From: Aaron Patterson <aaron@...>
Date: 2007-10-30T01:05:38+09:00
Subject: Re: Racc error recovery

On Mon, Oct 29, 2007 at 03:40:02AM +0900, barjunk wrote:
> On Oct 25, 10:29 pm, Aaron Patterson <aa...@tenderlovemaking.com>
> wrote:
> > On Thu, Oct 25, 2007 at 02:14:20PM +0900, Giles Bowkett wrote:
> > > > I'm writing a racc parser, and I need to recover from parse errors.
> > > ...
> > > > Unfortunately I can't seem to find any good documentation or examples on
> > > > error recovery.  I've read the Racc documentation about on_error and
> >
> > > Wild stab in the dark: is there any equivalent stuff in yacc you could
> > > use as a starting point?
> >
> > Yes there is.  I was hoping I could get an answer without digging up my
> > lex & yacc book. ;-)  Fortunately I found it and there is an error
> > recovery section that has definitely helped.
> >
> >
> >
> > > I realize it probably isn't modeled that closely, but that's the first
> > > thought I had.
> >
> > > Wild stab in the dark, part 2: tried using Sass instead? It's a CSS
> > > DSL. Or do you have to work with the CSS you've already got?
> >
> > No.  I'm trying to parse CSS, not generate it.
> >
> > --
> > Aaron Pattersonhttp://tenderlovemaking.com/
> 
> Aaron,
> 
> Would you mind sharing some of what you found in the book?  I'm more
> interested in what you were thinking originally and then what the book
> had to say to change your mind.

What I was thinking originally was that on_error could eat up tokens
until the grammar could be recovered.  Then, after reading the lex/yacc
book I found out that that is exactly what the "error" token does.

Take this grammar/script for example:

  class Tester
  
    token A LBRACE 
  
  rule
    a_a_lbrace
      : A A LBRACE { puts val }
      | A error LBRACE { puts "got error"; puts val }
      ;
  
  end

  require 'tester.tab'
  
  class Foo < Tester
    def initialize
      @tokens = [
        [ :A, 'a' ],
        [ :B, 'b' ],
        [ :LBRACE, '{' ],
      ]
    end
  
    def parse
      do_parse
    end
  
    def next_token
      @tokens.shift
    end
  end
  
  Foo.new.parse

The second rule gets called, and Racc uses LBRACE as a "pivot point"
(IIRC, that is what the yacc book called it).  The error rule will eat
up tokens until an LBRACE is found.  Then parsing can continue as usual.

Unfortunately, I'm not sure when on_error is supposed to be called.  I'm
also having a hard time figuring out where to place the error token in
some of my rules.  But I am going to start a new thread about that.

-- 
Aaron Patterson
http://tenderlovemaking.com/