On Mon, Mar 04, 2013 at 04:12:21AM -0800, Ovid wrote:
> From: Jesse Luehrs <d...@tozt.net>
> >To: Pedro Melo <m...@simplicidade.org>
> >Cc: moose <moose@perl.org>
> >Sent: Friday, 1 March 2013, 18:05
> >Subject: Re: Role1 with attribute matches requires on Role2
> >
> >On Fri, Mar 01, 2013 at 04:56:57PM +0000, Pedro Melo wrote:
> >> Hi,
> >>
> >> I keep getting bitten by this, and I was wondering if this is
> >> something that will be solved eventually, or its something that would
> >> be nice to solve but it is actually tricky to do, or just that I'm
> >> doing it wrong.
> >>
> >> I have two roles, R1 and R2. R1 defines a attribute ctx. R2 requires a
> >> 'ctx'.
> >>
> >> If I use the two roles on a class C, with a single "with 'R1', 'R2';" I get
> >>
> >> 'R1|R2' requires the method 'ctx' to be implemented by 'C'
> >>
> >> I can work around it with two "with 'R1'; with 'R2';". But I loose all
> >> the other protections wrt duplicate methods I might have in R1 and R2.
> >>
> >> Opinions?
> >
> >This is something that we'd like to fix - the problem is mainly that
> >role accessors aren't generated until they are composed into a class,
> >but role application composes roles together before composing them into
> >the class. What we really need to do is provide a way for roles to say
> >"I'm going to generate a method of this name", which we would be able to
> >query during the role summation process (we can't actually generate the
> >methods themselves because doing so requires an instance metaclass,
> >which only classes have).
> >
> >Patches to fix this would be appreciated.
>
>
> Out of curiosity, how is this handled in the internals? Does the role
> composition do a simple ->can() check for methods? If so, during composition,
> could has() fake a forward declaration?
>
> {
> package Foo;
> sub bar;
> }
> say "yes" if Foo->can('bar');
>
> You're not actually creating the method yet, but you're promising that it
> will be there.
Yes, that is another possibility. I'm not sure which way would be
simpler.
-doy
