Issue #5512: speed up the long division algorithm for Python longs.
[python.git] / Objects / intobject.c
blobd4532f49a67bb38c7e31aadf266151599e4e37bc
2 /* Integer object implementation */
4 #include "Python.h"
5 #include <ctype.h>
7 static PyObject *int_int(PyIntObject *v);
9 long
10 PyInt_GetMax(void)
12 return LONG_MAX; /* To initialize sys.maxint */
15 /* Integers are quite normal objects, to make object handling uniform.
16 (Using odd pointers to represent integers would save much space
17 but require extra checks for this special case throughout the code.)
18 Since a typical Python program spends much of its time allocating
19 and deallocating integers, these operations should be very fast.
20 Therefore we use a dedicated allocation scheme with a much lower
21 overhead (in space and time) than straight malloc(): a simple
22 dedicated free list, filled when necessary with memory from malloc().
24 block_list is a singly-linked list of all PyIntBlocks ever allocated,
25 linked via their next members. PyIntBlocks are never returned to the
26 system before shutdown (PyInt_Fini).
28 free_list is a singly-linked list of available PyIntObjects, linked
29 via abuse of their ob_type members.
32 #define BLOCK_SIZE 1000 /* 1K less typical malloc overhead */
33 #define BHEAD_SIZE 8 /* Enough for a 64-bit pointer */
34 #define N_INTOBJECTS ((BLOCK_SIZE - BHEAD_SIZE) / sizeof(PyIntObject))
36 struct _intblock {
37 struct _intblock *next;
38 PyIntObject objects[N_INTOBJECTS];
41 typedef struct _intblock PyIntBlock;
43 static PyIntBlock *block_list = NULL;
44 static PyIntObject *free_list = NULL;
46 static PyIntObject *
47 fill_free_list(void)
49 PyIntObject *p, *q;
50 /* Python's object allocator isn't appropriate for large blocks. */
51 p = (PyIntObject *) PyMem_MALLOC(sizeof(PyIntBlock));
52 if (p == NULL)
53 return (PyIntObject *) PyErr_NoMemory();
54 ((PyIntBlock *)p)->next = block_list;
55 block_list = (PyIntBlock *)p;
56 /* Link the int objects together, from rear to front, then return
57 the address of the last int object in the block. */
58 p = &((PyIntBlock *)p)->objects[0];
59 q = p + N_INTOBJECTS;
60 while (--q > p)
61 Py_TYPE(q) = (struct _typeobject *)(q-1);
62 Py_TYPE(q) = NULL;
63 return p + N_INTOBJECTS - 1;
66 #ifndef NSMALLPOSINTS
67 #define NSMALLPOSINTS 257
68 #endif
69 #ifndef NSMALLNEGINTS
70 #define NSMALLNEGINTS 5
71 #endif
72 #if NSMALLNEGINTS + NSMALLPOSINTS > 0
73 /* References to small integers are saved in this array so that they
74 can be shared.
75 The integers that are saved are those in the range
76 -NSMALLNEGINTS (inclusive) to NSMALLPOSINTS (not inclusive).
78 static PyIntObject *small_ints[NSMALLNEGINTS + NSMALLPOSINTS];
79 #endif
80 #ifdef COUNT_ALLOCS
81 Py_ssize_t quick_int_allocs;
82 Py_ssize_t quick_neg_int_allocs;
83 #endif
85 PyObject *
86 PyInt_FromLong(long ival)
88 register PyIntObject *v;
89 #if NSMALLNEGINTS + NSMALLPOSINTS > 0
90 if (-NSMALLNEGINTS <= ival && ival < NSMALLPOSINTS) {
91 v = small_ints[ival + NSMALLNEGINTS];
92 Py_INCREF(v);
93 #ifdef COUNT_ALLOCS
94 if (ival >= 0)
95 quick_int_allocs++;
96 else
97 quick_neg_int_allocs++;
98 #endif
99 return (PyObject *) v;
101 #endif
102 if (free_list == NULL) {
103 if ((free_list = fill_free_list()) == NULL)
104 return NULL;
106 /* Inline PyObject_New */
107 v = free_list;
108 free_list = (PyIntObject *)Py_TYPE(v);
109 PyObject_INIT(v, &PyInt_Type);
110 v->ob_ival = ival;
111 return (PyObject *) v;
114 PyObject *
115 PyInt_FromSize_t(size_t ival)
117 if (ival <= LONG_MAX)
118 return PyInt_FromLong((long)ival);
119 return _PyLong_FromSize_t(ival);
122 PyObject *
123 PyInt_FromSsize_t(Py_ssize_t ival)
125 if (ival >= LONG_MIN && ival <= LONG_MAX)
126 return PyInt_FromLong((long)ival);
127 return _PyLong_FromSsize_t(ival);
130 static void
131 int_dealloc(PyIntObject *v)
133 if (PyInt_CheckExact(v)) {
134 Py_TYPE(v) = (struct _typeobject *)free_list;
135 free_list = v;
137 else
138 Py_TYPE(v)->tp_free((PyObject *)v);
141 static void
142 int_free(PyIntObject *v)
144 Py_TYPE(v) = (struct _typeobject *)free_list;
145 free_list = v;
148 long
149 PyInt_AsLong(register PyObject *op)
151 PyNumberMethods *nb;
152 PyIntObject *io;
153 long val;
155 if (op && PyInt_Check(op))
156 return PyInt_AS_LONG((PyIntObject*) op);
158 if (op == NULL || (nb = Py_TYPE(op)->tp_as_number) == NULL ||
159 nb->nb_int == NULL) {
160 PyErr_SetString(PyExc_TypeError, "an integer is required");
161 return -1;
164 io = (PyIntObject*) (*nb->nb_int) (op);
165 if (io == NULL)
166 return -1;
167 if (!PyInt_Check(io)) {
168 if (PyLong_Check(io)) {
169 /* got a long? => retry int conversion */
170 val = PyLong_AsLong((PyObject *)io);
171 Py_DECREF(io);
172 if ((val == -1) && PyErr_Occurred())
173 return -1;
174 return val;
176 else
178 Py_DECREF(io);
179 PyErr_SetString(PyExc_TypeError,
180 "nb_int should return int object");
181 return -1;
185 val = PyInt_AS_LONG(io);
186 Py_DECREF(io);
188 return val;
191 Py_ssize_t
192 PyInt_AsSsize_t(register PyObject *op)
194 #if SIZEOF_SIZE_T != SIZEOF_LONG
195 PyNumberMethods *nb;
196 PyIntObject *io;
197 Py_ssize_t val;
198 #endif
200 if (op == NULL) {
201 PyErr_SetString(PyExc_TypeError, "an integer is required");
202 return -1;
205 if (PyInt_Check(op))
206 return PyInt_AS_LONG((PyIntObject*) op);
207 if (PyLong_Check(op))
208 return _PyLong_AsSsize_t(op);
209 #if SIZEOF_SIZE_T == SIZEOF_LONG
210 return PyInt_AsLong(op);
211 #else
213 if ((nb = Py_TYPE(op)->tp_as_number) == NULL ||
214 (nb->nb_int == NULL && nb->nb_long == 0)) {
215 PyErr_SetString(PyExc_TypeError, "an integer is required");
216 return -1;
219 if (nb->nb_long != 0)
220 io = (PyIntObject*) (*nb->nb_long) (op);
221 else
222 io = (PyIntObject*) (*nb->nb_int) (op);
223 if (io == NULL)
224 return -1;
225 if (!PyInt_Check(io)) {
226 if (PyLong_Check(io)) {
227 /* got a long? => retry int conversion */
228 val = _PyLong_AsSsize_t((PyObject *)io);
229 Py_DECREF(io);
230 if ((val == -1) && PyErr_Occurred())
231 return -1;
232 return val;
234 else
236 Py_DECREF(io);
237 PyErr_SetString(PyExc_TypeError,
238 "nb_int should return int object");
239 return -1;
243 val = PyInt_AS_LONG(io);
244 Py_DECREF(io);
246 return val;
247 #endif
250 unsigned long
251 PyInt_AsUnsignedLongMask(register PyObject *op)
253 PyNumberMethods *nb;
254 PyIntObject *io;
255 unsigned long val;
257 if (op && PyInt_Check(op))
258 return PyInt_AS_LONG((PyIntObject*) op);
259 if (op && PyLong_Check(op))
260 return PyLong_AsUnsignedLongMask(op);
262 if (op == NULL || (nb = Py_TYPE(op)->tp_as_number) == NULL ||
263 nb->nb_int == NULL) {
264 PyErr_SetString(PyExc_TypeError, "an integer is required");
265 return (unsigned long)-1;
268 io = (PyIntObject*) (*nb->nb_int) (op);
269 if (io == NULL)
270 return (unsigned long)-1;
271 if (!PyInt_Check(io)) {
272 if (PyLong_Check(io)) {
273 val = PyLong_AsUnsignedLongMask((PyObject *)io);
274 Py_DECREF(io);
275 if (PyErr_Occurred())
276 return (unsigned long)-1;
277 return val;
279 else
281 Py_DECREF(io);
282 PyErr_SetString(PyExc_TypeError,
283 "nb_int should return int object");
284 return (unsigned long)-1;
288 val = PyInt_AS_LONG(io);
289 Py_DECREF(io);
291 return val;
294 #ifdef HAVE_LONG_LONG
295 unsigned PY_LONG_LONG
296 PyInt_AsUnsignedLongLongMask(register PyObject *op)
298 PyNumberMethods *nb;
299 PyIntObject *io;
300 unsigned PY_LONG_LONG val;
302 if (op && PyInt_Check(op))
303 return PyInt_AS_LONG((PyIntObject*) op);
304 if (op && PyLong_Check(op))
305 return PyLong_AsUnsignedLongLongMask(op);
307 if (op == NULL || (nb = Py_TYPE(op)->tp_as_number) == NULL ||
308 nb->nb_int == NULL) {
309 PyErr_SetString(PyExc_TypeError, "an integer is required");
310 return (unsigned PY_LONG_LONG)-1;
313 io = (PyIntObject*) (*nb->nb_int) (op);
314 if (io == NULL)
315 return (unsigned PY_LONG_LONG)-1;
316 if (!PyInt_Check(io)) {
317 if (PyLong_Check(io)) {
318 val = PyLong_AsUnsignedLongLongMask((PyObject *)io);
319 Py_DECREF(io);
320 if (PyErr_Occurred())
321 return (unsigned PY_LONG_LONG)-1;
322 return val;
324 else
326 Py_DECREF(io);
327 PyErr_SetString(PyExc_TypeError,
328 "nb_int should return int object");
329 return (unsigned PY_LONG_LONG)-1;
333 val = PyInt_AS_LONG(io);
334 Py_DECREF(io);
336 return val;
338 #endif
340 PyObject *
341 PyInt_FromString(char *s, char **pend, int base)
343 char *end;
344 long x;
345 Py_ssize_t slen;
346 PyObject *sobj, *srepr;
348 if ((base != 0 && base < 2) || base > 36) {
349 PyErr_SetString(PyExc_ValueError,
350 "int() base must be >= 2 and <= 36");
351 return NULL;
354 while (*s && isspace(Py_CHARMASK(*s)))
355 s++;
356 errno = 0;
357 if (base == 0 && s[0] == '0') {
358 x = (long) PyOS_strtoul(s, &end, base);
359 if (x < 0)
360 return PyLong_FromString(s, pend, base);
362 else
363 x = PyOS_strtol(s, &end, base);
364 if (end == s || !isalnum(Py_CHARMASK(end[-1])))
365 goto bad;
366 while (*end && isspace(Py_CHARMASK(*end)))
367 end++;
368 if (*end != '\0') {
369 bad:
370 slen = strlen(s) < 200 ? strlen(s) : 200;
371 sobj = PyString_FromStringAndSize(s, slen);
372 if (sobj == NULL)
373 return NULL;
374 srepr = PyObject_Repr(sobj);
375 Py_DECREF(sobj);
376 if (srepr == NULL)
377 return NULL;
378 PyErr_Format(PyExc_ValueError,
379 "invalid literal for int() with base %d: %s",
380 base, PyString_AS_STRING(srepr));
381 Py_DECREF(srepr);
382 return NULL;
384 else if (errno != 0)
385 return PyLong_FromString(s, pend, base);
386 if (pend)
387 *pend = end;
388 return PyInt_FromLong(x);
391 #ifdef Py_USING_UNICODE
392 PyObject *
393 PyInt_FromUnicode(Py_UNICODE *s, Py_ssize_t length, int base)
395 PyObject *result;
396 char *buffer = (char *)PyMem_MALLOC(length+1);
398 if (buffer == NULL)
399 return PyErr_NoMemory();
401 if (PyUnicode_EncodeDecimal(s, length, buffer, NULL)) {
402 PyMem_FREE(buffer);
403 return NULL;
405 result = PyInt_FromString(buffer, NULL, base);
406 PyMem_FREE(buffer);
407 return result;
409 #endif
411 /* Methods */
413 /* Integers are seen as the "smallest" of all numeric types and thus
414 don't have any knowledge about conversion of other types to
415 integers. */
417 #define CONVERT_TO_LONG(obj, lng) \
418 if (PyInt_Check(obj)) { \
419 lng = PyInt_AS_LONG(obj); \
421 else { \
422 Py_INCREF(Py_NotImplemented); \
423 return Py_NotImplemented; \
426 /* ARGSUSED */
427 static int
428 int_print(PyIntObject *v, FILE *fp, int flags)
429 /* flags -- not used but required by interface */
431 long int_val = v->ob_ival;
432 Py_BEGIN_ALLOW_THREADS
433 fprintf(fp, "%ld", int_val);
434 Py_END_ALLOW_THREADS
435 return 0;
438 static PyObject *
439 int_repr(PyIntObject *v)
441 return _PyInt_Format(v, 10, 0);
444 static int
445 int_compare(PyIntObject *v, PyIntObject *w)
447 register long i = v->ob_ival;
448 register long j = w->ob_ival;
449 return (i < j) ? -1 : (i > j) ? 1 : 0;
452 static long
453 int_hash(PyIntObject *v)
455 /* XXX If this is changed, you also need to change the way
456 Python's long, float and complex types are hashed. */
457 long x = v -> ob_ival;
458 if (x == -1)
459 x = -2;
460 return x;
463 static PyObject *
464 int_add(PyIntObject *v, PyIntObject *w)
466 register long a, b, x;
467 CONVERT_TO_LONG(v, a);
468 CONVERT_TO_LONG(w, b);
469 x = a + b;
470 if ((x^a) >= 0 || (x^b) >= 0)
471 return PyInt_FromLong(x);
472 return PyLong_Type.tp_as_number->nb_add((PyObject *)v, (PyObject *)w);
475 static PyObject *
476 int_sub(PyIntObject *v, PyIntObject *w)
478 register long a, b, x;
479 CONVERT_TO_LONG(v, a);
480 CONVERT_TO_LONG(w, b);
481 x = a - b;
482 if ((x^a) >= 0 || (x^~b) >= 0)
483 return PyInt_FromLong(x);
484 return PyLong_Type.tp_as_number->nb_subtract((PyObject *)v,
485 (PyObject *)w);
489 Integer overflow checking for * is painful: Python tried a couple ways, but
490 they didn't work on all platforms, or failed in endcases (a product of
491 -sys.maxint-1 has been a particular pain).
493 Here's another way:
495 The native long product x*y is either exactly right or *way* off, being
496 just the last n bits of the true product, where n is the number of bits
497 in a long (the delivered product is the true product plus i*2**n for
498 some integer i).
500 The native double product (double)x * (double)y is subject to three
501 rounding errors: on a sizeof(long)==8 box, each cast to double can lose
502 info, and even on a sizeof(long)==4 box, the multiplication can lose info.
503 But, unlike the native long product, it's not in *range* trouble: even
504 if sizeof(long)==32 (256-bit longs), the product easily fits in the
505 dynamic range of a double. So the leading 50 (or so) bits of the double
506 product are correct.
508 We check these two ways against each other, and declare victory if they're
509 approximately the same. Else, because the native long product is the only
510 one that can lose catastrophic amounts of information, it's the native long
511 product that must have overflowed.
514 static PyObject *
515 int_mul(PyObject *v, PyObject *w)
517 long a, b;
518 long longprod; /* a*b in native long arithmetic */
519 double doubled_longprod; /* (double)longprod */
520 double doubleprod; /* (double)a * (double)b */
522 CONVERT_TO_LONG(v, a);
523 CONVERT_TO_LONG(w, b);
524 longprod = a * b;
525 doubleprod = (double)a * (double)b;
526 doubled_longprod = (double)longprod;
528 /* Fast path for normal case: small multiplicands, and no info
529 is lost in either method. */
530 if (doubled_longprod == doubleprod)
531 return PyInt_FromLong(longprod);
533 /* Somebody somewhere lost info. Close enough, or way off? Note
534 that a != 0 and b != 0 (else doubled_longprod == doubleprod == 0).
535 The difference either is or isn't significant compared to the
536 true value (of which doubleprod is a good approximation).
539 const double diff = doubled_longprod - doubleprod;
540 const double absdiff = diff >= 0.0 ? diff : -diff;
541 const double absprod = doubleprod >= 0.0 ? doubleprod :
542 -doubleprod;
543 /* absdiff/absprod <= 1/32 iff
544 32 * absdiff <= absprod -- 5 good bits is "close enough" */
545 if (32.0 * absdiff <= absprod)
546 return PyInt_FromLong(longprod);
547 else
548 return PyLong_Type.tp_as_number->nb_multiply(v, w);
552 /* Integer overflow checking for unary negation: on a 2's-complement
553 * box, -x overflows iff x is the most negative long. In this case we
554 * get -x == x. However, -x is undefined (by C) if x /is/ the most
555 * negative long (it's a signed overflow case), and some compilers care.
556 * So we cast x to unsigned long first. However, then other compilers
557 * warn about applying unary minus to an unsigned operand. Hence the
558 * weird "0-".
560 #define UNARY_NEG_WOULD_OVERFLOW(x) \
561 ((x) < 0 && (unsigned long)(x) == 0-(unsigned long)(x))
563 /* Return type of i_divmod */
564 enum divmod_result {
565 DIVMOD_OK, /* Correct result */
566 DIVMOD_OVERFLOW, /* Overflow, try again using longs */
567 DIVMOD_ERROR /* Exception raised */
570 static enum divmod_result
571 i_divmod(register long x, register long y,
572 long *p_xdivy, long *p_xmody)
574 long xdivy, xmody;
576 if (y == 0) {
577 PyErr_SetString(PyExc_ZeroDivisionError,
578 "integer division or modulo by zero");
579 return DIVMOD_ERROR;
581 /* (-sys.maxint-1)/-1 is the only overflow case. */
582 if (y == -1 && UNARY_NEG_WOULD_OVERFLOW(x))
583 return DIVMOD_OVERFLOW;
584 xdivy = x / y;
585 xmody = x - xdivy * y;
586 /* If the signs of x and y differ, and the remainder is non-0,
587 * C89 doesn't define whether xdivy is now the floor or the
588 * ceiling of the infinitely precise quotient. We want the floor,
589 * and we have it iff the remainder's sign matches y's.
591 if (xmody && ((y ^ xmody) < 0) /* i.e. and signs differ */) {
592 xmody += y;
593 --xdivy;
594 assert(xmody && ((y ^ xmody) >= 0));
596 *p_xdivy = xdivy;
597 *p_xmody = xmody;
598 return DIVMOD_OK;
601 static PyObject *
602 int_div(PyIntObject *x, PyIntObject *y)
604 long xi, yi;
605 long d, m;
606 CONVERT_TO_LONG(x, xi);
607 CONVERT_TO_LONG(y, yi);
608 switch (i_divmod(xi, yi, &d, &m)) {
609 case DIVMOD_OK:
610 return PyInt_FromLong(d);
611 case DIVMOD_OVERFLOW:
612 return PyLong_Type.tp_as_number->nb_divide((PyObject *)x,
613 (PyObject *)y);
614 default:
615 return NULL;
619 static PyObject *
620 int_classic_div(PyIntObject *x, PyIntObject *y)
622 long xi, yi;
623 long d, m;
624 CONVERT_TO_LONG(x, xi);
625 CONVERT_TO_LONG(y, yi);
626 if (Py_DivisionWarningFlag &&
627 PyErr_Warn(PyExc_DeprecationWarning, "classic int division") < 0)
628 return NULL;
629 switch (i_divmod(xi, yi, &d, &m)) {
630 case DIVMOD_OK:
631 return PyInt_FromLong(d);
632 case DIVMOD_OVERFLOW:
633 return PyLong_Type.tp_as_number->nb_divide((PyObject *)x,
634 (PyObject *)y);
635 default:
636 return NULL;
640 static PyObject *
641 int_true_divide(PyObject *v, PyObject *w)
643 /* If they aren't both ints, give someone else a chance. In
644 particular, this lets int/long get handled by longs, which
645 underflows to 0 gracefully if the long is too big to convert
646 to float. */
647 if (PyInt_Check(v) && PyInt_Check(w))
648 return PyFloat_Type.tp_as_number->nb_true_divide(v, w);
649 Py_INCREF(Py_NotImplemented);
650 return Py_NotImplemented;
653 static PyObject *
654 int_mod(PyIntObject *x, PyIntObject *y)
656 long xi, yi;
657 long d, m;
658 CONVERT_TO_LONG(x, xi);
659 CONVERT_TO_LONG(y, yi);
660 switch (i_divmod(xi, yi, &d, &m)) {
661 case DIVMOD_OK:
662 return PyInt_FromLong(m);
663 case DIVMOD_OVERFLOW:
664 return PyLong_Type.tp_as_number->nb_remainder((PyObject *)x,
665 (PyObject *)y);
666 default:
667 return NULL;
671 static PyObject *
672 int_divmod(PyIntObject *x, PyIntObject *y)
674 long xi, yi;
675 long d, m;
676 CONVERT_TO_LONG(x, xi);
677 CONVERT_TO_LONG(y, yi);
678 switch (i_divmod(xi, yi, &d, &m)) {
679 case DIVMOD_OK:
680 return Py_BuildValue("(ll)", d, m);
681 case DIVMOD_OVERFLOW:
682 return PyLong_Type.tp_as_number->nb_divmod((PyObject *)x,
683 (PyObject *)y);
684 default:
685 return NULL;
689 static PyObject *
690 int_pow(PyIntObject *v, PyIntObject *w, PyIntObject *z)
692 register long iv, iw, iz=0, ix, temp, prev;
693 CONVERT_TO_LONG(v, iv);
694 CONVERT_TO_LONG(w, iw);
695 if (iw < 0) {
696 if ((PyObject *)z != Py_None) {
697 PyErr_SetString(PyExc_TypeError, "pow() 2nd argument "
698 "cannot be negative when 3rd argument specified");
699 return NULL;
701 /* Return a float. This works because we know that
702 this calls float_pow() which converts its
703 arguments to double. */
704 return PyFloat_Type.tp_as_number->nb_power(
705 (PyObject *)v, (PyObject *)w, (PyObject *)z);
707 if ((PyObject *)z != Py_None) {
708 CONVERT_TO_LONG(z, iz);
709 if (iz == 0) {
710 PyErr_SetString(PyExc_ValueError,
711 "pow() 3rd argument cannot be 0");
712 return NULL;
716 * XXX: The original exponentiation code stopped looping
717 * when temp hit zero; this code will continue onwards
718 * unnecessarily, but at least it won't cause any errors.
719 * Hopefully the speed improvement from the fast exponentiation
720 * will compensate for the slight inefficiency.
721 * XXX: Better handling of overflows is desperately needed.
723 temp = iv;
724 ix = 1;
725 while (iw > 0) {
726 prev = ix; /* Save value for overflow check */
727 if (iw & 1) {
728 ix = ix*temp;
729 if (temp == 0)
730 break; /* Avoid ix / 0 */
731 if (ix / temp != prev) {
732 return PyLong_Type.tp_as_number->nb_power(
733 (PyObject *)v,
734 (PyObject *)w,
735 (PyObject *)z);
738 iw >>= 1; /* Shift exponent down by 1 bit */
739 if (iw==0) break;
740 prev = temp;
741 temp *= temp; /* Square the value of temp */
742 if (prev != 0 && temp / prev != prev) {
743 return PyLong_Type.tp_as_number->nb_power(
744 (PyObject *)v, (PyObject *)w, (PyObject *)z);
746 if (iz) {
747 /* If we did a multiplication, perform a modulo */
748 ix = ix % iz;
749 temp = temp % iz;
752 if (iz) {
753 long div, mod;
754 switch (i_divmod(ix, iz, &div, &mod)) {
755 case DIVMOD_OK:
756 ix = mod;
757 break;
758 case DIVMOD_OVERFLOW:
759 return PyLong_Type.tp_as_number->nb_power(
760 (PyObject *)v, (PyObject *)w, (PyObject *)z);
761 default:
762 return NULL;
765 return PyInt_FromLong(ix);
768 static PyObject *
769 int_neg(PyIntObject *v)
771 register long a;
772 a = v->ob_ival;
773 /* check for overflow */
774 if (UNARY_NEG_WOULD_OVERFLOW(a)) {
775 PyObject *o = PyLong_FromLong(a);
776 if (o != NULL) {
777 PyObject *result = PyNumber_Negative(o);
778 Py_DECREF(o);
779 return result;
781 return NULL;
783 return PyInt_FromLong(-a);
786 static PyObject *
787 int_abs(PyIntObject *v)
789 if (v->ob_ival >= 0)
790 return int_int(v);
791 else
792 return int_neg(v);
795 static int
796 int_nonzero(PyIntObject *v)
798 return v->ob_ival != 0;
801 static PyObject *
802 int_invert(PyIntObject *v)
804 return PyInt_FromLong(~v->ob_ival);
807 static PyObject *
808 int_lshift(PyIntObject *v, PyIntObject *w)
810 long a, b, c;
811 PyObject *vv, *ww, *result;
813 CONVERT_TO_LONG(v, a);
814 CONVERT_TO_LONG(w, b);
815 if (b < 0) {
816 PyErr_SetString(PyExc_ValueError, "negative shift count");
817 return NULL;
819 if (a == 0 || b == 0)
820 return int_int(v);
821 if (b >= LONG_BIT) {
822 vv = PyLong_FromLong(PyInt_AS_LONG(v));
823 if (vv == NULL)
824 return NULL;
825 ww = PyLong_FromLong(PyInt_AS_LONG(w));
826 if (ww == NULL) {
827 Py_DECREF(vv);
828 return NULL;
830 result = PyNumber_Lshift(vv, ww);
831 Py_DECREF(vv);
832 Py_DECREF(ww);
833 return result;
835 c = a << b;
836 if (a != Py_ARITHMETIC_RIGHT_SHIFT(long, c, b)) {
837 vv = PyLong_FromLong(PyInt_AS_LONG(v));
838 if (vv == NULL)
839 return NULL;
840 ww = PyLong_FromLong(PyInt_AS_LONG(w));
841 if (ww == NULL) {
842 Py_DECREF(vv);
843 return NULL;
845 result = PyNumber_Lshift(vv, ww);
846 Py_DECREF(vv);
847 Py_DECREF(ww);
848 return result;
850 return PyInt_FromLong(c);
853 static PyObject *
854 int_rshift(PyIntObject *v, PyIntObject *w)
856 register long a, b;
857 CONVERT_TO_LONG(v, a);
858 CONVERT_TO_LONG(w, b);
859 if (b < 0) {
860 PyErr_SetString(PyExc_ValueError, "negative shift count");
861 return NULL;
863 if (a == 0 || b == 0)
864 return int_int(v);
865 if (b >= LONG_BIT) {
866 if (a < 0)
867 a = -1;
868 else
869 a = 0;
871 else {
872 a = Py_ARITHMETIC_RIGHT_SHIFT(long, a, b);
874 return PyInt_FromLong(a);
877 static PyObject *
878 int_and(PyIntObject *v, PyIntObject *w)
880 register long a, b;
881 CONVERT_TO_LONG(v, a);
882 CONVERT_TO_LONG(w, b);
883 return PyInt_FromLong(a & b);
886 static PyObject *
887 int_xor(PyIntObject *v, PyIntObject *w)
889 register long a, b;
890 CONVERT_TO_LONG(v, a);
891 CONVERT_TO_LONG(w, b);
892 return PyInt_FromLong(a ^ b);
895 static PyObject *
896 int_or(PyIntObject *v, PyIntObject *w)
898 register long a, b;
899 CONVERT_TO_LONG(v, a);
900 CONVERT_TO_LONG(w, b);
901 return PyInt_FromLong(a | b);
904 static int
905 int_coerce(PyObject **pv, PyObject **pw)
907 if (PyInt_Check(*pw)) {
908 Py_INCREF(*pv);
909 Py_INCREF(*pw);
910 return 0;
912 return 1; /* Can't do it */
915 static PyObject *
916 int_int(PyIntObject *v)
918 if (PyInt_CheckExact(v))
919 Py_INCREF(v);
920 else
921 v = (PyIntObject *)PyInt_FromLong(v->ob_ival);
922 return (PyObject *)v;
925 static PyObject *
926 int_long(PyIntObject *v)
928 return PyLong_FromLong((v -> ob_ival));
931 static PyObject *
932 int_float(PyIntObject *v)
934 return PyFloat_FromDouble((double)(v -> ob_ival));
937 static PyObject *
938 int_oct(PyIntObject *v)
940 return _PyInt_Format(v, 8, 0);
943 static PyObject *
944 int_hex(PyIntObject *v)
946 return _PyInt_Format(v, 16, 0);
949 static PyObject *
950 int_subtype_new(PyTypeObject *type, PyObject *args, PyObject *kwds);
952 static PyObject *
953 int_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
955 PyObject *x = NULL;
956 int base = -909;
957 static char *kwlist[] = {"x", "base", 0};
959 if (type != &PyInt_Type)
960 return int_subtype_new(type, args, kwds); /* Wimp out */
961 if (!PyArg_ParseTupleAndKeywords(args, kwds, "|Oi:int", kwlist,
962 &x, &base))
963 return NULL;
964 if (x == NULL)
965 return PyInt_FromLong(0L);
966 if (base == -909)
967 return PyNumber_Int(x);
968 if (PyString_Check(x)) {
969 /* Since PyInt_FromString doesn't have a length parameter,
970 * check here for possible NULs in the string. */
971 char *string = PyString_AS_STRING(x);
972 if (strlen(string) != PyString_Size(x)) {
973 /* create a repr() of the input string,
974 * just like PyInt_FromString does */
975 PyObject *srepr;
976 srepr = PyObject_Repr(x);
977 if (srepr == NULL)
978 return NULL;
979 PyErr_Format(PyExc_ValueError,
980 "invalid literal for int() with base %d: %s",
981 base, PyString_AS_STRING(srepr));
982 Py_DECREF(srepr);
983 return NULL;
985 return PyInt_FromString(string, NULL, base);
987 #ifdef Py_USING_UNICODE
988 if (PyUnicode_Check(x))
989 return PyInt_FromUnicode(PyUnicode_AS_UNICODE(x),
990 PyUnicode_GET_SIZE(x),
991 base);
992 #endif
993 PyErr_SetString(PyExc_TypeError,
994 "int() can't convert non-string with explicit base");
995 return NULL;
998 /* Wimpy, slow approach to tp_new calls for subtypes of int:
999 first create a regular int from whatever arguments we got,
1000 then allocate a subtype instance and initialize its ob_ival
1001 from the regular int. The regular int is then thrown away.
1003 static PyObject *
1004 int_subtype_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
1006 PyObject *tmp, *newobj;
1007 long ival;
1009 assert(PyType_IsSubtype(type, &PyInt_Type));
1010 tmp = int_new(&PyInt_Type, args, kwds);
1011 if (tmp == NULL)
1012 return NULL;
1013 if (!PyInt_Check(tmp)) {
1014 ival = PyLong_AsLong(tmp);
1015 if (ival == -1 && PyErr_Occurred()) {
1016 Py_DECREF(tmp);
1017 return NULL;
1019 } else {
1020 ival = ((PyIntObject *)tmp)->ob_ival;
1023 newobj = type->tp_alloc(type, 0);
1024 if (newobj == NULL) {
1025 Py_DECREF(tmp);
1026 return NULL;
1028 ((PyIntObject *)newobj)->ob_ival = ival;
1029 Py_DECREF(tmp);
1030 return newobj;
1033 static PyObject *
1034 int_getnewargs(PyIntObject *v)
1036 return Py_BuildValue("(l)", v->ob_ival);
1039 static PyObject *
1040 int_getN(PyIntObject *v, void *context) {
1041 return PyInt_FromLong((Py_intptr_t)context);
1044 /* Convert an integer to the given base. Returns a string.
1045 If base is 2, 8 or 16, add the proper prefix '0b', '0o' or '0x'.
1046 If newstyle is zero, then use the pre-2.6 behavior of octal having
1047 a leading "0" */
1048 PyAPI_FUNC(PyObject*)
1049 _PyInt_Format(PyIntObject *v, int base, int newstyle)
1051 /* There are no doubt many, many ways to optimize this, using code
1052 similar to _PyLong_Format */
1053 long n = v->ob_ival;
1054 int negative = n < 0;
1055 int is_zero = n == 0;
1057 /* For the reasoning behind this size, see
1058 http://c-faq.com/misc/hexio.html. Then, add a few bytes for
1059 the possible sign and prefix "0[box]" */
1060 char buf[sizeof(n)*CHAR_BIT+6];
1062 /* Start by pointing to the end of the buffer. We fill in from
1063 the back forward. */
1064 char* p = &buf[sizeof(buf)];
1066 assert(base >= 2 && base <= 36);
1068 do {
1069 /* I'd use i_divmod, except it doesn't produce the results
1070 I want when n is negative. So just duplicate the salient
1071 part here. */
1072 long div = n / base;
1073 long mod = n - div * base;
1075 /* convert abs(mod) to the right character in [0-9, a-z] */
1076 char cdigit = (char)(mod < 0 ? -mod : mod);
1077 cdigit += (cdigit < 10) ? '0' : 'a'-10;
1078 *--p = cdigit;
1080 n = div;
1081 } while(n);
1083 if (base == 2) {
1084 *--p = 'b';
1085 *--p = '0';
1087 else if (base == 8) {
1088 if (newstyle) {
1089 *--p = 'o';
1090 *--p = '0';
1092 else
1093 if (!is_zero)
1094 *--p = '0';
1096 else if (base == 16) {
1097 *--p = 'x';
1098 *--p = '0';
1100 else if (base != 10) {
1101 *--p = '#';
1102 *--p = '0' + base%10;
1103 if (base > 10)
1104 *--p = '0' + base/10;
1106 if (negative)
1107 *--p = '-';
1109 return PyString_FromStringAndSize(p, &buf[sizeof(buf)] - p);
1112 static PyObject *
1113 int__format__(PyObject *self, PyObject *args)
1115 PyObject *format_spec;
1117 if (!PyArg_ParseTuple(args, "O:__format__", &format_spec))
1118 return NULL;
1119 if (PyBytes_Check(format_spec))
1120 return _PyInt_FormatAdvanced(self,
1121 PyBytes_AS_STRING(format_spec),
1122 PyBytes_GET_SIZE(format_spec));
1123 if (PyUnicode_Check(format_spec)) {
1124 /* Convert format_spec to a str */
1125 PyObject *result;
1126 PyObject *str_spec = PyObject_Str(format_spec);
1128 if (str_spec == NULL)
1129 return NULL;
1131 result = _PyInt_FormatAdvanced(self,
1132 PyBytes_AS_STRING(str_spec),
1133 PyBytes_GET_SIZE(str_spec));
1135 Py_DECREF(str_spec);
1136 return result;
1138 PyErr_SetString(PyExc_TypeError, "__format__ requires str or unicode");
1139 return NULL;
1142 static const unsigned char BitLengthTable[32] = {
1143 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
1144 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
1147 static PyObject *
1148 int_bit_length(PyIntObject *v)
1150 unsigned long n;
1151 long r = 0;
1153 if (v->ob_ival < 0)
1154 /* avoid undefined behaviour when v->ob_ival == -LONG_MAX-1 */
1155 n = 0U-(unsigned long)v->ob_ival;
1156 else
1157 n = (unsigned long)v->ob_ival;
1159 while (n >= 32) {
1160 r += 6;
1161 n >>= 6;
1163 r += (long)(BitLengthTable[n]);
1164 return PyInt_FromLong(r);
1167 PyDoc_STRVAR(int_bit_length_doc,
1168 "int.bit_length() -> int\n\
1170 Number of bits necessary to represent self in binary.\n\
1171 >>> bin(37)\n\
1172 '0b100101'\n\
1173 >>> (37).bit_length()\n\
1174 6");
1176 #if 0
1177 static PyObject *
1178 int_is_finite(PyObject *v)
1180 Py_RETURN_TRUE;
1182 #endif
1184 static PyMethodDef int_methods[] = {
1185 {"conjugate", (PyCFunction)int_int, METH_NOARGS,
1186 "Returns self, the complex conjugate of any int."},
1187 {"bit_length", (PyCFunction)int_bit_length, METH_NOARGS,
1188 int_bit_length_doc},
1189 #if 0
1190 {"is_finite", (PyCFunction)int_is_finite, METH_NOARGS,
1191 "Returns always True."},
1192 #endif
1193 {"__trunc__", (PyCFunction)int_int, METH_NOARGS,
1194 "Truncating an Integral returns itself."},
1195 {"__getnewargs__", (PyCFunction)int_getnewargs, METH_NOARGS},
1196 {"__format__", (PyCFunction)int__format__, METH_VARARGS},
1197 {NULL, NULL} /* sentinel */
1200 static PyGetSetDef int_getset[] = {
1201 {"real",
1202 (getter)int_int, (setter)NULL,
1203 "the real part of a complex number",
1204 NULL},
1205 {"imag",
1206 (getter)int_getN, (setter)NULL,
1207 "the imaginary part of a complex number",
1208 (void*)0},
1209 {"numerator",
1210 (getter)int_int, (setter)NULL,
1211 "the numerator of a rational number in lowest terms",
1212 NULL},
1213 {"denominator",
1214 (getter)int_getN, (setter)NULL,
1215 "the denominator of a rational number in lowest terms",
1216 (void*)1},
1217 {NULL} /* Sentinel */
1220 PyDoc_STRVAR(int_doc,
1221 "int(x[, base]) -> integer\n\
1223 Convert a string or number to an integer, if possible. A floating point\n\
1224 argument will be truncated towards zero (this does not include a string\n\
1225 representation of a floating point number!) When converting a string, use\n\
1226 the optional base. It is an error to supply a base when converting a\n\
1227 non-string. If base is zero, the proper base is guessed based on the\n\
1228 string content. If the argument is outside the integer range a\n\
1229 long object will be returned instead.");
1231 static PyNumberMethods int_as_number = {
1232 (binaryfunc)int_add, /*nb_add*/
1233 (binaryfunc)int_sub, /*nb_subtract*/
1234 (binaryfunc)int_mul, /*nb_multiply*/
1235 (binaryfunc)int_classic_div, /*nb_divide*/
1236 (binaryfunc)int_mod, /*nb_remainder*/
1237 (binaryfunc)int_divmod, /*nb_divmod*/
1238 (ternaryfunc)int_pow, /*nb_power*/
1239 (unaryfunc)int_neg, /*nb_negative*/
1240 (unaryfunc)int_int, /*nb_positive*/
1241 (unaryfunc)int_abs, /*nb_absolute*/
1242 (inquiry)int_nonzero, /*nb_nonzero*/
1243 (unaryfunc)int_invert, /*nb_invert*/
1244 (binaryfunc)int_lshift, /*nb_lshift*/
1245 (binaryfunc)int_rshift, /*nb_rshift*/
1246 (binaryfunc)int_and, /*nb_and*/
1247 (binaryfunc)int_xor, /*nb_xor*/
1248 (binaryfunc)int_or, /*nb_or*/
1249 int_coerce, /*nb_coerce*/
1250 (unaryfunc)int_int, /*nb_int*/
1251 (unaryfunc)int_long, /*nb_long*/
1252 (unaryfunc)int_float, /*nb_float*/
1253 (unaryfunc)int_oct, /*nb_oct*/
1254 (unaryfunc)int_hex, /*nb_hex*/
1255 0, /*nb_inplace_add*/
1256 0, /*nb_inplace_subtract*/
1257 0, /*nb_inplace_multiply*/
1258 0, /*nb_inplace_divide*/
1259 0, /*nb_inplace_remainder*/
1260 0, /*nb_inplace_power*/
1261 0, /*nb_inplace_lshift*/
1262 0, /*nb_inplace_rshift*/
1263 0, /*nb_inplace_and*/
1264 0, /*nb_inplace_xor*/
1265 0, /*nb_inplace_or*/
1266 (binaryfunc)int_div, /* nb_floor_divide */
1267 int_true_divide, /* nb_true_divide */
1268 0, /* nb_inplace_floor_divide */
1269 0, /* nb_inplace_true_divide */
1270 (unaryfunc)int_int, /* nb_index */
1273 PyTypeObject PyInt_Type = {
1274 PyVarObject_HEAD_INIT(&PyType_Type, 0)
1275 "int",
1276 sizeof(PyIntObject),
1278 (destructor)int_dealloc, /* tp_dealloc */
1279 (printfunc)int_print, /* tp_print */
1280 0, /* tp_getattr */
1281 0, /* tp_setattr */
1282 (cmpfunc)int_compare, /* tp_compare */
1283 (reprfunc)int_repr, /* tp_repr */
1284 &int_as_number, /* tp_as_number */
1285 0, /* tp_as_sequence */
1286 0, /* tp_as_mapping */
1287 (hashfunc)int_hash, /* tp_hash */
1288 0, /* tp_call */
1289 (reprfunc)int_repr, /* tp_str */
1290 PyObject_GenericGetAttr, /* tp_getattro */
1291 0, /* tp_setattro */
1292 0, /* tp_as_buffer */
1293 Py_TPFLAGS_DEFAULT | Py_TPFLAGS_CHECKTYPES |
1294 Py_TPFLAGS_BASETYPE | Py_TPFLAGS_INT_SUBCLASS, /* tp_flags */
1295 int_doc, /* tp_doc */
1296 0, /* tp_traverse */
1297 0, /* tp_clear */
1298 0, /* tp_richcompare */
1299 0, /* tp_weaklistoffset */
1300 0, /* tp_iter */
1301 0, /* tp_iternext */
1302 int_methods, /* tp_methods */
1303 0, /* tp_members */
1304 int_getset, /* tp_getset */
1305 0, /* tp_base */
1306 0, /* tp_dict */
1307 0, /* tp_descr_get */
1308 0, /* tp_descr_set */
1309 0, /* tp_dictoffset */
1310 0, /* tp_init */
1311 0, /* tp_alloc */
1312 int_new, /* tp_new */
1313 (freefunc)int_free, /* tp_free */
1317 _PyInt_Init(void)
1319 PyIntObject *v;
1320 int ival;
1321 #if NSMALLNEGINTS + NSMALLPOSINTS > 0
1322 for (ival = -NSMALLNEGINTS; ival < NSMALLPOSINTS; ival++) {
1323 if (!free_list && (free_list = fill_free_list()) == NULL)
1324 return 0;
1325 /* PyObject_New is inlined */
1326 v = free_list;
1327 free_list = (PyIntObject *)Py_TYPE(v);
1328 PyObject_INIT(v, &PyInt_Type);
1329 v->ob_ival = ival;
1330 small_ints[ival + NSMALLNEGINTS] = v;
1332 #endif
1333 return 1;
1337 PyInt_ClearFreeList(void)
1339 PyIntObject *p;
1340 PyIntBlock *list, *next;
1341 int i;
1342 int u; /* remaining unfreed ints per block */
1343 int freelist_size = 0;
1345 list = block_list;
1346 block_list = NULL;
1347 free_list = NULL;
1348 while (list != NULL) {
1349 u = 0;
1350 for (i = 0, p = &list->objects[0];
1351 i < N_INTOBJECTS;
1352 i++, p++) {
1353 if (PyInt_CheckExact(p) && p->ob_refcnt != 0)
1354 u++;
1356 next = list->next;
1357 if (u) {
1358 list->next = block_list;
1359 block_list = list;
1360 for (i = 0, p = &list->objects[0];
1361 i < N_INTOBJECTS;
1362 i++, p++) {
1363 if (!PyInt_CheckExact(p) ||
1364 p->ob_refcnt == 0) {
1365 Py_TYPE(p) = (struct _typeobject *)
1366 free_list;
1367 free_list = p;
1369 #if NSMALLNEGINTS + NSMALLPOSINTS > 0
1370 else if (-NSMALLNEGINTS <= p->ob_ival &&
1371 p->ob_ival < NSMALLPOSINTS &&
1372 small_ints[p->ob_ival +
1373 NSMALLNEGINTS] == NULL) {
1374 Py_INCREF(p);
1375 small_ints[p->ob_ival +
1376 NSMALLNEGINTS] = p;
1378 #endif
1381 else {
1382 PyMem_FREE(list);
1384 freelist_size += u;
1385 list = next;
1388 return freelist_size;
1391 void
1392 PyInt_Fini(void)
1394 PyIntObject *p;
1395 PyIntBlock *list;
1396 int i;
1397 int u; /* total unfreed ints per block */
1399 #if NSMALLNEGINTS + NSMALLPOSINTS > 0
1400 PyIntObject **q;
1402 i = NSMALLNEGINTS + NSMALLPOSINTS;
1403 q = small_ints;
1404 while (--i >= 0) {
1405 Py_XDECREF(*q);
1406 *q++ = NULL;
1408 #endif
1409 u = PyInt_ClearFreeList();
1410 if (!Py_VerboseFlag)
1411 return;
1412 fprintf(stderr, "# cleanup ints");
1413 if (!u) {
1414 fprintf(stderr, "\n");
1416 else {
1417 fprintf(stderr,
1418 ": %d unfreed int%s\n",
1419 u, u == 1 ? "" : "s");
1421 if (Py_VerboseFlag > 1) {
1422 list = block_list;
1423 while (list != NULL) {
1424 for (i = 0, p = &list->objects[0];
1425 i < N_INTOBJECTS;
1426 i++, p++) {
1427 if (PyInt_CheckExact(p) && p->ob_refcnt != 0)
1428 /* XXX(twouters) cast refcount to
1429 long until %zd is universally
1430 available
1432 fprintf(stderr,
1433 "# <int at %p, refcnt=%ld, val=%ld>\n",
1434 p, (long)p->ob_refcnt,
1435 p->ob_ival);
1437 list = list->next;