4 Commits

Include renames

Author	SHA1	Message	Date
bvandeusenandClaude Opus 5.5	e11011e977	fix: the backup package checks or explicitly drops every error the linter flags ... Closing a refused S3 response and a test file, and the fake bucket's writes, now say they ignore the error on purpose; handler parameters the tests don't use are named _. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-02 13:56:22 -05:00
bvandeusenandClaude Opus 5.5	b6ae205862	feat: Minstrel can put and list objects in a path-style S3 bucket, signed with SigV4 and no SDK (#6771) ... Bucket.Put stores a body under a key, signed over its SHA-256; Bucket.List pages through ListObjectsV2. A backup only adds and lists, so an append-only key is enough. Keys are encoded once, exactly as signed, so music paths with spaces, plus signs and accents reach the bucket intact. A refusal names S3's code in words ("the secret key is wrong"). The signer is pinned to AWS's published GET Object, PUT Object and ListObjects examples. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-02 13:51:02 -05:00
bvandeusenandClaude Opus 5.5	8be69b752a	feat: a backup is one uncompressed tar of per-table JSON Lines that stays the same bytes day to day (#6769) ... The file is minstrel-<UTC stamp>.tar: manifest.json first, then tables/<table>.jsonl, rows one per line in primary-key order. The run's time lives only in the manifest and member headers are fixed, so a deduplicating off-site copy stores unchanged tables once. WriteFile writes to a hidden .part file, syncs it and renames it, so a watched folder never shows half a backup. Read refuses anything that isn't a whole backup: another format, a newer format version, a cut-off file, or a member holding fewer rows than the manifest says. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-02 13:51:02 -05:00
bvandeusenandClaude Opus 5.5	a67897a6cd	feat: Minstrel can read its irreplaceable data out as JSON Lines and load it back into an empty instance (#6770) ... internal/backup names every table a backup holds, in parent-first order, and every table it leaves out with the reason: what the server computes from the music files, provider caches it fetches again, sign-ins and one-time links, and logs of past runs. The library's identity rows travel because every like, play and playlist entry hangs off a track id, and the sync change feed travels because the apps' first sync reads it. Sealed credentials are left out of the file, and so are the markers that say tag and art enrichment are done, so a restored server asks for the credentials again and redoes that work rather than believing it has it. Snapshot reads one REPEATABLE READ, read-only transaction, each table in primary-key order, keys sorted. Restore refuses an instance with accounts or tracks and a backup from a newer schema, replaces seeded settings, inserts by column intersection in batches, skips generated columns, and carries serial sequences on past the restored ids. Guards: every schema table is backed up or excluded; every backed-up table comes after what it references and references nothing excluded; every omitted column exists; a restore gives back the backup byte for byte. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-02 13:43:16 -05:00