move sections
[python/dscho.git] / Lib / test / test_long.py
blob412ec7c52517c96b5dfdb183d30c7f464be261dc
1 import unittest
2 from test import test_support
3 import sys
5 import random
6 import math
8 # Used for lazy formatting of failure messages
9 class Frm(object):
10 def __init__(self, format, *args):
11 self.format = format
12 self.args = args
14 def __str__(self):
15 return self.format % self.args
17 # SHIFT should match the value in longintrepr.h for best testing.
18 SHIFT = sys.long_info.bits_per_digit
19 BASE = 2 ** SHIFT
20 MASK = BASE - 1
21 KARATSUBA_CUTOFF = 70 # from longobject.c
23 # Max number of base BASE digits to use in test cases. Doubling
24 # this will more than double the runtime.
25 MAXDIGITS = 15
27 # build some special values
28 special = map(long, [0, 1, 2, BASE, BASE >> 1])
29 special.append(0x5555555555555555L)
30 special.append(0xaaaaaaaaaaaaaaaaL)
31 # some solid strings of one bits
32 p2 = 4L # 0 and 1 already added
33 for i in range(2*SHIFT):
34 special.append(p2 - 1)
35 p2 = p2 << 1
36 del p2
37 # add complements & negations
38 special = special + map(lambda x: ~x, special) + \
39 map(lambda x: -x, special)
41 L = [
42 ('0', 0),
43 ('1', 1),
44 ('9', 9),
45 ('10', 10),
46 ('99', 99),
47 ('100', 100),
48 ('314', 314),
49 (' 314', 314),
50 ('314 ', 314),
51 (' \t\t 314 \t\t ', 314),
52 (repr(sys.maxint), sys.maxint),
53 (' 1x', ValueError),
54 (' 1 ', 1),
55 (' 1\02 ', ValueError),
56 ('', ValueError),
57 (' ', ValueError),
58 (' \t\t ', ValueError)
60 if test_support.have_unicode:
61 L += [
62 (unicode('0'), 0),
63 (unicode('1'), 1),
64 (unicode('9'), 9),
65 (unicode('10'), 10),
66 (unicode('99'), 99),
67 (unicode('100'), 100),
68 (unicode('314'), 314),
69 (unicode(' 314'), 314),
70 (unicode('\u0663\u0661\u0664 ','raw-unicode-escape'), 314),
71 (unicode(' \t\t 314 \t\t '), 314),
72 (unicode(' 1x'), ValueError),
73 (unicode(' 1 '), 1),
74 (unicode(' 1\02 '), ValueError),
75 (unicode(''), ValueError),
76 (unicode(' '), ValueError),
77 (unicode(' \t\t '), ValueError),
78 (unichr(0x200), ValueError),
82 class LongTest(unittest.TestCase):
84 # Get quasi-random long consisting of ndigits digits (in base BASE).
85 # quasi == the most-significant digit will not be 0, and the number
86 # is constructed to contain long strings of 0 and 1 bits. These are
87 # more likely than random bits to provoke digit-boundary errors.
88 # The sign of the number is also random.
90 def getran(self, ndigits):
91 self.assertTrue(ndigits > 0)
92 nbits_hi = ndigits * SHIFT
93 nbits_lo = nbits_hi - SHIFT + 1
94 answer = 0L
95 nbits = 0
96 r = int(random.random() * (SHIFT * 2)) | 1 # force 1 bits to start
97 while nbits < nbits_lo:
98 bits = (r >> 1) + 1
99 bits = min(bits, nbits_hi - nbits)
100 self.assertTrue(1 <= bits <= SHIFT)
101 nbits = nbits + bits
102 answer = answer << bits
103 if r & 1:
104 answer = answer | ((1 << bits) - 1)
105 r = int(random.random() * (SHIFT * 2))
106 self.assertTrue(nbits_lo <= nbits <= nbits_hi)
107 if random.random() < 0.5:
108 answer = -answer
109 return answer
111 # Get random long consisting of ndigits random digits (relative to base
112 # BASE). The sign bit is also random.
