a list-form interface returned "ipaddr": ["127.0.0.1/8"] instead of the schema-declared string. fromUci now surfaces both forms additively: ipaddr is always the first address as a string (preserving the v1.0/v1.1 contract); a new ipaddrs array carries the full list. toUci prefers ipaddrs when present and falls back to ipaddr. validate accepts either.
• Makefile lint-emdash scanned build/sdk/ (OpenWrt feeds checkouts contain em-dashes in upstream package READMEs / test fixtures we don't author or control). grep --exclude-dir=sdk now skips it.

Added

• New error code 503 init_script_missing with a pre-flight check. Before any uci write, transaction() now confirms each /etc/init.d/<svc> listed in reload_services actually exists; missing → fail-fast with a 503 carrying the missing path in message, no uci change. Motivates this: live-router testing of v1.2.0 showed that POST /sqm/queues on a box without sqm-scripts produced 500 reload_failed_unrecovered. The cause: step-5 reload returned exit 127 (script not found); the snapshot-restore worked but the SECOND reload attempt also returned 127 (same missing script), so uapi recorded both errors and surfaced "unrecovered" - misleading: uci IS in a known state, only the daemon reload couldn't run because the daemon isn't installed. The new pre-flight makes the two scenarios distinct on the wire:

  • 503 init_script_missing: daemon not installed; uci state unchanged.
  • 500 reload_failed_restored: daemon installed but reload exited non-zero; uci state restored from snapshot.
  • 500 reload_failed_unrecovered: as before, only for the genuinely unrecoverable case (snapshot import / restore-reload both threw or returned errors).

  Step numbering in the atomic-transaction recipe shifts by one in CLAUDE.md: pre-flight is step 0, flock moves to step 1, etc.

Tests

• Unit: 422 → 431 (+9 covering ipaddr/ipaddrs semantics, init_script_missing pre-flight against absent paths and unsafe names, empty reload_services pass-through).
• Integration: +1 file (30_init_script_missing_test.sh) verifying live behavior on a router without sqm-scripts: 503 returned, uci unchanged, healthy resources unaffected.

Live verification

End-to-end run against a real OpenWrt 25.12.4 router (apk-tools 3.0.5) with the locally-built APK installed:

• 6 test phases, 131 assertions: 131 pass, 0 fail.
• GET /packages/installed now lists 188 packages (was 0).
• GET /network/interfaces/loopback returns "ipaddr": "127.0.0.1/8" (schema-conformant string) and "ipaddrs": ["127.0.0.1/8"].
• POST /sqm/queues on the unpatched router returned 500 reload_failed_unrecovered; on 1.2.1 it returns 503 init_script_missing with "init script /etc/init.d/sqm not found (is the daemon installed?)" and leaves uci untouched.