From: "Marcin Mielżyński" <lopx@...>
Date: 2008-01-13T11:15:25+09:00
Subject: Re: 1.9 significantly slower than 1.8 on Mac

Derek Chesterfield pisze:

> 
> $ ruby -v
>> ruby 1.8.6 (2007-09-24 patchlevel 111) [universal-darwin9.0]
> 
> $ ruby logreport2.rb
>>                   user     system      total        real
>> Simple:       7.960000   0.000000   7.960000 (  7.965567)
> 
> $ ruby19 -v
>> ruby 1.9.0 (2007-12-25 revision 14709) [i686-darwin9.1.0]
> 
> $ ruby19 logreport2.rb
>>                   user     system      total        real
>> Simple:      13.750000   0.010000  13.760000 ( 13.754742)
> 
> 

Ruby 1.9 uses Oniguruma which is an encoding agnostic engine.
e.g: /src: (.*?);/ compiles down to:

[exact5:src: ]
[mem-start:1]
[jump:(1)]
[anychar]
[push:(-6)]
[mem-end:1]
[exact1:;]
[end]

seemingly simple match:

/src: (.*?);/ =~ "src: 172.25.20.79;"

turns out to:

    0> "src: 17..."         [exact5:src: ]
    5> "172.25...."         [mem-start:1]
    5> "172.25...."         [jump:(1)]
    5> "172.25...."         [push:(-6)]
    5> "172.25...."         [mem-end:1]
    5> "172.25...."         [exact1:;]
    5> "172.25...."         [anychar]
    6> "72.25.2..."         [push:(-6)]
    6> "72.25.2..."         [mem-end:1]
    6> "72.25.2..."         [exact1:;]
    6> "72.25.2..."         [anychar]
    7> "2.25.20..."         [push:(-6)]
    7> "2.25.20..."         [mem-end:1]
    7> "2.25.20..."         [exact1:;]
    7> "2.25.20..."         [anychar]
    8> ".25.20...."         [push:(-6)]
    8> ".25.20...."         [mem-end:1]
    8> ".25.20...."         [exact1:;]
    8> ".25.20...."         [anychar]
    9> "25.20.7..."         [push:(-6)]
    9> "25.20.7..."         [mem-end:1]
    9> "25.20.7..."         [exact1:;]
    9> "25.20.7..."         [anychar]
   10> "5.20.79..."         [push:(-6)]
   10> "5.20.79..."         [mem-end:1]
   10> "5.20.79..."         [exact1:;]
   10> "5.20.79..."         [anychar]
   11> ".20.79;"            [push:(-6)]
   11> ".20.79;"            [mem-end:1]
   11> ".20.79;"            [exact1:;]
   11> ".20.79;"            [anychar]
   12> "20.79;"             [push:(-6)]
   12> "20.79;"             [mem-end:1]
   12> "20.79;"             [exact1:;]
   12> "20.79;"             [anychar]
   13> "0.79;"              [push:(-6)]
   13> "0.79;"              [mem-end:1]
   13> "0.79;"              [exact1:;]
   13> "0.79;"              [anychar]
   14> ".79;"               [push:(-6)]
   14> ".79;"               [mem-end:1]
   14> ".79;"               [exact1:;]
   14> ".79;"               [anychar]
   15> "79;"                [push:(-6)]
   15> "79;"                [mem-end:1]
   15> "79;"                [exact1:;]
   15> "79;"                [anychar]
   16> "9;"                 [push:(-6)]
   16> "9;"                 [mem-end:1]
   16> "9;"                 [exact1:;]
   16> "9;"                 [anychar]
   17> ";"                  [push:(-6)]
   17> ";"                  [mem-end:1]
   17> ";"                  [exact1:;]
   18> ""                   [end]
match at 0


The bottleneck here is the anychar (OP_ANYCHAR in regexec.c) instruction 
which uses 'enclen' and 'is_mbc_newline' (which are function pointers, 
so impossible to be inlined).

Joni goes a bit further and uses a bunch of singlebyte specialized 
opcodes when it detects singlebyte encoding being used.



lopex