Re: Extending attribute declaration syntax - best practice

On Tue, Aug 28, 2012 at 10:19:36AM -0700, Peter Shangov wrote:
> Hi, 
> 
> MoooseX::Has::Options (https://metacpan.org/module/MooseX::Has::Options) adds a different syntax for declaring Moose attribute options, i.e. instead of:
> 
>     has 'foo', is => 'ro', isa => 'Str', lazy_build => 1;
> 
> it allows you to say:
> 
>     has 'foo', qw(:ro :lazy_build), isa => 'Str';
> 
> The current implementation is a little naive and works by directly replacing the 'has' subroutine in the calling package. Instead I want to make this work providing a new attribute metaclass that knows how to deal with the column-option syntax. The problem is that Moose::has() checks that the options list is even-sized, and will die with an odd number of column options.
> 
> So which of the following imperfect approaches would you recommend that I should take:
> 
> 1. Replace the imported 'has' with a new one that does exactly the same but without the list size check. This means that the custom 'has' will go out of sync whenever the implementation of Moose::has() changes.
> 
> 2. Apply an 'around' method modifier to Moose::has() that adds a dummy column option if the options list is not even-sized, and make the attribute metaclass handle it as a no-op. This means that the end will get to see this dummy option (e.g. in stack traces) and possibly get confused.
> 
> 3. Apply an 'around' method modifier to Moose::has() that does all the column options pre-processing itself. This means not using an attribute metaclass at all and is only marginally better than the original approach.
> 
> Any ideas?

That check really doesn't belong in Moose::has - it should be done in
_process_options in Class::MOP::Attribute. A patch to move that over
would be quite appreciated.

-doy