[Bug tree-optimization/102586] [12 Regression] ICE in clear_padding_type, at gimple-fold.c:4798 since r12-3433-ga25e0b5e6ac8a77a

["jason at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102586

--- Comment #16 from Jason Merrill <jason at gcc dot gnu.org> ---
Created attachment 52410
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52410&action=edit
sketch of vbase handling

This is roughly what I had in mind, though it's algorithmically poor because it
walks all the bases each time.

But then it occurred to me that unless we know the complete object type, we
don't know whether we're dealing with a base subobject, and risk clearing too
much:

struct C1 { virtual void f() {} char d; }; // vfn to make non-layout-POD
struct C2: C1 { char c; };

__attribute__((noipa)) void
foo (C1 *q)
{
  __builtin_clear_padding (q);
}

int main()
{
  C2 c2;
  c2.c = 42;
  foo (&c2);
  if (c2.c != 42)
    __builtin_abort();
}

Because of this, I think we probably want __builtin_clear_padding to reject
arguments that point to non-trivially-copyable type.  And then we don't need to
worry about vbases.