Re^3: RE performance

Thanks!

Isn't it strange that the noCapture case is faster than the noParens??

Here's what use re 'debug' gets me for the 4 regex variants:

Compiling REx `(est|\d)'
size 10 first at 3
   1: OPEN1(3)
   3:   BRANCH(6)
   4:     EXACT <est>(8)
   6:   BRANCH(8)
   7:     DIGIT(8)
   8: CLOSE1(10)
  10: END(0)
minlen 1
Compiling REx `(?:est|\d)'
size 7
   1: BRANCH(4)
   2:   EXACT <est>(7)
   4: BRANCH(6)
   5:   DIGIT(7)
   6: TAIL(7)
   7: END(0)
minlen 1
Compiling REx `est|\d'
size 6
   1: BRANCH(4)
   2:   EXACT <est>(6)
   4: BRANCH(6)
   5:   DIGIT(6)
   6: END(0)
minlen 1
Compiling REx `est'
size 3 first at 1
   1: EXACT <est>(3)
   3: END(0)
anchored `est' at 0 (checking anchored isall) minlen 3
Compiling REx `\d'
size 2 first at 1
   1: DIGIT(2)
   2: END(0)
stclass `DIGIT' minlen 1
[download]

The TAIL(7) on the noCapture case looks redundant to me; maybe that's something they fixed in 5.8.0?

What happens if you look at those regexes under 5.8.0 with -Dr or use re 'debug', please?

( Edit: Zaxo did this just above, and got the same results, except he gets offsets. No idea what those mean...)
--
Mike

Comment on Re^3: RE performance Download Code

Replies are listed 'Best First'.
Re: Re^3: RE performance by japhy (Canon) on Sep 06, 2002 at 14:38 UTC
"TAIL" is the node that defines the end of a non-capturing parenthetical group or an if-then parenthetical group (like `(?(1)a\|b)`). That's about all. It gets optimized out for non-capturing parens if there is no "BRANCH" found (that is, no "\|" metacharacter), so that something like `/A(?:B)C/` can be optimized to an "EXACT" node matching "ABC". Basically, "TAIL" takes the place of having another "indented" layer. _____________________________________________________ Jeff`[japhy]`Pinyan: Perl, regex, and perl hacker, who'd like a job (NYC-area) `s++=END;++y(;-P)}y js++=;shajsj<++y(p-q)}?print:??;`	[reply] [d/l] [select]
Re: Re^3: RE performance by PodMaster (Abbot) on Sep 05, 2002 at 06:44 UTC
I think I got a line on that one (i finally read perldebguts), and I quote (cause I likes to ;) # Do nothing NOTHING no Match empty string. # A variant of above which delimits a group, thus stops optimizations TAIL no Match empty string. Can jump here from outside. So i guess this is because qr{} will generate patterns like that, and you normally wouldn't use /(?:foo\|\d)/, you'd do something like /$pat\|$QRed2/. This is because if perl sees 4: BRANCH(6) 5: DIGIT(6) which means there are no patterns after DIGIT , it can optimize (cause the digit in parentheses is the index to the next pattern, or something along those lines), and that's not neccessarily true if you got (?:). I think it's one of those overlooked optimizations (hopefully japhy will answer concretely ~ i'm not a re => god yet ;D). No idea what this offsets thing is, I don't get it , but then I don't have perl compiled with the -DDEBUGGING, so it might relate to that somehow. `____________________________________________________` ** The Third rule of perl club is a statement of fact: pod is sexy.	[reply]