[Bug tree-optimization/84470] test for address of member being null not eliminated

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84470

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2018-02-27
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, is p->a valid to compute if p is NULL?  Is it valid to compute p->b if p
is NULL for

struct A { char a[4]; char b[4]; };

?

Because you say "could" in

 if (p->a == 0)          // could only be true when p is null
    __builtin_abort ();   // can be eliminated

which implies it would invoke undefined behavior?

ISTR we still want to not break old offsetof implementations using &(T
*)0->member
style computation?

But yes, wrapping rules say that &p->m is never NULL if m is not a member
at offset zero.  See tree_single_nonzero_warnv_p for ADDR_EXPR handling.