# Problems with links using pdfpages and pax

I have to merge some HTML documentation into a LaTeX workflow. I am running into problems with internal links using the pax package.

I am using wkhtmltopdf to produce the pdf pages from the html. I use the pdfpages package along with pax to embed those pdf pages into the LaTeX source.

When I compile with pdflatex the links from the html-sourced pdf no longer work.

The links do work in the wkhtmltopdf-generated pdfs before they're included in the LaTeX document.

I extract the link info using java 1.6 with this command:

java -cp /usr/local/share/java/pax.jar:/usr/local/share/java/PDFBox-0.7.3.jar \
pax.PDFAnnotExtractor filename.pdf


I've also used the pdfannotextractor.pl script that comes with pax, with the same results: I get plenty of error messages like this:

!!! Warning: Annotation on page 1 not recognized!
java.lang.NullPointerException


Although at the end of those warnings, it says * Result: [ok] The debug information from the script looks like this:

PDFAnnotExtractor 0.1l, 2012/04/18 - Copyright (c) 2008, 2011, 2012 by Heiko Oberdiek.
* CLASSPATH: []
* is_win: [0]
* pax.jar: [/usr/local/share/java/pax.jar]
* pdfbox.jar: [/usr/local/share/java/PDFBox-0.7.3.jar]
* Which java: [/usr/local/bin/java]
* System: [java -cp /usr/local/share/java/pax.jar:/usr/local/share/java/PDFBox-0.7.3.jar pax.PDFAnnotExtractor logistic_python.pdf]


If I turn off internal links with wkhtmltopdf, I get no errors.

The pax file is still created but it looks non-informative--lines look like this:

\[{pagenum}{18}\\
\[{page}{1}{0 0 612 792}{}\\
DestLabel={1},
Border={[0 0 0]},
}\\


And sure enough, when I pdflatex the LaTeX document, the links in the in the generated PDF do not work.

I'm on FreeBSD, using TeXLive 2015, PDFBox-0.7.3. I create the pdfs from html on a Windows machine with the latest wkhtmltopdf.

edit

Thinking the difference between platforms might be the cause, I tried using the pax java program on windows as well as the wkhtmltopdf program (Win7). Same results. Also I am now using this simple html file for testing:

<html>
<body>
<h1>test</h1>
<p><a href="#myanchor">Click</a> me.</p>
<h2><a name="myanchor">Anchor</a></h4>
<p>text</p>
</body>
</html>


I get the null pointer exception even with this simple case.

wkhtmltopdf test.htm test.pdf
java -cp path\to\pax.jar;\path\to\pdfbox.jar test.pdf


Turns out this is not a problem with the pax package or its corresponding java program. The PDF that wkhtmltopdf creates has a non-standard structure (whatever standard really means when it comes to PDF).

I wrote a very specific python program to parse such PDFs and write out a pax file after the style of the pax package's java program. It works for me and I can "include" html files in my LaTeX workflow.

1. convert the html+css to pdf using wkhtmltopdf
2. use my python program to write out the pax file
3. use the pdfpages and pax packages to create the final pdf with LaTeX.

Result is the included PDF looks good (wkhtmltopdf understands css) and the links work. The pax package is awesome.

If anyone else comes across this problem, please see my github repo with the code:

https://github.com/tiarno/paxmaker