ePubLift/epublift web-v1.30.0

on GitHub

Added

• Choose a book's unique identifier in the Metadata form: keep it, take
  another of the book's identifiers, or make a new one (see cli-v4.6.0). A book
  with obfuscated fonts keeps its identifier, and the form says why. In all 13
  languages.

Changed

• zip 9.3.0 (was 9.1.3), both crates, floor raised. It carries reader
  fixes merged upstream that no advisory lists yet, so their severity is
  unknown. On the 562-book shelf no book that opened before is refused now.
• epubveri 0.37.1, epubsana 0.38.0, kepubverto 0.4.0, styloria 0.19, in the
  CLI and in the web's in-browser Validate and Repair (vendored WASM moved with
  the crates; VENDOR.md has the digests, and epubveri's third-party licence
  file now sits beside it). Measured on the 562-book shelf, old vs new, repair
  first: exit codes the same; check gives the same findings, except the
  reworded text of CSS-007 (info) in 28 books; repair -y books byte-identical
  in all 257 it changes; upgraded books and --kepub (50 measured) differ only
  in dcterms:modified. The browser build's validate equals the CLI's check
  in all 562.
• pulldown-cmark 0.13.6 for RUSTSEC-2026-0433 (wikilinks are never enabled
  here), crc32fast 1.5.4, and 20 indirect dependencies to their newest
  compatible versions.
• zenwebp 0.4.6 (was 0.4.4), the WebP encoder, with roxmltree 0.21.2,
  structured-zstd 0.2.6, uuid 1.29, webpki-roots 1.0.11,
  unicode-normalization 0.1.26, tokio 1.54.1 and brotli 9.1.1 (the last
  bench-only). zenwebp 0.4.6 fixes demuxer panics and hangs on crafted input
  and a lossless-encoder fault; epublift encodes lossy, so the lossless fix
  does not reach it. Measured on the 562-book shelf, old vs new: check
  findings identical in all 562; --webp output graded the same in all 562,
  WebP bytes −0.37% over 11,365 images, all decodable, quality against the
  source unchanged (295 sampled images: median PSNR 38.78 → 38.84 dB); in 6
  books one or a few images flip between converted and kept-original, as the
  never-larger rule decides; .eparc archives the same size, and every
  archive restores to the original's content with either version.

Fixed

• The web service starts when its log directory can't be written. An
  EPUBLIFT_LOG_DIR that exists but can't be written, such as a mounted volume
  owned by root, made it exit at startup. It now logs to stdout and warns. With
  the variable unset, the default image no longer warns on every start: that
  message is now INFO.
• PDF import keeps figures stored behind another filter. paginaverto
  0.1.4 undoes a filter in front of DCTDecode, so such a figure is a real JPEG
  and not still-deflated bytes under a JPEG label. Measured on 18 PDFs: the 22
  errors (OPF-029 and PKG-021, 11 figures in 2 books) are gone, those 11 entries
  are the only ones that changed, and the other 16 PDFs import as before.
• PDF import no longer writes text that breaks the XHTML. Some fonts map
  glyphs they have no character for to U+FFFE or U+FFFF, which XML does not
  allow, so the chapter was not well-formed (RSC-016) and readers could
  refuse it. They are now dropped, as control characters already were; the
  same holds for Markdown import. Measured on 18 local PDFs: 2 imports
  carried them (2–4 per chapter), both are now valid, and no other byte changed.

Run it

docker pull ghcr.io/epublift/epublift-web:1.30.0

Multi-arch (amd64 + arm64). The web service ships as a container image
only; for the command-line tool, see the latest cli-v* release.