I thought perl changed back and forth as needed.

Think about it like this:

A Perl-scalar is not something has only holds a single value, bit rather it consists of a number of "slots" that contain values for the different contexts in which the scalar can be used.

Normally these values are "the same" and are computed by Perl as needed (e.g. a scalar containing a 4 in it's integer-slot will get the value "4" (a string!) in it's string-slot when used in a string-context), but not always.

A good example for this are references. Observe:

my $ref = []; # reference to an array

my $a = $ref + 0; # force numeric context
my $b = "$ref"; # force string-contect

print "$a $b\n";
[download]

As you can see in numeric context the value becomes the memory-address whereas in string-context it also shows the type of the reference "ARRAY(0x.....)".