From: Eric Hodel <drbrain@...7.net>
Date: 2006-09-05T09:10:51+09:00
Subject: Re: Compiling Regexp only once

On Sep 4, 2006, at 4:29 PM, Rick DeNatale wrote:

> On 9/4/06, Robert Klemme <shortcutter@googlemail.com> wrote:
>
>> Note though that it's usually faster to have a regexp in line.  So in
>> your case you might have a method that does the line parsing (or
>> multiple line parsing) and that's where you can put the inline regexp
>> for max efficiency.
>
> Robert,
>
> I'm not sure what you mean by inline, do you mean a regexp literal  
> like
> /[ABC]/ vs. RegEx.new('/[ABC]/')  if so that's not clear.
>
> See my benchmarks on the thread about "Ordered contrast for String or
> Array," where changing a literal /./ to a constant set to
> RegEx.new('/./') caused a 200 fold performance increase.

Those aren't the same regular expression.  I expect that your  
benchmark is flawed.

Looking at it:

$ ruby
str = "abcdefghijklmnopqrstuvwxyz" * 5
RE = Regexp.new('/./')
p str.scan RE
-:3: warning: parenthesize argument(s) for future version
p str.scan /./
-:4: warning: parenthesize argument(s) for future version
[]
["a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m",  
"n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "a",  
"b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o",  
"p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "a", "b", "c",  
"d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q",  
"r", "s", "t", "u", "v", "w", "x", "y", "z", "a", "b", "c", "d", "e",  
"f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s",  
"t", "u", "v", "w", "x", "y", "z", "a", "b", "c", "d", "e", "f", "g",  
"h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u",  
"v", "w", "x", "y", "z"]

Also, you forgot about using String#split (but it is slower than  
unpack and scan).

$ ruby -e 'p Regexp.new("/./").source, /./.source'
"/./"
"."

The correct way to construct a Regexp from Regexp.new is to omit //  
in the pattern when it is a String.

> Not that you always want to do that, benchmarking always beats  
> rules of thumb.

Benchmarking and testing beat rules of thumb.

-- 
Eric Hodel - drbrain@segment7.net - http://blog.segment7.net
This implementation is HODEL-HASH-9600 compliant

http://trackmap.robotcoop.com