CGI->can

cgi_can.pl

use strict;
use warnings;
use Benchmark 'cmpthese';
use CGI;

package Cat;

sub walk {}

package main;

cmpthese(-1, {
    'CGI->can("new")' => sub {
        my $code = CGI->can('new'); # compiled by default
    },
    'CGI->can("cache")' => sub {
        my $code = CGI->can('cache'); # autoloaded on demand
    },
    'CGI->can("no_cache")' => sub {
        my $code = CGI->can('no_cache'); # doesn't exist
    },
    'Cat->can("walk")' => sub {
        my $code = Cat->can('walk');
    },
    'Cat->can("run")' => sub {
        my $code = Cat->can('run');
    },
});

__END__
                          Rate CGI->can("no_cache") CGI->can("new") CGI->can("cache") Cat->can("walk") Cat->can("run")
CGI->can("no_cache")    6516/s                   --            -98%              -98%            -100%           -100%
CGI->can("new")       267962/s                4012%              --               -1%             -79%            -83%
CGI->can("cache")     270491/s                4051%              1%                --             -79%            -83%
Cat->can("walk")     1303975/s               19911%            387%              382%               --            -20%
Cat->can("run")      1622942/s               24806%            506%              500%              24%              --

discussion.md

• CGI.pm overrides the UNIVERSAL->can method, and so CGI->can is basically slower than an ordinary module's. That's not so amazing.
• Though CGI.pm implements the cache method, the method is not compiled by default, but autoloaded on demand.
• CGI->can("no_cache") always returns false because the method is not implemented by CGI.pm.
• According to the above benchmark result, CGI->can("no_cache") is much slower than CGI->can("cache"). That's the problem.