ePubLift/epublift cli-v4.3.0

on GitHub

Fixed

• Test shelf, 562 books, repair -y then upgrade: 394 → 427 come out valid
  (cli-v4.2.0 → cli-v4.3.0), and no book has more errors than before. The
  changes below add up to that.
• Upgrading an EPUB 2 book no longer makes it invalid over layout attributes
  that EPUB 2 allowed and EPUB 3 does not: an image sized in percent
  (<img width="40%">), a table's cellspacing, cellpadding, border,
  width and height, width, height and valign on its cells, and a
  line break's clear (<br clear="all">). Each
  is written as the CSS the HTML Standard maps it to, as a class rule in
  epublift-presentational.css, wherever the book's own CSS doesn't already
  set that property and none of its selectors reads the attribute; the rest
  are left as written and counted in the report. See docs/usage.md.
• The upgrade renames the elements HTML made obsolete to their HTML
  counterparts, under the same checks: <strike> → <s>, <big> → span
  at font-size: larger, <center> around inline content → div at
  text-align: center, <font face size> → span with font-family and
  the legacy size keyword. <font color> and size="7" are left as they are.
  Test shelf: all 14 books holding these elements lose 162 → 67 findings, and
  4 come out valid.
• The upgrade's closing epubsana pass also points the package's
  unique-identifier at the book's one real identifier when the element it
  names is empty (fix.package_identifier; declined when the book holds two
  candidates, a UUID and an ISBN say). Eight books on the test shelf that
  stayed invalid after repair now come out of the upgrade valid.
• The upgrade's closing pass also renames an id the upgrade itself made
  invalid, with every link to it (fix.content_document_invalid_id). XHTML
  1.x trimmed the spaces from an id through its DTD, HTML does not, so a
  valid EPUB 2 book with id="k10 " came out invalid. One book on the test
  shelf; 426 → 427 valid, none worse.
• repair (CLI and web) fixes two more defects, from epubsana 0.34.0: a
  list placed directly inside another list instead of inside its item
  (<ul><li>…</li><ul>…</ul> — common in tables of contents) moves into
  the item before it, and an image declared as one standard type but holding
  another (a PNG declared image/jpeg) gets its declaration corrected.
  Test shelf, repair then upgrade: 418 → 426 valid books, none worse.

Changed

• veripublica conventions v0.7 (a long option name means one thing in
  every family tool). Three meta spellings change; the old ones still
  work in 4.x, print a warning on stderr, and go in the next major:
  • meta enrich --lang → --assume-language (--language writes
    dc:language; this one only steers the lookup);
  • meta enrich --cover → --include-cover (meta set --cover takes an
    image path);
  • meta show --json → meta show --format metadata (--json is the
    shared envelope in check).
• The check and repair JSON envelopes now say "convention": "0.7".
• Optimize writes out the named characters an XHTML 1.x DOCTYPE declared
  ( , ’, …) when it replaces that DOCTYPE with EPUB 3's,
  instead of leaving them undeclared for the final repair pass. Same output,
  much faster on books with many documents: an 805-document book took 11
  minutes on the command line and timed out on the web; it now takes 5
  seconds.
• epubveri 0.31.0, epubsana 0.34.0 and kepubverto 0.2.3 (crates and the
  web's Validate and Repair modules together), on styloria 0.15.

Security

• HTTPS (metadata lookup, OCR model download, Smart Import): rustls-webpki
  0.103.16 and webpki-roots 1.0.10 (RUSTSEC-2026-0382, RUSTSEC-2026-0381:
  distrusted intermediates could be accepted).

Full Changelog: cli-v4.2.0...cli-v4.3.0