Issue #4258: Make it possible to use 30-bit digits for PyLongs:

- new configure option --enable-big-digits
 - new structseq sys.int_info giving information about the internal format
By default, 30-bit digits are enabled on 64-bit machines but
disabled on 32-bit machines.

Doc/library/sys.rst

@@ -413,6 +413,23 @@ always available.
    same information.
 
 
+.. data:: int_info
+
+   A struct sequence that holds information about Python's
+   internal representation of integers.  The attributes are read only.
+
+   +-------------------------+----------------------------------------------+
+   | attribute               | explanation                                  |
+   +=========================+==============================================+
+   | :const:`bits_per_digit` | number of bits held in each digit.  Python   |
+   |                         | integers are stored internally in base       |
+   |                         | ``2**int_info.bits_per_digit``               |
+   +-------------------------+----------------------------------------------+
+   | :const:`sizeof_digit`   | size in bytes of the C type used to          |
+   |                         | represent a digit                            |
+   +-------------------------+----------------------------------------------+
+
+
 .. function:: intern(string)
 
    Enter *string* in the table of "interned" strings and return the interned string

Doc/whatsnew/3.1.rst

@@ -87,5 +87,28 @@ Some smaller changes made to the core Python language are:
 
   (Contributed by Fredrik Johansson and Victor Stinner; :issue:`3439`.)
 
+* Integers are now stored internally either in base 2**15 or in base
+  2**30, the base being determined at build time.  Previously, they
+  were always stored in base 2**15.  Using base 2**30 gives
+  significant performance improvements on 64-bit machines, but
+  benchmark results on 32-bit machines have been mixed.  Therefore,
+  the default is to use base 2**30 on 64-bit machines and base 2**15
+  on 32-bit machines; on Unix, there's a new configure option
+  --enable-big-digits that can be used to override this default.
+
+  Apart from the performance improvements this change should be
+  invisible to end users, with one exception: for testing and
+  debugging purposes there's a new structseq ``sys.int_info`` that
+  provides information about the internal format, giving the number of
+  bits per digit and the size in bytes of the C type used to store
+  each digit::
+
+     >>> import sys
+     >>> sys.int_info
+     sys.int_info(bits_per_digit=30, sizeof_digit=4)
+
+
+  (Contributed by Mark Dickinson; :issue:`4258`.)
+
 
 .. ======================================================================

Include/longintrepr.h

@@ -7,24 +7,62 @@ extern "C" {
 
 /* This is published for the benefit of "friend" marshal.c only. */
 
-/* Parameters of the long integer representation.
+/* Parameters of the long integer representation.  There are two different
-   These shouldn't have to be changed as C should guarantee that a short
+   sets of parameters: one set for 30-bit digits, stored in an unsigned 32-bit
-   contains at least 16 bits, but it's made changeable anyway.
+   integer type, and one set for 15-bit digits with each digit stored in an
-   Note: 'digit' should be able to hold 2*MASK+1, and 'twodigits'
+   unsigned short.  The value of PYLONG_BITS_IN_DIGIT, defined either at
-   should be able to hold the intermediate results in 'mul'
+   configure time or in pyport.h, is used to decide which digit size to use.
-   (at most (BASE-1)*(2*BASE+1) == MASK*(2*MASK+3)).
-   Also, x_sub assumes that 'digit' is an unsigned type, and overflow
-   is handled by taking the result mod 2**N for some N > SHIFT.
-   And, at some places it is assumed that MASK fits in an int, as well.
-   long_pow() requires that SHIFT be divisible by 5. */
 
+   Type 'digit' should be able to hold 2*PyLong_BASE-1, and type 'twodigits'
+   should be an unsigned integer type able to hold all integers up to
+   PyLong_BASE*PyLong_BASE-1.  x_sub assumes that 'digit' is an unsigned type,
+   and that overflow is handled by taking the result modulo 2**N for some N >
+   PyLong_SHIFT.  The majority of the code doesn't care about the precise
+   value of PyLong_SHIFT, but there are some notable exceptions:
+
+   - long_pow() requires that PyLong_SHIFT be divisible by 5
+
+   - PyLong_{As,From}ByteArray require that PyLong_SHIFT be at least 8
+
+   - long_hash() requires that PyLong_SHIFT is *strictly* less than the number
+     of bits in an unsigned long, as do the PyLong <-> long (or unsigned long)
+     conversion functions
+
+   - the long <-> size_t/Py_ssize_t conversion functions expect that
+     PyLong_SHIFT is strictly less than the number of bits in a size_t
+
+   - the marshal code currently expects that PyLong_SHIFT is a multiple of 15
+
+   - NSMALLNEGINTS and NSMALLPOSINTS should be small enough to fit in a single
+     digit; with the current values this forces PyLong_SHIFT >= 9
+
+  The values 15 and 30 should fit all of the above requirements, on any
+  platform.
+*/
+
+#if HAVE_STDINT_H
+#include <stdint.h>
+#endif
+
+#if PYLONG_BITS_IN_DIGIT == 30
+#if !(defined HAVE_UINT64_T && defined HAVE_UINT32_T &&          \
+      defined HAVE_INT64_T && defined HAVE_INT32_T)
+#error "30-bit long digits requested, but the necessary types are not available on this platform"
+#endif
+typedef PY_UINT32_T digit;
+typedef PY_INT32_T sdigit; /* signed variant of digit */
+typedef PY_UINT64_T twodigits;
+typedef PY_INT64_T stwodigits; /* signed variant of twodigits */
+#define PyLong_SHIFT	30
+#elif PYLONG_BITS_IN_DIGIT == 15
 typedef unsigned short digit;
-typedef short sdigit;                   /* signed variant of digit */
+typedef short sdigit; /* signed variant of digit */
-#define BASE_TWODIGITS_TYPE long
+typedef unsigned long twodigits;
-typedef unsigned BASE_TWODIGITS_TYPE twodigits;
+typedef long stwodigits; /* signed variant of twodigits */
-typedef BASE_TWODIGITS_TYPE stwodigits; /* signed variant of twodigits */
-
 #define PyLong_SHIFT	15
+#else
+#error "PYLONG_BITS_IN_DIGIT should be 15 or 30"
+#endif
 #define PyLong_BASE	((digit)1 << PyLong_SHIFT)
 #define PyLong_MASK	((digit)(PyLong_BASE - 1))
 

