From: Robert Dober <robert.dober@...>
Date: 2006-07-27T06:03:04+09:00
Subject: Re: gsub

------=_Part_32169_32286249.1153947781493
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

Sorry about yet another post, but  I think this  is noteworthy...

Well that is nice ;)
> actually that would be something like
>
> s.sub( %r{#{Regexp.escape(r)}(?!.*#{Regexp.escape(r)})},"")
>
> but I am afraid that this is much more expensive than the
> reverse.sub.reverse trick and much less readable
>
> Robert
>
Variable length lookahead assertions are a killer, just look at this, I have
replaced the last < in this page with ***
that is a string of length 2369, look at the benchmark please:

robert@roma:~/log/ruby/theory/regex$ cat test2.rb && ./test2.rb
#!/usr/bin/env ruby
#
#

require 'benchmark'
string = nil
File.open("gsub.html"){ | file | string = file.read }
s1=s2=nil
r1 = Regexp.new( "<" )
r2 = Regexp.new( "<(?!.*<)", Regexp::MULTILINE )

n = 50000
Benchmark.bm do |x|
      x.report("reverse.sub.reverse") {
        n.times do
                s1 = string.reverse.sub(r1,"***").reverse
        end
        }
      x.report("lookahead") {
        n.times do
                s2 = string.sub(r2,"***")
        end
        }
end
      user     system      total        real
reverse.sub.reverse  1.470000   0.010000   1.480000 (  1.483691)
lookahead 48.100000   0.060000  48.160000 ( 52.608063)

As far as I know fixed with lookahead is about ok, though.
Cheers
Robert

------=_Part_32169_32286249.1153947781493--