Comments on: Benchmarking: Testing a string against a single composite Regular Expressions, or an array of Regexps http://www.doboism.com/blog/2009/12/21/benchmarking-testing-a-string-against-a-single-composite-regular-expressions-or-an-array-of-regexps/ Software and Elseware Sun, 28 Feb 2010 23:27:47 +0300 http://wordpress.org/?v=2.9 hourly 1 By: Amitay Dobo http://www.doboism.com/blog/2009/12/21/benchmarking-testing-a-string-against-a-single-composite-regular-expressions-or-an-array-of-regexps/comment-page-1/#comment-177 Amitay Dobo Mon, 21 Dec 2009 21:08:07 +0000 http://www.doboism.com/blog/?p=81#comment-177 Thanks for the comment, Ori, and for correcting the inaccuracies above. I've found (and didn't finish reading) an interesting article discussing NFA implementations and comparing them with common backtracking algorithms at http://swtch.com/~rsc/regexp/regexp1.html . It seems that even when the "regExp" is an actual Regular Expression (i.e no backreferenes) libraries such as used by SpiderMonkey indeed do not use NFA/DFA. I wonder how much of the improvements i've seen comes down to staying in the compiled C context of the regExp Engine when backtracking on the outermost "|" operators, as opposed to iterating in javascript and going back to the engine for each test. Thanks for the comment, Ori, and for correcting the inaccuracies above. I’ve found (and didn’t finish reading) an interesting article discussing NFA implementations and comparing them with common backtracking algorithms at http://swtch.com/~rsc/regexp/regexp1.html . It seems that even when the “regExp” is an actual Regular Expression (i.e no backreferenes) libraries such as used by SpiderMonkey indeed do not use NFA/DFA.
I wonder how much of the improvements i’ve seen comes down to staying in the compiled C context of the regExp Engine when backtracking on the outermost “|” operators, as opposed to iterating in javascript and going back to the engine for each test.

]]> By: Ori Berger http://www.doboism.com/blog/2009/12/21/benchmarking-testing-a-string-against-a-single-composite-regular-expressions-or-an-array-of-regexps/comment-page-1/#comment-176 Ori Berger Mon, 21 Dec 2009 10:30:56 +0000 http://www.doboism.com/blog/?p=81#comment-176 Well, it's been very long since regexps were actually compiled to NFAs or DFA, or for that matter, since they have been regular expressions in the CS sense of the word. There are a few matchers around that actually do NFA/DFA stuff, but none of the standard libraries (in Perl, Python, JavaScript, Java, ...) do -- because that would mean no backreferences which all like to support following Perl. Well, it’s been very long since regexps were actually compiled to NFAs or DFA, or for that matter, since they have been regular expressions in the CS sense of the word.

There are a few matchers around that actually do NFA/DFA stuff, but none of the standard libraries (in Perl, Python, JavaScript, Java, …) do — because that would mean no backreferences which all like to support following Perl.

]]>