Include/longobject.h

@@ -26,6 +26,7 @@ PyAPI_FUNC(Py_ssize_t) PyLong_AsSsize_t(PyObject *);
 PyAPI_FUNC(size_t) PyLong_AsSize_t(PyObject *);
 PyAPI_FUNC(unsigned long) PyLong_AsUnsignedLong(PyObject *);
 PyAPI_FUNC(unsigned long) PyLong_AsUnsignedLongMask(PyObject *);
+PyAPI_FUNC(PyObject *) PyLong_GetInfo(void);
 
 /* It may be useful in the future. I've added it in the PyInt -> PyLong
    cleanup to keep the extra information. [CH] */

Include/pyport.h

@@ -69,6 +69,57 @@ Used in:  PY_LONG_LONG
 #endif
 #endif /* HAVE_LONG_LONG */
 
+/* a build with 30-bit digits for Python long integers needs an exact-width
+ * 32-bit unsigned integer type to store those digits.  (We could just use
+ * type 'unsigned long', but that would be wasteful on a system where longs
+ * are 64-bits.)  On Unix systems, the autoconf macro AC_TYPE_UINT32_T defines
+ * uint32_t to be such a type unless stdint.h or inttypes.h defines uint32_t.
+ * However, it doesn't set HAVE_UINT32_T, so we do that here.
+ */
+#if (defined UINT32_MAX || defined uint32_t)
+#ifndef PY_UINT32_T
+#define HAVE_UINT32_T 1
+#define PY_UINT32_T uint32_t
+#endif
+#endif
+
+/* Macros for a 64-bit unsigned integer type; used for type 'twodigits' in the
+ * long integer implementation, when 30-bit digits are enabled.
+ */
+#if (defined UINT64_MAX || defined uint64_t)
+#ifndef PY_UINT64_T
+#define HAVE_UINT64_T 1
+#define PY_UINT64_T uint64_t
+#endif
+#endif
+
+/* Signed variants of the above */
+#if (defined INT32_MAX || defined int32_t)
+#ifndef PY_INT32_T
+#define HAVE_INT32_T 1
+#define PY_INT32_T int32_t
+#endif
+#endif
+#if (defined INT64_MAX || defined int64_t)
+#ifndef PY_INT64_T
+#define HAVE_INT64_T 1
+#define PY_INT64_T int64_t
+#endif
+#endif
+
+/* If PYLONG_BITS_IN_DIGIT is not defined then we'll use 30-bit digits if all
+   the necessary integer types are available, and we're on a 64-bit platform
+   (as determined by SIZEOF_VOID_P); otherwise we use 15-bit digits. */
+
+#ifndef PYLONG_BITS_IN_DIGIT
+#if (defined HAVE_UINT64_T && defined HAVE_INT64_T && \
+     defined HAVE_UINT32_T && defined HAVE_INT32_T && SIZEOF_VOID_P >= 8)
+#define PYLONG_BITS_IN_DIGIT 30
+#else
+#define PYLONG_BITS_IN_DIGIT 15
+#endif
+#endif
+
 /* uintptr_t is the C9X name for an unsigned integral type such that a
  * legitimate void* can be cast to uintptr_t and then back to void* again
  * without loss of information.  Similarly for intptr_t, wrt a signed

Lib/test/test_long.py

@@ -15,7 +15,7 @@ class Frm(object):
         return self.format % self.args
 
 # SHIFT should match the value in longintrepr.h for best testing.
-SHIFT = 15
+SHIFT = sys.int_info.bits_per_digit
 BASE = 2 ** SHIFT
 MASK = BASE - 1
 KARATSUBA_CUTOFF = 70   # from longobject.c
@@ -120,6 +120,35 @@ class LongTest(unittest.TestCase):
                 y = self.getran(leny) or 1
                 self.check_division(x, y)
 
+        # specific numbers chosen to exercise corner cases of the
+        # current long division implementation
+
+        # 30-bit cases involving a quotient digit estimate of BASE+1
+        self.check_division(1231948412290879395966702881,
+                            1147341367131428698)
+        self.check_division(815427756481275430342312021515587883,
+                       707270836069027745)
+        self.check_division(627976073697012820849443363563599041,
+                       643588798496057020)
+        self.check_division(1115141373653752303710932756325578065,
+                       1038556335171453937726882627)
+        # 30-bit cases that require the post-subtraction correction step
+        self.check_division(922498905405436751940989320930368494,
+                       949985870686786135626943396)
+        self.check_division(768235853328091167204009652174031844,
+                       1091555541180371554426545266)
+
+        # 15-bit cases involving a quotient digit estimate of BASE+1
+        self.check_division(20172188947443, 615611397)
+        self.check_division(1020908530270155025, 950795710)
+        self.check_division(128589565723112408, 736393718)
+        self.check_division(609919780285761575, 18613274546784)
+        # 15-bit cases that require the post-subtraction correction step
+        self.check_division(710031681576388032, 26769404391308)
+        self.check_division(1933622614268221, 30212853348836)
+
+
+
     def test_karatsuba(self):
         digits = list(range(1, 5)) + list(range(KARATSUBA_CUTOFF,
                                                 KARATSUBA_CUTOFF + 10))

