PATCH RFC: Warn about pointer wraparound with -Wstrict-overflow

[Richard Guenther]

On Tue, Apr 8, 2008 at 10:56 PM, Ian Lance Taylor <iant@google.com> wrote:
> Jakub Jelinek <jakub@redhat.com> writes:
>
>  > I agree, both that it is a good idea to make -fwrapv/-ftrapv affect
>  > even pointer arithmetics and that we want this in 4.3 and maybe even
>  > in 4.2.  I'd just be interested to know how many warnings it will generate
>  > on a few popular packages to see whether the warnings might be actually
>  > useful for security investigators or whether there will be just too many
>  > to make them useless.
>
>  I ran the new compiler with -Wstrict-overflow=3 over an old set of cc1
>  .i files.  I got 95 warnings.  Of those, exactly one was the new
>  pointer wraparound warning.
>
>  ../../trunk/gcc/sched-vis.c:49: warning: assuming pointer wraparound does not occur when comparing P +- C1 with P +- C2
>
>  It was for the "if (cur > end)" line in this function:
>
>  static char *
>  safe_concat (char *buf, char *cur, const char *str)
>  {
>   char *end = buf + BUF_LEN - 2;        /* Leave room for null.  */
>   int c;
>
>   if (cur > end)
>     {
>       *end = '\0';
>       return end;
>     }
>
>
>  It is happening because, of course, this is being inlined, and cur is
>  based on buf.  This is a false positive.  It's easy to avoid by
>  writing the code as, e.g.,:
>   if (cur - buf > BUF_LEN - 2)
>
>
>  For an aggressive warning like -Wstrict-overflow=3, I think that one
>  false positive on all of cc1 is not too bad.  Especially when you
>  consider that this is one false positive amidst 94 existing false
>  positives.

How about some C++ code using libstdc++?  (of course #pragma system_header
may save our day here)

Richard.