mirror of https://github.com/python/cpython
1452 lines
51 KiB
Python
1452 lines
51 KiB
Python
from test import test_support
|
|
import unittest
|
|
import codecs
|
|
import sys, _testcapi, io
|
|
|
|
class Queue(object):
|
|
"""
|
|
queue: write bytes at one end, read bytes from the other end
|
|
"""
|
|
def __init__(self, buffer):
|
|
self._buffer = buffer
|
|
|
|
def write(self, chars):
|
|
self._buffer += chars
|
|
|
|
def read(self, size=-1):
|
|
if size<0:
|
|
s = self._buffer
|
|
self._buffer = self._buffer[:0] # make empty
|
|
return s
|
|
else:
|
|
s = self._buffer[:size]
|
|
self._buffer = self._buffer[size:]
|
|
return s
|
|
|
|
class MixInCheckStateHandling:
|
|
def check_state_handling_decode(self, encoding, u, s):
|
|
for i in range(len(s)+1):
|
|
d = codecs.getincrementaldecoder(encoding)()
|
|
part1 = d.decode(s[:i])
|
|
state = d.getstate()
|
|
self.assert_(isinstance(state[1], int))
|
|
# Check that the condition stated in the documentation for
|
|
# IncrementalDecoder.getstate() holds
|
|
if not state[1]:
|
|
# reset decoder to the default state without anything buffered
|
|
d.setstate((state[0][:0], 0))
|
|
# Feeding the previous input may not produce any output
|
|
self.assert_(not d.decode(state[0]))
|
|
# The decoder must return to the same state
|
|
self.assertEqual(state, d.getstate())
|
|
# Create a new decoder and set it to the state
|
|
# we extracted from the old one
|
|
d = codecs.getincrementaldecoder(encoding)()
|
|
d.setstate(state)
|
|
part2 = d.decode(s[i:], True)
|
|
self.assertEqual(u, part1+part2)
|
|
|
|
def check_state_handling_encode(self, encoding, u, s):
|
|
for i in range(len(u)+1):
|
|
d = codecs.getincrementalencoder(encoding)()
|
|
part1 = d.encode(u[:i])
|
|
state = d.getstate()
|
|
d = codecs.getincrementalencoder(encoding)()
|
|
d.setstate(state)
|
|
part2 = d.encode(u[i:], True)
|
|
self.assertEqual(s, part1+part2)
|
|
|
|
class ReadTest(unittest.TestCase, MixInCheckStateHandling):
|
|
def check_partial(self, input, partialresults):
|
|
# get a StreamReader for the encoding and feed the bytestring version
|
|
# of input to the reader byte by byte. Read every available from
|
|
# the StreamReader and check that the results equal the appropriate
|
|
# entries from partialresults.
|
|
q = Queue(b"")
|
|
r = codecs.getreader(self.encoding)(q)
|
|
result = ""
|
|
for (c, partialresult) in zip(input.encode(self.encoding), partialresults):
|
|
q.write(bytes([c]))
|
|
result += r.read()
|
|
self.assertEqual(result, partialresult)
|
|
# check that there's nothing left in the buffers
|
|
self.assertEqual(r.read(), "")
|
|
self.assertEqual(r.bytebuffer, b"")
|
|
self.assertEqual(r.charbuffer, "")
|
|
|
|
# do the check again, this time using a incremental decoder
|
|
d = codecs.getincrementaldecoder(self.encoding)()
|
|
result = ""
|
|
for (c, partialresult) in zip(input.encode(self.encoding), partialresults):
|
|
result += d.decode(bytes([c]))
|
|
self.assertEqual(result, partialresult)
|
|
# check that there's nothing left in the buffers
|
|
self.assertEqual(d.decode(b"", True), "")
|
|
self.assertEqual(d.buffer, b"")
|
|
|
|
