Also keep in mind that the first scenario specifies a reference to a variable, other_hash.   So, you now have two or more references to the same chunk of storage.   (Fortran calls that EQUIVALENCE.)

Data::Dumper makes this clear.   Consider this:

use strict; use warnings; 
use Data::Dumper; 
my @a = (1, 2); my $b = { "c"=>\@a }; 
print Dumper(\@a, $b); 
$b->{"c"}[0]=4; 
print Dumper(\@a, $b);'
[download]

$VAR1 = [
          1,
          2
        ];
$VAR2 = {
          'c' => $VAR1
        };
$VAR1 = [
          4,
          2
        ];
$VAR2 = {
          'c' => $VAR1
        };
[download]

The dumper knows that the hash-key points to the same block of storage that the array-variable does, so it refers to it in the second case by its name.   And you can see how a change, made by means of the hash-key, changed the value in the array variable, because the two are the same.   This is a very easy mistake to make in a production setting.

Also note that “references” will work in this way, whether-or-not a particular reference comes into existence by means of the “autovivification” feature.