The /p modifier in Perl5.10 regexps

casiano has asked for the wisdom of the Perl Monks concerning the following question:

Dear Monks

The /p modifier introduced in 5.10 makes ${^PREMATCH}, {$^MATCH}, and ${^POSTMATCH} available for use after matching. It is assumed it is faster than using $`, $& and $'. But when I benchmark this code:

pl@nereida:~/src/perl/perltesting$ cat ./timeregexpwithp.pl
#!/usr/local/lib/perl/5.10.1/bin//perl5.10.1 -w
use strict;
use Benchmark qw{:all};

cmpthese(
  1000000,
  {
    p => q{'hola juan' =~ /ju/p; my ($a, $b, $c) = (${^PREMATCH}, ${^M
+ATCH}, ${^POSTMATCH} ) },
    oldway => q{'hola juan' =~ /ju/; my ($a, $b, $c) = ($`, $&, $') }
  }
);

cmpthese(
  10000000,
  {
    pnoassign => q{'hola juan' =~ /ju/p; },
    oldwaynoassign => q{'hola juan' =~ /ju/; }
  }
);
[download]

I've got:

pl@nereida:~/src/perl/perltesting$ ./timeregexpwithp.pl
           Rate      p oldway
p      892857/s     --    -6%
oldway 952381/s     7%     --
                    Rate oldwaynoassign      pnoassign
oldwaynoassign 3184713/s             --            -1%
pnoassign      3225806/s             1%             --
[download]

It seems that the old way is faster. Any explanations?

Thanks

Comment on The /p modifier in Perl5.10 regexps Select or Download Code

Replies are listed 'Best First'.
Re: The /p modifier in Perl5.10 regexps by lodin (Hermit) on Aug 30, 2009 at 12:01 UTC
The point is that $` with friends slow down every other pattern as well. If you use /p you will still have a comparatively slow match, but it won't effect the other matches. use 5.010; use strict; use warnings; use Benchmark qw(cmpthese timethese); my $r1 = timethese( -1, { p_before => sub { 'hola juan' =~ /ju/p }, no_p_before => sub { 'hola juan' =~ /ju/ }, }, 'none'); eval '$`'; my $r2 = timethese( -1, { p_after => sub { 'hola juan' =~ /ju/p }, no_p_after => sub { 'hola juan' =~ /ju/ }, }, 'none'); cmpthese({ %$r1, %$r2 }); __END__ Rate p_before no_p_after p_after no_p_before p_before 1067899/s -- -2% -2% -80% no_p_after 1084733/s 2% -- -1% -80% p_after 1091288/s 2% 1% -- -79% no_p_before 5318863/s 398% 390% 387% -- [download] As you see, the pattern with /p and the patterns executed after $` has been seen are equally fast, and that makes sense since they do the same work. The pattern without /p and before $` is however significantly faster, since it does not have to compute the extra match variables. Hope this helps, lodin	[reply] [d/l] [select]
Re: The /p modifier in Perl5.10 regexps by casiano (Pilgrim) on Aug 30, 2009 at 11:26 UTC
Sorry, let me clarify a bit the question I mean, the difference is not significant when no assignment is involved and there is a more significant difference when I store the results Casiano	[reply]


Welcome to the Monastery
	PerlMonks