Dr. Franz Fehringer wrote:
On http://www.cuj.com/boost/ (Boost Corner) the first article is I/O System: dynamic_any campaign by Alexander Nasonov. Is dynamic_any really scheduled for inclusion in boost (1.32 say)?
Not really. It's not even scheduled for review yet. Actually, I'm working on multiple dispatch in the library. This means that call/call_ex scheme will be replaced with something new. My current favorite is nested definition class (like spirit::grammar's definition). Implementation for a function with one anyT argument is easy and done already (I should think it over, though). Multi-methods come into play when there are more then one anyT in function's signature. My current vision is based on proposal N1529: Draft proposal for adding Multimethods to C++ Document number: WG21/N1529 = J16/03-0112 Author: Julian Smith http://open-std.org/jtc1/sc22/wg21/docs/papers/2003/n1529.html // Example is taken from N1529 struct shape {...}; struct square : shape {...}; struct triangle : shape {...}; bool overlap( virtual shape& a, virtual shape& b); bool overlap( static square& a, static triangle& b) {...} bool overlap( static triangle& a, static square& b) {...} bool overlap( static shape& a, static square& b) {...} bool overlap( static square& a, static shape& b) {...} I'm trying to be as close to this as possible. Compare: struct overlap : function<overlap, bool(anyT&, anyT&)> { struct definition { bool operator()(overload_id<0>, square& a, triangle& b) const {...} bool operator()(overload_id<1>, triangle& a, square& b) const {...} bool operator()(overload_id<2>, shape& a, square& b) const {...} bool operator()(overload_id<3>, square& a, shape& b) const {...} definition(overlap const& ) {} BOOST_STATIC_CONSTAT(int, max_overload_id = 3); }; }; New syntax for virtual arguments isn't required because anyT arguments imply virtuality. In order to enumerate all arguments' combinations at compile-time, overload_id<N> is introduced. I hope to finish it within a couple of months. -- Alexander Nasonov