From: Matt Armstrong <matt@...>
Date: 2002-12-14T06:39:51+09:00
Subject: Re: [RCR] Global Regexp Match Mechanism (//g)

"Austin Ziegler" <austin@halostatue.ca> writes:

> I'm not sure I agree with you about the costs; however, I agree
> completely about the thread-unsafe behaviour. This is why I suggest
> in my RCR that the MatchData contain the position data. It's not as
> flexible (read: "unsafe") as the Perl version, but I think it will
> do the job. I will try to write more detail this weekend.

The MatchData already contains the character position of the end of
the match (MatchData#end).  If Regexp#match accepts an index offset,
that position can be extracted from the MatchData and passed in.

Originally you had:

    while md = /foo/g.match(foostr)
      puts md.to_s
      foostr = md.post_match
    end

With Nobu's patched version of Regexp#match, that loop becomes:

    pos = 0
    while md = /\Gfoo/.match(foostr, pos)
        puts md.to_s
        pos = md.end
    end

The inefficient "foostr = md.post_match" is gone.