Lib/test/test_sys.py

@@ -333,6 +333,9 @@ class SysModuleTest(unittest.TestCase):
         self.assert_(isinstance(sys.executable, str))
         self.assertEqual(len(sys.float_info), 11)
         self.assertEqual(sys.float_info.radix, 2)
+        self.assertEqual(len(sys.int_info), 2)
+        self.assert_(sys.int_info.bits_per_digit % 5 == 0)
+        self.assert_(sys.int_info.sizeof_digit >= 1)
         self.assert_(isinstance(sys.hexversion, int))
         self.assert_(isinstance(sys.maxsize, int))
         self.assert_(isinstance(sys.maxunicode, int))
@@ -437,6 +440,7 @@ class SizeofTest(unittest.TestCase):
         if hasattr(sys, "gettotalrefcount"):
             self.header += '2P'
             self.vheader += '2P'
+        self.longdigit = sys.int_info.sizeof_digit
         import _testcapi
         self.gc_headsize = _testcapi.SIZEOF_PYGC_HEAD
         self.file = open(test.support.TESTFN, 'wb')
@@ -471,7 +475,7 @@ class SizeofTest(unittest.TestCase):
         size = self.calcsize
         gc_header_size = self.gc_headsize
         # bool objects are not gc tracked
-        self.assertEqual(sys.getsizeof(True), size(vh) + self.H)
+        self.assertEqual(sys.getsizeof(True), size(vh) + self.longdigit)
         # but lists are
         self.assertEqual(sys.getsizeof([]), size(vh + 'PP') + gc_header_size)
 
@@ -479,8 +483,8 @@ class SizeofTest(unittest.TestCase):
         h = self.header
         vh = self.vheader
         size = self.calcsize
-        self.assertEqual(sys.getsizeof(True), size(vh) + self.H)
+        self.assertEqual(sys.getsizeof(True), size(vh) + self.longdigit)
-        self.assertEqual(sys.getsizeof(True, -1), size(vh) + self.H)
+        self.assertEqual(sys.getsizeof(True, -1), size(vh) + self.longdigit)
 
     def test_objecttypes(self):
         # check all types defined in Objects/
@@ -489,7 +493,7 @@ class SizeofTest(unittest.TestCase):
         size = self.calcsize
         check = self.check_sizeof
         # bool
-        check(True, size(vh) + self.H)
+        check(True, size(vh) + self.longdigit)
         # buffer
         # XXX
         # builtin_function_or_method
@@ -607,11 +611,12 @@ class SizeofTest(unittest.TestCase):
         check(reversed([]), size(h + 'lP'))
         # long
         check(0, size(vh))
-        check(1, size(vh) + self.H)
+        check(1, size(vh) + self.longdigit)
-        check(-1, size(vh) + self.H)
+        check(-1, size(vh) + self.longdigit)
-        check(32768, size(vh) + 2*self.H)
+        PyLong_BASE = 2**sys.int_info.bits_per_digit
-        check(32768*32768-1, size(vh) + 2*self.H)
+        check(PyLong_BASE, size(vh) + 2*self.longdigit)
-        check(32768*32768, size(vh) + 3*self.H)
+        check(PyLong_BASE**2-1, size(vh) + 2*self.longdigit)
+        check(PyLong_BASE**2, size(vh) + 3*self.longdigit)
         # memory
         check(memoryview(b''), size(h + 'P PP2P2i7P'))
         # module

Misc/NEWS

@@ -12,6 +12,13 @@ What's New in Python 3.1 alpha 2?
 Core and Builtins
 -----------------
 
+- Issue #4258: Make it possible to use base 2**30 instead of base
+  2**15 for the internal representation of integers, for performance
+  reasons.  Base 2**30 is enabled by default on 64-bit machines.  Add
+  --enable-big-digits option to configure, which overrides the
+  default.  Add sys.int_info structseq to provide information about
+  the internal format.
+
 - Issue #4474: PyUnicode_FromWideChar now converts characters outside
   the BMP to surrogate pairs, on systems with sizeof(wchar_t) == 4
   and sizeof(Py_UNICODE) == 2.

Objects/longobject.c

@@ -4,6 +4,7 @@
 
 #include "Python.h"
 #include "longintrepr.h"
+#include "structseq.h"
 
 #include <ctype.h>
 #include <stddef.h>
@@ -204,6 +205,7 @@ PyLong_FromLong(long ival)
 		return (PyObject*)v;
 	}
 
+#if PyLONG_SHIFT==15
 	/* 2 digits */
 	if (!(abs_ival >> 2*PyLong_SHIFT)) {
 		v = _PyLong_New(2);
@@ -216,6 +218,7 @@ PyLong_FromLong(long ival)
 		}
 		return (PyObject*)v;
 	}
+#endif
 
 	/* Larger numbers: loop to determine number of digits */
 	t = abs_ival;
@@ -2864,10 +2867,20 @@ long_mul(PyLongObject *a, PyLongObject *b)
 
