1058 lines
39 KiB
Python
1058 lines
39 KiB
Python
"""Unit tests for the bytes and bytearray types.
|
|
|
|
XXX This is a mess. Common tests should be moved to buffer_tests.py,
|
|
which itself ought to be unified with string_tests.py (and the latter
|
|
should be modernized).
|
|
"""
|
|
|
|
import os
|
|
import re
|
|
import sys
|
|
import copy
|
|
import functools
|
|
import pickle
|
|
import tempfile
|
|
import unittest
|
|
import test.test_support
|
|
import test.string_tests
|
|
import test.buffer_tests
|
|
|
|
|
|
if sys.flags.bytes_warning:
|
|
def check_bytes_warnings(func):
|
|
@functools.wraps(func)
|
|
def wrapper(*args, **kw):
|
|
with test.test_support.check_warnings(('', BytesWarning)):
|
|
return func(*args, **kw)
|
|
return wrapper
|
|
else:
|
|
# no-op
|
|
def check_bytes_warnings(func):
|
|
return func
|
|
|
|
|
|
class Indexable:
|
|
def __init__(self, value=0):
|
|
self.value = value
|
|
def __index__(self):
|
|
return self.value
|
|
|
|
|
|
class BaseBytesTest(unittest.TestCase):
|
|
|
|
def test_basics(self):
|
|
b = self.type2test()
|
|
self.assertEqual(type(b), self.type2test)
|
|
self.assertEqual(b.__class__, self.type2test)
|
|
|
|
def test_empty_sequence(self):
|
|
b = self.type2test()
|
|
self.assertEqual(len(b), 0)
|
|
self.assertRaises(IndexError, lambda: b[0])
|
|
self.assertRaises(IndexError, lambda: b[1])
|
|
self.assertRaises(IndexError, lambda: b[sys.maxint])
|
|
self.assertRaises(IndexError, lambda: b[sys.maxint+1])
|
|
self.assertRaises(IndexError, lambda: b[10**100])
|
|
self.assertRaises(IndexError, lambda: b[-1])
|
|
self.assertRaises(IndexError, lambda: b[-2])
|
|
self.assertRaises(IndexError, lambda: b[-sys.maxint])
|
|
self.assertRaises(IndexError, lambda: b[-sys.maxint-1])
|
|
self.assertRaises(IndexError, lambda: b[-sys.maxint-2])
|
|
self.assertRaises(IndexError, lambda: b[-10**100])
|
|
|
|
def test_from_list(self):
|
|
ints = list(range(256))
|
|
b = self.type2test(i for i in ints)
|
|
self.assertEqual(len(b), 256)
|
|
self.assertEqual(list(b), ints)
|
|
|
|
def test_from_index(self):
|
|
b = self.type2test([Indexable(), Indexable(1), Indexable(254),
|
|
Indexable(255)])
|
|
self.assertEqual(list(b), [0, 1, 254, 255])
|
|
self.assertRaises(ValueError, self.type2test, [Indexable(-1)])
|
|
self.assertRaises(ValueError, self.type2test, [Indexable(256)])
|
|
|
|
def test_from_ssize(self):
|
|
self.assertEqual(self.type2test(0), b'')
|
|
self.assertEqual(self.type2test(1), b'\x00')
|
|
self.assertEqual(self.type2test(5), b'\x00\x00\x00\x00\x00')
|
|
self.assertRaises(ValueError, self.type2test, -1)
|
|
|
|
self.assertEqual(self.type2test('0', 'ascii'), b'0')
|
|
self.assertEqual(self.type2test(b'0'), b'0')
|
|
self.assertRaises(OverflowError, self.type2test, sys.maxsize + 1)
|
|
|
|
def test_constructor_type_errors(self):
|
|
self.assertRaises(TypeError, self.type2test, 0.0)
|
|
class C:
|
|
pass
|
|
# allowed in 2.x
|
|
#self.assertRaises(TypeError, self.type2test, ["0"])
|
|
self.assertRaises(TypeError, self.type2test, [0.0])
|
|
self.assertRaises(TypeError, self.type2test, [None])
|
|
self.assertRaises(TypeError, self.type2test, [C()])
|
|
|
|
def test_constructor_value_errors(self):
|
|
self.assertRaises(ValueError, self.type2test, [-1])
|
|
self.assertRaises(ValueError, self.type2test, [-sys.maxint])
|
|
self.assertRaises(ValueError, self.type2test, [-sys.maxint-1])
|
|
self.assertRaises(ValueError, self.type2test, [-sys.maxint-2])
|
|
self.assertRaises(ValueError, self.type2test, [-10**100])
|
|
self.assertRaises(ValueError, self.type2test, [256])
|
|
self.assertRaises(ValueError, self.type2test, [257])
|
|
self.assertRaises(ValueError, self.type2test, [sys.maxint])
|
|
self.assertRaises(ValueError, self.type2test, [sys.maxint+1])
|
|
self.assertRaises(ValueError, self.type2test, [10**100])
|
|
|
|
def test_compare(self):
|
|
b1 = self.type2test([1, 2, 3])
|
|
b2 = self.type2test([1, 2, 3])
|
|
b3 = self.type2test([1, 3])
|
|
|
|
self.assertEqual(b1, b2)
|
|
self.assertTrue(b2 != b3)
|
|
self.assertTrue(b1 <= b2)
|
|
self.assertTrue(b1 <= b3)
|
|
self.assertTrue(b1 < b3)
|
|
self.assertTrue(b1 >= b2)
|
|
self.assertTrue(b3 >= b2)
|
|
self.assertTrue(b3 > b2)
|
|
|
|
self.assertFalse(b1 != b2)
|
|
self.assertFalse(b2 == b3)
|
|
self.assertFalse(b1 > b2)
|
|
self.assertFalse(b1 > b3)
|
|
self.assertFalse(b1 >= b3)
|
|
self.assertFalse(b1 < b2)
|
|
self.assertFalse(b3 < b2)
|
|
self.assertFalse(b3 <= b2)
|
|
|
|
@check_bytes_warnings
|
|
def test_compare_to_str(self):
