On Mar 26, 2021, at 3:32 PM, Barry Revzin <barry.revzin@gmail.com> wrote:

On Fri, Mar 26, 2021 at 2:00 PM David Rector via SG7 <sg7@lists.isocpp.org> wrote:
Attached is a new paper addressing how the P2237/P2230 facilities might be expanded to be able handle the `Sum<T,U>` example discussed awhile back, and more generally subsume the capabilities of the "string-injection" approach, while avoiding its drawbacks. 

Hi David,

I didn't understand the Sum<T, U> example before and I still don't. There's not really a description of it in the paper, just a lot of code. What problem is it demonstrating?


Thanks for this feedback, I should add an example and context to the paper.

This was originally presented awhile back on this list; https://lists.isocpp.org/sg7/2021/02/0186.php is where I give a crude description of its possible utility, which Ville improves upon by suggesting this could be implemented as a generalization of the Mediator-Observer pattern (https://lists.isocpp.org/sg7/2021/02/0187.php; following that there is some good discussion about the how design patterns in general are good examples for metaprogramming, a good way to identify deficiencies.

The crude implementation in the paper might work as follows:
#include SumTU.h // Sum<T,U>

template <class T, class U> 
class operator_plus_exists {...}; //sfinae detector 

template<typename T, typename U, 
         typename = std::enable_if_t<!operator_plus_exists <T, U>::value>> 
Sum<T, U> operator+(T t, U u) { 
  return Sum<T, U>(t, u); 

struct A {
  int f;
  int g;
struct B {
  int g;
  A h;
struct C { 
  int f; 
  int g; 
  B h; //NB different type from B::h

int main() { 
  A a1{.f = 1, .g = 2}; 
  A a2{.f = 3, .g = 6};
  B b{         .g = 3, .h = A{.f = 4, .g = 5}} 
  C c( .f = 7, .g = 1, .h = B{        .g = 3, .h = A{.f = 1, .g = 2}); 
  Sum<Sum<Sum<A,A>, B>, C> res = a1 + a2 + b + c; 

  assert(res.f = 11);
  assert(res.g = 12);
  Sum<A,B> res2 = res.h;

  assert(res2.g == 8);