汉字课与速记课的表现

当有疑问时,基准!

\d+ 速度确实更快。基督教青年会。

use strict; use warnings;
use Benchmark qw(:all);
use feature "switch";

my $r1='\d+';
my $r2='[0-9]+';
my $r3='[[:digit:]]+';

sub test {
    my @lines = <DATA>;
    $_=shift;
    my $RegEx=(caller(0))[3];
    given($_) {
        when(1) { $RegEx=$r1; }
        when(2) { $RegEx=$r2; }
        when(3) { $RegEx=$r3; }
        default { die "$RegEx can't deal with $_\n"; }
    }   
    my $ln;
    my $total;
    my @numbers;
    foreach my $line (@lines) {
        $total=0;
        @numbers=$line=~/($RegEx)/g;
        $total+=$_ for (@numbers) ;
        $ln=$numbers[$#numbers];
        $total-=$ln;
        if ($ln != $total) {
            print "Bad RegEx result: Last Num != total in line!\n";
            print "Total=$total, ln=$ln, line:$line";
            die;
        }       
    }   
}

cmpthese(-10, {$r1=>'test(1)', $r2=>'test(2)', $r3=>'test(3)'});


__DATA__
Clip clap clock 1 mouse ran up the clock with 3 hands. The total here is 4.
The mouse with 2 ears followed. The total here is 2.
After that, the 6 wiskered mouse did dances with 14 second timing. 20.
It is hard to make up 5 lines with 2 or 3 numbers in each line. 10.
You start thinking about nurserey rhymes with 1 o 2 or 3 number. 6.
1 12 13 123 23 13 55 66 21 45 1 373

                 Rate [[:digit:]]+       [0-9]+          \d+
[[:digit:]]+ 200781/s           --          -1%          -2%
[0-9]+       202831/s           1%           --          -1%
\d+          204605/s           2%           1%           --

在带有Perl 5.10.1的Ubuntu 10.04上:

                 Rate [[:digit:]]+       [0-9]+          \d+
[[:digit:]]+ 264412/s           --          -3%          -6%
[0-9]+       273202/s           3%           --          -3%
\d+          280541/s           6%           3%           --

好吧,不知道实现(也不想在Reflector中遍历数百行代码),很难确定,所以让我们尝试一些疯狂的推测。

• 假设A)“[J-P]”编码如下 "JKLMNOP".Contains(chr)
• 假设B)搜索数字的速度可以通过 Char.IsDigit(chr)

鉴于此,这归结为两个问题:

• Regex会治疗吗 \d
• Regex是否有可能将[0-9]识别为不同于[J-P]的特例并使用它 IsDigit 而不是 Contains

所以,我想说有可能\d比[0-9]快。

(注意:我使用C#/.NET作为例子,但是基本原则应该是相同的,无论平台如何)