 	CHECK_BINOP(a, b);
 
+	/* fast path for single-digit multiplication */
 	if (ABS(Py_SIZE(a)) <= 1 && ABS(Py_SIZE(b)) <= 1) {
-		PyObject *r;
+		stwodigits v = (stwodigits)(MEDIUM_VALUE(a)) * MEDIUM_VALUE(b);
-		r = PyLong_FromLong(MEDIUM_VALUE(a)*MEDIUM_VALUE(b));
+#ifdef HAVE_LONG_LONG
-		return r;
+		return PyLong_FromLongLong((PY_LONG_LONG)v);
+#else
+		/* if we don't have long long then we're almost certainly
+		   using 15-bit digits, so v will fit in a long.  In the
+		   unlikely event that we're using 30-bit digits on a platform
+		   without long long, a large v will just cause us to fall
+		   through to the general multiplication code below. */
+		if (v >= LONG_MIN && v <= LONG_MAX)
+			return PyLong_FromLong((long)v);
+#endif
 	}
 
 	z = k_mul(a, b);
@@ -3991,6 +4004,45 @@ PyTypeObject PyLong_Type = {
 	PyObject_Del,				/* tp_free */
 };
 
+static PyTypeObject Int_InfoType;
+
+PyDoc_STRVAR(int_info__doc__,
+"sys.int_info\n\
+\n\
+A struct sequence that holds information about Python's\n\
+internal representation of integers.  The attributes are read only.");
+
+static PyStructSequence_Field int_info_fields[] = {
+	{"bits_per_digit", "size of a digit in bits"},
+	{"sizeof_digit", "size in bytes of the C type used to "
+	                 "represent a digit"},
+	{NULL, NULL}
+};
+
+static PyStructSequence_Desc int_info_desc = {
+	"sys.int_info",   /* name */
+	int_info__doc__,  /* doc */
+	int_info_fields,  /* fields */
+	2                 /* number of fields */
+};
+
+PyObject *
+PyLong_GetInfo(void)
+{
+	PyObject* int_info;
+	int field = 0;
+	int_info = PyStructSequence_New(&Int_InfoType);
+	if (int_info == NULL)
+		return NULL;
+	PyStructSequence_SET_ITEM(int_info, field++, PyLong_FromLong(PyLong_SHIFT));
+	PyStructSequence_SET_ITEM(int_info, field++, PyLong_FromLong(sizeof(digit)));
+	if (PyErr_Occurred()) {
+		Py_CLEAR(int_info);
+		return NULL;
+	}
+	return int_info;
+}
+
 int
 _PyLong_Init(void)
 {
@@ -4023,6 +4075,10 @@ _PyLong_Init(void)
 		v->ob_digit[0] = abs(ival);
 	}
 #endif
+	/* initialize int_info */
+	if (Int_InfoType.tp_name == 0)
+		PyStructSequence_InitType(&Int_InfoType, &int_info_desc);
+
 	return 1;
 }
 

PC/pyconfig.h

@@ -396,6 +396,42 @@ Py_NO_ENABLE_SHARED to find out.  Also support MS_NO_COREDLL for b/w compat */
 
 #endif
 
+/* define signed and unsigned exact-width 32-bit and 64-bit types, used in the
+   implementation of Python long integers. */
+#ifndef PY_UINT32_T
+#if SIZEOF_INT == 4
+#define HAVE_UINT32_T 1
+#define PY_UINT32_T unsigned int
+#elif SIZEOF_LONG == 4
+#define HAVE_UINT32_T 1
+#define PY_UINT32_T unsigned long
+#endif
+#endif
+
+#ifndef PY_UINT64_T
+#if SIZEOF_LONG_LONG == 8
+#define HAVE_UINT64_T 1
+#define PY_UINT64_T unsigned PY_LONG_LONG
+#endif
+#endif
+
+#ifndef PY_INT32_T
+#if SIZEOF_INT == 4
+#define HAVE_INT32_T 1
+#define PY_INT32_T int
+#elif SIZEOF_LONG == 4
+#define HAVE_INT32_T 1
+#define PY_INT32_T long
+#endif
+#endif
+
+#ifndef PY_INT64_T
+#if SIZEOF_LONG_LONG == 8
+#define HAVE_INT64_T 1
+#define PY_INT64_T PY_LONG_LONG
+#endif
+#endif
+
 /* Fairly standard from here! */
 
 /* Define to 1 if you have the `copysign' function. */

Python/marshal.c

@@ -11,6 +11,8 @@
 #include "code.h"
 #include "marshal.h"
 
+#define ABS(x) ((x) < 0 ? -(x) : (x))
+
 /* High water mark to determine when the marshalled object is dangerously deep
  * and risks coring the interpreter.  When the object stack gets this deep,
  * raise an exception instead of continuing.
@@ -122,6 +124,56 @@ w_long64(long x, WFILE *p)
 }
 #endif
 
