Re: implementing NDR parsing method into bison

[Sylvain SCHMITZ]

jonhattan wrote:
> hi all,
> After I finish my project we want to continue maintaining the code and
> perhaps(surely) include DR method that was left aside of main bison in
> the previous project. "We" is the compilers department of the ULPGC to
> which J.Fortes and Miguel Angel Perez Aguiar (my tutors) belongs and me.
You can already count on me for any help on the writing of the code 
itself and its maintainance later.

I've worked on another version of DR(1) in bison 1.875 CVS as of 
beginning of september, though less complete than that of Carlos Javier 
Borges, since I did not write any skeleton.  The code could be improved, 
it didn't go through any extensive testing and debugging phase, but it 
already showed dramatic improvements in performances in the generation 
of the tables.  On this subject, for people familiar with LALR(1) 
computation in bison, the DR(1) method is a *lot* more costly for tables 
generations ---I had benchmarks in gprof showing negligeable time (0.01 
second) for the computation of the LALR(1) tables of a C++ grammar with 
bison versus 0.59 second for the DR(1) tables--- so improvements on this 
matter are welcome.  We need data structures allowing faster matching of 
equivalent DR(1) items.

The code can be downloaded with CVS from module "bison" in:
:pserver:address@hidden:/repository
the password being "anoncvs".  I can provide you with a CVS write access 
on this repository if you want.

I have used doxygen to document my code, and here are some HTML and PS 
outputs:
http://charybde.i3s.unice.fr/~schmitz/bison-doc/files.html
http://charybde.i3s.unice.fr/~schmitz/bison-doc/refman.ps.gz

> Finally, documentation will be generated in both spanish and english.
The lack of an english version was the main reason for me to try to 
recode it, as I don't understand a word of spanish.

We can meet in person until sunday in José Fortes Galvez' office if you 
want.

--
  Sylvain Schmitz