Commit graph

2 commits

Author SHA1 Message Date
mithral
1b11867efe Phase 6a: directory grouping, audiobooks and music playback
A book is a folder of chapters, not a file. The scanner gained a grouping
mode so a directory of tracks becomes one item with ordered files.

- registry: MediaType.Group (file | directory). Audiobooks and music group by
  directory; everything else stays one item per file.
- scanner: builds groups, upserts one item per group, marks items missing when
  their directory disappears, and keeps the file-level added/updated/missing
  accounting. Tracks are ordered naturally, so chapter 2 precedes chapter 10.
- natsort: one shared natural-order comparator, now used by both the scanner
  and the comics module (which had its own copy).
- media: GET /items/{id}/files/{fileID} streams one specific track. It
  verifies the file belongs to the item, so an item id cannot be paired with
  another item's file id.

No new module was needed for playback: the polymorphic progress table already
keys on file_id, so per-chapter resume fell out of the existing design. The
web player resumes the exact track and offset.

Verified: 3 chapters grouped into one book titled "Mistborn", natural order,
range request returned 206, mismatched file id 404s, per-track progress
round-trips.
2026-10-09 21:20:47 -04:00
mithral
5c63932419 Phase 3: EPUB module, file streaming, per-user progress
- stream: serving goes through http.ServeContent, so Range/If-Range/
  If-Modified-Since are correct for every media type by construction
- progress: the polymorphic store. One row per user/item/file; the PK includes
  the file so multi-track audiobooks resume per track later. Finished is
  inferred when position reaches duration.
- media: generic item endpoints (detail, file, cover, progress, recent). Every
  one routes through auth.CanAccessLibrary; an unauthorised item 404s.
- covers: a per-kind extractor registry. Extraction is lazy and recorded in
  the assets table. A missing cover is a placeholder, never a 500.
- modules/ebooks: EPUB parsing (container.xml -> OPF -> metadata, spine, cover
  with three fallback strategies), cover extraction to /config, and lazy
  metadata parsing so parser fixes need no rescan.

Verified against a generated EPUB: title/2 authors/language/publisher/spine,
cover served as PNG, range request returned 206 with correct Content-Range,
progress round-tripped.
2026-10-09 16:36:27 -04:00