+/* We assume that Python longs are stored internally in base some power of
+   2**15; for the sake of portability we'll always read and write them in base
+   exactly 2**15. */
+
+#define PyLong_MARSHAL_SHIFT 15
+#define PyLong_MARSHAL_BASE ((short)1 << PyLong_MARSHAL_SHIFT)
+#define PyLong_MARSHAL_MASK (PyLong_MARSHAL_BASE - 1)
+#if PyLong_SHIFT % PyLong_MARSHAL_SHIFT != 0
+#error "PyLong_SHIFT must be a multiple of PyLong_MARSHAL_SHIFT"
+#endif
+#define PyLong_MARSHAL_RATIO (PyLong_SHIFT / PyLong_MARSHAL_SHIFT)
+
+static void
+w_PyLong(const PyLongObject *ob, WFILE *p)
+{
+	Py_ssize_t i, j, n, l;
+	digit d;
+
+	w_byte(TYPE_LONG, p);
+	if (Py_SIZE(ob) == 0) {
+		w_long((long)0, p);
+		return;
+	}
+
+	/* set l to number of base PyLong_MARSHAL_BASE digits */
+	n = ABS(Py_SIZE(ob));
+	l = (n-1) * PyLong_MARSHAL_RATIO;
+	d = ob->ob_digit[n-1];
+	assert(d != 0); /* a PyLong is always normalized */
+	do {
+		d >>= PyLong_MARSHAL_SHIFT;
+		l++;
+	} while (d != 0);
+	w_long((long)(Py_SIZE(ob) > 0 ? l : -l), p);
+
+	for (i=0; i < n-1; i++) {
+		d = ob->ob_digit[i];
+		for (j=0; j < PyLong_MARSHAL_RATIO; j++) {
+			w_short(d & PyLong_MARSHAL_MASK, p);
+			d >>= PyLong_MARSHAL_SHIFT;
+		}
+		assert (d == 0);
+	}
+	d = ob->ob_digit[n-1];
+	do {
+		w_short(d & PyLong_MARSHAL_MASK, p);
+		d >>= PyLong_MARSHAL_SHIFT;
+	} while (d != 0);
+}
+
 static void
 w_object(PyObject *v, WFILE *p)
 {
@@ -155,14 +207,8 @@ w_object(PyObject *v, WFILE *p)
 		if ((x == -1)  && PyErr_Occurred()) {
 			PyLongObject *ob = (PyLongObject *)v;
 			PyErr_Clear();
-			w_byte(TYPE_LONG, p);
+			w_PyLong(ob, p);
-			n = Py_SIZE(ob);
+		}
-			w_long((long)n, p);
-			if (n < 0)
-				n = -n;
-			for (i = 0; i < n; i++)
-				w_short(ob->ob_digit[i], p);
-		} 
 		else {
 #if SIZEOF_LONG > 4
 			long y = Py_ARITHMETIC_RIGHT_SHIFT(long, x, 31);
@@ -480,6 +526,56 @@ r_long64(RFILE *p)
 #endif
 }
 
+static PyObject *
+r_PyLong(RFILE *p)
+{
+	PyLongObject *ob;
+	int size, i, j, md;
+	long n;
+	digit d;
+
+	n = r_long(p);
+	if (n == 0)
+		return (PyObject *)_PyLong_New(0);
+	if (n < -INT_MAX || n > INT_MAX) {
+		PyErr_SetString(PyExc_ValueError,
+			       "bad marshal data (long size out of range)");
+		return NULL;
+	}
+
+	size = 1 + (ABS(n)-1) / PyLong_MARSHAL_RATIO;
+	ob = _PyLong_New(size);
+	if (ob == NULL)
+		return NULL;
+	Py_SIZE(ob) = n > 0 ? size : -size;
+
+	for (i = 0; i < size-1; i++) {
+		d = 0;
+		for (j=0; j < PyLong_MARSHAL_RATIO; j++) {
+			md = r_short(p);
+			if (md < 0 || md > PyLong_MARSHAL_BASE)
+				goto bad_digit;
+			d += (digit)md << j*PyLong_MARSHAL_SHIFT;
+		}
+		ob->ob_digit[i] = d;
+	}
+	d = 0;
+	for (j=0; j < (ABS(n)-1)%PyLong_MARSHAL_RATIO + 1; j++) {
+		md = r_short(p);
+		if (md < 0 || md > PyLong_MARSHAL_BASE)
+			goto bad_digit;
+		d += (digit)md << j*PyLong_MARSHAL_SHIFT;
+	}
+	ob->ob_digit[size-1] = d;
+	return (PyObject *)ob;
+  bad_digit:
+	Py_DECREF(ob);
+	PyErr_SetString(PyExc_ValueError,
+			"bad marshal data (digit out of range in long)");
+	return NULL;
+}
+
+
 static PyObject *
 r_object(RFILE *p)
 {
@@ -544,38 +640,8 @@ r_object(RFILE *p)
 		break;
 
 	case TYPE_LONG:
-		{
+		retval = r_PyLong(p);
-			int size;
+		break;
-			PyLongObject *ob;
-			n = r_long(p);
-			if (n < -INT_MAX || n > INT_MAX) {
-				PyErr_SetString(PyExc_ValueError,
-						"bad marshal data (long size out of range)");
-				retval = NULL;
-				break;
-			}
-			size = n<0 ? -n : n;
-			ob = _PyLong_New(size);
-			if (ob == NULL) {
-				retval = NULL;
-				break;
-			}
-			Py_SIZE(ob) = n;
-			for (i = 0; i < size; i++) {
-				int digit = r_short(p);
-				if (digit < 0) {
-					Py_DECREF(ob);
-					PyErr_SetString(PyExc_ValueError,
-							"bad marshal data (negative digit in long)");
-					ob = NULL;
-					break;
-				}
-				if (ob != NULL)
-					ob->ob_digit[i] = digit;
-			}
-			retval = (PyObject *)ob;
-			break;
-		}
 
