From: nobu.nokada@...
Date: 2002-01-09T03:31:18+09:00
Subject: Re: Meaning of \G (was: Re: Efficient "lexing" in Ruby (maybe an RCR?))

At Tue, 8 Jan 2002 02:01:13 +0900,
Matt Armstrong <matt+dated+1010854868.2fb314@lickey.com> wrote:
> >> I liked your idea at first, but this method is actually slower in my
> >> case.  Why?  String#index with an offset does not support \A or ^ in
> >> the regexp.
> >
> > Use \G instead.  \A always matches the beginning of the string.
> 
> Yes, this works, thanks.  Strangely, the "String#index with \G"
> version is still a little bit slower than a version that keeps
> chopping up the string with $POSTMATCH and using Regexp#~= with \A.
> I'll have to investigate.

Hmmm, you might test with -K option?

$ cat /tmp/linecount_postmatch.rb
src = IO.read(ARGV.shift)
cnt = 0
while /\A.*\n/ =~ src
  cnt += 1
  src = $'
end
p cnt

$ cat /tmp/linecount_index.rb
src = IO.read(ARGV.shift)
cnt = 0
pos = 0
while pos = src.index(/\G.*\n/, pos)
  cnt += 1
  pos += $&.length
end
p cnt

$ time ruby /tmp/linecount_postmatch.rb eval.c
9170

real	0m11.477s
user	0m8.210s
sys	0m3.280s

$ time ruby /tmp/linecount_index.rb eval.c
9170

real	0m0.200s
user	0m0.200s
sys	0m0.010s

$ time ruby -Ke /tmp/linecount_postmatch.rb eval.c
9170

real	0m11.680s
user	0m8.660s
sys	0m3.010s

$ time ruby -Ke /tmp/linecount_index.rb eval.c
9170

real	0m17.721s
user	0m17.570s
sys	0m0.040s

I guess regex.c:re_adjust_startpos() is a bottleneck.  Since
String#scan doesn't adjust start position on the way, it's fast
as String#index without -K option.  It needs backward scan on
multibyte string.

> But the meaning of \G is not clear.  Matz' book says "match point
> where last match finished" -- but for String#index that definition is
> confusing.  Does \G for String#index just mean "anchor at beginning of
> offset?"

It can be said, "match point where last match finished, or
start position for first match".

-- 
Nobu Nakada