[lug] Advice on writing a config file (probably XML)

[Bear Giles]

Timothy C. Klein wrote:
> I don't really know what advantages XML would have

XML parsers are standard with many languages and a low-cost way of 
reading in structured data.  With some tools your program can work 
exclusively with the DOM and let the library handle all import and 
export work.

> but LEX and YACC
> (flex and bison versions from GNU) are tailor made for this job. 

Ha ha ha ha!

No they aren't.  LALR(1) grammars are fairly expressive - you can 
write a surprisingly decent NLP parsers in yacc.  That's the 
technical reason why I have long felt that programming languages 
*are* a human language to communicate to other programmers as much 
as they're instructions to a machine - limited vocabulary and 
sentence structures, but if we're honest the vast bulk of our 
spoken words are LALR(1).  (Mapping the sentence diagramming you 
were taught in school to yacc grammars is left as an exercise for 
the reader.)

But config files are simple enumerations of facts, and you don't 
need a LALR(1) grammar for that.  A PDA is powerful enough to 
handle arbitrary nesting, and a fraction of the complexity of a 
LALR(1) parser.  That's what XML is.

Even if you write your own parser, if you want extensibility 
you'll end up with something similiar in feel to XML anyway.  I 
know, I've been there when redesigning file formats in pre-XML 
format.  We called it nesting name-value pairs, e.g.,

  (foo => bar)
  (baz => (Boulder => Udall, Denver => DeGette))

but it's easily recognized as attribute-less XML with different 
coloring.