 	case TYPE_FLOAT:
 		{

Python/sysmodule.c

@@ -1012,6 +1012,7 @@ PyDoc_STR(
 Static objects:\n\
 \n\
 float_info -- a dict with information about the float implementation.\n\
+int_info -- a struct sequence with information about the int implementation.\n\
 maxsize -- the largest supported length of containers.\n\
 maxunicode -- the largest supported character\n\
 builtin_module_names -- tuple of module names built into this interpreter\n\
@@ -1375,6 +1376,8 @@ _PySys_Init(void)
 			    PyLong_FromSsize_t(PY_SSIZE_T_MAX));
 	SET_SYS_FROM_STRING("float_info",
 			    PyFloat_GetInfo());
+	SET_SYS_FROM_STRING("int_info",
+			    PyLong_GetInfo());
 	SET_SYS_FROM_STRING("maxunicode",
 			    PyLong_FromLong(PyUnicode_GetMax()));
 	SET_SYS_FROM_STRING("builtin_module_names",

configure

@@ -1320,6 +1320,8 @@ Optional Features:
   --enable-profiling      enable C-level code profiling
   --enable-ipv6           Enable ipv6 (with ipv4) support
   --disable-ipv6          Disable ipv6 support
+  --enable-big-digits[=BITS]
+                          use big digits for Python longs [BITS=30]
 
 Optional Packages:
   --with-PACKAGE[=ARG]    use PACKAGE [ARG=yes]
@@ -6965,6 +6967,386 @@ _ACEOF
 
