Re: [PATCH] Don't issue array bound warnings on zero-length arrays

[Richard Biener <richard dot guenther at gmail dot com>]

On Fri, Aug 30, 2013 at 5:13 PM, Meador Inge <meadori@codesourcery.com> wrote:
> Hi All,
>
> This patch fixes a minor issue that can occur when issuing array bounds
> warnings.  In GNU C mode we allow empty lists and their upper bound is
> initialized to -1.  This confuses the array bound analysis in VRP and
> in some cases we end up issuing false positives.  This patch fixes
> the issue by bailing out when a zero-length array is encountered.
>
> OK for trunk?
>
> gcc/
>
> 2013-08-30  Meador Inge  <meadori@codesourcery.com>
>
>         * tree-vrp.c (check_array_ref): Bail out no emtpy arrays.
>
> gcc/testsuite/
>
> 2013-08-30  Meador Inge  <meadori@codesourcery.com>
>
>         * gcc.dg/Warray-bounds-11.c: New testcase.
>
> Index: gcc/testsuite/gcc.dg/Warray-bounds-11.c
> ===================================================================
> --- gcc/testsuite/gcc.dg/Warray-bounds-11.c     (revision 0)
> +++ gcc/testsuite/gcc.dg/Warray-bounds-11.c     (revision 0)
> @@ -0,0 +1,12 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -Warray-bounds -std=gnu99" } */
> +/* Test zero-length arrays for GNU C.  */
> +
> +unsigned int a[] = { };
> +unsigned int size_a;
> +
> +int test(void)
> +{
> +  /* This should not warn.  */
> +  return size_a ? a[0] : 0;
> +}
> Index: gcc/tree-vrp.c
> ===================================================================
> --- gcc/tree-vrp.c      (revision 202088)
> +++ gcc/tree-vrp.c      (working copy)
> @@ -6137,9 +6137,10 @@ check_array_ref (location_t location, tr
>    low_sub = up_sub = TREE_OPERAND (ref, 1);
>    up_bound = array_ref_up_bound (ref);
>
> -  /* Can not check flexible arrays.  */
> +  /* Can not check flexible arrays or zero-length arrays.  */
>    if (!up_bound
> -      || TREE_CODE (up_bound) != INTEGER_CST)
> +      || TREE_CODE (up_bound) != INTEGER_CST
> +      || tree_int_cst_equal (up_bound, integer_minus_one_node))

That doesn't look correct - what if the lower bound is -10?  That can
easily happen
for Ada, so please revert the patch.  And I fail to see why the testcase should
not warn.  Clearly you have a definition of a here and it doesn't have
an element
so the access is out of bounds.

Richard.

>      return;
>
>    /* Accesses to trailing arrays via pointers may access storage