Add const_pointer_cast(global_ptr<U>)

[Amir Kamil]

We currently allow explicit conversions between global_ptr<T> and global_ptr<char>. I propose we remove this and instead follow the pattern of std::shared_ptr to introduce a full set of pointer casts (http://en.cppreference.com/w/cpp/memory/shared_ptr/pointer_cast). This would also enable polymorphic casts.

Proposed API:

namespace upcxx {

  template<typename T, typename U>
  global_ptr<T> static_pointer_cast(global_ptr<U> ptr);

  template<typename T, typename U>
  global_ptr<T> dynamic_pointer_cast(global_ptr<U> ptr);

  template<typename T, typename U>
  global_ptr<T> const_pointer_cast(global_ptr<U> ptr);

  template<typename T, typename U>
  global_ptr<T> reinterpret_pointer_cast(global_ptr<U> ptr);

} // namespace upcxx

We would specify similar semantics and constraints as in http://en.cppreference.com/w/cpp/memory/shared_ptr/pointer_cast. For dynamic_pointer_cast, we would additionally require ptr.is_local().

On a tangential note, we should also add a type member global_ptr<T>::element_type that aliases T.

[Amir Kamil]

Removed conversions between global_ptr<T> and global_ptr<char> and added reinterpret_pointer_cast in dbe1a3a.

[Amir Kamil]

Added global_ptr<T>::element_type in 01a04cc.

[Dan Bonachea]

This issue was triaged at the 2018-06-13 Pagoda meeting and assigned a new milestone/priority.

[Amir Kamil]

I'm not in favor of providing dynamic_pointer_cast, as it cannot be done locally. The others are doable without any communication.

[Dan Bonachea]

const_pointer_cast doesn't seem terribly useful until we resolve issue #51, as users would be wise to avoid global_ptr<const T> entirely until then

[Amir Kamil]

Merged static_pointer_cast in d57cd44.

[Dan Bonachea]

The only remaining conversion we might implement is const_pointer_cast(global_ptr<U>), once issue #51 is resolved to make global_ptr<const T> usable in practice.

[Dan Bonachea]

This issue was triaged at the 2019-07-24 Pagoda issue meeting and assigned a new milestone.

[Amir Kamil]

Resolved by PR #42.