114 def getran2(ndigits):
115 answer = 0L
116 for i in xrange(ndigits):
117 answer = (answer << SHIFT) | random.randint(0, MASK)
118 if random.random() < 0.5:
119 answer = -answer
120 return answer
122 def check_division(self, x, y):
123 eq = self.assertEqual
124 q, r = divmod(x, y)
125 q2, r2 = x//y, x%y
126 pab, pba = x*y, y*x
127 eq(pab, pba, Frm("multiplication does not commute for %r and %r", x, y))
128 eq(q, q2, Frm("divmod returns different quotient than / for %r and %r", x, y))
129 eq(r, r2, Frm("divmod returns different mod than %% for %r and %r", x, y))
130 eq(x, q*y + r, Frm("x != q*y + r after divmod on x=%r, y=%r", x, y))
131 if y > 0:
132 self.assertTrue(0 <= r < y, Frm("bad mod from divmod on %r and %r", x, y))
133 else:
134 self.assertTrue(y < r <= 0, Frm("bad mod from divmod on %r and %r", x, y))
136 def test_division(self):
137 digits = range(1, MAXDIGITS+1) + range(KARATSUBA_CUTOFF,
138 KARATSUBA_CUTOFF + 14)
139 digits.append(KARATSUBA_CUTOFF * 3)
140 for lenx in digits:
141 x = self.getran(lenx)
142 for leny in digits:
143 y = self.getran(leny) or 1L
144 self.check_division(x, y)
146 # specific numbers chosen to exercise corner cases of the
147 # current long division implementation
149 # 30-bit cases involving a quotient digit estimate of BASE+1
150 self.check_division(1231948412290879395966702881L,
151 1147341367131428698L)
152 self.check_division(815427756481275430342312021515587883L,
153 707270836069027745L)
154 self.check_division(627976073697012820849443363563599041L,
155 643588798496057020L)
156 self.check_division(1115141373653752303710932756325578065L,
157 1038556335171453937726882627L)
158 # 30-bit cases that require the post-subtraction correction step
159 self.check_division(922498905405436751940989320930368494L,
160 949985870686786135626943396L)
161 self.check_division(768235853328091167204009652174031844L,
162 1091555541180371554426545266L)
164 # 15-bit cases involving a quotient digit estimate of BASE+1
165 self.check_division(20172188947443L, 615611397L)
166 self.check_division(1020908530270155025L, 950795710L)
167 self.check_division(128589565723112408L, 736393718L)
168 self.check_division(609919780285761575L, 18613274546784L)
169 # 15-bit cases that require the post-subtraction correction step
170 self.check_division(710031681576388032L, 26769404391308L)
171 self.check_division(1933622614268221L, 30212853348836L)
175 def test_karatsuba(self):
176 digits = range(1, 5) + range(KARATSUBA_CUTOFF, KARATSUBA_CUTOFF + 10)
177 digits.extend([KARATSUBA_CUTOFF * 10, KARATSUBA_CUTOFF * 100])
179 bits = [digit * SHIFT for digit in digits]
181 # Test products of long strings of 1 bits -- (2**x-1)*(2**y-1) ==
182 # 2**(x+y) - 2**x - 2**y + 1, so the proper result is easy to check.
183 for abits in bits:
184 a = (1L << abits) - 1
185 for bbits in bits:
186 if bbits < abits:
187 continue
188 b = (1L << bbits) - 1
189 x = a * b
190 y = ((1L << (abits + bbits)) -
191 (1L << abits) -
192 (1L << bbits) +
194 self.assertEqual(x, y,
195 Frm("bad result for a*b: a=%r, b=%r, x=%r, y=%r", a, b, x, y))
197 def check_bitop_identities_1(self, x):
198 eq = self.assertEqual
199 eq(x & 0, 0, Frm("x & 0 != 0 for x=%r", x))
200 eq(x | 0, x, Frm("x | 0 != x for x=%r", x))
201 eq(x ^ 0, x, Frm("x ^ 0 != x for x=%r", x))
202 eq(x & -1, x, Frm("x & -1 != x for x=%r", x))
203 eq(x | -1, -1, Frm("x | -1 != -1 for x=%r", x))
204 eq(x ^ -1, ~x, Frm("x ^ -1 != ~x for x=%r", x))
205 eq(x, ~~x, Frm("x != ~~x for x=%r", x))
206 eq(x & x, x, Frm("x & x != x for x=%r", x))
207 eq(x | x, x, Frm("x | x != x for x=%r", x))
208 eq(x ^ x, 0, Frm("x ^ x != 0 for x=%r", x))
209 eq(x & ~x, 0, Frm("x & ~x != 0 for x=%r", x))
210 eq(x | ~x, -1, Frm("x | ~x != -1 for x=%r", x))
211 eq(x ^ ~x, -1, Frm("x ^ ~x != -1 for x=%r", x))
212 eq(-x, 1 + ~x, Frm("not -x == 1 + ~x for x=%r", x))
213 eq(-x, ~(x-1), Frm("not -x == ~(x-1) forx =%r", x))
