Avoid undefined behaviour due to overflow in i_divmod (Objects/intobject.c).

diff --git a/Objects/intobject.c b/Objects/intobject.c
index 43dedf25485e31f4e65bcf3a95f6033bc8424c23..b03ef703cbcc60bb132e8ed6bc3e38f154aa2f5e 100644 (file)
--- a/Objects/intobject.c
+++ b/Objects/intobject.c
@@ -580,7 +580,16 @@ i_divmod(register long x, register long y,
        if (y == -1 && UNARY_NEG_WOULD_OVERFLOW(x))
                return DIVMOD_OVERFLOW;
        xdivy = x / y;
-       xmody = x - xdivy * y;
+       /* xdiv*y can overflow on platforms where x/y gives floor(x/y)
+        * for x and y with differing signs. (This is unusual
+        * behaviour, and C99 prohibits it, but it's allowed by C89;
+        * for an example of overflow, take x = LONG_MIN, y = 5 or x =
+        * LONG_MAX, y = -5.)  However, x - xdivy*y is always
+        * representable as a long, since it lies strictly between
+        * -abs(y) and abs(y).  We add casts to avoid intermediate
+        * overflow.
+        */
+       xmody = (long)(x - (unsigned long)xdivy * y);
        /* If the signs of x and y differ, and the remainder is non-0,
         * C89 doesn't define whether xdivy is now the floor or the
         * ceiling of the infinitely precise quotient.  We want the floor,