|
|
# Byte comparisons with unicode should always fail!
|
|
# Test this for all expected byte orders and Unicode character sizes
|
|
self.assertEqual(self.type2test(b"\0a\0b\0c") == u"abc", False)
|
|
self.assertEqual(self.type2test(b"\0\0\0a\0\0\0b\0\0\0c") == u"abc", False)
|
|
self.assertEqual(self.type2test(b"a\0b\0c\0") == u"abc", False)
|
|
self.assertEqual(self.type2test(b"a\0\0\0b\0\0\0c\0\0\0") == u"abc", False)
|
|
self.assertEqual(self.type2test() == unicode(), False)
|
|
self.assertEqual(self.type2test() != unicode(), True)
|
|
|
|
def test_reversed(self):
|
|
input = list(map(ord, "Hello"))
|
|
b = self.type2test(input)
|
|
output = list(reversed(b))
|
|
input.reverse()
|
|
self.assertEqual(output, input)
|
|
|
|
def test_getslice(self):
|
|
def by(s):
|
|
return self.type2test(map(ord, s))
|
|
b = by("Hello, world")
|
|
|
|
self.assertEqual(b[:5], by("Hello"))
|
|
self.assertEqual(b[1:5], by("ello"))
|
|
self.assertEqual(b[5:7], by(", "))
|
|
self.assertEqual(b[7:], by("world"))
|
|
self.assertEqual(b[7:12], by("world"))
|
|
self.assertEqual(b[7:100], by("world"))
|
|
|
|
self.assertEqual(b[:-7], by("Hello"))
|
|
self.assertEqual(b[-11:-7], by("ello"))
|
|
self.assertEqual(b[-7:-5], by(", "))
|
|
self.assertEqual(b[-5:], by("world"))
|
|
self.assertEqual(b[-5:12], by("world"))
|
|
self.assertEqual(b[-5:100], by("world"))
|
|
self.assertEqual(b[-100:5], by("Hello"))
|
|
|
|
def test_extended_getslice(self):
|
|
# Test extended slicing by comparing with list slicing.
|
|
L = list(range(255))
|
|
b = self.type2test(L)
|
|
indices = (0, None, 1, 3, 19, 100, -1, -2, -31, -100)
|
|
for start in indices:
|
|
for stop in indices:
|
|
# Skip step 0 (invalid)
|
|
for step in indices[1:]:
|
|
self.assertEqual(b[start:stop:step], self.type2test(L[start:stop:step]))
|
|
|
|
def test_encoding(self):
|
|
sample = u"Hello world\n\u1234\u5678\u9abc\udef0"
|
|
for enc in ("utf8", "utf16"):
|
|
b = self.type2test(sample, enc)
|
|
self.assertEqual(b, self.type2test(sample.encode(enc)))
|
|
self.assertRaises(UnicodeEncodeError, self.type2test, sample, "latin1")
|
|
b = self.type2test(sample, "latin1", "ignore")
|
|
self.assertEqual(b, self.type2test(sample[:-4], "utf-8"))
|
|
|
|
def test_decode(self):
|
|
sample = u"Hello world\n\u1234\u5678\u9abc\def0\def0"
|
|
for enc in ("utf8", "utf16"):
|
|
b = self.type2test(sample, enc)
|
|
self.assertEqual(b.decode(enc), sample)
|
|
sample = u"Hello world\n\x80\x81\xfe\xff"
|
|
b = self.type2test(sample, "latin1")
|
|
self.assertRaises(UnicodeDecodeError, b.decode, "utf8")
|
|
self.assertEqual(b.decode("utf8", "ignore"), "Hello world\n")
|
|
self.assertEqual(b.decode(errors="ignore", encoding="utf8"),
|
|
"Hello world\n")
|
|
|
|
def test_from_int(self):
|
|
b = self.type2test(0)
|
|
self.assertEqual(b, self.type2test())
|
|
b = self.type2test(10)
|
|
self.assertEqual(b, self.type2test([0]*10))
|
|
b = self.type2test(10000)
|
|
self.assertEqual(b, self.type2test([0]*10000))
|
|
|
|
def test_concat(self):
|
|
b1 = self.type2test(b"abc")
|
|
b2 = self.type2test(b"def")
|
|
self.assertEqual(b1 + b2, b"abcdef")
|
|
self.assertEqual(b1 + bytes(b"def"), b"abcdef")
|
|
self.assertEqual(bytes(b"def") + b1, b"defabc")
|
|
self.assertRaises(TypeError, lambda: b1 + u"def")
|
|
self.assertRaises(TypeError, lambda: u"abc" + b2)
|
|
|
|
def test_repeat(self):
|
|
for b in b"abc", self.type2test(b"abc"):
|
|
self.assertEqual(b * 3, b"abcabcabc")
|
|
self.assertEqual(b * 0, b"")
|
|
self.assertEqual(b * -1, b"")
|
|
self.assertRaises(TypeError, lambda: b * 3.14)
|
|
self.assertRaises(TypeError, lambda: 3.14 * b)
|
|
# XXX Shouldn't bytes and bytearray agree on what to raise?
|
|
self.assertRaises((OverflowError, MemoryError),
|
|
lambda: b * sys.maxsize)
|
|
|
|
def test_repeat_1char(self):
|
|
self.assertEqual(self.type2test(b'x')*100, self.type2test([ord('x')]*100))
|
|
|
|
def test_contains(self):
|
|
b = self.type2test(b"abc")
|
|
self.assertIn(ord('a'), b)
|
|
self.assertIn(int(ord('a')), b)
|
|
self.assertNotIn(200, b)
|
|
self.assertRaises(ValueError, lambda: 300 in b)
|
|
self.assertRaises(ValueError, lambda: -1 in b)
|
|
self.assertRaises(TypeError, lambda: None in b)
|
|
self.assertRaises(TypeError, lambda: float(ord('a')) in b)
|
|
self.assertRaises(TypeError, lambda: u"a" in b)
|
|
for f in bytes, bytearray:
|
|
self.assertIn(f(b""), b)
|
|
self.assertIn(f(b"a"), b)
|
|
self.assertIn(f(b"b"), b)
|
|
self.assertIn(f(b"c"), b)
|
|
self.assertIn(f(b"ab"), b)
|
|
self.assertIn(f(b"bc"), b)
|
|
self.assertIn(f(b"abc"), b)
|
|
self.assertNotIn(f(b"ac"), b)
|
|
self.assertNotIn(f(b"d"), b)
|
|
self.assertNotIn(f(b"dab"), b)
|
|
self.assertNotIn(f(b"abd"), b)
|
|
|
|
def test_fromhex(self):
|
|
self.assertRaises(TypeError, self.type2test.fromhex)
|
|
self.assertRaises(TypeError, self.type2test.fromhex, 1)
|
|
self.assertEquals(self.type2test.fromhex(u''), self.type2test())
|
|
b = bytearray([0x1a, 0x2b, 0x30])
|
|
self.assertEquals(self.type2test.fromhex(u'1a2B30'), b)
|
|
self.assertEquals(self.type2test.fromhex(u' 1A 2B 30 '), b)
|
|
self.assertEquals(self.type2test.fromhex(u'0000'), b'\0\0')
|
|
self.assertRaises(ValueError, self.type2test.fromhex, u'a')
|
|
self.assertRaises(ValueError, self.type2test.fromhex, u'rt')
|
|
self.assertRaises(ValueError, self.type2test.fromhex, u'1a b cd')
|
|
self.assertRaises(ValueError, self.type2test.fromhex, u'\x00')
|
|
self.assertRaises(ValueError, self.type2test.fromhex, u'12 \x00 34')
|
|
|
|
def test_join(self):
|
|
self.assertEqual(self.type2test(b"").join([]), b"")
|
|
self.assertEqual(self.type2test(b"").join([b""]), b"")
|
|
for lst in [[b"abc"], [b"a", b"bc"], [b"ab", b"c"], [b"a", b"b", b"c"]]:
|
|
lst = list(map(self.type2test, lst))
|
|
self.assertEqual(self.type2test(b"").join(lst), b"abc")
|
|
self.assertEqual(self.type2test(b"").join(tuple(lst)), b"abc")
|
|
self.assertEqual(self.type2test(b"").join(iter(lst)), b"abc")
|
|
self.assertEqual(self.type2test(b".").join([b"ab", b"cd"]), b"ab.cd")