214 for n in xrange(2*SHIFT):
215 p2 = 2L ** n
216 eq(x << n >> n, x,
217 Frm("x << n >> n != x for x=%r, n=%r", (x, n)))
218 eq(x // p2, x >> n,
219 Frm("x // p2 != x >> n for x=%r n=%r p2=%r", (x, n, p2)))
220 eq(x * p2, x << n,
221 Frm("x * p2 != x << n for x=%r n=%r p2=%r", (x, n, p2)))
222 eq(x & -p2, x >> n << n,
223 Frm("not x & -p2 == x >> n << n for x=%r n=%r p2=%r", (x, n, p2)))
224 eq(x & -p2, x & ~(p2 - 1),
225 Frm("not x & -p2 == x & ~(p2 - 1) for x=%r n=%r p2=%r", (x, n, p2)))
227 def check_bitop_identities_2(self, x, y):
228 eq = self.assertEqual
229 eq(x & y, y & x, Frm("x & y != y & x for x=%r, y=%r", (x, y)))
230 eq(x | y, y | x, Frm("x | y != y | x for x=%r, y=%r", (x, y)))
231 eq(x ^ y, y ^ x, Frm("x ^ y != y ^ x for x=%r, y=%r", (x, y)))
232 eq(x ^ y ^ x, y, Frm("x ^ y ^ x != y for x=%r, y=%r", (x, y)))
233 eq(x & y, ~(~x | ~y), Frm("x & y != ~(~x | ~y) for x=%r, y=%r", (x, y)))
234 eq(x | y, ~(~x & ~y), Frm("x | y != ~(~x & ~y) for x=%r, y=%r", (x, y)))
235 eq(x ^ y, (x | y) & ~(x & y),
236 Frm("x ^ y != (x | y) & ~(x & y) for x=%r, y=%r", (x, y)))
237 eq(x ^ y, (x & ~y) | (~x & y),
238 Frm("x ^ y == (x & ~y) | (~x & y) for x=%r, y=%r", (x, y)))
239 eq(x ^ y, (x | y) & (~x | ~y),
240 Frm("x ^ y == (x | y) & (~x | ~y) for x=%r, y=%r", (x, y)))
242 def check_bitop_identities_3(self, x, y, z):
243 eq = self.assertEqual
244 eq((x & y) & z, x & (y & z),
245 Frm("(x & y) & z != x & (y & z) for x=%r, y=%r, z=%r", (x, y, z)))
246 eq((x | y) | z, x | (y | z),
247 Frm("(x | y) | z != x | (y | z) for x=%r, y=%r, z=%r", (x, y, z)))
248 eq((x ^ y) ^ z, x ^ (y ^ z),
249 Frm("(x ^ y) ^ z != x ^ (y ^ z) for x=%r, y=%r, z=%r", (x, y, z)))
250 eq(x & (y | z), (x & y) | (x & z),
251 Frm("x & (y | z) != (x & y) | (x & z) for x=%r, y=%r, z=%r", (x, y, z)))
252 eq(x | (y & z), (x | y) & (x | z),
253 Frm("x | (y & z) != (x | y) & (x | z) for x=%r, y=%r, z=%r", (x, y, z)))
255 def test_bitop_identities(self):
256 for x in special:
257 self.check_bitop_identities_1(x)
258 digits = xrange(1, MAXDIGITS+1)
259 for lenx in digits:
260 x = self.getran(lenx)
261 self.check_bitop_identities_1(x)
262 for leny in digits:
263 y = self.getran(leny)
264 self.check_bitop_identities_2(x, y)
265 self.check_bitop_identities_3(x, y, self.getran((lenx + leny)//2))
267 def slow_format(self, x, base):
268 if (x, base) == (0, 8):
269 # this is an oddball!
270 return "0L"
271 digits = []
272 sign = 0
273 if x < 0:
274 sign, x = 1, -x
275 while x:
276 x, r = divmod(x, base)
277 digits.append(int(r))
278 digits.reverse()
279 digits = digits or [0]
280 return '-'[:sign] + \
281 {8: '0', 10: '', 16: '0x'}[base] + \
282 "".join(map(lambda i: "0123456789abcdef"[i], digits)) + "L"
284 def check_format_1(self, x):
285 for base, mapper in (8, oct), (10, repr), (16, hex):
286 got = mapper(x)
287 expected = self.slow_format(x, base)
288 msg = Frm("%s returned %r but expected %r for %r",
289 mapper.__name__, got, expected, x)
290 self.assertEqual(got, expected, msg)
291 self.assertEqual(long(got, 0), x, Frm('long("%s", 0) != %r', got, x))
292 # str() has to be checked a little differently since there's no
293 # trailing "L"
294 got = str(x)
295 expected = self.slow_format(x, 10)[:-1]
296 msg = Frm("%s returned %r but expected %r for %r",
297 mapper.__name__, got, expected, x)
298 self.assertEqual(got, expected, msg)
300 def test_format(self):
301 for x in special:
302 self.check_format_1(x)
303 for i in xrange(10):
304 for lenx in xrange(1, MAXDIGITS+1):
305 x = self.getran(lenx)
306 self.check_format_1(x)
308 def test_long(self):
309 self.assertEqual(long(314), 314L)
310 self.assertEqual(long(3.14), 3L)
311 self.assertEqual(long(314L), 314L)
312 # Check that long() of basic types actually returns a long
313 self.assertEqual(type(long(314)), long)
314 self.assertEqual(type(long(3.14)), long)
315 self.assertEqual(type(long(314L)), long)
316 # Check that conversion from float truncates towards zero
317 self.assertEqual(long(-3.14), -3L)
318 self.assertEqual(long(3.9), 3L)
319 self.assertEqual(long(-3.9), -3L)
320 self.assertEqual(long(3.5), 3L)
321 self.assertEqual(long(-3.5), -3L)
322 self.assertEqual(long("-3"), -3L)
323 self.assertEqual(long("0b10", 2), 2L)
324 self.assertEqual(long("0o10", 8), 8L)
325 self.assertEqual(long("0x10", 16), 16L)
326 if test_support.have_unicode:
327 self.assertEqual(long(unicode("-3")), -3L)
328 # Different base:
329 self.assertEqual(long("10",16), 16L)
330 if test_support.have_unicode:
331 self.assertEqual(long(unicode("10"),16), 16L)
332 # Check conversions from string (same test set as for int(), and then some)
333 LL = [
334 ('1' + '0'*20, 10L**20),
335 ('1' + '0'*100, 10L**100)
337 L2 = L[:]
338 if test_support.have_unicode:
339 L2 += [
340 (unicode('1') + unicode('0')*20, 10L**20),
341 (unicode('1') + unicode('0')*100, 10L**100),
343 for s, v in L2 + LL:
344 for sign in "", "+", "-":
345 for prefix in "", " ", "\t", " \t\t ":
346 ss = prefix + sign + s
347 vv = v
348 if sign == "-" and v is not ValueError:
349 vv = -v
350 try:
351 self.assertEqual(long(ss), long(vv))
352 except v:
353 pass
355 self.assertRaises(ValueError, long, '123\0')
356 self.assertRaises(ValueError, long, '53', 40)
357 self.assertRaises(TypeError, long, 1, 12)
359 # tests with base 0
360 self.assertEqual(long(' 0123 ', 0), 83)
361 self.assertEqual(long(' 0123 ', 0), 83)
362 self.assertEqual(long('000', 0), 0)
363 self.assertEqual(long('0o123', 0), 83)
364 self.assertEqual(long('0x123', 0), 291)
365 self.assertEqual(long('0b100', 0), 4)
366 self.assertEqual(long(' 0O123 ', 0), 83)
367 self.assertEqual(long(' 0X123 ', 0), 291)
368 self.assertEqual(long(' 0B100 ', 0), 4)
369 self.assertEqual(long('0', 0), 0)
370 self.assertEqual(long('+0', 0), 0)
371 self.assertEqual(long('-0', 0), 0)
372 self.assertEqual(long('00', 0), 0)
373 self.assertRaises(ValueError, long, '08', 0)
374 self.assertRaises(ValueError, long, '-012395', 0)
376 # SF patch #1638879: embedded NULs were not detected with
377 # explicit base
378 self.assertRaises(ValueError, long, '123\0', 10)
379 self.assertRaises(ValueError, long, '123\x00 245', 20)
381 self.assertEqual(long('100000000000000000000000000000000', 2),
382 4294967296)
383 self.assertEqual(long('102002022201221111211', 3), 4294967296)
384 self.assertEqual(long('10000000000000000', 4), 4294967296)
385 self.assertEqual(long('32244002423141', 5), 4294967296)
386 self.assertEqual(long('1550104015504', 6), 4294967296)
387 self.assertEqual(long('211301422354', 7), 4294967296)
388 self.assertEqual(long('40000000000', 8), 4294967296)
389 self.assertEqual(long('12068657454', 9), 4294967296)
390 self.assertEqual(long('4294967296', 10), 4294967296)
391 self.assertEqual(long('1904440554', 11), 4294967296)
392 self.assertEqual(long('9ba461594', 12), 4294967296)
393 self.assertEqual(long('535a79889', 13), 4294967296)
394 self.assertEqual(long('2ca5b7464', 14), 4294967296)
395 self.assertEqual(long('1a20dcd81', 15), 4294967296)
396 self.assertEqual(long('100000000', 16), 4294967296)
397 self.assertEqual(long('a7ffda91', 17), 4294967296)
398 self.assertEqual(long('704he7g4', 18), 4294967296)
399 self.assertEqual(long('4f5aff66', 19), 4294967296)
400 self.assertEqual(long('3723ai4g', 20), 4294967296)
401 self.assertEqual(long('281d55i4', 21), 4294967296)
402 self.assertEqual(long('1fj8b184', 22), 4294967296)
403 self.assertEqual(long('1606k7ic', 23), 4294967296)
404 self.assertEqual(long('mb994ag', 24), 4294967296)
405 self.assertEqual(long('hek2mgl', 25), 4294967296)
406 self.assertEqual(long('dnchbnm', 26), 4294967296)
407 self.assertEqual(long('b28jpdm', 27), 4294967296)
408 self.assertEqual(long('8pfgih4', 28), 4294967296)
409 self.assertEqual(long('76beigg', 29), 4294967296)
410 self.assertEqual(long('5qmcpqg', 30), 4294967296)
