On Sun, May 31, 2015 at 9:53 AM, Marek Polacek <pola...@redhat.com> wrote:
> The code here was assuming that get_maxval_strlen returns an INTEGER_CST.
> As the testcase demonstrates, that is not always true, so check for that
> before calling compare_tree_int. E.g. gimple_fold_builtin_strncpy already
> has this check.
>
> Bootstrapped/regtested on x86_64-linux, ok for trunk and 5?
Ok.
Thanks,
Richard.
> 2015-05-31 Marek Polacek <pola...@redhat.com>
>
> PR middle-end/66345
> * gimple-fold.c (gimple_fold_builtin_snprintf): Return false if
> get_maxval_strlen does not produce an INTEGER_CST.
>
> * gcc.dg/torture/pr66345.c: New test.
>
> diff --git gcc/gimple-fold.c gcc/gimple-fold.c
> index c91f218..b2ce851 100644
> --- gcc/gimple-fold.c
> +++ gcc/gimple-fold.c
> @@ -2530,7 +2530,7 @@ gimple_fold_builtin_snprintf (gimple_stmt_iterator *gsi)
> return false;
>
> tree orig_len = get_maxval_strlen (orig, 0);
> - if (!orig_len)
> + if (!orig_len || TREE_CODE (orig_len) != INTEGER_CST)
> return false;
>
> /* We could expand this as
> diff --git gcc/testsuite/gcc.dg/torture/pr66345.c
> gcc/testsuite/gcc.dg/torture/pr66345.c
> index e69de29..1687987 100644
> --- gcc/testsuite/gcc.dg/torture/pr66345.c
> +++ gcc/testsuite/gcc.dg/torture/pr66345.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile } */
> +
> +extern int snprintf (char *, unsigned long, const char *, ...);
> +const char a[] = "";
> +int b;
> +void
> +get_bar ()
> +{
> + snprintf (0, 0, "%s", &a[b]);
> +}
>
> Marek