|
|
# XXX more...
|
|
|
|
def test_count(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.count(b'i'), 4)
|
|
self.assertEqual(b.count(b'ss'), 2)
|
|
self.assertEqual(b.count(b'w'), 0)
|
|
|
|
def test_startswith(self):
|
|
b = self.type2test(b'hello')
|
|
self.assertFalse(self.type2test().startswith(b"anything"))
|
|
self.assertTrue(b.startswith(b"hello"))
|
|
self.assertTrue(b.startswith(b"hel"))
|
|
self.assertTrue(b.startswith(b"h"))
|
|
self.assertFalse(b.startswith(b"hellow"))
|
|
self.assertFalse(b.startswith(b"ha"))
|
|
|
|
def test_endswith(self):
|
|
b = self.type2test(b'hello')
|
|
self.assertFalse(bytearray().endswith(b"anything"))
|
|
self.assertTrue(b.endswith(b"hello"))
|
|
self.assertTrue(b.endswith(b"llo"))
|
|
self.assertTrue(b.endswith(b"o"))
|
|
self.assertFalse(b.endswith(b"whello"))
|
|
self.assertFalse(b.endswith(b"no"))
|
|
|
|
def test_find(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.find(b'ss'), 2)
|
|
self.assertEqual(b.find(b'ss', 3), 5)
|
|
self.assertEqual(b.find(b'ss', 1, 7), 2)
|
|
self.assertEqual(b.find(b'ss', 1, 3), -1)
|
|
self.assertEqual(b.find(b'w'), -1)
|
|
self.assertEqual(b.find(b'mississippian'), -1)
|
|
|
|
def test_rfind(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.rfind(b'ss'), 5)
|
|
self.assertEqual(b.rfind(b'ss', 3), 5)
|
|
self.assertEqual(b.rfind(b'ss', 0, 6), 2)
|
|
self.assertEqual(b.rfind(b'w'), -1)
|
|
self.assertEqual(b.rfind(b'mississippian'), -1)
|
|
|
|
def test_index(self):
|
|
b = self.type2test(b'world')
|
|
self.assertEqual(b.index(b'w'), 0)
|
|
self.assertEqual(b.index(b'orl'), 1)
|
|
self.assertRaises(ValueError, b.index, b'worm')
|
|
self.assertRaises(ValueError, b.index, b'ldo')
|
|
|
|
def test_rindex(self):
|
|
# XXX could be more rigorous
|
|
b = self.type2test(b'world')
|
|
self.assertEqual(b.rindex(b'w'), 0)
|
|
self.assertEqual(b.rindex(b'orl'), 1)
|
|
self.assertRaises(ValueError, b.rindex, b'worm')
|
|
self.assertRaises(ValueError, b.rindex, b'ldo')
|
|
|
|
def test_replace(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.replace(b'i', b'a'), b'massassappa')
|
|
self.assertEqual(b.replace(b'ss', b'x'), b'mixixippi')
|
|
|
|
def test_split(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.split(b'i'), [b'm', b'ss', b'ss', b'pp', b''])
|
|
self.assertEqual(b.split(b'ss'), [b'mi', b'i', b'ippi'])
|
|
self.assertEqual(b.split(b'w'), [b])
|
|
|
|
def test_split_whitespace(self):
|
|
for b in (b' arf barf ', b'arf\tbarf', b'arf\nbarf', b'arf\rbarf',
|
|
b'arf\fbarf', b'arf\vbarf'):
|
|
b = self.type2test(b)
|
|
self.assertEqual(b.split(), [b'arf', b'barf'])
|
|
self.assertEqual(b.split(None), [b'arf', b'barf'])
|
|
self.assertEqual(b.split(None, 2), [b'arf', b'barf'])
|
|
for b in (b'a\x1Cb', b'a\x1Db', b'a\x1Eb', b'a\x1Fb'):
|
|
b = self.type2test(b)
|
|
self.assertEqual(b.split(), [b])
|
|
self.assertEqual(self.type2test(b' a bb c ').split(None, 0), [b'a bb c '])
|
|
self.assertEqual(self.type2test(b' a bb c ').split(None, 1), [b'a', b'bb c '])
|
|
self.assertEqual(self.type2test(b' a bb c ').split(None, 2), [b'a', b'bb', b'c '])
|
|
self.assertEqual(self.type2test(b' a bb c ').split(None, 3), [b'a', b'bb', b'c'])
|
|
|
|
def test_split_string_error(self):
|
|
self.assertRaises(TypeError, self.type2test(b'a b').split, u' ')
|
|
|
|
def test_split_unicodewhitespace(self):
|
|
b = self.type2test(b"\x09\x0A\x0B\x0C\x0D\x1C\x1D\x1E\x1F")
|
|
self.assertEqual(b.split(), [b'\x1c\x1d\x1e\x1f'])
|
|
|
|
def test_rsplit(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.rsplit(b'i'), [b'm', b'ss', b'ss', b'pp', b''])
|
|
self.assertEqual(b.rsplit(b'ss'), [b'mi', b'i', b'ippi'])
|
|
