Comments on: Rhapsody Greasemonkey Script: Optimizing Text Manipulation in Javascript with Regular Expressions http://www.embracingchaos.com/2007/04/rhapsody_grease-2.html Analysis of Trends in Technology, Business, Society Wed, 25 Jan 2012 18:52:12 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: leodirac http://www.embracingchaos.com/2007/04/rhapsody_grease-2.html/comment-page-1#comment-367 leodirac Wed, 13 Jun 2007 21:48:37 +0000 http://wp.embracingchaos.com/2007/04/rhapsody_grease-2.html#comment-367 <p>Thanks for pointing that our Eric, and for the detailed off-line explanation. I should really run some performance tests on FF to check how they've implemented it.</p> Thanks for pointing that our Eric, and for the detailed off-line explanation. I should really run some performance tests on FF to check how they've implemented it.

]]> By: Eric Lippert http://www.embracingchaos.com/2007/04/rhapsody_grease-2.html/comment-page-1#comment-366 Eric Lippert Wed, 23 May 2007 06:46:55 +0000 http://wp.embracingchaos.com/2007/04/rhapsody_grease-2.html#comment-366 <p>I don't doubt that for typical data the regexp algorithm is O(N), or that it is typically faster than the O(MN) naive solution. </p> <p>However I am surprised to learn that Firefox compiles regular expressions into state machines. (As someone who used to develop a competing product and is still tangentially involved with it, I am probably still legally prevented from looking at the firefox implementation, so I cannot check myself.)</p> <p>The JScript implementation of regular expressions compiles into a bytecode language which is then run through a custom bytecode interpreter. This interpreter is allowed to do backtracking. This is the usual approach these days for so-called "regular" expressions, because regular expressions are not actually regular anymore. There is not a 1-1 mapping between a finite state machine and a modern regular expression. Modern regular expressions require at least a stack, which makes them equivalent to something as strong or stronger than push-down automata, not state machines.</p> <p>This means that (at least in our implementation, I cannot speak for anyone else's) you cannot rely on matching being O(N), because of the potential backtracking. It is certainly possible to build regular expressions which take polynomial time to match and I believe it is possible to build a regular expression which takes exponential time to match.</p> <p>Yours is probably not an example of such though.<br /> </p> I don't doubt that for typical data the regexp algorithm is O(N), or that it is typically faster than the O(MN) naive solution.

However I am surprised to learn that Firefox compiles regular expressions into state machines. (As someone who used to develop a competing product and is still tangentially involved with it, I am probably still legally prevented from looking at the firefox implementation, so I cannot check myself.)

The JScript implementation of regular expressions compiles into a bytecode language which is then run through a custom bytecode interpreter. This interpreter is allowed to do backtracking. This is the usual approach these days for so-called "regular" expressions, because regular expressions are not actually regular anymore. There is not a 1-1 mapping between a finite state machine and a modern regular expression. Modern regular expressions require at least a stack, which makes them equivalent to something as strong or stronger than push-down automata, not state machines.

This means that (at least in our implementation, I cannot speak for anyone else's) you cannot rely on matching being O(N), because of the potential backtracking. It is certainly possible to build regular expressions which take polynomial time to match and I believe it is possible to build a regular expression which takes exponential time to match.

Yours is probably not an example of such though.

]]>