{
    my %working_dir_of;
}
sub get {
    ...
}
[download]

won't work, but

my %working_dir_of;

sub get {
   ...
}
[download]

does: accessor and declaration have to be in the same scope. In principle, the string eval technique works, as seen with this snippet

#!/usr/bin/perl
use strict;
my %hash = ( foo => 'bar');

sub meth {
    my ($h,$k) = @_;
    my $string = sprintf 'print $%s{%s},"\n"', $h, $k;
    eval $string;
};  
meth('hash','foo');
__END__
bar
[download]

but it looks a bit whacky and eval is costly, too. Why is it deemed good practice to name an attribute e.g. dbh and the corresponding inside out attribute hash %dbh_of ? Looks to me like calling for trouble.

If you want to stick with that "good practice", I'd recommend to set up a property hash keyed by the real attribute names, and having as values references to the attribute hashes:

my( %dbh_of, %dir_of, ...);

my %props = (
    dbh => \%dbh_of,
    dir => \%dir_of,
    ...
);

sub get {
    my($self,$attr) = @_;
    my $return_val;
  
    # if valid attribute, return the object's value
    if( defined $props {$attr} ) {
        return $props {$attr} -> { refaddr $self};
    }
    # otherwise, carp a warning
    else {
        carp("Invalid attribute '$attr'");
    }
}
[download]

Note that I've dropped the return value in case of failure; returning nothing at all in this case is better than returning an undefined value IMHO.

Have a look at Alter for a simpler, faster and cleaner approach to insiteoudish object data encapsulation (e.g. no hassle with DESTROY).

_($_=" "x(1<<5)."?\n".q·/)Oo.  G°\        /
                              /\_¯/(q    /
----------------------------  \__(m.====·.(_("always off the crowd"))."·
");sub _{s./.($e="'Itrs `mnsgdq Gdbj O`qkdq")=~y/"-y/#-z/;$e.e && print}