From: "Radosław Bułat" <radek.bulat@...>
Date: 2008-02-12T05:41:21+09:00
Subject: Re: very bad character performance on ruby1.9

I have something similar to this problem.

#========
require "benchmark"

def longest_repeated_substring(s)
  found = nil
  1.upto(s.size / 2) do |n|
    (s.size - n).times do |i|
      sub = s.slice(i, n)
      if s.index(sub, i + n) # > marks[i] ? i + n : marks[i])
        found = sub
        break
      end
    end
    if found.nil? || found.size < n
      break
    end
  end
  return found
end

system("wget http://www.gnu.org/licenses/gpl-2.0.txt") unless
File.exists?("gpl-2.0.txt")

Benchmark.bmbm do |make|
  make.report do
    longest_repeated_substring(ARGV[0] || File.read("gpl-2.0.txt"))
  end
end

#=======end

Benchmarks shows that running it on ruby1.9 is about 15x slower that ruby1.8.
There is outputs (I removed ) from profiling it (-rprofile):

#ruby1.8
  %   cumulative   self              self     total
 time   seconds   seconds    calls  ms/call  ms/call  name
 56.50     3.00      3.00      120    25.00    44.00  Integer#times
 29.38     4.56      1.56    42250     0.04     0.04  String#index
  7.72     4.97      0.41    42250     0.01     0.01  String#slice
  5.84     5.28      0.31    42252     0.01     0.01  Fixnum#+
  0.38     5.30      0.02        2    10.00  2650.00  Integer#upto
  0.19     5.31      0.01       21     0.48     0.48  String#gsub!

#ruby1.9
  %   cumulative   self              self     total
 time   seconds   seconds    calls  ms/call  ms/call  name
  0.13     0.06      0.03      242     0.12     0.12  String#size
  1.05     0.30      0.24    42252     0.01     0.01  Fixnum#+
  5.71     1.61      1.31        2   655.00 10375.00  Integer#times
 43.79    11.65     10.04    42250     0.24     0.24  String#slice
 48.76    22.83     11.18    42250     0.26     0.26  String#index

It looks like the problem is in String#slice and String#index methods.

-- 
Rados�aw Bu�at

http://radarek.jogger.pl - m�j blog