self.assertEqual(b.rsplit(b'w'), [b])
|
|
|
|
def test_rsplit_whitespace(self):
|
|
for b in (b' arf barf ', b'arf\tbarf', b'arf\nbarf', b'arf\rbarf',
|
|
b'arf\fbarf', b'arf\vbarf'):
|
|
b = self.type2test(b)
|
|
self.assertEqual(b.rsplit(), [b'arf', b'barf'])
|
|
self.assertEqual(b.rsplit(None), [b'arf', b'barf'])
|
|
self.assertEqual(b.rsplit(None, 2), [b'arf', b'barf'])
|
|
self.assertEqual(self.type2test(b' a bb c ').rsplit(None, 0), [b' a bb c'])
|
|
self.assertEqual(self.type2test(b' a bb c ').rsplit(None, 1), [b' a bb', b'c'])
|
|
self.assertEqual(self.type2test(b' a bb c ').rsplit(None, 2), [b' a', b'bb', b'c'])
|
|
self.assertEqual(self.type2test(b' a bb c ').rsplit(None, 3), [b'a', b'bb', b'c'])
|
|
|
|
def test_rsplit_string_error(self):
|
|
self.assertRaises(TypeError, self.type2test(b'a b').rsplit, u' ')
|
|
|
|
def test_rsplit_unicodewhitespace(self):
|
|
b = self.type2test(b"\x09\x0A\x0B\x0C\x0D\x1C\x1D\x1E\x1F")
|
|
self.assertEqual(b.rsplit(), [b'\x1c\x1d\x1e\x1f'])
|
|
|
|
def test_partition(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.partition(b'ss'), (b'mi', b'ss', b'issippi'))
|
|
self.assertEqual(b.partition(b'w'), (b'mississippi', b'', b''))
|
|
|
|
def test_rpartition(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.rpartition(b'ss'), (b'missi', b'ss', b'ippi'))
|
|
self.assertEqual(b.rpartition(b'i'), (b'mississipp', b'i', b''))
|
|
self.assertEqual(b.rpartition(b'w'), (b'', b'', b'mississippi'))
|
|
|
|
def test_pickling(self):
|
|
for proto in range(pickle.HIGHEST_PROTOCOL + 1):
|
|
for b in b"", b"a", b"abc", b"\xffab\x80", b"\0\0\377\0\0":
|
|
b = self.type2test(b)
|
|
ps = pickle.dumps(b, proto)
|
|
q = pickle.loads(ps)
|
|
self.assertEqual(b, q)
|
|
|
|
def test_strip(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.strip(b'i'), b'mississipp')
|
|
self.assertEqual(b.strip(b'm'), b'ississippi')
|
|
self.assertEqual(b.strip(b'pi'), b'mississ')
|
|
self.assertEqual(b.strip(b'im'), b'ssissipp')
|
|
self.assertEqual(b.strip(b'pim'), b'ssiss')
|
|
self.assertEqual(b.strip(b), b'')
|
|
|
|
def test_lstrip(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.lstrip(b'i'), b'mississippi')
|
|
self.assertEqual(b.lstrip(b'm'), b'ississippi')
|
|
self.assertEqual(b.lstrip(b'pi'), b'mississippi')
|
|
self.assertEqual(b.lstrip(b'im'), b'ssissippi')
|
|
self.assertEqual(b.lstrip(b'pim'), b'ssissippi')
|
|
|
|
def test_rstrip(self):
|
|
b = self.type2test(b'mississippi')
|
|
self.assertEqual(b.rstrip(b'i'), b'mississipp')
|
|
self.assertEqual(b.rstrip(b'm'), b'mississippi')
|
|
self.assertEqual(b.rstrip(b'pi'), b'mississ')
|
|
self.assertEqual(b.rstrip(b'im'), b'mississipp')
|
|
self.assertEqual(b.rstrip(b'pim'), b'mississ')
|
|
|
|
def test_strip_whitespace(self):
|
|
b = self.type2test(b' \t\n\r\f\vabc \t\n\r\f\v')
|
|
self.assertEqual(b.strip(), b'abc')
|
|
self.assertEqual(b.lstrip(), b'abc \t\n\r\f\v')
|
|
self.assertEqual(b.rstrip(), b' \t\n\r\f\vabc')
|
|
|
|
def test_strip_bytearray(self):
|
|
self.assertEqual(self.type2test(b'abc').strip(memoryview(b'ac')), b'b')
|
|
self.assertEqual(self.type2test(b'abc').lstrip(memoryview(b'ac')), b'bc')
|
|
self.assertEqual(self.type2test(b'abc').rstrip(memoryview(b'ac')), b'ab')
|
|
|
|
def test_strip_string_error(self):
|
|
self.assertRaises(TypeError, self.type2test(b'abc').strip, u'b')
|
|
self.assertRaises(TypeError, self.type2test(b'abc').lstrip, u'b')
|
|
self.assertRaises(TypeError, self.type2test(b'abc').rstrip, u'b')
|
|
|
|
def test_ord(self):
|
|
b = self.type2test(b'\0A\x7f\x80\xff')
|
|
self.assertEqual([ord(b[i:i+1]) for i in range(len(b))],
|
|
[0, 65, 127, 128, 255])
|
|
|
|
|
|
class ByteArrayTest(BaseBytesTest):
