patches/overflow.diff

Fix incorrect overflow test, that itself overflowed.

"(unsigned long)(x * y)" will multiply ints x and y into a new int, then
converts the int into an unsigned long. Replace it with
"(unsigned long)(x) * (unsigned long)(y)" which performs the multiplication in a
larger space.

Converted from cl/147088677.

--- a/mpn/generic/get_d.c
+++ b/mpn/generic/get_d.c
@@ -151,8 +151,8 @@ mpn_get_d (mp_srcptr up, mp_size_t size,
   /* Adjust exp to a radix point just above {up,size}, guarding against
      overflow.  After this exp can of course be reduced to anywhere within
      the {up,size} region without underflow.  */
-  if (UNLIKELY ((unsigned long) (GMP_NUMB_BITS * size)
-		> ((unsigned long) LONG_MAX - exp)))
+  if (UNLIKELY ((unsigned long) (GMP_NUMB_BITS) * (unsigned long) (size)
+		> (LONG_MAX - (unsigned long) exp)))
     {
 #if _GMP_IEEE_FLOATS
       goto ieee_infinity;