ePubLift/epublift cli-v4.0.0

on GitHub

Security

• rustls 0.23.46 (RUSTSEC-2026-0346 to 0349, published on release day),
  used for the metadata lookups' HTTPS connections.

Changed

• Breaking: images are kept as they are by default. epublift -i book.epub upgrades the structure and leaves JPEG and PNG untouched; add
  --webp to re-encode them to WebP (size-safe, as before). Many readers,
  Kobo e-ink among them, show WebP blank, so keeping images is the default
  that works everywhere. --keep-images still works (it is now the default),
  --target 3.4 still re-encodes to its formats unless --keep-images is
  given, and --kepub --webp is the same as --kepub --kepub-webp. restore
  follows the same rules, with --webp for WebP on re-target.
• Breaking: the default report goes beside the output, not beside the
  input. With -o elsewhere/, a run no longer writes <input>_report.txt
  into the input's directory, which may be read-only or shared. Without -o
  nothing changes; -r/--report still sets the path.
• epublift check runs epubveri 0.28.0 and epublift repair runs
  epubsana 0.29.0. epubveri checks <style>, <script>, <meta> and
  <base> in EPUB 2 against XHTML 1.1's attribute lists, grades the encoding
  of every XML file, checks MathML attribute names per element (<mi title>
  is now RSC-005, as in epubcheck) and names the rule and the element path
  on SVG findings. epubsana adds safe fixes, applied without asking: SVG
  attribute and element names back to their camelCase (viewbox →
  viewBox, lineargradient → linearGradient), and dir="LTR"/"RTL"
  lowercased. It now asks before giving a document with no <title> one,
  before giving an NCX navPoint with no id one, and an empty title falls
  back to the book's title. It drops an empty optional metadata element in
  an EPUB 3 package too, where it is an error, not only in EPUB 2. Measured
  on 562 books, epubveri 0.26.0 against 0.28.0: the same findings, statuses
  and exit codes on every book.
• --kepub runs kepubverto 0.2.0, the family's KEPUB converter, in place
  of epublift's own transform. The span ids are kepubify's, as before, but
  kepubverto inserts its markup without re-serializing the document, so
  every other byte stays as it was, and its Kobo style element is valid.
  Measured on 562 books, graded by epubveri 0.28.0 against the same
  conversion without --kepub: the old transform added findings to every
  book (4,686 RSC-005 errors, 22,128 RSC-036 usage notes); kepubverto
  adds none. --kepub with the experimental Zstandard packaging is now
  refused (Kobo reads only a conformant EPUB).
• The gen-sample test-book binary needs the gen-sample feature
  (cargo run --features gen-sample --bin gen-sample), so installing the
  crate installs epublift alone. The package sent to crates.io holds only
  the code, README, LICENSE and CHANGELOG.

Fixed

• An upgraded book stays valid far more often. Measured on 562 books
  (epubveri 0.28.0): of 252 valid inputs, 249 come out valid (54 with
  cli-v3.3.0); 44 invalid inputs come out valid. What changed:
  • EPUB 2 metadata is translated to EPUB 3 instead of carrying
    attributes EPUB 3 forbids. opf:role, opf:file-as and an
    identifier's opf:scheme become <meta refines> (role, with
    scheme="marc:relators" for a MARC code; file-as;
    identifier-type); an empty element just loses them. Of several
    dc:dates one stays (publication first); a modification date gives way
    to dcterms:modified, a creation date becomes dcterms:created.
  • The upgrade finishes with epubsana's EPUB 3 fixes, run on the
    packaged book. Its safe fixes are applied — named entities such as
      become the characters they stand for (the new DOCTYPE declares
    none), inline SVG gets the manifest's svg property — and three fixes
    that EPUB 3 requires of a book EPUB 2 accepted: an empty <title>
    filled from the book's table of contents, heading or title, the
    Content-Type meta spelled as EPUB 3 wants it, an empty metadata
    element dropped. Other fixes that need a decision are left to
    epublift repair.
  • The navigation document generated from toc.ncx escapes its text.
    A label such as A & B was written back as a bare &, which made
    the whole book invalid (RSC-016, fatal).
  • An SVG 1.1 DOCTYPE is dropped (OPF-073: EPUB 3 allows no external
    identifier on SVG), unless it declares entities the file needs.
• A conversion that makes the book larger says so: "Size grew from 0.86
  MB to 0.92 MB (+6.7%)" instead of "Size reduced … (-6.7% savings)". The
  first line of --help no longer says images are converted to WebP.
• -V no longer says .dirty for a clean cargo install --git build
  (Cargo's untracked .cargo-ok marker counted as a change), and takes no
  hash from a repository the crate merely sits inside.
• Help texts: meta enrich --lang applies even when the book has a
  dc:language and is never written into the book; meta set --isbn writes
  a dc:identifier, not dc:source; -q/--quality covers AVIF/JPEG XL too.

Full Changelog: cli-v3.3.0...cli-v4.0.0