Hi all,
  I'm working on (basic) devirtualization, and am wondering how immutable the vptr is in D. There is no clear spec on this.
I am currently assuming that an object's vptr is completely immutable and that use of the object after changes to its vptr are UB (in contrast to the more complicated situation in C++).

Please help me find an example where this assumption does not hold.

```
class A {
  void foo();
}

void may_do_anything(A a); // e.g. delete, destroy, ...

void my_current_assumption(A a) {
  a.foo();
  auto tempvptr = a.__vptr;
  // The assumption is that if the vptr changes, that use of `a` afterwards is UB.
  may_do_anything(a);
  a.foo(); // either UB or identical to "tempvptr.foo(a)" (pardon the notation)
}

void test_vptr(A a) {
  a.foo();
  may_do_anything(a);
  if (a.__vptr) { // use of `a`, so the check is UB if vptr has changed
     //...
  }
}
```

-Johan

On Saturday, 11 November 2017 at 12:52:09 UTC, Johan Engelen wrote:
> Hi all,
>   I'm working on (basic) devirtualization, and am wondering how immutable the vptr is in D. There is no clear spec on this.
> I am currently assuming that an object's vptr is completely immutable and that use of the object after changes to its vptr are UB (in contrast to the more complicated situation in C++).
>
> Please help me find an example where this assumption does not hold.
>
> ```
> class A {
>   void foo();
> }

IIRC entry 0 of the vtable is for TypeInfo*, so at least casting and destruction won't work correctly.