|
|
type2test = bytearray
|
|
|
|
def test_nohash(self):
|
|
self.assertRaises(TypeError, hash, bytearray())
|
|
|
|
def test_bytearray_api(self):
|
|
short_sample = b"Hello world\n"
|
|
sample = short_sample + b"\0"*(20 - len(short_sample))
|
|
tfn = tempfile.mktemp()
|
|
try:
|
|
# Prepare
|
|
with open(tfn, "wb") as f:
|
|
f.write(short_sample)
|
|
# Test readinto
|
|
with open(tfn, "rb") as f:
|
|
b = bytearray(20)
|
|
n = f.readinto(b)
|
|
self.assertEqual(n, len(short_sample))
|
|
# Python 2.x
|
|
b_sample = (ord(s) for s in sample)
|
|
self.assertEqual(list(b), list(b_sample))
|
|
# Test writing in binary mode
|
|
with open(tfn, "wb") as f:
|
|
f.write(b)
|
|
with open(tfn, "rb") as f:
|
|
self.assertEqual(f.read(), sample)
|
|
# Text mode is ambiguous; don't test
|
|
finally:
|
|
try:
|
|
os.remove(tfn)
|
|
except os.error:
|
|
pass
|
|
|
|
def test_reverse(self):
|
|
b = bytearray(b'hello')
|
|
self.assertEqual(b.reverse(), None)
|
|
self.assertEqual(b, b'olleh')
|
|
b = bytearray(b'hello1') # test even number of items
|
|
b.reverse()
|
|
self.assertEqual(b, b'1olleh')
|
|
b = bytearray()
|
|
b.reverse()
|
|
self.assertFalse(b)
|
|
|
|
def test_regexps(self):
|
|
def by(s):
|
|
return bytearray(map(ord, s))
|
|
b = by("Hello, world")
|
|
self.assertEqual(re.findall(r"\w+", b), [by("Hello"), by("world")])
|
|
|
|
def test_setitem(self):
|
|
b = bytearray([1, 2, 3])
|
|
b[1] = 100
|
|
self.assertEqual(b, bytearray([1, 100, 3]))
|
|
b[-1] = 200
|
|
self.assertEqual(b, bytearray([1, 100, 200]))
|
|
b[0] = Indexable(10)
|
|
self.assertEqual(b, bytearray([10, 100, 200]))
|
|
try:
|
|
b[3] = 0
|
|
self.fail("Didn't raise IndexError")
|
|
except IndexError:
|
|
pass
|
|
try:
|
|
b[-10] = 0
|
|
self.fail("Didn't raise IndexError")
|
|
except IndexError:
|
|
pass
|
|
try:
|
|
b[0] = 256
|
|
self.fail("Didn't raise ValueError")
|
|
except ValueError:
|
|
pass
|
|
try:
|
|
b[0] = Indexable(-1)
|
|
self.fail("Didn't raise ValueError")
|
|
except ValueError:
|
|
pass
|
|
try:
|
|
b[0] = None
|
|
self.fail("Didn't raise TypeError")
|
|
except TypeError:
|
|
pass
|
|
|
|
def test_delitem(self):
|
|
b = bytearray(range(10))
|
|
del b[0]
|
|
self.assertEqual(b, bytearray(range(1, 10)))
|
|
del b[-1]
|
|
self.assertEqual(b, bytearray(range(1, 9)))
|
|
del b[4]
|
|
self.assertEqual(b, bytearray([1, 2, 3, 4, 6, 7, 8]))
|
|
|
|
def test_setslice(self):
|
|
b = bytearray(range(10))
|
|
self.assertEqual(list(b), list(range(10)))
|
|
|
|
b[0:5] = bytearray([1, 1, 1, 1, 1])
|
|
self.assertEqual(b, bytearray([1, 1, 1, 1, 1, 5, 6, 7, 8, 9]))
|
|
|
|
del b[0:-5]
|
|
self.assertEqual(b, bytearray([5, 6, 7, 8, 9]))
|
|
|
|
b[0:0] = bytearray([0, 1, 2, 3, 4])
|
|
self.assertEqual(b, bytearray(range(10)))
|
|
|
|
b[-7:-3] = bytearray([100, 101])
|
|
self.assertEqual(b, bytearray([0, 1, 2, 100, 101, 7, 8, 9]))
|
|
|
|
b[3:5] = [3, 4, 5, 6]
|
|
self.assertEqual(b, bytearray(range(10)))
|
|
|
|
b[3:0] = [42, 42, 42]
|
|
self.assertEqual(b, bytearray([0, 1, 2, 42, 42, 42, 3, 4, 5, 6, 7, 8, 9]))
|
|
|
|
def test_extended_set_del_slice(self):
|
|
indices = (0, None, 1, 3, 19, 300, 1<<333, -1, -2, -31, -300)
|
|
for start in indices:
|
|
for stop in indices:
|
|
# Skip invalid step 0
|
|
for step in indices[1:]:
|
|
L = list(range(255))
|
|
b = bytearray(L)
|
|
# Make sure we have a slice of exactly the right length,
|
|
# but with different data.
|
|
data = L[start:stop:step]
|
|
data.reverse()
|
|
L[start:stop:step] = data
|
|
b[start:stop:step] = data
|
|
self.assertEquals(b, bytearray(L))
|
|
|
|
del L[start:stop:step]
|
|
del b[start:stop:step]
|
|
self.assertEquals(b, bytearray(L))
|
|
|
|
def test_setslice_trap(self):
