[Bug tree-optimization/86572] unsafe strlen folding of const arguments with non-const offset

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86572

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=86434

--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
See also bug 86434 which discusses another couple of problems with this
transformation.  The solution in this case isn't necessarily to defer the
folding until later but do it in a way that avoids these excessive results for
undefined input.  For example, returning zero would be safer and in line with
what GCC does in other cases, such as in some instances of out-of-bounds array
accesses:

$ cat c.c && gcc -O2 -Wall -Wextra -fdump-tree-optimized=/dev/stdout c.c
const char a[4] = { 1, 2, 3 };

int f (void)
{
  return a[8];   // missing -Warray-bounds, folded to zero
}

;; Function f (f, funcdef_no=0, decl_uid=1899, cgraph_uid=1, symbol_order=1)

f ()
{
  <bb 2> [local count: 1073741825]:
  return 0;

}