Fixed Layout EPUB to PDF Failure

[martinmcu]

Hello,

i am pulling my hair out trying to get this book to convert. It should never have been created as an EPUB in the first place, as every page has a background image that covers 100% of the page, and all the text is designed to precisely overlay those graphics. As a result, it renders horrifically on my Kindle (jailbroken with KOReader). Rant aside, please help!

I have attached an epub that renders correctly in Calibre 8.7.0 that contains only the table of contents page. I have edited this using the built-in epub editor, and running "Check book" comes back error free. Reading this epub in Calibre produces the following page:



However, the PDF output is atrocious (But it does seem to match the rendering inside of KOReader):



I am attempting the conversion from the command line using the following command:

ebook-convert 'Minecraft_ Guide to Combat - Mojang Ab.epub' .pdf -u devicepixel --custom-size 838x1188 --disable-font-rescaling --embed-all-fonts -vv -d ./Debug --pdf-page-margin-right=0 --pdf-page-margin-left=0 --pdf-page-margin-bottom=0 --pdf-page-margin-top=0 --pdf-no-cover --uncompressed-pdf

The desired output would be a PDF that is an exact representation of that page, with no border around the outside. The page size was chosen based on the following portion of the CSS:

#spread_6-div-1 {
  -webkit-transform: translate(0, 0) rotate(0deg) skew(0deg) scale(1, 1);
  -webkit-transform-origin: 0% 0%;
  height: 1188px;
  left: 0;
  position: absolute;
  top: 0;
  transform: translate(0, 0) rotate(0deg) skew(0deg) scale(1, 1);
  transform-origin: 0% 0%;
  width: 838px;
}

There was something similar for every page in the full epub, with consistent width/height dimensions.

In looking through the debug output, the html generated at the processed step renders the same in Calibre as the input epub. I converted to epub as follows and imported it into Calibre

cd Debug/processed
zip -r processed.epub *

[kovidgoyal]

calibre does not support fixed layout books.

[martinmcu]

In what way does Calibre not support them? The viewer is rendering the book correctly, and all of the processing steps up until PDF output seem to be preserving the properly formatted EPUB. What is the rendering engine for this final step?

Thanks for your prompt initial reply! I'm really hoping to find a way to make this book usable for my kid.

[kovidgoyal]

It doesnt support them in any way. If it happens to work in the viewer, that is entirely accidental.

[martinmcu]

Ahhh, understood. I will enjoy the happy accident of it working in the viewer then. Thanks so much! Calibre is one of the greatest pieces of software ever, and makes having a library of ebooks possible and enjoyable.

[martinmcu]

Solved!! Admittedly, not entirely within the framework of Calibre, but worth documenting for others. Here are the steps that I took:

1. Edit the epub inside of Calibre. The CSS needed to be updated to define the page size properly. These were the lines that were added to force a precise page size with no margins:
   

   @page {
     size: 838px 1188px;
     margin: 0;
   }

   After the CSS was updated, I made sure that "Check book" came back clean
2. Unzip the epub to a temporary working folder.
3. (Optional) Optimize background images. While the CSS defined a page size of 838x1188 pixels, the actual images were much larger. Using Imagemagick on Linux made this trivial:
   

   cd /tmp/book/OEBPS/images
   for i in $(ls -1 page*jpg); do convert $i -resize 838x1188 small_$i; mv small_$i $i; done

4. Remove links from table of contents page. Unfortunately, I could not figure out a way to keep the links valid. After converting, they were referring to the absolute paths to the expanded xhtml files on my computer. In this case, a simple sed script cleaned this out:
   

   cd /tmp/book/OEBPS/xhtml
   sed -re 's:<a href=\"spread_[0-9]+.xhtml\">::g;s:</a>::g' -i spread_6.xhtml

5. Convert each individual xhtml file to PDF using WeasyPrint:
   

   cd /tmp/book/OEBPS/xhtml
   for htm in $(ls -1 *xhtml); do weasyprint --optimize-images --full-fonts --hinting $htm ${htm}.pdf; done
   rename 's/_(\d)\./_0\1./' *pdf

   The rename command was necessary to put the pages in the correct order to join them into one PDF. Otherwise, page 5 would have wound up between pages 49 and 50.
6. Create a single PDF from the individual pages
   

   pdfunite $(ls -1 *pdf | xargs) book.pdf

7. Create a table of contents for the PDF. This was done using pdftk by dumping the PDF data, writing a list of bookmarks, and then merging it back in:
   

   pdftk book.pdf dump_data > book.info

   Lines like the following were added to the book.info file
   

   BookmarkBegin
   BookmarkTitle: Contents
   BookmarkLevel: 1
   BookmarkPageNumber: 6
   BookmarkBegin
   BookmarkTitle: Before You Begin
   BookmarkLevel: 2
   BookmarkPageNumber: 8
   BookmarkBegin
   BookmarkTitle: Be Prepared
   BookmarkLevel: 1
   BookmarkPageNumber: 11
   BookmarkBegin
   BookmarkTitle: Making Sense of the Screen
   BookmarkLevel: 2
   BookmarkPageNumber: 12

   This was slightly painful, as I couldn't find a good way to automate the generation of these lines. Still, there weren't that many that were needed. Once the bookmarks were written, it was time to merge them back into the PDF:
   

   pdftk book.pdf update_info book.info output book2.pdf

8. Finally, this PDF could be brought back into Calibre by "Add books > Add files to selected book records"