RE: [boost] Re: [Variant] Binary visitation and return type

From: Hartmut Kaiser (hartmutkaiser_at_[hidden])
Date: 2004-12-08 11:22:16

Stefan Slapeta wrote:

> > is there a means of using the binary visitation pattern provided by 
> > the Boost.Variant library and make the return type of the 
> visitation 
> > process dependent on the types currently stored inside the visited 
> > variant object instances?
> > 
> 
> When I look at the current apply_visitor interface, I can't 
> think of any:
> 
> template<typename Visitor, typename Variant>
>    typename Visitor::result_type
>    apply_visitor(Visitor & visitor, Variant & operand);
> 
> 
> You would need something like the lazy function return type 
> deduction in
> Phoenix:
> 
> struct Visitor
> {
>      template <typename Arg1, typename Arg2>
>      struct result
>      {
>          typedef ... result_type;
>      };
> 
>      ...
> };
> 
> and
> 
> template<typename Visitor, typename Variant>
>    typename Visitor::result<???>::result_type
>    apply_visitor(Visitor & visitor, Variant & operand);
> 
> I really don't know if ??? (the real argument types) can be 
> determined at compile time.

Sure. Actually I've expected to have something like that used in the binary
visiting code...

> >     variant_type term1 = 1L;
> >     variant_type term2 = 2UL;
> >     variant_type term3 = -2L;
> > 
> >     variant_type unsigned_result = 
> >         boost::apply_visitor(add_visitor(), term1, term2);
> >     variant_type signed_result = 
> >         boost::apply_visitor(add_visitor(), term1, term3);
> > 
> > Is this possible with the current implementation?
> > 
> 
> That's something different: here you assign again to a 
> variant! Why don't you simply use boost::variant as return 
> type for the visitor?

Yes, you're right. I've simply choose a bad sample for what I was looking
for. 
But it seems, you've got my point anyway.

Regards Hartmut