Re: RFC: Fix is_subtype_of to handle not-yet-defined role

On Wed, Jul 04, 2012 at 11:16:02AM -0500, Christopher J. Madsen wrote:
> After MooseX-Types 0.35 was released, I started getting failure reports
> for a number of my modules that use it.  This turned out to be caused by
> this issue:
> 
>   use Moose::Util::TypeConstraints;
>   # { package Foo::Role; use Moose::Role; }
>   my $type = role_type('Foo::Role');
>   $type->is_subtype_of('Bar');
> 
> This dies: Can't call method "does_role" on an undefined value at
> Moose/Meta/TypeConstraint/Role.pm line 101.
> 
> Uncommenting the definition of Foo::Role will remove the error.  But
> it's often necessary to define types (and subtypes) involving roles
> before the role itself is defined.  For instance, in a tree structure, a
> role might declare attributes using such types.
> 
> It would be possible to work around this in MooseX-Types, but it seems
> more straightforward to fix the root issue in
> Moose::Meta::TypeConstraint::Role, and that's what I did.
> 
> Please review branch rfc/undef-role.  It adds a test to demonstrate the
> problem and then fixes it by checking the return value of
> Class::MOP::class_of($self->role) before calling methods on it.
> 
> I've run "MOOSE_TEST_MD=MooseX prove -b t/author-test-my-dependents.t"
> on it, and don't get any additional failures (over the ones that happen
> on master).
> 
> Related RT tickets:
>   https://rt.cpan.org/Ticket/Display.html?id=77299
>   https://rt.cpan.org/Ticket/Display.html?id=77303
> 
> Thanks.

Merged, thanks.

-doy