 fi
 
+
+  { echo "$as_me:$LINENO: checking for uint32_t" >&5
+echo $ECHO_N "checking for uint32_t... $ECHO_C" >&6; }
+if test "${ac_cv_c_uint32_t+set}" = set; then
+  echo $ECHO_N "(cached) $ECHO_C" >&6
+else
+  ac_cv_c_uint32_t=no
+     for ac_type in 'uint32_t' 'unsigned int' 'unsigned long int' \
+	 'unsigned long long int' 'unsigned short int' 'unsigned char'; do
+       cat >conftest.$ac_ext <<_ACEOF
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+$ac_includes_default
+int
+main ()
+{
+static int test_array [1 - 2 * !(($ac_type) -1 >> (32 - 1) == 1)];
+test_array [0] = 0
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.$ac_objext
+if { (ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5
+  (eval "$ac_compile") 2>conftest.er1
+  ac_status=$?
+  grep -v '^ *+' conftest.er1 >conftest.err
+  rm -f conftest.er1
+  cat conftest.err >&5
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then
+  case $ac_type in
+  uint32_t) ac_cv_c_uint32_t=yes ;;
+  *) ac_cv_c_uint32_t=$ac_type ;;
+esac
+
+else
+  echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+
+fi
+
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+       test "$ac_cv_c_uint32_t" != no && break
+     done
+fi
+{ echo "$as_me:$LINENO: result: $ac_cv_c_uint32_t" >&5
+echo "${ECHO_T}$ac_cv_c_uint32_t" >&6; }
+  case $ac_cv_c_uint32_t in #(
+  no|yes) ;; #(
+  *)
+
+cat >>confdefs.h <<\_ACEOF
+#define _UINT32_T 1
+_ACEOF
+
+
+cat >>confdefs.h <<_ACEOF
+#define uint32_t $ac_cv_c_uint32_t
+_ACEOF
+;;
+  esac
+
+
+  { echo "$as_me:$LINENO: checking for uint64_t" >&5
+echo $ECHO_N "checking for uint64_t... $ECHO_C" >&6; }
+if test "${ac_cv_c_uint64_t+set}" = set; then
+  echo $ECHO_N "(cached) $ECHO_C" >&6
+else
+  ac_cv_c_uint64_t=no
+     for ac_type in 'uint64_t' 'unsigned int' 'unsigned long int' \
+	 'unsigned long long int' 'unsigned short int' 'unsigned char'; do
+       cat >conftest.$ac_ext <<_ACEOF
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+$ac_includes_default
+int
+main ()
+{
+static int test_array [1 - 2 * !(($ac_type) -1 >> (64 - 1) == 1)];
+test_array [0] = 0
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.$ac_objext
+if { (ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5
+  (eval "$ac_compile") 2>conftest.er1
+  ac_status=$?
+  grep -v '^ *+' conftest.er1 >conftest.err
+  rm -f conftest.er1
+  cat conftest.err >&5
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then
+  case $ac_type in
+  uint64_t) ac_cv_c_uint64_t=yes ;;
+  *) ac_cv_c_uint64_t=$ac_type ;;
+esac
+
+else
+  echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+
+fi
+
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+       test "$ac_cv_c_uint64_t" != no && break
+     done
+fi
+{ echo "$as_me:$LINENO: result: $ac_cv_c_uint64_t" >&5
+echo "${ECHO_T}$ac_cv_c_uint64_t" >&6; }
+  case $ac_cv_c_uint64_t in #(
+  no|yes) ;; #(
+  *)
+
+cat >>confdefs.h <<\_ACEOF
+#define _UINT64_T 1
+_ACEOF
+
+
+cat >>confdefs.h <<_ACEOF
+#define uint64_t $ac_cv_c_uint64_t
+_ACEOF
+;;
+  esac
+
+
+  { echo "$as_me:$LINENO: checking for int32_t" >&5
+echo $ECHO_N "checking for int32_t... $ECHO_C" >&6; }
+if test "${ac_cv_c_int32_t+set}" = set; then
+  echo $ECHO_N "(cached) $ECHO_C" >&6
+else
+  ac_cv_c_int32_t=no
+     for ac_type in 'int32_t' 'int' 'long int' \
+	 'long long int' 'short int' 'signed char'; do
+       cat >conftest.$ac_ext <<_ACEOF
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+$ac_includes_default
+int
+main ()
+{
+static int test_array [1 - 2 * !(0 < ($ac_type) (((($ac_type) 1 << (32 - 2)) - 1) * 2 + 1))];
+test_array [0] = 0
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.$ac_objext
+if { (ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5
+  (eval "$ac_compile") 2>conftest.er1
+  ac_status=$?
+  grep -v '^ *+' conftest.er1 >conftest.err
+  rm -f conftest.er1
+  cat conftest.err >&5
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then
+  cat >conftest.$ac_ext <<_ACEOF
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+$ac_includes_default
+int
+main ()
+{
+static int test_array [1 - 2 * !(($ac_type) (((($ac_type) 1 << (32 - 2)) - 1) * 2 + 1)
+	         < ($ac_type) (((($ac_type) 1 << (32 - 2)) - 1) * 2 + 2))];
+test_array [0] = 0
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.$ac_objext
+if { (ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5
+  (eval "$ac_compile") 2>conftest.er1
+  ac_status=$?
+  grep -v '^ *+' conftest.er1 >conftest.err
+  rm -f conftest.er1
+  cat conftest.err >&5
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then
+  :
+else
+  echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+	case $ac_type in
+  int32_t) ac_cv_c_int32_t=yes ;;
+  *) ac_cv_c_int32_t=$ac_type ;;
+esac
+
+fi
+
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+else
+  echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+
+fi
+
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+       test "$ac_cv_c_int32_t" != no && break
+     done
+fi
+{ echo "$as_me:$LINENO: result: $ac_cv_c_int32_t" >&5
+echo "${ECHO_T}$ac_cv_c_int32_t" >&6; }
+  case $ac_cv_c_int32_t in #(
+  no|yes) ;; #(
+  *)
+
+cat >>confdefs.h <<_ACEOF
+#define int32_t $ac_cv_c_int32_t
+_ACEOF
+;;
+  esac
+
+
+  { echo "$as_me:$LINENO: checking for int64_t" >&5
+echo $ECHO_N "checking for int64_t... $ECHO_C" >&6; }
+if test "${ac_cv_c_int64_t+set}" = set; then
+  echo $ECHO_N "(cached) $ECHO_C" >&6
+else
+  ac_cv_c_int64_t=no
+     for ac_type in 'int64_t' 'int' 'long int' \
+	 'long long int' 'short int' 'signed char'; do
+       cat >conftest.$ac_ext <<_ACEOF
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+$ac_includes_default
+int
+main ()
+{
+static int test_array [1 - 2 * !(0 < ($ac_type) (((($ac_type) 1 << (64 - 2)) - 1) * 2 + 1))];
+test_array [0] = 0
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.$ac_objext
+if { (ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5
+  (eval "$ac_compile") 2>conftest.er1
+  ac_status=$?
+  grep -v '^ *+' conftest.er1 >conftest.err
+  rm -f conftest.er1
+  cat conftest.err >&5
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then
+  cat >conftest.$ac_ext <<_ACEOF
+/* confdefs.h.  */
+_ACEOF
+cat confdefs.h >>conftest.$ac_ext
+cat >>conftest.$ac_ext <<_ACEOF
+/* end confdefs.h.  */
+$ac_includes_default
+int
+main ()
+{
+static int test_array [1 - 2 * !(($ac_type) (((($ac_type) 1 << (64 - 2)) - 1) * 2 + 1)
+	         < ($ac_type) (((($ac_type) 1 << (64 - 2)) - 1) * 2 + 2))];
+test_array [0] = 0
+
+  ;
+  return 0;
+}
+_ACEOF
+rm -f conftest.$ac_objext
+if { (ac_try="$ac_compile"
+case "(($ac_try" in
+  *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;;
+  *) ac_try_echo=$ac_try;;
+esac
+eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5
+  (eval "$ac_compile") 2>conftest.er1
+  ac_status=$?
+  grep -v '^ *+' conftest.er1 >conftest.err
+  rm -f conftest.er1
+  cat conftest.err >&5
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); } && {
+	 test -z "$ac_c_werror_flag" ||
+	 test ! -s conftest.err
+       } && test -s conftest.$ac_objext; then
+  :
+else
+  echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+	case $ac_type in
+  int64_t) ac_cv_c_int64_t=yes ;;
+  *) ac_cv_c_int64_t=$ac_type ;;
+esac
+
+fi
+
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+else
+  echo "$as_me: failed program was:" >&5
+sed 's/^/| /' conftest.$ac_ext >&5
+
+
+fi
+
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+       test "$ac_cv_c_int64_t" != no && break
+     done
+fi
+{ echo "$as_me:$LINENO: result: $ac_cv_c_int64_t" >&5
+echo "${ECHO_T}$ac_cv_c_int64_t" >&6; }
+  case $ac_cv_c_int64_t in #(
+  no|yes) ;; #(
+  *)
+
+cat >>confdefs.h <<_ACEOF
+#define int64_t $ac_cv_c_int64_t
+_ACEOF
+;;
+  esac
+
 { echo "$as_me:$LINENO: checking for ssize_t" >&5
 echo $ECHO_N "checking for ssize_t... $ECHO_C" >&6; }
 if test "${ac_cv_type_ssize_t+set}" = set; then
@@ -21687,6 +22069,37 @@ fi
 
