redesign.txt

Reimplementation of the bibtex2web program
==========================================

The bibtex2web program
(http://homes.cs.washington.edu/~mernst/software/bibtex2web.html) is
extremely useful, and is widely used.  However, bibtex2web is brittle and
hard to maintain, and it lacks some features.  It seems that the best
approach for features and long-term stability would be to rewrite.

Currently bibtex2web is written in perl and uses the bp library for parsing
BibTeX files.  The rewrite would give an opportunity to use a nicer
programming language as well.  (Search Google for "bibtex parser" to find a
new library to build upon.  Java seems like the best choice, as of December
2010.)  As a minor change, the generated webpages should be XHTML 1.0
Strict rather than XHTML 1.0 Transitional.

The current design (from a usage point of view) seems to work pretty well,
and can be largely maintained but also improved.

The user adds extra BibTeX fields and writes HTML template files with
special markers.  For instance, it might look like
  <bibtex2web project="speculation" bibfiles="/homes/gws/mernst/bib/*.bib">
  </bibtex2web>
The begin marker includes information about what papers to include (see
below for info about the predicate).  A static program processes the HTML
templates into regular static HTML pages.  It leaves the markers there, but
replaces anything in between them; the result is that running the program
twice in a row has no further effect, but if the BibTeX files change, then
running the program again updates the webpage.

This feels somewhat like server-side HTML commands.  However, the
ahead-of-time processing makes it easier to notify the user when there is
an error, rather than just producing a bad HTML page.  Also, the existence
of static HTML makes it easier to run other tools, such as HTML validators
and link checkers.

Two desired changes to the current bibtex2web design are:
 * User writes whole HTML page templates, not just headers and footers.
   This means that a single webpage can easily have multiple lists of
   papers, which is currently difficult to achieve.  This also means that
   the user can run HTML validation on the template files.
 * The inline markers can include a predicate (expressed as a boolean
   formula over BibTeX fields) indicating which subset of fields to
   include.  The exact syntax is to be determined.  One possibility is
   plain ASCII:
     <bibtex2web predicate="project='speculation' and author='Michael D. Ernst'">
     </bibtex2web>
   Another possibility is XML:
     <bibtex2web>
       <predicate>
       <and>
         <project="speculation" />
         <author="Michael D. Ernst" />
       </and>
       </predicate>
     </bibtex2web>

Because the requirements are somewhat vague, the first step in this project
would be defining the problem, and the best way to do that is by writing
the new manual.  After that, the system can be built, tested, and deployed.


The most important use cases are the local ones:
  ~mernst/sync/pag-pubs-src
  ~mernst/www/pubs-sources