in reply to Performance penalty of using qr//

The problem is actually to do with (potential) captures. Currently, capture information (the information that is used to create the values of $1 et al when accessed) is stored as part of the regex object. This model doesn't work well when the same regex can be used in multiple places. This code: 
$r = qr/(.)/;
"a" =~ $r;
print "$1\n";
{ "b" =~ $r; print "$1\n"; }
print "$1\n";

[download]
outputs: 
a
b
a

[download]
Thus a single regex object has to be associated with multiple capture sets, which change as scopes are exited.

The current workaround for this is to duplicate the qr// object each time it's executed, which is sub-optimal. Unfortunately fixing this properly is non-trivial.

Dave.

Replies are listed 'Best First'.
Re^2: Performance penalty of using qr//
by Anonymous Monk on Dec 21, 2017 at 13:22 UTC
    Um ... so ... when is the sometime the optimization kicks in? Can it be measured?
[reply]
      Um ... so ... when is the sometime the optimization kicks in
      What optimisation are you referring to?

      Dave.

[reply]

        I'm guessing the optimization of using qr// over a plain string (ie: "Since Perl may compile the pattern at the moment of execution of the qr() operator, using qr() may have speed advantages in some situations ...")

        From your previous post I would say that this happens when there is a big compilation overhead, so I thought about this, and used this list of words for testing: 

        use strict;
use warnings;
use Benchmark qw( cmpthese timethese );

open my $words, "<", "linuxwords.txt" or die "$!";
my @words = <$words>;
chomp @words;

my @search = @words[0..10];

$" = "|";

my $re = qr/^(?:@words)$/;
my $str = "^(?:@words)\$";

my $r = timethese
(
    -5,
    {
        use_qr  => sub { map /$re/, @search },
        use_str => sub { map /$str/, @search },
        use_re  => sub { map /^(?:@words)$/, @search },
    }
);
cmpthese $r;

        [download]
        Benchmark: running use_qr, use_re, use_str for at least 5 CPU seconds.
+..
    use_qr:  5 wallclock secs ( 5.23 usr +  0.00 sys =  5.23 CPU) @ 98
+736.51/s (n=515997)
    use_re:  5 wallclock secs ( 5.33 usr +  0.00 sys =  5.33 CPU) @ 23
+.99/s (n=128)
   use_str:  5 wallclock secs ( 5.23 usr +  0.00 sys =  5.23 CPU) @ 22
+68.47/s (n=11855)
           Rate  use_re use_str  use_qr
use_re   24.0/s      --    -99%   -100%
use_str  2268/s   9355%      --    -98%
use_qr  98737/s 411431%   4253%      --

        [download]
        The re case is pretty bad because of the systematic interpolation, but I can't help but feel like I might be missing something because of how absurd the difference between qr and str is? But if this is correct, then qr is a clear winner for dictionary search.

[reply]
[d/l]
[select]

In Section Seekers of Perl Wisdom