# Check whether the reset method works properly
|
|
d.reset()
|
|
result = ""
|
|
for (c, partialresult) in zip(input.encode(self.encoding), partialresults):
|
|
result += d.decode(bytes([c]))
|
|
self.assertEqual(result, partialresult)
|
|
# check that there's nothing left in the buffers
|
|
self.assertEqual(d.decode(b"", True), "")
|
|
self.assertEqual(d.buffer, b"")
|
|
|
|
# check iterdecode()
|
|
encoded = input.encode(self.encoding)
|
|
self.assertEqual(
|
|
input,
|
|
"".join(codecs.iterdecode([bytes([c]) for c in encoded], self.encoding))
|
|
)
|
|
|
|
def test_readline(self):
|
|
def getreader(input):
|
|
stream = io.BytesIO(input.encode(self.encoding))
|
|
return codecs.getreader(self.encoding)(stream)
|
|
|
|
def readalllines(input, keepends=True, size=None):
|
|
reader = getreader(input)
|
|
lines = []
|
|
while True:
|
|
line = reader.readline(size=size, keepends=keepends)
|
|
if not line:
|
|
break
|
|
lines.append(line)
|
|
return "|".join(lines)
|
|
|
|
s = "foo\nbar\r\nbaz\rspam\u2028eggs"
|
|
sexpected = "foo\n|bar\r\n|baz\r|spam\u2028|eggs"
|
|
sexpectednoends = "foo|bar|baz|spam|eggs"
|
|
self.assertEqual(readalllines(s, True), sexpected)
|
|
self.assertEqual(readalllines(s, False), sexpectednoends)
|
|
self.assertEqual(readalllines(s, True, 10), sexpected)
|
|
self.assertEqual(readalllines(s, False, 10), sexpectednoends)
|
|
|
|
# Test long lines (multiple calls to read() in readline())
|
|
vw = []
|
|
vwo = []
|
|
for (i, lineend) in enumerate("\n \r\n \r \u2028".split()):
|
|
vw.append((i*200)*"\3042" + lineend)
|
|
vwo.append((i*200)*"\3042")
|
|
self.assertEqual(readalllines("".join(vw), True), "".join(vw))
|
|
self.assertEqual(readalllines("".join(vw), False),"".join(vwo))
|
|
|
|
# Test lines where the first read might end with \r, so the
|
|
# reader has to look ahead whether this is a lone \r or a \r\n
|
|
for size in range(80):
|
|
for lineend in "\n \r\n \r \u2028".split():
|
|
s = 10*(size*"a" + lineend + "xxx\n")
|
|
reader = getreader(s)
|
|
for i in range(10):
|
|
self.assertEqual(
|
|
reader.readline(keepends=True),
|
|
size*"a" + lineend,
|
|
)
|
|
reader = getreader(s)
|
|
for i in range(10):
|
|
self.assertEqual(
|
|
reader.readline(keepends=False),
|
|
size*"a",
|
|
)
|
|
|
|
def test_bug1175396(self):
|
|
s = [
|
|
'<%!--===================================================\r\n',
|
|
' BLOG index page: show recent articles,\r\n',
|
|
' today\'s articles, or articles of a specific date.\r\n',
|
|
'========================================================--%>\r\n',
|
|
'<%@inputencoding="ISO-8859-1"%>\r\n',
|
|
'<%@pagetemplate=TEMPLATE.y%>\r\n',
|
|
'<%@import=import frog.util, frog%>\r\n',
|
|
'<%@import=import frog.objects%>\r\n',
|
|
'<%@import=from frog.storageerrors import StorageError%>\r\n',
|
|
'<%\r\n',
|
|
'\r\n',
|
|
'import logging\r\n',
|
|
'log=logging.getLogger("Snakelets.logger")\r\n',
|
|
'\r\n',
|
|
'\r\n',
|
|
'user=self.SessionCtx.user\r\n',
|
|
'storageEngine=self.SessionCtx.storageEngine\r\n',
|
|
'\r\n',
|
|
'\r\n',
|
|
'def readArticlesFromDate(date, count=None):\r\n',
|
|
' entryids=storageEngine.listBlogEntries(date)\r\n',
