Perl: Why is it slower to declare (my) variables inside a loop?

Question

What's the difference, from the interpreter's POV, between the following the following programs:

#!/usr/bin/perl -w

use strict;

for (1..10000000) {
    my $jimmy = $_**2;
}

and

#!/usr/bin/perl -w

use strict;

my $jimmy;
for (1..10000000) {
    $jimmy = $_**2;
}

"time" reports for the first program:

real    0m1.519s
user    0m1.513s
sys     0m0.004s

and for the second:

real    0m1.023s
user    0m1.012s
sys     0m0.002s

Answer 1

The my declaration in Perl has two primary effects; a compile-time one (wherein it allocates a slot on the containing sub's scratchpad, and makes sure that all references to that name within the proper scope are resolved to that particular scratchpad slot), and a runtime one (wherein it resets the value of that pad slot to undef, or to some particular value if you wrote my $var = foo).

The compile-time portion of course has zero amortized runtime cost, but the runtime portion is run once each time execution passes the my declaration. As others have pointed out, your two examples have different performance because they have different semantics in general -- one clears the variable every time through the loop, and the other doesn't.

Answer 2

Since the example programs you have given do not really do anything it is hard to give you a specific reason why one type of declaration would be better than the other. As many other posters have pointed out, declaring the variable in the loop creates a new variable each time. In your examples that creation is redundant, but consider the following examples using closures.

my @closures;
my $jimmy;

for (1 .. 10) {
    $jimmy = $_** 2;
    push @closures, sub {print "$jimmy\n"};
}

and this one:

my @closures;

for (1 .. 10) {
    my $jimmy = $_** 2;
    push @closures, sub {print "$jimmy\n"};
}

In each case the code builds up a series of code references, but in the first example since all the code refs refer to the same $jimmy each one will print 100 when called. In the second example each code ref will print a different number (1, 4, 9, 16, 25, ...)

So in this case the time difference does not really matter since the two blocks of code do very different things.