epublift 5.3.0 (command line) and 1.34.0 (web page) are out, and
https://epublift.itpax.net is running them. This covers 5.1 to 5.3, everything since my post on 5.0. As before, the book you start from stays as it is, and Validate and Repair run in your browser without uploading it.
Validate is stricter: some books that passed will now fail
Validate and
epublift check now run epubveri 0.40. Where the EPUB specification and epubcheck disagree, it follows the specification; before, it followed epubcheck. So a book can now pass epubcheck and fail here, or the other way round. Nothing in your book changed, only the verdict.
The case you are most likely to meet is an image with no alt in an EPUB 3 book. EPUB 2 does not require alt, so an EPUB 2 book upgraded by Optimize can now be reported invalid for that alone. Optimize does not write alt="" for you: that marks an image as decorative, which is wrong for a picture that carries content. Describing the image is an editor's job.
On my own test shelf of 562 books, 2 originals and 24 upgraded books that passed before now fail. Apart from that, what Repair, Optimize and kepub write is the same as in 5.2, with one fix: an image alone in a figure with a caption needs no alt, and kepub no longer wraps it in a span that made it an error. If a retailer asks for epubcheck by name, run epubcheck as well.
PDF import
- OCR reads Turkish and the other Latin-script languages, including letters with a cedilla, a breve or a dotless i. The first scan downloads 8 MB more model data. A scan in a script the model cannot read, such as Cyrillic, Greek or Arabic, gets a warning that its text will be wrong or missing.
- Scans in JBIG2 and JPEG 2000 are read instead of refused.
- Chapters follow the PDF's own outline, or its printed contents page, and italics stay italic.
- When page 1 is a cover picture, it becomes the book's cover.
- Images are at most 1,920 pixels on the longest side, and 2,560 for the cover. Nothing is enlarged, and a JPEG keeps its own quality.
- Words broken by a soft hyphen at a line end are joined again.
- Pages that could not be read are named in a warning, and the rest of the book is written.
- The title comes from the PDF when it has a real one. Otherwise the title and author come from the file name, without the site names a download adds to it.
- There is no made-up "Section 1" heading any more, and every imported book has a style sheet.
The web page's Import PDF now reads scans by OCR as well, up to 50 pages. A longer scan needs the command line.
PeterT, crates.io is still blocked by the same dependency. The install command takes the new tag:
Code:
cargo install --locked --git https://github.com/ePubLift/epublift --tag cli-v5.3.0 epublift --features epub34,metadata,pdf-ocr,markdown,validate
Full release notes:
cli-v5.3.0 and
web-v1.34.0; 5.1 and 5.2 are on the
releases page. If a book comes out of Optimize with more errors than it went in with, please tell me here. Validate, epubcheck or your own editor will show the difference, and your original is untouched.