|
|
' entryids.reverse() # descending\r\n',
|
|
' if count:\r\n',
|
|
' entryids=entryids[:count]\r\n',
|
|
' try:\r\n',
|
|
' return [ frog.objects.BlogEntry.load(storageEngine, date, Id) for Id in entryids ]\r\n',
|
|
' except StorageError,x:\r\n',
|
|
' log.error("Error loading articles: "+str(x))\r\n',
|
|
' self.abort("cannot load articles")\r\n',
|
|
'\r\n',
|
|
'showdate=None\r\n',
|
|
'\r\n',
|
|
'arg=self.Request.getArg()\r\n',
|
|
'if arg=="today":\r\n',
|
|
' #-------------------- TODAY\'S ARTICLES\r\n',
|
|
' self.write("<h2>Today\'s articles</h2>")\r\n',
|
|
' showdate = frog.util.isodatestr() \r\n',
|
|
' entries = readArticlesFromDate(showdate)\r\n',
|
|
'elif arg=="active":\r\n',
|
|
' #-------------------- ACTIVE ARTICLES redirect\r\n',
|
|
' self.Yredirect("active.y")\r\n',
|
|
'elif arg=="login":\r\n',
|
|
' #-------------------- LOGIN PAGE redirect\r\n',
|
|
' self.Yredirect("login.y")\r\n',
|
|
'elif arg=="date":\r\n',
|
|
' #-------------------- ARTICLES OF A SPECIFIC DATE\r\n',
|
|
' showdate = self.Request.getParameter("date")\r\n',
|
|
' self.write("<h2>Articles written on %s</h2>"% frog.util.mediumdatestr(showdate))\r\n',
|
|
' entries = readArticlesFromDate(showdate)\r\n',
|
|
'else:\r\n',
|
|
' #-------------------- RECENT ARTICLES\r\n',
|
|
' self.write("<h2>Recent articles</h2>")\r\n',
|
|
' dates=storageEngine.listBlogEntryDates()\r\n',
|
|
' if dates:\r\n',
|
|
' entries=[]\r\n',
|
|
' SHOWAMOUNT=10\r\n',
|
|
' for showdate in dates:\r\n',
|
|
' entries.extend( readArticlesFromDate(showdate, SHOWAMOUNT-len(entries)) )\r\n',
|
|
' if len(entries)>=SHOWAMOUNT:\r\n',
|
|
' break\r\n',
|
|
' \r\n',
|
|
]
|
|
stream = io.BytesIO("".join(s).encode(self.encoding))
|
|
reader = codecs.getreader(self.encoding)(stream)
|
|
for (i, line) in enumerate(reader):
|
|
self.assertEqual(line, s[i])
|
|
|
|
def test_readlinequeue(self):
|
|
q = Queue(b"")
|
|
writer = codecs.getwriter(self.encoding)(q)
|
|
reader = codecs.getreader(self.encoding)(q)
|
|
|
|
# No lineends
|
|
writer.write("foo\r")
|
|
self.assertEqual(reader.readline(keepends=False), "foo")
|
|
writer.write("\nbar\r")
|
|
self.assertEqual(reader.readline(keepends=False), "")
|
|
self.assertEqual(reader.readline(keepends=False), "bar")
|
|
writer.write("baz")
|
|
self.assertEqual(reader.readline(keepends=False), "baz")
|
|
self.assertEqual(reader.readline(keepends=False), "")
|
|
|
|
# Lineends
|
|
writer.write("foo\r")
|
|
self.assertEqual(reader.readline(keepends=True), "foo\r")
|
|
writer.write("\nbar\r")
|
|
self.assertEqual(reader.readline(keepends=True), "\n")
|
|
self.assertEqual(reader.readline(keepends=True), "bar\r")
|
|
writer.write("baz")
|
|
self.assertEqual(reader.readline(keepends=True), "baz")
|
|
self.assertEqual(reader.readline(keepends=True), "")
|
|
writer.write("foo\r\n")
|
|
self.assertEqual(reader.readline(keepends=True), "foo\r\n")
|
|
|
|
def test_bug1098990_a(self):
|
|
s1 = "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy\r\n"
|
|
s2 = "offending line: ladfj askldfj klasdj fskla dfzaskdj fasklfj laskd fjasklfzzzzaa%whereisthis!!!\r\n"
|
|
s3 = "next line.\r\n"
|
|
|
|
s = (s1+s2+s3).encode(self.encoding)
|
|
stream = io.BytesIO(s)
|
|
reader = codecs.getreader(self.encoding)(stream)
|
|
self.assertEqual(reader.readline(), s1)
|
|
self.assertEqual(reader.readline(), s2)
|
|
self.assertEqual(reader.readline(), s3)
|
|
self.assertEqual(reader.readline(), "")
|
|
|
|
def test_bug1098990_b(self):
|
|
s1 = "aaaaaaaaaaaaaaaaaaaaaaaa\r\n"
|
|
s2 = "bbbbbbbbbbbbbbbbbbbbbbbb\r\n"
|
|
s3 = "stillokay:bbbbxx\r\n"
|
|
s4 = "broken!!!!badbad\r\n"
|
|
s5 = "againokay.\r\n"
|
|
|
|
s = (s1+s2+s3+s4+s5).encode(self.encoding)
|
|
stream = io.BytesIO(s)
|
|
reader = codecs.getreader(self.encoding)(stream)
|
|
self.assertEqual(reader.readline(), s1)
|
|
self.assertEqual(reader.readline(), s2)
|
|
self.assertEqual(reader.readline(), s3)
|
|
self.assertEqual(reader.readline(), s4)
|
|
self.assertEqual(reader.readline(), s5)
|
|
self.assertEqual(reader.readline(), "")
|
|
|
|
class UTF32Test(ReadTest):
|
|
encoding = "utf-32"
|
|
|
|
spamle = (b'\xff\xfe\x00\x00'
|
|
b's\x00\x00\x00p\x00\x00\x00a\x00\x00\x00m\x00\x00\x00'
|
|
b's\x00\x00\x00p\x00\x00\x00a\x00\x00\x00m\x00\x00\x00')
|
|
spambe = (b'\x00\x00\xfe\xff'
|
|
b'\x00\x00\x00s\x00\x00\x00p\x00\x00\x00a\x00\x00\x00m'
|
|
b'\x00\x00\x00s\x00\x00\x00p\x00\x00\x00a\x00\x00\x00m')
|
|
|
|
def test_only_one_bom(self):
|
|
_,_,reader,writer = codecs.lookup(self.encoding)
|
|
# encode some stream
|
|
s = io.BytesIO()
|
|
f = writer(s)
|
|
f.write("spam")
|
|
f.write("spam")
|
|
d = s.getvalue()
|
|
# check whether there is exactly one BOM in it
|
|
self.assert_(d == self.spamle or d == self.spambe)
|
|
# try to read it back
|
|
s = io.BytesIO(d)
|
|
f = reader(s)
|
|
self.assertEquals(f.read(), "spamspam")
|
|
|
|
def test_badbom(self):
|
|
s = io.BytesIO(4*b"\xff")
|
|
f = codecs.getreader(self.encoding)(s)
|
|
self.assertRaises(UnicodeError, f.read)
|
|
|
|
s = io.BytesIO(8*b"\xff")
|
|
f = codecs.getreader(self.encoding)(s)
|
|
self.assertRaises(UnicodeError, f.read)
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u0100\uffff",
|
|
[
|
|
"", # first byte of BOM read
|
|
"", # second byte of BOM read
|
|
"", # third byte of BOM read
|
|
"", # fourth byte of BOM read => byteorder known
|
|
"",
|
|
"",
|
|
"",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100\uffff",
|
|
]
|
|
)
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_32_decode,
|
|
b"\xff", "strict", True)
|
|
|
|
def test_decoder_state(self):
|
|
self.check_state_handling_decode(self.encoding,
|
|
"spamspam", self.spamle)
|
|
self.check_state_handling_decode(self.encoding,
|
|
"spamspam", self.spambe)
|
|
|
|
class UTF32LETest(ReadTest):
|
|
encoding = "utf-32-le"
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u0100\uffff",
|
|
[
|
|
"",
|
|
"",
|
|
"",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100\uffff",
|
|
]
|
|
)
|
|
|
|
def test_simple(self):
|
|
self.assertEqual("\U00010203".encode(self.encoding), b"\x03\x02\x01\x00")
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_32_le_decode,
|
|
b"\xff", "strict", True)
|
|
|
|
class UTF32BETest(ReadTest):
|
|
encoding = "utf-32-be"
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u0100\uffff",
|
|
[
|
|
"",
|
|
"",
|
|
"",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100\uffff",
|
|
]
|
|
)
|
|
|
|
def test_simple(self):
|
|
self.assertEqual("\U00010203".encode(self.encoding), b"\x00\x01\x02\x03")
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_32_be_decode,
|
|
b"\xff", "strict", True)
|
|
|
|
class UTF16Test(ReadTest):
|
|
encoding = "utf-16"
|
|
|
|
spamle = b'\xff\xfes\x00p\x00a\x00m\x00s\x00p\x00a\x00m\x00'
|
|
spambe = b'\xfe\xff\x00s\x00p\x00a\x00m\x00s\x00p\x00a\x00m'
|
|
|
|
def test_only_one_bom(self):
|
|
_,_,reader,writer = codecs.lookup(self.encoding)
|
|
# encode some stream
|
|
s = io.BytesIO()
|
|
f = writer(s)
|
|
f.write("spam")
|
|
f.write("spam")
|
|
d = s.getvalue()
|
|
# check whether there is exactly one BOM in it
|
|
self.assert_(d == self.spamle or d == self.spambe)
|
|
# try to read it back
|
|
s = io.BytesIO(d)
|
|
f = reader(s)
|
|
self.assertEquals(f.read(), "spamspam")
|
|
|
|
def test_badbom(self):
|
|
s = io.BytesIO(b"\xff\xff")
|
|
f = codecs.getreader(self.encoding)(s)
|
|
self.assertRaises(UnicodeError, f.read)
|
|
|
|
s = io.BytesIO(b"\xff\xff\xff\xff")
|
|
f = codecs.getreader(self.encoding)(s)
|
|
self.assertRaises(UnicodeError, f.read)
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u0100\uffff",
|
|
[
|
|
"", # first byte of BOM read
|
|
"", # second byte of BOM read => byteorder known
|
|
"",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100\uffff",
|
|
]
|
|
)
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_16_decode,
|
|
b"\xff", "strict", True)
|
|
|
|
def test_decoder_state(self):
|
|
self.check_state_handling_decode(self.encoding,
|
|
"spamspam", self.spamle)
|
|
self.check_state_handling_decode(self.encoding,
|
|
"spamspam", self.spambe)
|
|
|
|
class UTF16LETest(ReadTest):
|
|
encoding = "utf-16-le"
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u0100\uffff",
|
|
[
|
|
"",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100\uffff",
|
|
]
|
|
)
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_16_le_decode,
|
|
b"\xff", "strict", True)
|
|
|
|
class UTF16BETest(ReadTest):
|
|
encoding = "utf-16-be"
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u0100\uffff",
|
|
[
|
|
"",
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100",
|
|
"\x00\xff\u0100\uffff",
|
|
]
|
|
)
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_16_be_decode,
|
|
b"\xff", "strict", True)
|
|
|
|
class UTF8Test(ReadTest):
|
|
encoding = "utf-8"
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\x00\xff\u07ff\u0800\uffff",
|
|
[
|
|
"\x00",
|
|
"\x00",
|
|
"\x00\xff",
|
|
"\x00\xff",
|
|
"\x00\xff\u07ff",
|
|
"\x00\xff\u07ff",
|
|
"\x00\xff\u07ff",
|
|
"\x00\xff\u07ff\u0800",
|
|
"\x00\xff\u07ff\u0800",
|
|
"\x00\xff\u07ff\u0800",
|
|
"\x00\xff\u07ff\u0800\uffff",
|
|
]
|
|
)
|
|
|
|
def test_decoder_state(self):
|
|
u = "\x00\x7f\x80\xff\u0100\u07ff\u0800\uffff\U0010ffff"
|
|
self.check_state_handling_decode(self.encoding,
|
|
u, u.encode(self.encoding))
|
|
|
|
class UTF7Test(ReadTest):
|
|
encoding = "utf-7"
|
|
|
|
# No test_partial() yet, because UTF-7 doesn't support it.
|
|
|
|
class UTF16ExTest(unittest.TestCase):
|
|
|
|
def test_errors(self):
|
|
self.assertRaises(UnicodeDecodeError, codecs.utf_16_ex_decode, b"\xff", "strict", 0, True)
|
|
|
|
def test_bad_args(self):
|
|
self.assertRaises(TypeError, codecs.utf_16_ex_decode)
|
|
|
|
class ReadBufferTest(unittest.TestCase):
|
|
|
|
def test_array(self):
|
|
import array
|
|
self.assertEqual(
|
|
codecs.readbuffer_encode(array.array("b", b"spam")),
|
|
(b"spam", 4)
|
|
)
|
|
|
|
def test_empty(self):
|
|
self.assertEqual(codecs.readbuffer_encode(""), (b"", 0))
|
|
|
|
def test_bad_args(self):
|
|
self.assertRaises(TypeError, codecs.readbuffer_encode)
|
|
self.assertRaises(TypeError, codecs.readbuffer_encode, 42)
|
|
|
|
class CharBufferTest(unittest.TestCase):
|
|
|
|
def test_string(self):
|
|
self.assertEqual(codecs.charbuffer_encode(b"spam"), (b"spam", 4))
|
|
|
|
def test_empty(self):
|
|
self.assertEqual(codecs.charbuffer_encode(b""), (b"", 0))
|
|
|
|
def test_bad_args(self):
|
|
self.assertRaises(TypeError, codecs.charbuffer_encode)
|
|
self.assertRaises(TypeError, codecs.charbuffer_encode, 42)
|
|
|
|
class UTF8SigTest(ReadTest):
|
|
encoding = "utf-8-sig"
|
|
|
|
def test_partial(self):
|
|
self.check_partial(
|
|
"\ufeff\x00\xff\u07ff\u0800\uffff",
|
|
[
|
|
"",
|
|
"",
|
|
"", # First BOM has been read and skipped
|
|
"",
|
|
"",
|
|
"\ufeff", # Second BOM has been read and emitted
|
|
"\ufeff\x00", # "\x00" read and emitted
|
|
"\ufeff\x00", # First byte of encoded "\xff" read
|
|
"\ufeff\x00\xff", # Second byte of encoded "\xff" read
|
|
"\ufeff\x00\xff", # First byte of encoded "\u07ff" read
|
|
"\ufeff\x00\xff\u07ff", # Second byte of encoded "\u07ff" read
|
|
"\ufeff\x00\xff\u07ff",
|
|
"\ufeff\x00\xff\u07ff",
|
|
"\ufeff\x00\xff\u07ff\u0800",
|
|
"\ufeff\x00\xff\u07ff\u0800",
|
|
"\ufeff\x00\xff\u07ff\u0800",
|
|
"\ufeff\x00\xff\u07ff\u0800\uffff",
|
|
]
|
|
)
|
|
|
|
def test_bug1601501(self):
|
|
# SF bug #1601501: check that the codec works with a buffer
|
|
str(b"\xef\xbb\xbf", "utf-8-sig")
|
|
|
|