|
|
# This test verifies that we correctly handle assigning self
|
|
# to a slice of self (the old Lambert Meertens trap).
|
|
b = bytearray(range(256))
|
|
b[8:] = b
|
|
self.assertEqual(b, bytearray(list(range(8)) + list(range(256))))
|
|
|
|
def test_iconcat(self):
|
|
b = bytearray(b"abc")
|
|
b1 = b
|
|
b += b"def"
|
|
self.assertEqual(b, b"abcdef")
|
|
self.assertEqual(b, b1)
|
|
self.assertTrue(b is b1)
|
|
b += b"xyz"
|
|
self.assertEqual(b, b"abcdefxyz")
|
|
try:
|
|
b += u""
|
|
except TypeError:
|
|
pass
|
|
else:
|
|
self.fail("bytes += unicode didn't raise TypeError")
|
|
|
|
def test_irepeat(self):
|
|
b = bytearray(b"abc")
|
|
b1 = b
|
|
b *= 3
|
|
self.assertEqual(b, b"abcabcabc")
|
|
self.assertEqual(b, b1)
|
|
self.assertTrue(b is b1)
|
|
|
|
def test_irepeat_1char(self):
|
|
b = bytearray(b"x")
|
|
b1 = b
|
|
b *= 100
|
|
self.assertEqual(b, b"x"*100)
|
|
self.assertEqual(b, b1)
|
|
self.assertTrue(b is b1)
|
|
|
|
def test_alloc(self):
|
|
b = bytearray()
|
|
alloc = b.__alloc__()
|
|
self.assertTrue(alloc >= 0)
|
|
seq = [alloc]
|
|
for i in range(100):
|
|
b += b"x"
|
|
alloc = b.__alloc__()
|
|
self.assertTrue(alloc >= len(b))
|
|
if alloc not in seq:
|
|
seq.append(alloc)
|
|
|
|
def test_extend(self):
|
|
orig = b'hello'
|
|
a = bytearray(orig)
|
|
a.extend(a)
|
|
self.assertEqual(a, orig + orig)
|
|
self.assertEqual(a[5:], orig)
|
|
a = bytearray(b'')
|
|
# Test iterators that don't have a __length_hint__
|
|
a.extend(map(ord, orig * 25))
|
|
a.extend(ord(x) for x in orig * 25)
|
|
self.assertEqual(a, orig * 50)
|
|
self.assertEqual(a[-5:], orig)
|
|
a = bytearray(b'')
|
|
a.extend(iter(map(ord, orig * 50)))
|
|
self.assertEqual(a, orig * 50)
|
|
self.assertEqual(a[-5:], orig)
|
|
a = bytearray(b'')
|
|
a.extend(list(map(ord, orig * 50)))
|
|
self.assertEqual(a, orig * 50)
|
|
self.assertEqual(a[-5:], orig)
|
|
a = bytearray(b'')
|
|
self.assertRaises(ValueError, a.extend, [0, 1, 2, 256])
|
|
self.assertRaises(ValueError, a.extend, [0, 1, 2, -1])
|
|
self.assertEqual(len(a), 0)
|
|
a = bytearray(b'')
|
|
a.extend([Indexable(ord('a'))])
|
|
self.assertEqual(a, b'a')
|
|
|
|
def test_remove(self):
|
|
b = bytearray(b'hello')
|
|
b.remove(ord('l'))
|
|
self.assertEqual(b, b'helo')
|
|
b.remove(ord('l'))
|
|
self.assertEqual(b, b'heo')
|
|
self.assertRaises(ValueError, lambda: b.remove(ord('l')))
|
|
self.assertRaises(ValueError, lambda: b.remove(400))
|
|
self.assertRaises(TypeError, lambda: b.remove(u'e'))
|
|
# remove first and last
|
|
b.remove(ord('o'))
|
|
b.remove(ord('h'))
|
|
self.assertEqual(b, b'e')
|
|
self.assertRaises(TypeError, lambda: b.remove(u'e'))
|
|
b.remove(Indexable(ord('e')))
|
|
self.assertEqual(b, b'')
|
|
|
|
def test_pop(self):
|
|
b = bytearray(b'world')
|
|
self.assertEqual(b.pop(), ord('d'))
|
|
self.assertEqual(b.pop(0), ord('w'))
|
|
self.assertEqual(b.pop(-2), ord('r'))
|
|
self.assertRaises(IndexError, lambda: b.pop(10))
|
|
self.assertRaises(OverflowError, lambda: bytearray().pop())
|
|
# test for issue #6846
|
|
self.assertEqual(bytearray(b'\xff').pop(), 0xff)
|
|
|
|
def test_nosort(self):
|
|
self.assertRaises(AttributeError, lambda: bytearray().sort())
|
|
|
|
def test_append(self):
|
|
b = bytearray(b'hell')
|
|
b.append(ord('o'))
|
|
self.assertEqual(b, b'hello')
|
|
self.assertEqual(b.append(100), None)
|
|
b = bytearray()
|
|
b.append(ord('A'))
|
|
self.assertEqual(len(b), 1)
|
|
self.assertRaises(TypeError, lambda: b.append(u'o'))
|
|
b = bytearray()
|
|
b.append(Indexable(ord('A')))
|
|
self.assertEqual(b, b'A')
|
|
|
|
def test_insert(self):
|
|
b = bytearray(b'msssspp')
|
|
b.insert(1, ord('i'))
|
|
b.insert(4, ord('i'))
|
|
b.insert(-2, ord('i'))
|
|
b.insert(1000, ord('i'))
|
|
self.assertEqual(b, b'mississippi')
|
|
# allowed in 2.x
|
|
#self.assertRaises(TypeError, lambda: b.insert(0, b'1'))
|
|
b = bytearray()
|
|
b.insert(0, Indexable(ord('A')))
|
|
self.assertEqual(b, b'A')
|
|
|
|
def test_copied(self):
|
|
# Issue 4348. Make sure that operations that don't mutate the array
|
|
# copy the bytes.
|
|
b = bytearray(b'abc')
|
|
self.assertFalse(b is b.replace(b'abc', b'cde', 0))
|
|
|
|
t = bytearray([i for i in range(256)])
|
|
x = bytearray(b'')
|
|
self.assertFalse(x is x.translate(t))
|
|
|
|
def test_partition_bytearray_doesnt_share_nullstring(self):
|
|
a, b, c = bytearray(b"x").partition(b"y")
|
|
self.assertEqual(b, b"")
|
|
self.assertEqual(c, b"")
|
|
self.assertTrue(b is not c)
|
|
b += b"!"
|
|
self.assertEqual(c, b"")
|
|
a, b, c = bytearray(b"x").partition(b"y")
|
|
self.assertEqual(b, b"")
|
|
self.assertEqual(c, b"")
|
|
# Same for rpartition
|
|
b, c, a = bytearray(b"x").rpartition(b"y")
|
|
self.assertEqual(b, b"")
|
|
self.assertEqual(c, b"")
|
|
self.assertTrue(b is not c)
|
|
b += b"!"
|
|
self.assertEqual(c, b"")
|
|
c, b, a = bytearray(b"x").rpartition(b"y")
|
|
self.assertEqual(b, b"")
|
|
self.assertEqual(c, b"")
|
|
|
|
def test_resize_forbidden(self):
|
|
# #4509: can't resize a bytearray when there are buffer exports, even
|
|
# if it wouldn't reallocate the underlying buffer.
|
|
# Furthermore, no destructive changes to the buffer may be applied
|
|
# before raising the error.
|
|
b = bytearray(range(10))
|
|
v = memoryview(b)
|
|
def resize(n):
|
|
b[1:-1] = range(n + 1, 2*n - 1)
|
|
resize(10)
|
|
orig = b[:]
|
|
self.assertRaises(BufferError, resize, 11)
|
|
self.assertEquals(b, orig)
|
|
self.assertRaises(BufferError, resize, 9)
|
|
self.assertEquals(b, orig)
|
|
self.assertRaises(BufferError, resize, 0)
|
|
self.assertEquals(b, orig)
|
|
# Other operations implying resize
|
|
self.assertRaises(BufferError, b.pop, 0)
|
|
self.assertEquals(b, orig)
|
|
self.assertRaises(BufferError, b.remove, b[1])
|
|
self.assertEquals(b, orig)
|
|
def delitem():
|
|
del b[1]
|
|
self.assertRaises(BufferError, delitem)
|
|
self.assertEquals(b, orig)
|
|
# deleting a non-contiguous slice
|
|
def delslice():
|
|
b[1:-1:2] = b""
|
|
self.assertRaises(BufferError, delslice)
|
|
self.assertEquals(b, orig)
|
|
|
|
def test_empty_bytearray(self):
|
|
# Issue #7561: operations on empty bytearrays could crash in many
|
|
# situations, due to a fragile implementation of the
|
|
# PyByteArray_AS_STRING() C macro.
|
|
self.assertRaises(ValueError, int, bytearray(b''))
|
|
|
|
|
|
class AssortedBytesTest(unittest.TestCase):
|
|
#
|
|
# Test various combinations of bytes and bytearray
|
|
#
|
|
|
|
@check_bytes_warnings
|
|
def test_repr_str(self):
|
|
for f in str, repr:
|
|
self.assertEqual(f(bytearray()), "bytearray(b'')")
|
|
self.assertEqual(f(bytearray([0])), "bytearray(b'\\x00')")
|
|
self.assertEqual(f(bytearray([0, 1, 254, 255])),
|
|
"bytearray(b'\\x00\\x01\\xfe\\xff')")
|
|
self.assertEqual(f(b"abc"), "b'abc'")
|
|
self.assertEqual(f(b"'"), '''b"'"''') # '''
|
|
self.assertEqual(f(b"'\""), r"""b'\'"'""") # '
|
|
|
|
def test_compare_bytes_to_bytearray(self):
|
|
self.assertEqual(b"abc" == bytes(b"abc"), True)
|
|
self.assertEqual(b"ab" != bytes(b"abc"), True)
|
|
self.assertEqual(b"ab" <= bytes(b"abc"), True)
|
|
self.assertEqual(b"ab" < bytes(b"abc"), True)
|
|
self.assertEqual(b"abc" >= bytes(b"ab"), True)
|
|
self.assertEqual(b"abc" > bytes(b"ab"), True)
|
|
|
|
self.assertEqual(b"abc" != bytes(b"abc"), False)
|
|
self.assertEqual(b"ab" == bytes(b"abc"), False)
|
|
self.assertEqual(b"ab" > bytes(b"abc"), False)
|
|
self.assertEqual(b"ab" >= bytes(b"abc"), False)
|
|
self.assertEqual(b"abc" < bytes(b"ab"), False)
|
|
self.assertEqual(b"abc" <= bytes(b"ab"), False)
|
|
|
|
self.assertEqual(bytes(b"abc") == b"abc", True)
|
|
self.assertEqual(bytes(b"ab") != b"abc", True)
|
|
self.assertEqual(bytes(b"ab") <= b"abc", True)
|
|
self.assertEqual(bytes(b"ab") < b"abc", True)
|
|
self.assertEqual(bytes(b"abc") >= b"ab", True)
|
|
self.assertEqual(bytes(b"abc") > b"ab", True)
|
|
|
|
self.assertEqual(bytes(b"abc") != b"abc", False)
|
|
self.assertEqual(bytes(b"ab") == b"abc", False)
|
|
self.assertEqual(bytes(b"ab") > b"abc", False)
|
|
self.assertEqual(bytes(b"ab") >= b"abc", False)
|
|
self.assertEqual(bytes(b"abc") < b"ab", False)
|
|
self.assertEqual(bytes(b"abc") <= b"ab", False)
|
|
|
|
def test_doc(self):
|
|
self.assertIsNotNone(bytearray.__doc__)
|
|
self.assertTrue(bytearray.__doc__.startswith("bytearray("), bytearray.__doc__)
|
|
self.assertIsNotNone(bytes.__doc__)
|
|
self.assertTrue(bytes.__doc__.startswith("bytes("), bytes.__doc__)
|
|
|
|
def test_from_bytearray(self):
|
|
sample = bytes(b"Hello world\n\x80\x81\xfe\xff")
|
|
buf = memoryview(sample)
|
|
b = bytearray(buf)
|
|
self.assertEqual(b, bytearray(sample))
|
|
|
|
@check_bytes_warnings
|
|
def test_to_str(self):
|
|
self.assertEqual(str(b''), "b''")
|
|
self.assertEqual(str(b'x'), "b'x'")
|
|
self.assertEqual(str(b'\x80'), "b'\\x80'")
|
|
self.assertEqual(str(bytearray(b'')), "bytearray(b'')")
|
|
self.assertEqual(str(bytearray(b'x')), "bytearray(b'x')")
|
|
self.assertEqual(str(bytearray(b'\x80')), "bytearray(b'\\x80')")
|
|
|
|
def test_literal(self):
|
|
tests = [
|
|
(b"Wonderful spam", "Wonderful spam"),
|
|
(br"Wonderful spam too", "Wonderful spam too"),
|
|
(b"\xaa\x00\000\200", "\xaa\x00\000\200"),
|
|
(br"\xaa\x00\000\200", r"\xaa\x00\000\200"),
|
|
]
|
|
for b, s in tests:
|
|
self.assertEqual(b, bytearray(s, 'latin-1'))
|
|
for c in range(128, 256):
|
|
self.assertRaises(SyntaxError, eval,
|
|
'b"%s"' % chr(c))
|
|
|
|
def test_translate(self):
|
|
b = b'hello'
|
|
ba = bytearray(b)
|
|
rosetta = bytearray(range(0, 256))
|
|
rosetta[ord('o')] = ord('e')
|
|
c = b.translate(rosetta, b'l')
|
|
self.assertEqual(b, b'hello')
|
|
self.assertEqual(c, b'hee')
|
|
c = ba.translate(rosetta, b'l')
|
|
self.assertEqual(ba, b'hello')
|
|
self.assertEqual(c, b'hee')
|
|
c = b.translate(None, b'e')
|
|
self.assertEqual(c, b'hllo')
|
|
c = ba.translate(None, b'e')
|
|
self.assertEqual(c, b'hllo')
|
|
self.assertRaises(TypeError, b.translate, None, None)
|
|
self.assertRaises(TypeError, ba.translate, None, None)
|
|
|
|
def test_split_bytearray(self):
|
|
self.assertEqual(b'a b'.split(memoryview(b' ')), [b'a', b'b'])
|
|
|
|
def test_rsplit_bytearray(self):
|
|
self.assertEqual(b'a b'.rsplit(memoryview(b' ')), [b'a', b'b'])
