From: Ryan Davis <ryand-ruby@...>
Date: 2002-09-21T05:18:10+09:00
Subject: Re: Speed up suggestions (regex obj vs plain usage)


On Friday, September 20, 2002, at 12:54 PM, Ryan Davis wrote:

> outside the loop:
>
>   comment_re = /^#/
>
> inside the loop:
>
>   next if line =~ comment_re
>
>   if line =~ /[^:]+:([^:]+):/ then
>     pom = $1
>     if pom =~ r
>       outf.write( line + "\n" )

WOW! I just found something out that is pretty interesting... too bad I 
don't understand why yet... By sheer luck I noticed in my email that I 
didn't extract the above regex to be instantiated outside the loop. So 
I did it. And it slowed down, a lot! here's the profile:

BEFORE:

time = 0.03045900 sec
   %   cumulative   self              self     total
  time   seconds   seconds    calls  ms/call  ms/call  name
  50.00     0.01      0.01        1     7.81    15.62  IO#each_line
  50.00     0.02      0.01       86     0.09     0.09  String#=~

AFTER:

time = 0.04492900 sec
   %   cumulative   self              self     total
  time   seconds   seconds    calls  ms/call  ms/call  name
  66.67     0.03      0.03        1    31.25    46.88  IO#each_line
  33.33     0.05      0.02      127     0.12     0.12  String#=~

So I made a new version and got rid of all the regex instances (diff 
follows) and the profile looks like:

time = 0.00842100 sec
   %   cumulative   self              self     total
  time   seconds   seconds    calls  ms/call  ms/call  name
100.00     0.01      0.01        1     7.81     7.81  IO#each_line

(we are now so small and fast (file sould be bigger for better 
sampling) that the profile has no relevant data... subsequent runs show 
the sampling jumping all over but the total time is the same)

<528> diff parse[34].rb
4,6c4,6
< r = /xxx\\xxxxxxxx/i
< second_re = /[^:]+:([^:]+):/
< comment_re = /^#/
---
 >
 >
 >
10c10
<   next if line =~ comment_re
---
 >   next if line =~ /^#/
12c12
<   if line =~ second_re then
---
 >   if line =~ /[^:]+:([^:]+):/ then
14c14
<     if pom =~ r
---
 >     if pom =~ /xxx\\xxxxxxxx/i

NOW, can someone tell me why regexen in place are much faster than 
instantiated regexen?