From: Steven Jenkins <steven.jenkins@...>
Date: 2005-11-13T01:56:33+09:00
Subject: Re: DocBook to PDF

Bob Hutchison wrote:
> The typical XML way to do this is DocBook -> XSL:FO -> pdf because  
> you'll be able to use Norm Walsh's stuff. If you do this and you have  
> input documents bigger than, say, two pages, you'll be wanting the  
> *fastest* XML processors you can get. I did this a couple of years  ago 
> and found libxml2 and libxslt2 the way to go. Getting a good  XSL:FO 
> processor was quite a trick then, and I have not been keeping  up to 
> date, so keep that in mind. The only free one that I could find  was the 
> Apache FOP processor, and it wasn't all that good  (incomplete). There 
> are a couple of commercial processors that are  apparently very good, 
> but I wasn't willing to spend a couple hundred  bucks on them.

I eventually went with LaTeX because I'm not just trying to make marks 
on paper, but trying to make beautiful documents that measure up to high 
standards of typesetting. All the out-of-the-box DocBook/XSL stuff I 
tried produced ugly output. Maybe things have gotten better.

The real appeal of LaTeX for me is that it operates on document objects 
at approximately the same level of abstraction as DocBook itself. It's 
fairly straightforward to translate between the two, and then use styles 
and macros to control the output formatting.

> An alternative that is worked quite well was to use the SGML  processors 
> for DocBook.
> 
> As I said, I've not been keeping up, but apparently both the Apache  XML 
> processors and Saxon have become much faster in the last couple  of 
> years. Both of those are Java. Unfortunately I don't think Ruby  has a 
> remote chance of being useful as an XML processor for this kind  of 
> application until Ruby gets much faster.

It depends on the application. I'm using a brute force REXML parser and 
it's plenty fast enough for what I need to do. My test data is about 
120k (37 pages typeset), and is fairly complex structurally: nested 
sections, variablelists, EPS figures, etc. I can convert it on an old P3 
in about 10 seconds. LaTeX is blazingly fast, so I can afford a little 
slowness upstream.

> Putting Ruby into that pipeline for processing the stream sounds more  
> reasonable, but still, that stream is going to get *very* long once  
> you've got XSL:FO.
> 
> If you really want to do this yourself brace yourself for a lot of  
> work. Maybe you should choose a subset of DocBook (isn't there a small 
> (ish) subset already defined?)

You don't need a very big subset for many documents. Mine handles

appendix
article
articleinfo
biblioid
blockquote
caption
colspec
emphasis
entry
figure
formalpara
imagedata
itemizedlist
listitem
mediaobject
orderedlist
para
pubdate
row
section
simpara
table
term
tgroup
thead
title
variablelist
xref

in about 350 lines of Ruby. It also handles profiles.

Steve