Eric Niebler wrote:
Don't feel stupid; there are contexts in which you actually *can* use wrap in type computations within wrap itself, so it's easy to get confused.
template<class Expr> struct wrap : proto::extends<Expr, wrap<Expr> > { template<struct Sig> struct result { typedef typename some_fun<wrap>::type type; //OK! }; };
I ended with that anyway as the result_type of a NT² expression depends on how you access them as we support floating point indexing that perform linear interpolation, ranges and more. So in the end , the geenral value_type was a bad idea whiel result_of compliance was OK.
The difference is that by the time result gets instantiated somewhere else in your program, wrap will be complete. This will work fine until you try to instantiate result from within wrap itself (e.g. to compute the return type of a nullary operator() overload). Then it blows up with very mysterious errors. It's tricky I saw that :€