sub build_lazy
{
    my ($var, $callback, @params) = @_;

    
    eval <<__EVAL__;
{
    my \$$var;
    sub $var {
        \\\$var = $callback->(@params) unless defined \$$var;
        \$$var;
    };
}
__EVAL__
}

# Later ...

build_lazy( 'dbh', \&build_dbh, @dbh_params );
build_lazy( 'foo', \&build_foo );
[download]

Of course, you might want some dwimmery for hashes vs. arrays vs. scalars, unless you're ok with everything as a reference. I would be, but that's just me. And, you might want to throw BEGIN or CHECK around the calls to build_lazy(), but that's probably overkill.

Well, if we're going with the sub approach, the eval isn't needed:

use strict;
sub build_lazy {
    my ($name, $callback, @params) = @_;
    
    my $var;
    no strict 'refs';
    *$name = sub {
        $var = $callback->(@params) unless defined $var;
        $var;
    };
}

build_lazy( 'dbh', sub { print @_ }, 'hi');
print dbh(), dbh(); # hi11
[download]

Its not that I don't like this approach (its still a hell of a lot better than what I have now), I just feel that... I don't know, just that its not the be can be done. Then again, I guess I have a weird mental condition where I'm unsatisfied with my code unless it takes on *exactly* the form thats in my mind, and I become uncomfortable when it isn't. I'm kind of weird. Thanks for the idea though.

Yeah, my little class in the root node would do the same thing, since it just checks for definedness. I just want a pretty tied interface to it. I guess I'll just write my own class, and maybe make it general enough for CPAN...