411 self.assertEqual(long('4q0jto4', 31), 4294967296)
412 self.assertEqual(long('4000000', 32), 4294967296)
413 self.assertEqual(long('3aokq94', 33), 4294967296)
414 self.assertEqual(long('2qhxjli', 34), 4294967296)
415 self.assertEqual(long('2br45qb', 35), 4294967296)
416 self.assertEqual(long('1z141z4', 36), 4294967296)
418 self.assertEqual(long('100000000000000000000000000000001', 2),
419 4294967297)
420 self.assertEqual(long('102002022201221111212', 3), 4294967297)
421 self.assertEqual(long('10000000000000001', 4), 4294967297)
422 self.assertEqual(long('32244002423142', 5), 4294967297)
423 self.assertEqual(long('1550104015505', 6), 4294967297)
424 self.assertEqual(long('211301422355', 7), 4294967297)
425 self.assertEqual(long('40000000001', 8), 4294967297)
426 self.assertEqual(long('12068657455', 9), 4294967297)
427 self.assertEqual(long('4294967297', 10), 4294967297)
428 self.assertEqual(long('1904440555', 11), 4294967297)
429 self.assertEqual(long('9ba461595', 12), 4294967297)
430 self.assertEqual(long('535a7988a', 13), 4294967297)
431 self.assertEqual(long('2ca5b7465', 14), 4294967297)
432 self.assertEqual(long('1a20dcd82', 15), 4294967297)
433 self.assertEqual(long('100000001', 16), 4294967297)
434 self.assertEqual(long('a7ffda92', 17), 4294967297)
435 self.assertEqual(long('704he7g5', 18), 4294967297)
436 self.assertEqual(long('4f5aff67', 19), 4294967297)
437 self.assertEqual(long('3723ai4h', 20), 4294967297)
438 self.assertEqual(long('281d55i5', 21), 4294967297)
439 self.assertEqual(long('1fj8b185', 22), 4294967297)
440 self.assertEqual(long('1606k7id', 23), 4294967297)
441 self.assertEqual(long('mb994ah', 24), 4294967297)
442 self.assertEqual(long('hek2mgm', 25), 4294967297)
443 self.assertEqual(long('dnchbnn', 26), 4294967297)
444 self.assertEqual(long('b28jpdn', 27), 4294967297)
445 self.assertEqual(long('8pfgih5', 28), 4294967297)
446 self.assertEqual(long('76beigh', 29), 4294967297)
447 self.assertEqual(long('5qmcpqh', 30), 4294967297)
448 self.assertEqual(long('4q0jto5', 31), 4294967297)
449 self.assertEqual(long('4000001', 32), 4294967297)
450 self.assertEqual(long('3aokq95', 33), 4294967297)
451 self.assertEqual(long('2qhxjlj', 34), 4294967297)
452 self.assertEqual(long('2br45qc', 35), 4294967297)
453 self.assertEqual(long('1z141z5', 36), 4294967297)
456 def test_conversion(self):
457 # Test __long__()
458 class ClassicMissingMethods:
459 pass
460 self.assertRaises(AttributeError, long, ClassicMissingMethods())
462 class MissingMethods(object):
463 pass
464 self.assertRaises(TypeError, long, MissingMethods())
466 class Foo0:
467 def __long__(self):
468 return 42L
470 class Foo1(object):
471 def __long__(self):
472 return 42L
474 class Foo2(long):
475 def __long__(self):
476 return 42L
478 class Foo3(long):
479 def __long__(self):
480 return self
482 class Foo4(long):
483 def __long__(self):
484 return 42
486 class Foo5(long):
487 def __long__(self):
488 return 42.
490 self.assertEqual(long(Foo0()), 42L)
491 self.assertEqual(long(Foo1()), 42L)
492 self.assertEqual(long(Foo2()), 42L)
493 self.assertEqual(long(Foo3()), 0)
494 self.assertEqual(long(Foo4()), 42)
495 self.assertRaises(TypeError, long, Foo5())
497 class Classic:
498 pass
499 for base in (object, Classic):
500 class LongOverridesTrunc(base):
501 def __long__(self):
502 return 42
503 def __trunc__(self):
504 return -12
505 self.assertEqual(long(LongOverridesTrunc()), 42)
507 class JustTrunc(base):
508 def __trunc__(self):
509 return 42
510 self.assertEqual(long(JustTrunc()), 42)
512 for trunc_result_base in (object, Classic):
513 class Integral(trunc_result_base):
514 def __int__(self):
515 return 42
517 class TruncReturnsNonLong(base):
518 def __trunc__(self):
519 return Integral()
520 self.assertEqual(long(TruncReturnsNonLong()), 42)
522 class NonIntegral(trunc_result_base):
523 def __trunc__(self):
524 # Check that we avoid infinite recursion.
