[class.union.general] Unclear selection of variant member activated by placement new

[frederick-vs-ja]

Full name of submitter (unless configured in github; will be published with the issue): Jiang An

Reference (section label): [class.union.general], [expr.static.cast]

Link to reflector thread (if any):

Issue description:

May be a part of CWG2675. Consider the following example:

#include <new>

union U {
  int a;
  int b;
};

static_assert(sizeof(U) == sizeof(int)); // passes on most platforms

int main() {
  U u; // no active member until C++26; u.a is active since C++26
  ::new (&u) int{};                         // #1
  ::new (reinterpret_cast<int*>(&u)) int{}; // #2
  ::new (&u.a) int{};                       // #3
}

On most platforms, u, u.a, and u.b occupy the same storage. So according to [intro.object] p2, all of #1, #2, #3 should create a subobject of u. But it is unclear which subobject is replaced.

It seems that we should at least guarantee #3 to replace u.a, which is necessary for std::variant and std::expected.

For #2, it's even unclear whether reinterpret_cast<int*>(&u) points to u.a or u.b. I guess it's better to introduce angelic indeterminism for such selections, i.e. to select an unspecified subobject in a way maximizing defined prefix of the execution (as suggested by P1839 and updated by CWG3039).

Suggested resolution: