mirror of
https://github.com/python/cpython.git
synced 2024-11-25 19:03:49 +08:00
d50ade68ec
pack_float, pack_double, save_float: All the routines for creating IEEE-format packed representations of floats and doubles simply ignored that rounding can (in rare cases) propagate out of a long string of 1 bits. At worst, the end-off carry can (by mistake) interfere with the exponent value, and then unpacking yields a result wrong by a factor of 2. In less severe cases, it can end up losing more low-order bits than intended, or fail to catch overflow *caused* by rounding. Bugfix candidate, but I already backported this to 2.2. In 2.3, this code remains in severe need of refactoring.
439 lines
15 KiB
Python
439 lines
15 KiB
Python
from test.test_support import TestFailed, verbose, verify
|
|
import struct
|
|
|
|
import sys
|
|
ISBIGENDIAN = sys.byteorder == "big"
|
|
del sys
|
|
verify((struct.pack('=i', 1)[0] == chr(0)) == ISBIGENDIAN,
|
|
"bigendian determination appears wrong")
|
|
|
|
def string_reverse(s):
|
|
chars = list(s)
|
|
chars.reverse()
|
|
return "".join(chars)
|
|
|
|
def bigendian_to_native(value):
|
|
if ISBIGENDIAN:
|
|
return value
|
|
else:
|
|
return string_reverse(value)
|
|
|
|
def simple_err(func, *args):
|
|
try:
|
|
func(*args)
|
|
except struct.error:
|
|
pass
|
|
else:
|
|
raise TestFailed, "%s%s did not raise struct.error" % (
|
|
func.__name__, args)
|
|
|
|
def any_err(func, *args):
|
|
try:
|
|
func(*args)
|
|
except (struct.error, OverflowError, TypeError):
|
|
pass
|
|
else:
|
|
raise TestFailed, "%s%s did not raise error" % (
|
|
func.__name__, args)
|
|
|
|
|
|
simple_err(struct.calcsize, 'Z')
|
|
|
|
sz = struct.calcsize('i')
|
|
if sz * 3 != struct.calcsize('iii'):
|
|
raise TestFailed, 'inconsistent sizes'
|
|
|
|
fmt = 'cbxxxxxxhhhhiillffd'
|
|
fmt3 = '3c3b18x12h6i6l6f3d'
|
|
sz = struct.calcsize(fmt)
|
|
sz3 = struct.calcsize(fmt3)
|
|
if sz * 3 != sz3:
|
|
raise TestFailed, 'inconsistent sizes (3*%s -> 3*%d = %d, %s -> %d)' % (
|
|
`fmt`, sz, 3*sz, `fmt3`, sz3)
|
|
|
|
simple_err(struct.pack, 'iii', 3)
|
|
simple_err(struct.pack, 'i', 3, 3, 3)
|
|
simple_err(struct.pack, 'i', 'foo')
|
|
simple_err(struct.unpack, 'd', 'flap')
|
|
s = struct.pack('ii', 1, 2)
|
|
simple_err(struct.unpack, 'iii', s)
|
|
simple_err(struct.unpack, 'i', s)
|
|
|
|
c = 'a'
|
|
b = 1
|
|
h = 255
|
|
i = 65535
|
|
l = 65536
|
|
f = 3.1415
|
|
d = 3.1415
|
|
|
|
for prefix in ('', '@', '<', '>', '=', '!'):
|
|
for format in ('xcbhilfd', 'xcBHILfd'):
|
|
format = prefix + format
|
|
if verbose:
|
|
print "trying:", format
|
|
s = struct.pack(format, c, b, h, i, l, f, d)
|
|
cp, bp, hp, ip, lp, fp, dp = struct.unpack(format, s)
|
|
if (cp != c or bp != b or hp != h or ip != i or lp != l or
|
|
int(100 * fp) != int(100 * f) or int(100 * dp) != int(100 * d)):
|
|