525 return NonIntegral()
527 class TruncReturnsNonIntegral(base):
528 def __trunc__(self):
529 return NonIntegral()
530 try:
531 long(TruncReturnsNonIntegral())
532 except TypeError as e:
533 self.assertEquals(str(e),
534 "__trunc__ returned non-Integral"
535 " (type NonIntegral)")
536 else:
537 self.fail("Failed to raise TypeError with %s" %
538 ((base, trunc_result_base),))
540 def test_misc(self):
542 # check the extremes in int<->long conversion
543 hugepos = sys.maxint
544 hugeneg = -hugepos - 1
545 hugepos_aslong = long(hugepos)
546 hugeneg_aslong = long(hugeneg)
547 self.assertEqual(hugepos, hugepos_aslong, "long(sys.maxint) != sys.maxint")
548 self.assertEqual(hugeneg, hugeneg_aslong,
549 "long(-sys.maxint-1) != -sys.maxint-1")
551 # long -> int should not fail for hugepos_aslong or hugeneg_aslong
552 x = int(hugepos_aslong)
553 try:
554 self.assertEqual(x, hugepos,
555 "converting sys.maxint to long and back to int fails")
556 except OverflowError:
557 self.fail("int(long(sys.maxint)) overflowed!")
558 if not isinstance(x, int):
559 self.fail("int(long(sys.maxint)) should have returned int")
560 x = int(hugeneg_aslong)
561 try:
562 self.assertEqual(x, hugeneg,
563 "converting -sys.maxint-1 to long and back to int fails")
564 except OverflowError:
565 self.fail("int(long(-sys.maxint-1)) overflowed!")
566 if not isinstance(x, int):
567 self.fail("int(long(-sys.maxint-1)) should have returned int")
568 # but long -> int should overflow for hugepos+1 and hugeneg-1
569 x = hugepos_aslong + 1
570 try:
571 y = int(x)
572 except OverflowError:
573 self.fail("int(long(sys.maxint) + 1) mustn't overflow")
574 self.assertIsInstance(y, long,
575 "int(long(sys.maxint) + 1) should have returned long")
577 x = hugeneg_aslong - 1
578 try:
579 y = int(x)
580 except OverflowError:
581 self.fail("int(long(-sys.maxint-1) - 1) mustn't overflow")
582 self.assertIsInstance(y, long,
583 "int(long(-sys.maxint-1) - 1) should have returned long")
585 class long2(long):
586 pass
587 x = long2(1L<<100)
588 y = int(x)
589 self.assertTrue(type(y) is long,
590 "overflowing int conversion must return long not long subtype")
592 # long -> Py_ssize_t conversion
593 class X(object):
594 def __getslice__(self, i, j):
595 return i, j
597 with test_support.check_py3k_warnings():
598 self.assertEqual(X()[-5L:7L], (-5, 7))
599 # use the clamping effect to test the smallest and largest longs
600 # that fit a Py_ssize_t
601 slicemin, slicemax = X()[-2L**100:2L**100]
602 self.assertEqual(X()[slicemin:slicemax], (slicemin, slicemax))
604 # ----------------------------------- tests of auto int->long conversion
606 def test_auto_overflow(self):
607 special = [0, 1, 2, 3, sys.maxint-1, sys.maxint, sys.maxint+1]
608 sqrt = int(math.sqrt(sys.maxint))
609 special.extend([sqrt-1, sqrt, sqrt+1])
610 special.extend([-i for i in special])
612 def checkit(*args):
613 # Heavy use of nested scopes here!
614 self.assertEqual(got, expected,
615 Frm("for %r expected %r got %r", args, expected, got))
617 for x in special:
618 longx = long(x)
620 expected = -longx
621 got = -x
622 checkit('-', x)
624 for y in special:
625 longy = long(y)
627 expected = longx + longy
628 got = x + y
629 checkit(x, '+', y)
631 expected = longx - longy
632 got = x - y
633 checkit(x, '-', y)
635 expected = longx * longy
636 got = x * y
637 checkit(x, '*', y)
639 if y:
640 with test_support.check_py3k_warnings():
641 expected = longx / longy
642 got = x / y
643 checkit(x, '/', y)
645 expected = longx // longy
646 got = x // y
647 checkit(x, '//', y)
649 expected = divmod(longx, longy)
650 got = divmod(longx, longy)
651 checkit(x, 'divmod', y)
653 if abs(y) < 5 and not (x == 0 and y < 0):
654 expected = longx ** longy
655 got = x ** y
656 checkit(x, '**', y)
658 for z in special:
659 if z != 0 :
660 if y >= 0:
661 expected = pow(longx, longy, long(z))
662 got = pow(x, y, z)
663 checkit('pow', x, y, '%', z)
664 else:
665 self.assertRaises(TypeError, pow,longx, longy, long(z))
667 @unittest.skipUnless(float.__getformat__("double").startswith("IEEE"),
668 "test requires IEEE 754 doubles")
669 def test_float_conversion(self):
670 import sys
671 DBL_MAX = sys.float_info.max
672 DBL_MAX_EXP = sys.float_info.max_exp
