Perl is indeed reluctant to release "unused" memory. Of course, your definition of "unused" may be different from Perl's.

In your second script, you indeed need at least 2GB to keep the data in $t and $mem.

And you are right that in your first script, Perl first builds the string and then copies it to the variable. After this assignment, the memory used to build the string can be re-used, but Perl rather reserves another batch of memory when you copy the first variable. I *think* it is because the memory footprint of the variable is slightly more than the memory used for the building of the string.

Consider the following two scripts:

$mem1 = 'a' x 1024 ** 2;
$mem2 = 'b' x 1024 ** 2;
[download]

and

$mem1 = 'a' x 1024 ** 2;
$mem2 = 'b' x 1024 ** 2;
$mem1 = $mem2;
[download]

Both scripts use almost exactly the same amount of memory, but of course, both variables will perfectly fit into each other.