# ^^^ calculate only to two decimal places
|
|
raise TestFailed, "unpack/pack not transitive (%s, %s)" % (
|
|
str(format), str((cp, bp, hp, ip, lp, fp, dp)))
|
|
|
|
# Test some of the new features in detail
|
|
|
|
# (format, argument, big-endian result, little-endian result, asymmetric)
|
|
tests = [
|
|
('c', 'a', 'a', 'a', 0),
|
|
('xc', 'a', '\0a', '\0a', 0),
|
|
('cx', 'a', 'a\0', 'a\0', 0),
|
|
('s', 'a', 'a', 'a', 0),
|
|
('0s', 'helloworld', '', '', 1),
|
|
('1s', 'helloworld', 'h', 'h', 1),
|
|
('9s', 'helloworld', 'helloworl', 'helloworl', 1),
|
|
('10s', 'helloworld', 'helloworld', 'helloworld', 0),
|
|
('11s', 'helloworld', 'helloworld\0', 'helloworld\0', 1),
|
|
('20s', 'helloworld', 'helloworld'+10*'\0', 'helloworld'+10*'\0', 1),
|
|
('b', 7, '\7', '\7', 0),
|
|
('b', -7, '\371', '\371', 0),
|
|
('B', 7, '\7', '\7', 0),
|
|
('B', 249, '\371', '\371', 0),
|
|
('h', 700, '\002\274', '\274\002', 0),
|
|
('h', -700, '\375D', 'D\375', 0),
|
|
('H', 700, '\002\274', '\274\002', 0),
|
|
('H', 0x10000-700, '\375D', 'D\375', 0),
|
|
('i', 70000000, '\004,\035\200', '\200\035,\004', 0),
|
|
('i', -70000000, '\373\323\342\200', '\200\342\323\373', 0),
|
|
('I', 70000000L, '\004,\035\200', '\200\035,\004', 0),
|
|
('I', 0x100000000L-70000000, '\373\323\342\200', '\200\342\323\373', 0),
|
|
('l', 70000000, '\004,\035\200', '\200\035,\004', 0),
|
|
('l', -70000000, '\373\323\342\200', '\200\342\323\373', 0),
|
|
('L', 70000000L, '\004,\035\200', '\200\035,\004', 0),
|
|
('L', 0x100000000L-70000000, '\373\323\342\200', '\200\342\323\373', 0),
|
|
('f', 2.0, '@\000\000\000', '\000\000\000@', 0),
|
|
('d', 2.0, '@\000\000\000\000\000\000\000',
|
|
'\000\000\000\000\000\000\000@', 0),
|
|
('f', -2.0, '\300\000\000\000', '\000\000\000\300', 0),
|
|
('d', -2.0, '\300\000\000\000\000\000\000\000',
|
|
'\000\000\000\000\000\000\000\300', 0),
|
|
]
|
|
|
|
for fmt, arg, big, lil, asy in tests:
|
|
if verbose:
|
|
print `fmt`, `arg`, `big`, `lil`
|
|
for (xfmt, exp) in [('>'+fmt, big), ('!'+fmt, big), ('<'+fmt, lil),
|
|
('='+fmt, ISBIGENDIAN and big or lil)]:
|
|
res = struct.pack(xfmt, arg)
|
|
if res != exp:
|
|
raise TestFailed, "pack(%s, %s) -> %s # expected %s" % (
|
|
`fmt`, `arg`, `res`, `exp`)
|
|
n = struct.calcsize(xfmt)
|
|
if n != len(res):
|
|
raise TestFailed, "calcsize(%s) -> %d # expected %d" % (
|
|
`xfmt`, n, len(res))
|
|
rev = struct.unpack(xfmt, res)[0]
|
|
if rev != arg and not asy:
|
|
raise TestFailed, "unpack(%s, %s) -> (%s,) # expected (%s,)" % (
|
|
`fmt`, `res`, `rev`, `arg`)
|
|
|
|
###########################################################################
|
|
# Simple native q/Q tests.
|
|
|
|
has_native_qQ = 1
|
|
try:
|
|
struct.pack("q", 5)
|
|
except struct.error:
|
|
has_native_qQ = 0
|
|
|
|
if verbose:
|
|
print "Platform has native q/Q?", has_native_qQ and "Yes." or "No."
|
|
|
|
any_err(struct.pack, "Q", -1) # can't pack -1 as unsigned regardless
|
|
simple_err(struct.pack, "q", "a") # can't pack string as 'q' regardless
|
|
simple_err(struct.pack, "Q", "a") # ditto, but 'Q'
|
|
|
|
def test_native_qQ():
|
|
bytes = struct.calcsize('q')
|
|
# The expected values here are in big-endian format, primarily because
|
|
# I'm on a little-endian machine and so this is the clearest way (for
|
|
# me) to force the code to get exercised.
|
|
for format, input, expected in (
|
|
('q', -1, '\xff' * bytes),
|
|
('q', 0, '\x00' * bytes),
|
|
('Q', 0, '\x00' * bytes),
|
|
('q', 1L, '\x00' * (bytes-1) + '\x01'),
|
|
('Q', (1L << (8*bytes))-1, '\xff' * bytes),
|
|
('q', (1L << (8*bytes-1))-1, '\x7f' + '\xff' * (bytes - 1))):
|
|
got = struct.pack(format, input)
|
|
native_expected = bigendian_to_native(expected)
|
|
verify(got == native_expected,
|
|
"%r-pack of %r gave %r, not %r" %
|
|
(format, input, got, native_expected))
|
|
retrieved = struct.unpack(format, got)[0]
|
|
verify(retrieved == input,
|
|
"%r-unpack of %r gave %r, not %r" %
|
|
(format, got, retrieved, input))
|
|
|
|
if has_native_qQ:
|
|
test_native_qQ()
|
|
|
|
###########################################################################
|
|
# Standard integer tests (bBhHiIlLqQ).
|
|
|
|
import binascii
|
|
|
|
class IntTester:
|
|
|
|
# XXX Most std integer modes fail to test for out-of-range.
|
|
# The "i" and "l" codes appear to range-check OK on 32-bit boxes, but
|
|
# fail to check correctly on some 64-bit ones (Tru64 Unix + Compaq C
|
|
# reported by Mark Favas).
|
|
BUGGY_RANGE_CHECK = "bBhHiIlL"
|
|
|
|
def __init__(self, formatpair, bytesize):
|
|
assert len(formatpair) == 2
|
|
self.formatpair = formatpair
|
|
for direction in "<>!=":
|
|
for code in formatpair:
|
|
format = direction + code
|
|
verify(struct.calcsize(format) == bytesize)
|
|
self.bytesize = bytesize
|
|
self.bitsize = bytesize * 8
|
|
self.signed_code, self.unsigned_code = formatpair
|
|
self.unsigned_min = 0
|
|
self.unsigned_max = 2L**self.bitsize - 1
|
|
self.signed_min = -(2L**(self.bitsize-1))
|
|
self.signed_max = 2L**(self.bitsize-1) - 1
|
|
|
|
def test_one(self, x, pack=struct.pack,
|
|
unpack=struct.unpack,
|
|
unhexlify=binascii.unhexlify):
|
|
if verbose:
|
|
print "trying std", self.formatpair, "on", x, "==", hex(x)
|
|
|
|
# Try signed.
|
|
code = self.signed_code
|
|
if self.signed_min <= x <= self.signed_max:
|
|
# Try big-endian.
|
|
expected = long(x)
|
|
if x < 0:
|
|
expected += 1L << self.bitsize
|
|
assert expected > 0
|
|
expected = hex(expected)[2:-1] # chop "0x" and trailing 'L'
|
|
if len(expected) & 1:
|
|
expected = "0" + expected
|
|
expected = unhexlify(expected)
|
|
expected = "\x00" * (self.bytesize - len(expected)) + expected
|
|
|
|
# Pack work?
|
|
format = ">" + code
|
|
got = pack(format, x)
|
|
verify(got == expected,
|
|
"'%s'-pack of %r gave %r, not %r" %
|
|
(format, x, got, expected))
|
|
|
|
# Unpack work?
|
|
retrieved = unpack(format, got)[0]
|
|
verify(x == retrieved,
|
|
"'%s'-unpack of %r gave %r, not %r" %
|
|
(format, got, retrieved, x))
|
|
|
|
# Adding any byte should cause a "too big" error.
|
|
any_err(unpack, format, '\x01' + got)
|
|
|
|
# Try little-endian.
|
|
format = "<" + code
|
|
expected = string_reverse(expected)
|
|
|
|
# Pack work?
|
|
got = pack(format, x)
|
|
verify(got == expected,
|
|
"'%s'-pack of %r gave %r, not %r" %
|
|
(format, x, got, expected))
|
|
|
|
# Unpack work?
|
|
retrieved = unpack(format, got)[0]
|
|
verify(x == retrieved,
|
|
"'%s'-unpack of %r gave %r, not %r" %
|
|
(format, got, retrieved, x))
|
|
|
|
# Adding any byte should cause a "too big" error.
|
|
any_err(unpack, format, '\x01' + got)
|
|
|
|
else:
|
|
# x is out of range -- verify pack realizes that.
|
|
if code in self.BUGGY_RANGE_CHECK:
|
|
if verbose:
|
|
print "Skipping buggy range check for code", code
|
|
else:
|
|
any_err(pack, ">" + code, x)
|
|
any_err(pack, "<" + code, x)
|
|
|
|
# Much the same for unsigned.
|
|
code = self.unsigned_code
|
|
if self.unsigned_min <= x <= self.unsigned_max:
|
|
# Try big-endian.
|
|
format = ">" + code
|
|
expected = long(x)
|
|
expected = hex(expected)[2:-1] # chop "0x" and trailing 'L'
|
|
if len(expected) & 1:
|
|
expected = "0" + expected
|
|
expected = unhexlify(expected)
|
|
expected = "\x00" * (self.bytesize - len(expected)) + expected
|
|
|
|
# Pack work?
|
|
got = pack(format, x)
|
|
verify(got == expected,
|
|
"'%s'-pack of %r gave %r, not %r" %
|
|
(format, x, got, expected))
|
|
|
|
# Unpack work?
|
|
retrieved = unpack(format, got)[0]
|
|
verify(x == retrieved,
|
|
"'%s'-unpack of %r gave %r, not %r" %
|
|
(format, got, retrieved, x))
|
|
|
|
# Adding any byte should cause a "too big" error.
|
|
any_err(unpack, format, '\x01' + got)
|
|
|
|
# Try little-endian.
|
|
format = "<" + code
|
|
expected = string_reverse(expected)
|
|
|
|
# Pack work?
|
|
got = pack(format, x)
|
|
verify(got == expected,
|
|
"'%s'-pack of %r gave %r, not %r" %
|
|
(format, x, got, expected))
|
|
|
|
# Unpack work?
|
|
retrieved = unpack(format, got)[0]
|
|
verify(x == retrieved,
|
|
"'%s'-unpack of %r gave %r, not %r" %
|
|
(format, got, retrieved, x))
|
|
|
|
# Adding any byte should cause a "too big" error.
|
|
any_err(unpack, format, '\x01' + got)
|
|
|
|
else:
|
|
# x is out of range -- verify pack realizes that.
|
|
if code in self.BUGGY_RANGE_CHECK:
|
|
if verbose:
|
|
print "Skipping buggy range check for code", code
|
|
else:
|
|
any_err(pack, ">" + code, x)
|
|
any_err(pack, "<" + code, x)
|
|
|
|
def run(self):
|
|
from random import randrange
|
|
|
|
# Create all interesting powers of 2.
|
|
values = []
|
|
for exp in range(self.bitsize + 3):
|
|
values.append(1L << exp)
|
|
|
|
# Add some random values.
|
|
for i in range(self.bitsize):
|
|
val = 0L
|
|
for j in range(self.bytesize):
|
|
val = (val << 8) | randrange(256)
|
|
values.append(val)
|
|
|
|
# Try all those, and their negations, and +-1 from them. Note
|
|
# that this tests all power-of-2 boundaries in range, and a few out
|
|
# of range, plus +-(2**n +- 1).
|
|
for base in values:
|
|
for val in -base, base:
|
|
for incr in -1, 0, 1:
|
|
x = val + incr
|
|
try:
|
|
x = int(x)
|
|
except OverflowError:
|
|
pass
|
|
self.test_one(x)
|
|
|
|
# Some error cases.
|
|
for direction in "<>":
|
|
for code in self.formatpair:
|
|
for badobject in "a string", 3+42j, randrange:
|
|
any_err(struct.pack, direction + code, badobject)
|
|
|
|
for args in [("bB", 1),
|
|
("hH", 2),
|
|
("iI", 4),
|
|
("lL", 4),
|
|
("qQ", 8)]:
|
|
t = IntTester(*args)
|
|
t.run()
|
|
|
|
|
|
###########################################################################
|
|
# The p ("Pascal string") code.
|
|
|
|
def test_p_code():
|
|
for code, input, expected, expectedback in [
|
|
('p','abc', '\x00', ''),
|
|
('1p', 'abc', '\x00', ''),
|
|
('2p', 'abc', '\x01a', 'a'),
|
|
('3p', 'abc', '\x02ab', 'ab'),
|
|
('4p', 'abc', '\x03abc', 'abc'),
|
|
('5p', 'abc', '\x03abc\x00', 'abc'),
|
|
('6p', 'abc', '\x03abc\x00\x00', 'abc'),
|
|
('1000p', 'x'*1000, '\xff' + 'x'*999, 'x'*255)]:
|
|
got = struct.pack(code, input)
|
|
if got != expected:
|
|
raise TestFailed("pack(%r, %r) == %r but expected %r" %
|
|
(code, input, got, expected))
|
|
(got,) = struct.unpack(code, got)
|
|
if got != expectedback:
|
|
raise TestFailed("unpack(%r, %r) == %r but expected %r" %
|
|
(code, input, got, expectedback))
|
|
|
|
test_p_code()
|
|
|
|
|
|
###########################################################################
|
|
# SF bug 705836. "<f" and ">f" had a severe rounding bug, where a carry
|
|
# from the low-order discarded bits could propagate into the exponent
|
|
# field, causing the result to be wrong by a factor of 2.
|
|
|
|
def test_705836():
|
|
import math
|
|
|
|
for base in range(1, 33):
|
|
# smaller <- largest representable float less than base.
|
|
delta = 0.5
|
|
while base - delta / 2.0 != base:
|
|
delta /= 2.0
|
|
smaller = base - delta
|
|
# Packing this rounds away a solid string of trailing 1 bits.
|
|
packed = struct.pack("<f", smaller)
|
|
unpacked = struct.unpack("<f", packed)[0]
|
|
# This failed at base = 2, 4, and 32, with unpacked = 1, 2, and
|
|
# 16, respectively.
|
|
verify(base == unpacked)
|
|
bigpacked = struct.pack(">f", smaller)
|
|
verify(bigpacked == string_reverse(packed),
|
|
">f pack should be byte-reversal of <f pack")
|
|
unpacked = struct.unpack(">f", bigpacked)[0]
|
|
verify(base == unpacked)
|
|
|
|
# Largest finite IEEE single.
|
|
big = (1 << 24) - 1
|
|
big = math.ldexp(big, 127 - 23)
|
|
packed = struct.pack(">f", big)
|
|
unpacked = struct.unpack(">f", packed)[0]
|
|
verify(big == unpacked)
|
|
|
|
# The same, but tack on a 1 bit so it rounds up to infinity.
|
|
big = (1 << 25) - 1
|
|
big = math.ldexp(big, 127 - 24)
|
|
try:
|
|
packed = struct.pack(">f", big)
|
|
except OverflowError:
|
|
pass
|
|
else:
|
|
TestFailed("expected OverflowError")
|
|
|
|
test_705836()
|