Boost logo

Boost :

From: Eric Niebler (neric_at_[hidden])
Date: 2002-10-25 11:07:52


On Fri, 25 Oct 2002 12:50:03 +0100, John Maddock wrote:
>
> See
> http://ourworld.compuserve.com/homepages/john_maddock/proposals/exregex.htm
> In short it depends on what you are trying to do (as ever). BTW I have
> never got GRETA to compile with anything other than VC7, but maybe Eric
> can jump in and correct me there...

I agree with John's tests here, but you must use care when interpreting
them. They show that when matching strings 19 Mb long, or 50 Kb long,
that Boost performs quite well, and this is true. If, on the other hand,
you are matching against strings 100's of chars long or less (more common
scenario?), then GRETA is often many times faster than Boost. It depends
on what you're doing. John, do so many people match 19 Mb strings that
this test case deserves top billing?

I encourage people to try regex++ and greta with their own patterns and on
their own data. That's they only way to tell if a particular regex engine
is the best for your needs.

John's experimental code is quite exciting, but it shouldn't be surprising
that is performs as well as GRETA even for shorter strings. It is using
the same algorithm GRETA uses (depth-first recursive search), modulo some
changes to make it fit boost's interface. It should be noted that
recursive algorithms are not safe (potential for stack overflow). It
remains to be seen if John will be able to remove the recursion from his
implementation (and how that will affect performance), or offer a safer,
iterative algorithm as an alternative, as GRETA does.

Before I release a new version of GRETA, I test it on my linux box with
the latest version of gcc. I know that GRETA 2.5.9 compiles with gcc 3.2.
 John, what version of gcc are you using? If you send me the compiler
errors, I can look into it.

My sincerest apologies to those how have tried to get GRETA in the past
few days and been unable to. The web guys picked the wrong time to update
the website, and they didn't do a very good job. Here's hoping that they
get the problem straightened out some time today.

Eric


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk