in reply to Problems I've had with array based objects

Perhaps I'm oversimplifying, but wouldn't this work as a base class for array based objects? 
package foo;

use strict;

sub new {
    bless [], ref $_[0] || $_[0] || __PACKAGE__;
}

sub declare {
    my $self = shift;

    ## auto vivification fails to make the correct indexes
    $self->[0] = {} unless $self->[0];

    no strict 'refs';

    foreach my $name (@_) {
        ## create the index reference
        $self->[0]{$name} = @$self;

        ## create a nicely named index variable
        ${ref($self) . '::' . $name . '_idx'} = @$self;

        ## create the accessor/mutator methods
        (my $code = 'sub { $_[0][VALUE] }') =~ s/VALUE/scalar @$self/e
+;
        *{ref($self) . '::get_' . $name} = eval $code;
        ($code = 'sub { $_[0][VALUE] = $_[1] }') =~ s/VALUE/scalar @$s
+elf/e;
        *{ref($self) . '::set_' . $name} = eval $code;

        ## create the (unitialized) attribute itself
        push @$self, undef;
    }
}

[download]
It trades space (possibly a lot of space) for time and convenience, but that's often a good trade. I'm sure there's a (perhaps non-trivial) way to move the index references back into package space to save memory, but I'll leave that as an exercise.

Replies are listed 'Best First'.
Re^2: Problems I've had with array based objects
by ikegami (Patriarch) on Aug 04, 2006 at 00:39 UTC

    Did you intend to support multiple inheritance? If so, you need to remove 

    ## create a nicely named index variable
${ref($self) . '::' . $name . '_idx'} = @$self;

    [download]

    Consider the case wheren Bar isa Foo, and where Baz isa Moo and Foo. The index of a particular attribute of Foo is object-specific, but the code to remove behaves as if the index is class-specific.

[reply]
[d/l]
      Absolutely correct. I didn't like making indexes object specific to begin with, and hence must have gotten ahead of myself. I think the right way to do this would put the indexes into the class namespace, but I was lazy.

      UPDATE: In fact, the same argument means that the accessor/mutator creation won't work either because new objects will clobber the old closures. I think the solution is to make everything package scoped, but clearly I was, in fact, over-simplifying.

[reply]

In Section Meditations