673 DBL_MANT_DIG = sys.float_info.mant_dig
675 exact_values = [0L, 1L, 2L,
676 long(2**53-3),
677 long(2**53-2),
678 long(2**53-1),
679 long(2**53),
680 long(2**53+2),
681 long(2**54-4),
682 long(2**54-2),
683 long(2**54),
684 long(2**54+4)]
685 for x in exact_values:
686 self.assertEqual(long(float(x)), x)
687 self.assertEqual(long(float(-x)), -x)
689 # test round-half-even
690 for x, y in [(1, 0), (2, 2), (3, 4), (4, 4), (5, 4), (6, 6), (7, 8)]:
691 for p in xrange(15):
692 self.assertEqual(long(float(2L**p*(2**53+x))), 2L**p*(2**53+y))
694 for x, y in [(0, 0), (1, 0), (2, 0), (3, 4), (4, 4), (5, 4), (6, 8),
695 (7, 8), (8, 8), (9, 8), (10, 8), (11, 12), (12, 12),
696 (13, 12), (14, 16), (15, 16)]:
697 for p in xrange(15):
698 self.assertEqual(long(float(2L**p*(2**54+x))), 2L**p*(2**54+y))
700 # behaviour near extremes of floating-point range
701 long_dbl_max = long(DBL_MAX)
702 top_power = 2**DBL_MAX_EXP
703 halfway = (long_dbl_max + top_power)//2
704 self.assertEqual(float(long_dbl_max), DBL_MAX)
705 self.assertEqual(float(long_dbl_max+1), DBL_MAX)
706 self.assertEqual(float(halfway-1), DBL_MAX)
707 self.assertRaises(OverflowError, float, halfway)
708 self.assertEqual(float(1-halfway), -DBL_MAX)
709 self.assertRaises(OverflowError, float, -halfway)
710 self.assertRaises(OverflowError, float, top_power-1)
711 self.assertRaises(OverflowError, float, top_power)
712 self.assertRaises(OverflowError, float, top_power+1)
713 self.assertRaises(OverflowError, float, 2*top_power-1)
714 self.assertRaises(OverflowError, float, 2*top_power)
715 self.assertRaises(OverflowError, float, top_power*top_power)
717 for p in xrange(100):
718 x = long(2**p * (2**53 + 1) + 1)
719 y = long(2**p * (2**53+ 2))
720 self.assertEqual(long(float(x)), y)
722 x = long(2**p * (2**53 + 1))
723 y = long(2**p * 2**53)
724 self.assertEqual(long(float(x)), y)
726 def test_float_overflow(self):
727 for x in -2.0, -1.0, 0.0, 1.0, 2.0:
728 self.assertEqual(float(long(x)), x)
730 shuge = '12345' * 120
731 huge = 1L << 30000
732 mhuge = -huge
733 namespace = {'huge': huge, 'mhuge': mhuge, 'shuge': shuge, 'math': math}
734 for test in ["float(huge)", "float(mhuge)",
735 "complex(huge)", "complex(mhuge)",
736 "complex(huge, 1)", "complex(mhuge, 1)",
737 "complex(1, huge)", "complex(1, mhuge)",
738 "1. + huge", "huge + 1.", "1. + mhuge", "mhuge + 1.",
739 "1. - huge", "huge - 1.", "1. - mhuge", "mhuge - 1.",
740 "1. * huge", "huge * 1.", "1. * mhuge", "mhuge * 1.",
741 "1. // huge", "huge // 1.", "1. // mhuge", "mhuge // 1.",
742 "1. / huge", "huge / 1.", "1. / mhuge", "mhuge / 1.",
743 "1. ** huge", "huge ** 1.", "1. ** mhuge", "mhuge ** 1.",
744 "math.sin(huge)", "math.sin(mhuge)",
745 "math.sqrt(huge)", "math.sqrt(mhuge)", # should do better
746 "math.floor(huge)", "math.floor(mhuge)"]:
748 self.assertRaises(OverflowError, eval, test, namespace)
750 # XXX Perhaps float(shuge) can raise OverflowError on some box?
751 # The comparison should not.
752 self.assertNotEqual(float(shuge), int(shuge),
753 "float(shuge) should not equal int(shuge)")
755 def test_logs(self):
756 LOG10E = math.log10(math.e)
758 for exp in range(10) + [100, 1000, 10000]:
759 value = 10 ** exp
760 log10 = math.log10(value)
761 self.assertAlmostEqual(log10, exp)
763 # log10(value) == exp, so log(value) == log10(value)/log10(e) ==
764 # exp/LOG10E
765 expected = exp / LOG10E
766 log = math.log(value)
767 self.assertAlmostEqual(log, expected)
769 for bad in -(1L << 10000), -2L, 0L:
770 self.assertRaises(ValueError, math.log, bad)
771 self.assertRaises(ValueError, math.log10, bad)
773 def test_mixed_compares(self):
774 eq = self.assertEqual
776 # We're mostly concerned with that mixing floats and longs does the
777 # right stuff, even when longs are too large to fit in a float.
778 # The safest way to check the results is to use an entirely different
779 # method, which we do here via a skeletal rational class (which
780 # represents all Python ints, longs and floats exactly).
781 class Rat:
782 def __init__(self, value):
783 if isinstance(value, (int, long)):
784 self.n = value
785 self.d = 1
786 elif isinstance(value, float):
787 # Convert to exact rational equivalent.
