From: djberg96@... (Daniel Berger)
Date: 2004-09-14T13:35:39+09:00
Subject: Re: Ruby regex engine behavior question

ts <decoux@moulon.inra.fr> wrote in message news:<200409131620.i8DGK1r18648@moulon.inra.fr>...
> >>>>> "D" == Daniel Berger <djberg96@hotmail.com> writes:
> 
> D> "[In the Ruby 1.6 regex engine] \G doesn't prohibit regex bump-along
>                                       ^^^^^^^
> 
>  are you sure of this ?
> 
> D> (it's 'start of current match' rather than 'end of last match'), which
> D> makes relatively useless to write complex parsers with."
> 
> 
> Guy Decoux

The OP has further clarified.  To quote:

When trying to match abcde with /\Gx?/g, the first match is
successful, because no x is found but the question mark allows zero
characters to be consumed. This match ends after zero characters into
the string � at start-of-string. In order to avoid infinite loops on a
zero-length matches, the engine then retries the match one position
down the string.

In Perl, \G means end-of-last-match, and since end-of-last-match was
at start-of-string, \G can't possibly match at one character into the
string:

    $ perl -le'$_="abcde"; s/\Gx?/!/; print'
    !abcde 

In Ruby (both 1.6 and 1.8, I found), \G merely means
start-of-current-match, which, of course, is satisfiable at that
point:

    $ ruby1.6 -e'puts "abcde".gsub(/\Gx?/,"!")'
    !a!b!c!d!e!
    $ ruby1.8 -e'puts "abcde".gsub(/\Gx?/,"!")'
    !a!b!c!d!e! 

Perl's \G is a powerful tool to write parsers because the regex engine
is prohibited from skipping characters to find a match � you can work
your way through a string with a multitude of patterns using /c (to
avoid resetting the end-of-last-match on match failure) applied
against the same string in turn, without them sabotaging each other.

End quote.

Thoughts?

Dan