Issue #1772851. Alters long.__hash__ from being *almost* completely

predictable to being completely predictable.  The value of hash(n)
is unchanged for any n that's small enough to be representable as an
int, and also unchanged for the vast majority of long integers n of
reasonable size.

diff --git a/Lib/test/test_hash.py b/Lib/test/test_hash.py
index 3d6c9d1b67c73a3b3126dfa191912ef589951b86..07b65da9bdf131b5a78bedca06111f8d231d3f54 100644 (file)
--- a/Lib/test/test_hash.py
+++ b/Lib/test/test_hash.py
@@ -18,10 +18,21 @@ class HashEqualityTestCase(unittest.TestCase):
 
     def test_numeric_literals(self):
         self.same_hash(1, 1L, 1.0, 1.0+0.0j)
+        self.same_hash(0, 0L, 0.0, 0.0+0.0j)
+        self.same_hash(-1, -1L, -1.0, -1.0+0.0j)
+        self.same_hash(-2, -2L, -2.0, -2.0+0.0j)
 
     def test_coerced_integers(self):
         self.same_hash(int(1), long(1), float(1), complex(1),
                        int('1'), float('1.0'))
+        self.same_hash(int(-2**31), long(-2**31), float(-2**31))
+        self.same_hash(int(1-2**31), long(1-2**31), float(1-2**31))
+        self.same_hash(int(2**31-1), long(2**31-1), float(2**31-1))
+        # for 64-bit platforms
+        self.same_hash(int(2**31), long(2**31), float(2**31))
+        self.same_hash(int(-2**63), long(-2**63), float(-2**63))
+        self.same_hash(int(1-2**63), long(1-2**63))
+        self.same_hash(int(2**63-1), long(2**63-1))
 
     def test_coerced_floats(self):
         self.same_hash(long(1.23e300), float(1.23e300))

diff --git a/Objects/longobject.c b/Objects/longobject.c
index abb69b07167871aff1309c10e76e2f7a3a2a84ed..f14f298513b63479f1fe249423be15c0ccc9e43a 100644 (file)
--- a/Objects/longobject.c
+++ b/Objects/longobject.c
@@ -1937,10 +1937,18 @@ long_hash(PyLongObject *v)
                i = -(i);
        }
 #define LONG_BIT_SHIFT (8*sizeof(long) - SHIFT)
+       /* The following loop produces a C long x such that (unsigned long)x
+          is congruent to the absolute value of v modulo ULONG_MAX.  The
+          resulting x is nonzero if and only if v is. */
        while (--i >= 0) {
                /* Force a native long #-bits (32 or 64) circular shift */
                x = ((x << SHIFT) & ~MASK) | ((x >> LONG_BIT_SHIFT) & MASK);
                x += v->ob_digit[i];
+               /* If the addition above overflowed (thinking of x as
+                  unsigned), we compensate by incrementing.  This preserves
+                  the value modulo ULONG_MAX. */
+               if ((unsigned long)x < v->ob_digit[i])
+                       x++;
        }
 #undef LONG_BIT_SHIFT
        x = x * sign;