This seems to me to be a bug, I just wanted to see if anyone had seen it before and if it reproduces for them.

#!/usr/bin/perl                                   
use strict;                                          
use warnings FATAL => qw(all);                       
                                                     
my $t = test->new();                                 
print "The End.\n";                                  
                                                     
{ package test;                                      
                                                     
    sub new {                                        
        my $class = shift;                           
        my $self = {                                 
            list => [ 'a', 'b', 'c' ]                
        };                                           
        bless $self, $class;                         
    }                                                
                                                     
    sub DESTROY {                                    
        my $self = shift;                            
        print STDERR "Bye!".$self->{list}->{4}."\n"; 
    }                                                
                                                     
1;                                                   
}
[download]

Using perl "(v5.18.4) built for x86_64-linux-thread-multi", rather than the nice runtime warning:

The End.
        (in cleanup) Not a HASH reference at ./test.pl line 20.

I get:

The End.
shell: Job 1, “./test.pl” terminated by signal SIGSEGV (Address boundary error)

Which is tedious to then actually pinpoint if the code is more elaborate. Fortunately the deciding factor is:

use warnings FATAL => qw(all);
[download]

Just plain `warnings` or `-w` won't do it. The backtrace starts in malloc.c and leads to an endless loop involving "Perl_ck_warner()" where it seems to have been trying to generate the "(in cleanup)" error message.