[tex-live] beamer vs. hyperref

Jonathan Kew jonathan_kew at sil.org
Fri Jul 11 22:30:23 CEST 2008

It has just come to my attention (though it may not actually be a new  
issue) that beamer and hyperref don't get along perfectly when used  
with xelatex.

First, consider a simple test using the "article" class:

   \title{A Test}
   \author{A. N. Author}

When this is run with xelatex, hyperref recognizes that it should use  
a dvipdfm-style driver:

   This is XeTeXk, Version 3.1415926-2.2-0.999.2 (Web2C 7.5.6)
   LaTeX2e <2005/12/01>
   Babel <v3.8j> and hyphenation patterns for english, usenglishmax,  
   Document Class: article 2005/09/16 v1.4f Standard LaTeX document  
   *hyperref using default driver hdvipdfm*

and the result is that hyperref features work as expected.

However, changing the document class to "beamer" leads to a surprising  

   This is XeTeXk, Version 3.1415926-2.2-0.999.2 (Web2C 7.5.6)
   Document Class: beamer 2007/03/11 cvs version 3.07 A class for  
typesetting pres
   entations (rcs-revision 1.70)

   Package hyperref Warning: Option `pdfpagelabels' is turned off
   (hyperref)                because \thepage is undefined.

   Hyperref stopped early
   *hyperref using default driver hdvips*

Hyperref has chosen the dvips driver, with the result that xetex's  
output driver xdvipdfmx generates a mass of warnings for all the dvips- 
specific \specials that it doesn't recognize.

So beamer is interfering with hyperref's attempt to choose the  
appropriate driver by default, which otherwise works well. This  
happens because beamer.cls actually defines a \count register  
\pdfoutput, if the pdfTeX primitive is not present:

   \ifx\pdfoutput\@undefined \newcount\pdfoutput \fi
   \ifx\pdfoutput\relax \newcount\pdfoutput \fi

(beamer.cls, lines 25-26.)

I'm not sure what else in beamer depends on having defined \pdfoutput;  
in a minimal test, removing these lines solves the hyperref problem  
and the output looks fine. However, there are probably many more  
implications that need to be considered. Alternatively, it would be  
possible to revise the tests in hyperref.sty to check for XeTeX before  
testing for the existence of \pdfoutput, so that it is not "fooled" by  
beamer's tricks. In any case, I suspect that defining \pdfoutput is a  
risky strategy, as it may mislead other packages besides hyperref, so  
I hope that beamer can be updated to eliminate this.

In the meantime, I suggest that in TeX Live, we add a hyperref.cfg  
file to the texmf-dist/tex/xelatex/xetexconfig/ directory, which will  
check that it is running under xelatex and then define  
\Hy at defaultdriver appropriately. Something like this:

     [2008/07/11 v1.2 hyperref configuration for XeLaTeX]
   \ifx\XeTeXversion\undefined \else
     % Change default driver to "dvipdfm" instead of "hypertex",
     % requires hyperref 2002/06/05 v6.72s
     \providecommand*{\Hy at defaultdriver}{hdvipdfm}%

By placing this file in a tex/xelatex subtree, it will be found in  
preference to the existing file under tex/latex *only* when running  
under xelatex, and so other engines will not be affected.

Unless there are objections (or better suggestions), I propose to add  
this to the xetexconfig directory shortly.

Comments welcome....  JK

More information about the tex-live mailing list