[Bug middle-end/50460] [4.7 Regression] __builtin___strcpy_chk/__builtin_object_size don't work

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50460

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-09-25 11:11:21 UTC ---
(In reply to comment #4)
> Looking at:
> const char *str1 = "JIHGFEDCBA";
> #define strcpy(x,y) __builtin___strcpy_chk (x, y, __builtin_object_size (x, 1))
> 
> int
> f1 (void)
> {
>   struct A { char buf1[9]; char buf2[1]; } a;
>   strcpy (a.buf1 + (0 + 4), str1 + 5);
>   return 0;
> }
> 
> int
> f2 (void)
> {
>   struct A { char buf1[9]; char buf2[1]; } a;
>   strcpy ((char *) &a + (0 + 4), str1 + 5);
>   return 0;
> }
> 
> int
> f3 (void)
> {
>   struct A { char buf1[9]; char buf2[1]; } a;
>   char *p = (char *) &a;
>   strcpy (p + (0 + 4), str1 + 5);
>   return 0;
> }
> 
> int
> f4 (void)
> {
>   struct A { char buf0; char buf1[9]; char buf2[1]; } a;
>   char *p = (char *) &a;
>   strcpy (p + (0 + 5), str1 + 5);
>   return 0;
> }
> 
> int
> f5 (void)
> {
>   struct A { char buf0; char buf1[9]; char buf2[1]; } a;
>   strcpy ((char *) &a + (0 + 5), str1 + 5);
>   return 0;
> }
> 
> with GCC 4.4, seems we have always reconstructed it into &a.buf1[4].
> So likely we want to reconstruct it from the MEM_REF in the *.objsz pass then.
> If there is union involved, we probably want to reconstruct it to the
> alternative with the largest possible __builtin_object_size (X, 1) resp.
> smallest possible __builtin_object_size (X, 3).

I'm not sure.  What's the C / fortify difference of a.buf1 + 9 vs. a.buf2?
Both would be MEM[&a, 9].  I suppose we didn't re-construct array-refs in
4.4 from

 void *p = a.buf1;
 char *q = p + 4;

so, did we fail with 4.4 here, too?