 LIBS=$LIBS_SAVE
 
+# determine what size digit to use for Python's longs
+{ echo "$as_me:$LINENO: checking digit size for Python's longs" >&5
+echo $ECHO_N "checking digit size for Python's longs... $ECHO_C" >&6; }
+# Check whether --enable-big-digits was given.
+if test "${enable_big_digits+set}" = set; then
+  enableval=$enable_big_digits; case $enable_big_digits in
+yes)
+  enable_big_digits=30 ;;
+no)
+  enable_big_digits=15 ;;
+15|30)
+  ;;
+*)
+  { { echo "$as_me:$LINENO: error: bad value $enable_big_digits for --enable-big-digits; value should be 15 or 30" >&5
+echo "$as_me: error: bad value $enable_big_digits for --enable-big-digits; value should be 15 or 30" >&2;}
+   { (exit 1); exit 1; }; } ;;
+esac
+{ echo "$as_me:$LINENO: result: $enable_big_digits" >&5
+echo "${ECHO_T}$enable_big_digits" >&6; }
+
+cat >>confdefs.h <<_ACEOF
+#define PYLONG_BITS_IN_DIGIT $enable_big_digits
+_ACEOF
+
+
+else
+  { echo "$as_me:$LINENO: result: no value specified" >&5
+echo "${ECHO_T}no value specified" >&6; }
+fi
+
+
 # check for wchar.h
 if test "${ac_cv_header_wchar_h+set}" = set; then
   { echo "$as_me:$LINENO: checking for wchar.h" >&5

configure.in

@@ -1288,6 +1288,10 @@ AC_TYPE_PID_T
 AC_TYPE_SIGNAL
 AC_TYPE_SIZE_T
 AC_TYPE_UID_T
+AC_TYPE_UINT32_T
+AC_TYPE_UINT64_T
+AC_TYPE_INT32_T
+AC_TYPE_INT64_T
 AC_CHECK_TYPE(ssize_t,
   AC_DEFINE(HAVE_SSIZE_T, 1, Define if your compiler provides ssize_t),,)
 
@@ -3125,6 +3129,25 @@ AC_CHECK_DECLS([isinf, isnan, isfinite], [], [], [[#include <math.h>]])
 
 LIBS=$LIBS_SAVE
 
+# determine what size digit to use for Python's longs
+AC_MSG_CHECKING([digit size for Python's longs])
+AC_ARG_ENABLE(big-digits,
+AC_HELP_STRING([--enable-big-digits@<:@=BITS@:>@],[use big digits for Python longs [[BITS=30]]]),
+[case $enable_big_digits in
+yes)
+  enable_big_digits=30 ;;
+no)
+  enable_big_digits=15 ;;
+[15|30])
+  ;;
+*)
+  AC_MSG_ERROR([bad value $enable_big_digits for --enable-big-digits; value should be 15 or 30]) ;;
+esac
+AC_MSG_RESULT($enable_big_digits)
+AC_DEFINE_UNQUOTED(PYLONG_BITS_IN_DIGIT, $enable_big_digits, [Define as the preferred size in bits of long digits])
+],
+[AC_MSG_RESULT(no value specified)])
+
 # check for wchar.h
 AC_CHECK_HEADER(wchar.h, [
   AC_DEFINE(HAVE_WCHAR_H, 1, 

pyconfig.h.in

@@ -855,6 +855,9 @@
 /* Defined if PTHREAD_SCOPE_SYSTEM supported. */
 #undef PTHREAD_SYSTEM_SCHED_SUPPORTED
 
+/* Define as the preferred size in bits of long digits */
+#undef PYLONG_BITS_IN_DIGIT
+
 /* Define to printf format modifier for Py_ssize_t */
 #undef PY_FORMAT_SIZE_T
 
@@ -1036,6 +1039,16 @@
 /* Define to force use of thread-safe errno, h_errno, and other functions */
 #undef _REENTRANT
 
+/* Define for Solaris 2.5.1 so the uint32_t typedef from <sys/synch.h>,
+   <pthread.h>, or <semaphore.h> is not used. If the typedef was allowed, the
+   #define below would cause a syntax error. */
+#undef _UINT32_T
+
+/* Define for Solaris 2.5.1 so the uint64_t typedef from <sys/synch.h>,
+   <pthread.h>, or <semaphore.h> is not used. If the typedef was allowed, the
+   #define below would cause a syntax error. */
+#undef _UINT64_T
+
 /* Define to the level of X/Open that your system supports */
 #undef _XOPEN_SOURCE
 
@@ -1062,6 +1075,14 @@
 /* Define to `int' if <sys/types.h> doesn't define. */
 #undef gid_t
 
+/* Define to the type of a signed integer type of width exactly 32 bits if
+   such a type exists and the standard includes do not define it. */
+#undef int32_t
+
+/* Define to the type of a signed integer type of width exactly 64 bits if
+   such a type exists and the standard includes do not define it. */
+#undef int64_t
+
 /* Define to `int' if <sys/types.h> does not define. */
 #undef mode_t
 
@@ -1083,6 +1104,14 @@
 /* Define to `int' if <sys/types.h> doesn't define. */
 #undef uid_t
 
+/* Define to the type of an unsigned integer type of width exactly 32 bits if
+   such a type exists and the standard includes do not define it. */
+#undef uint32_t
+
+/* Define to the type of an unsigned integer type of width exactly 64 bits if
+   such a type exists and the standard includes do not define it. */
+#undef uint64_t
+
 /* Define to empty if the keyword does not work. */
 #undef volatile