@zygoloid: PTAL

I spoke with zygoloid, and the conclusion is two-fold: 1) is that we should rephrase this paragraph to say "If X, the behaviour is undefined", but 2) is that we should clarify the wording to make it clear that the following is not allowed:

struct A {
  const int n;
  ~A() { printf("%d", n");
};

void f() {
  A a = {4};
  a.~A();
  new (&a) A{5};
}

It should do so by being clear that a.~A() refers to the "original pointer" (and not a laundered version thereof). With that in place, the paragraph in question might become redundant.

We should perhaps file a CWG issue for this.

This has been converted to a core issue.