Should Moose::Meta::Method::Accessor be using inline_slot_access?

[Stefan O'Rear]

It is very easy to implement Class::MOP::Instance's
inline_[sg]et_slot_value functions in terms of inline_slot_access,
but very hard to do the reverse; because of this, it seems to me
that Moose::Meta::Method::Accessor should always use the more
specific, easier to implement variant.  Right now, the implementations
of read/write accessors and lazy attributes both use inline_slot_access
in fairly easy to eliminate ways; is this a bug?

If so, should code outside subclasses of Class::MOP::Instance be using
inline_slot_access at all?

For context and clarification, here is a mockup of what I'm currently
trying to do.  This file fails to compile at the end when it attempts
to inline an accessor for a lazy attribute, and so uses
inline_slot_access in a non-essential way.

-sorear

#! /usr/bin/perl
package SomeAwesomeDB;

sub new_row { }
sub read { }
sub write { }

package MooseX::SomeAwesomeDBFields;

use Moose::Role;

sub inline_create_instance {
    my ($self, $classvar) = @_;

    "bless SomeAwesomeDB::new_row(), $classvar";
}

sub inline_get_slot_value {
    my ($self, $invar, $slot) = @_;

    "SomeAwesomeDB::read($invar, \"$slot\")";
}

sub inline_set_slot_value {
    my ($self, $invar, $slot, $valexp) = @_;

    "SomeAwesomeDB::write($invar, \"$slot\", $valexp)";
}

sub inline_slot_access {
    die "SomeAwesomeDB fields are not lvalues, do you expect to tie them?";
}

# implementation of methods not called in the example deliberately
# omitted

package Toy;

use Moose;
use Moose::Util::MetaRole;

Moose::Util::MetaRole::apply_metaclass_roles
    (for_class  => __PACKAGE__,
     instance_metaclass_roles => [ 'MooseX::SomeAwesomeDBFields' ]);

has breaker => (
    is  => 'ro',
    isa => 'bool',
    lazy => 1,
    default => sub { 0 },
);

__PACKAGE__->meta->make_immutable;

signature.asc
Description: Digital signature