use Benchmark qw(cmpthese);

## take the results with a grain of salt, mmmkay?
cmpthese(-15, {
    predeclare => sub { my $x = {}; $x->{foo} = "bar" x 20 },
    nodeclare  => sub { my $x; $x->{foo} = "bar" x 20 }
});

__output__

Benchmark: running nodeclare, predeclare, each for at least 15 CPU seconds...
 nodeclare: 24 wallclock secs (15.33 usr +  0.00 sys = 15.33 CPU) @ 110465.75/s (n=1693440)
predeclare: 18 wallclock secs (15.58 usr +  0.02 sys = 15.60 CPU) @ 89620.83/s (n=1398085)
               Rate predeclare  nodeclare
predeclare  89621/s         --       -19%
nodeclare  110466/s        23%         --

##</code><code>##

my $foo = "a string";
print "\$foo is $foo\n";

## this $foo only lasts to the end of the for loop scope
for my $foo (1..3) {
    print "\$foo is $foo\n";
}

print "\$foo is $foo\n";

__output__

$foo is a string
$foo is 1
$foo is 2
$foo is 3
$foo is a string