788 f, e = math.frexp(abs(value))
789 assert f == 0 or 0.5 <= f < 1.0
790 # |value| = f * 2**e exactly
792 # Suck up CHUNK bits at a time; 28 is enough so that we suck
793 # up all bits in 2 iterations for all known binary double-
794 # precision formats, and small enough to fit in an int.
795 CHUNK = 28
796 top = 0
797 # invariant: |value| = (top + f) * 2**e exactly
798 while f:
799 f = math.ldexp(f, CHUNK)
800 digit = int(f)
801 assert digit >> CHUNK == 0
802 top = (top << CHUNK) | digit
803 f -= digit
804 assert 0.0 <= f < 1.0
805 e -= CHUNK
807 # Now |value| = top * 2**e exactly.
808 if e >= 0:
809 n = top << e
810 d = 1
811 else:
812 n = top
813 d = 1 << -e
814 if value < 0:
815 n = -n
816 self.n = n
817 self.d = d
818 assert float(n) / float(d) == value
819 else:
820 raise TypeError("can't deal with %r" % value)
822 def __cmp__(self, other):
823 if not isinstance(other, Rat):
824 other = Rat(other)
825 return cmp(self.n * other.d, self.d * other.n)
827 cases = [0, 0.001, 0.99, 1.0, 1.5, 1e20, 1e200]
828 # 2**48 is an important boundary in the internals. 2**53 is an
829 # important boundary for IEEE double precision.
830 for t in 2.0**48, 2.0**50, 2.0**53:
831 cases.extend([t - 1.0, t - 0.3, t, t + 0.3, t + 1.0,
832 long(t-1), long(t), long(t+1)])
833 cases.extend([0, 1, 2, sys.maxint, float(sys.maxint)])
834 # 1L<<20000 should exceed all double formats. long(1e200) is to
835 # check that we get equality with 1e200 above.
836 t = long(1e200)
837 cases.extend([0L, 1L, 2L, 1L << 20000, t-1, t, t+1])
838 cases.extend([-x for x in cases])
839 for x in cases:
840 Rx = Rat(x)
841 for y in cases:
842 Ry = Rat(y)
843 Rcmp = cmp(Rx, Ry)
844 xycmp = cmp(x, y)
845 eq(Rcmp, xycmp, Frm("%r %r %d %d", x, y, Rcmp, xycmp))
846 eq(x == y, Rcmp == 0, Frm("%r == %r %d", x, y, Rcmp))
847 eq(x != y, Rcmp != 0, Frm("%r != %r %d", x, y, Rcmp))
848 eq(x < y, Rcmp < 0, Frm("%r < %r %d", x, y, Rcmp))
849 eq(x <= y, Rcmp <= 0, Frm("%r <= %r %d", x, y, Rcmp))
850 eq(x > y, Rcmp > 0, Frm("%r > %r %d", x, y, Rcmp))
851 eq(x >= y, Rcmp >= 0, Frm("%r >= %r %d", x, y, Rcmp))
853 def test_nan_inf(self):
854 self.assertRaises(OverflowError, long, float('inf'))
855 self.assertRaises(OverflowError, long, float('-inf'))
856 self.assertRaises(ValueError, long, float('nan'))
858 def test_bit_length(self):
859 tiny = 1e-10
860 for x in xrange(-65000, 65000):
861 x = long(x)
862 k = x.bit_length()
863 # Check equivalence with Python version
864 self.assertEqual(k, len(bin(x).lstrip('-0b')))
865 # Behaviour as specified in the docs
866 if x != 0:
867 self.assertTrue(2**(k-1) <= abs(x) < 2**k)
868 else:
869 self.assertEqual(k, 0)
870 # Alternative definition: x.bit_length() == 1 + floor(log_2(x))
871 if x != 0:
872 # When x is an exact power of 2, numeric errors can
873 # cause floor(log(x)/log(2)) to be one too small; for
874 # small x this can be fixed by adding a small quantity
875 # to the quotient before taking the floor.
876 self.assertEqual(k, 1 + math.floor(
877 math.log(abs(x))/math.log(2) + tiny))
879 self.assertEqual((0L).bit_length(), 0)
880 self.assertEqual((1L).bit_length(), 1)
881 self.assertEqual((-1L).bit_length(), 1)
882 self.assertEqual((2L).bit_length(), 2)
883 self.assertEqual((-2L).bit_length(), 2)
884 for i in [2, 3, 15, 16, 17, 31, 32, 33, 63, 64, 234]:
885 a = 2L**i
886 self.assertEqual((a-1).bit_length(), i)
887 self.assertEqual((1-a).bit_length(), i)
888 self.assertEqual((a).bit_length(), i+1)
889 self.assertEqual((-a).bit_length(), i+1)
890 self.assertEqual((a+1).bit_length(), i+1)
891 self.assertEqual((-a-1).bit_length(), i+1)
894 def test_main():
895 test_support.run_unittest(LongTest)
897 if __name__ == "__main__":
898 test_main()