Force zlib.crc32 and zlib.adler32 to return a signed integer on all platforms

regardless of the native sizeof(long) used in the integer object.

This somewhat odd behavior of returning a signed is maintained in 2.x for
compatibility reasons of always returning an integer rather than a long object.

Fixes Issue1202 for Python 2.6

Doc/library/zlib.rst

@@ -42,6 +42,12 @@ The available exception and functions in this module are:
    the algorithm is designed for use as a checksum algorithm, it is not suitable
    for use as a general hash algorithm.
 
+   This function always returns an integer object.
+
+.. versionchanged:: 2.6
+   For consistent cross-platform behavior we always return a signed integer.
+   ie: Results in the (2**31)...(2**32-1) range will be negative.
+
 
 .. function:: compress(string[, level])
 
@@ -74,6 +80,12 @@ The available exception and functions in this module are:
    the algorithm is designed for use as a checksum algorithm, it is not suitable
    for use as a general hash algorithm.
 
+   This function always returns an integer object.
+
+.. versionchanged:: 2.6
+   For consistent cross-platform behavior we always return a signed integer.
+   ie: Results in the (2**31)...(2**32-1) range will be negative.
+
 
 .. function:: decompress(string[, wbits[, bufsize]])
 

Lib/test/test_zlib.py

@@ -38,6 +38,15 @@ class ChecksumTestCase(unittest.TestCase):
         self.assertEqual(zlib.crc32("penguin"), zlib.crc32("penguin", 0))
         self.assertEqual(zlib.adler32("penguin"),zlib.adler32("penguin",1))
 
+    def test_abcdefghijklmnop(self):
+        """test issue1202 compliance: signed crc32, adler32 in 2.x"""
+        foo = 'abcdefghijklmnop'
+        # explicitly test signed behavior
+        self.assertEqual(zlib.crc32(foo), -1808088941)
+        self.assertEqual(zlib.crc32('spam'), 1138425661)
+        self.assertEqual(zlib.adler32(foo+foo), -721416943)
+        self.assertEqual(zlib.adler32('spam'), 72286642)
+
 
 
 class ExceptionTestCase(unittest.TestCase):

Misc/NEWS

@@ -27,6 +27,10 @@ Core and builtins
 Library
 -------
 
+- Issue #1202: zlib.crc32 and zlib.adler32 no longer return different values
+  on 32-bit vs. 64-bit python interpreters.  Both were correct, but they now
+  both return a signed integer object for consistency.
+
 - Issue #1158: add %f format (fractions of a second represented as
   microseconds) to datetime objects.  Understood by both strptime and
   strftime.

Modules/zlibmodule.c

@@ -884,37 +884,46 @@ PyDoc_STRVAR(adler32__doc__,
 "adler32(string[, start]) -- Compute an Adler-32 checksum of string.\n"
 "\n"
 "An optional starting value can be specified.  The returned checksum is\n"
-"an integer.");
+"a signed integer.");
 
 static PyObject *
 PyZlib_adler32(PyObject *self, PyObject *args)
 {
     uLong adler32val = adler32(0L, Z_NULL, 0);
     Byte *buf;
-    int len;
+    int len, signed_val;
 
     if (!PyArg_ParseTuple(args, "s#|k:adler32", &buf, &len, &adler32val))
 	return NULL;
-    adler32val = adler32(adler32val, buf, len);
-    return PyInt_FromLong(adler32val);
+    /* In Python 2.x we return a signed integer regardless of native platform
+     * long size (the 32bit unsigned long is treated as 32-bit signed and sign
+     * extended into a 64-bit long inside the integer object).  3.0 does the
+     * right thing and returns unsigned. http://bugs.python.org/issue1202 */
+    signed_val = adler32(adler32val, buf, len);
+    return PyInt_FromLong(signed_val);
 }
 
 PyDoc_STRVAR(crc32__doc__,
 "crc32(string[, start]) -- Compute a CRC-32 checksum of string.\n"
 "\n"
 "An optional starting value can be specified.  The returned checksum is\n"
-"an integer.");
+"a signed integer.");
 
 static PyObject *
 PyZlib_crc32(PyObject *self, PyObject *args)
 {
     uLong crc32val = crc32(0L, Z_NULL, 0);
     Byte *buf;
-    int len;
+    int len, signed_val;
+
     if (!PyArg_ParseTuple(args, "s#|k:crc32", &buf, &len, &crc32val))
 	return NULL;
-    crc32val = crc32(crc32val, buf, len);
-    return PyInt_FromLong(crc32val);
+    /* In Python 2.x we return a signed integer regardless of native platform
+     * long size (the 32bit unsigned long is treated as 32-bit signed and sign
+     * extended into a 64-bit long inside the integer object).  3.0 does the
+     * right thing and returns unsigned. http://bugs.python.org/issue1202 */
+    signed_val = crc32(crc32val, buf, len);
+    return PyInt_FromLong(signed_val);
 }