|
|
|
|
# Optimizations:
|
|
# __iter__? (optimization)
|
|
# __reversed__? (optimization)
|
|
|
|
# XXX More string methods? (Those that don't use character properties)
|
|
|
|
# There are tests in string_tests.py that are more
|
|
# comprehensive for things like split, partition, etc.
|
|
# Unfortunately they are all bundled with tests that
|
|
# are not appropriate for bytes
|
|
|
|
# I've started porting some of those into bytearray_tests.py, we should port
|
|
# the rest that make sense (the code can be cleaned up to use modern
|
|
# unittest methods at the same time).
|
|
|
|
class BytearrayPEP3137Test(unittest.TestCase,
|
|
test.buffer_tests.MixinBytesBufferCommonTests):
|
|
def marshal(self, x):
|
|
return bytearray(x)
|
|
|
|
def test_returns_new_copy(self):
|
|
val = self.marshal(b'1234')
|
|
# On immutable types these MAY return a reference to themselves
|
|
# but on mutable types like bytearray they MUST return a new copy.
|
|
for methname in ('zfill', 'rjust', 'ljust', 'center'):
|
|
method = getattr(val, methname)
|
|
newval = method(3)
|
|
self.assertEqual(val, newval)
|
|
self.assertTrue(val is not newval,
|
|
methname+' returned self on a mutable object')
|
|
for expr in ('val.split()[0]', 'val.rsplit()[0]',
|
|
'val.partition(".")[0]', 'val.rpartition(".")[2]',
|
|
'val.splitlines()[0]', 'val.replace("", "")'):
|
|
newval = eval(expr)
|
|
self.assertEqual(val, newval)
|
|
self.assertTrue(val is not newval,
|
|
expr+' returned val on a mutable object')
|
|
|
|
class FixedStringTest(test.string_tests.BaseTest):
|
|
|
|
def fixtype(self, obj):
|
|
if isinstance(obj, str):
|
|
return obj.encode("utf-8")
|
|
return super(FixedStringTest, self).fixtype(obj)
|
|
|
|
# Currently the bytes containment testing uses a single integer
|
|
# value. This may not be the final design, but until then the
|
|
# bytes section with in a bytes containment not valid
|
|
def test_contains(self):
|
|
pass
|
|
def test_expandtabs(self):
|
|
pass
|
|
def test_upper(self):
|
|
pass
|
|
def test_lower(self):
|
|
pass
|
|
def test_hash(self):
|
|
# XXX check this out
|
|
pass
|
|
|
|
|
|
class ByteArrayAsStringTest(FixedStringTest):
|
|
type2test = bytearray
|
|
|
|
|
|
class ByteArraySubclass(bytearray):
|
|
pass
|
|
|
|
class ByteArraySubclassTest(unittest.TestCase):
|
|
|
|
def test_basic(self):
|
|
self.assertTrue(issubclass(ByteArraySubclass, bytearray))
|
|
self.assertIsInstance(ByteArraySubclass(), bytearray)
|
|
|
|
a, b = b"abcd", b"efgh"
|
|
_a, _b = ByteArraySubclass(a), ByteArraySubclass(b)
|
|
|
|
# test comparison operators with subclass instances
|
|
self.assertTrue(_a == _a)
|
|
self.assertTrue(_a != _b)
|
|
self.assertTrue(_a < _b)
|
|
self.assertTrue(_a <= _b)
|
|
self.assertTrue(_b >= _a)
|
|
self.assertTrue(_b > _a)
|
|
self.assertTrue(_a is not a)
|
|
|
|
# test concat of subclass instances
|
|
self.assertEqual(a + b, _a + _b)
|
|
self.assertEqual(a + b, a + _b)
|
|
self.assertEqual(a + b, _a + b)
|
|
|
|
# test repeat
|
|
self.assertTrue(a*5 == _a*5)
|
|
|
|
def test_join(self):
|
|
# Make sure join returns a NEW object for single item sequences
|
|
# involving a subclass.
|
|
# Make sure that it is of the appropriate type.
|
|
s1 = ByteArraySubclass(b"abcd")
|
|
s2 = bytearray().join([s1])
|
|
self.assertTrue(s1 is not s2)
|
|
self.assertTrue(type(s2) is bytearray, type(s2))
|
|
|
|
# Test reverse, calling join on subclass
|
|
s3 = s1.join([b"abcd"])
|
|
self.assertTrue(type(s3) is bytearray)
|
|
|
|
def test_pickle(self):
|
|
a = ByteArraySubclass(b"abcd")
|
|
a.x = 10
|
|
a.y = ByteArraySubclass(b"efgh")
|
|
for proto in range(pickle.HIGHEST_PROTOCOL + 1):
|
|
b = pickle.loads(pickle.dumps(a, proto))
|
|
self.assertNotEqual(id(a), id(b))
|
|
self.assertEqual(a, b)
|
|
self.assertEqual(a.x, b.x)
|
|
self.assertEqual(a.y, b.y)
|
|
self.assertEqual(type(a), type(b))
|
|
self.assertEqual(type(a.y), type(b.y))
|
|
|
|
def test_copy(self):
|
|
a = ByteArraySubclass(b"abcd")
|
|
a.x = 10
|
|
a.y = ByteArraySubclass(b"efgh")
|
|
for copy_method in (copy.copy, copy.deepcopy):
|
|
b = copy_method(a)
|
|
self.assertNotEqual(id(a), id(b))
|
|
self.assertEqual(a, b)
|
|
self.assertEqual(a.x, b.x)
|
|
self.assertEqual(a.y, b.y)
|
|
self.assertEqual(type(a), type(b))
|
|
self.assertEqual(type(a.y), type(b.y))
|
|
|
|
def test_init_override(self):
|
|
class subclass(bytearray):
|
|
def __init__(self, newarg=1, *args, **kwargs):
|
|
bytearray.__init__(self, *args, **kwargs)
|
|
x = subclass(4, source=b"abcd")
|
|
self.assertEqual(x, b"abcd")
|
|
x = subclass(newarg=4, source=b"abcd")
|
|
self.assertEqual(x, b"abcd")
|
|
|
|
def test_main():
|
|
#test.test_support.run_unittest(BytesTest)
|
|
#test.test_support.run_unittest(AssortedBytesTest)
|
|
#test.test_support.run_unittest(BytesAsStringTest)
|
|
test.test_support.run_unittest(
|
|
ByteArrayTest,
|
|
ByteArrayAsStringTest,
|
|
ByteArraySubclassTest,
|
|
BytearrayPEP3137Test)
|
|
|
|
if __name__ == "__main__":
|
|
test_main()
|