def test_bom(self):
|
|
d = codecs.getincrementaldecoder("utf-8-sig")()
|
|
s = "spam"
|
|
self.assertEqual(d.decode(s.encode("utf-8-sig")), s)
|
|
|
|
def test_decoder_state(self):
|
|
u = "\x00\x7f\x80\xff\u0100\u07ff\u0800\uffff\U0010ffff"
|
|
self.check_state_handling_decode(self.encoding,
|
|
u, u.encode(self.encoding))
|
|
|
|
class RecodingTest(unittest.TestCase):
|
|
def test_recoding(self):
|
|
f = io.BytesIO()
|
|
f2 = codecs.EncodedFile(f, "unicode_internal", "utf-8")
|
|
f2.write("a")
|
|
f2.close()
|
|
# Python used to crash on this at exit because of a refcount
|
|
# bug in _codecsmodule.c
|
|
|
|
# From RFC 3492
|
|
punycode_testcases = [
|
|
# A Arabic (Egyptian):
|
|
("\u0644\u064A\u0647\u0645\u0627\u0628\u062A\u0643\u0644"
|
|
"\u0645\u0648\u0634\u0639\u0631\u0628\u064A\u061F",
|
|
b"egbpdaj6bu4bxfgehfvwxn"),
|
|
# B Chinese (simplified):
|
|
("\u4ED6\u4EEC\u4E3A\u4EC0\u4E48\u4E0D\u8BF4\u4E2D\u6587",
|
|
b"ihqwcrb4cv8a8dqg056pqjye"),
|
|
# C Chinese (traditional):
|
|
("\u4ED6\u5011\u7232\u4EC0\u9EBD\u4E0D\u8AAA\u4E2D\u6587",
|
|
b"ihqwctvzc91f659drss3x8bo0yb"),
|
|
# D Czech: Pro<ccaron>prost<ecaron>nemluv<iacute><ccaron>esky
|
|
("\u0050\u0072\u006F\u010D\u0070\u0072\u006F\u0073\u0074"
|
|
"\u011B\u006E\u0065\u006D\u006C\u0075\u0076\u00ED\u010D"
|
|
"\u0065\u0073\u006B\u0079",
|
|
b"Proprostnemluvesky-uyb24dma41a"),
|
|
# E Hebrew:
|
|
("\u05DC\u05DE\u05D4\u05D4\u05DD\u05E4\u05E9\u05D5\u05D8"
|
|
"\u05DC\u05D0\u05DE\u05D3\u05D1\u05E8\u05D9\u05DD\u05E2"
|
|
"\u05D1\u05E8\u05D9\u05EA",
|
|
b"4dbcagdahymbxekheh6e0a7fei0b"),
|
|
# F Hindi (Devanagari):
|
|
("\u092F\u0939\u0932\u094B\u0917\u0939\u093F\u0928\u094D"
|
|
"\u0926\u0940\u0915\u094D\u092F\u094B\u0902\u0928\u0939"
|
|
"\u0940\u0902\u092C\u094B\u0932\u0938\u0915\u0924\u0947"
|
|
"\u0939\u0948\u0902",
|
|
b"i1baa7eci9glrd9b2ae1bj0hfcgg6iyaf8o0a1dig0cd"),
|
|
|
|
#(G) Japanese (kanji and hiragana):
|
|
("\u306A\u305C\u307F\u3093\u306A\u65E5\u672C\u8A9E\u3092"
|
|
"\u8A71\u3057\u3066\u304F\u308C\u306A\u3044\u306E\u304B",
|
|
b"n8jok5ay5dzabd5bym9f0cm5685rrjetr6pdxa"),
|
|
|
|
# (H) Korean (Hangul syllables):
|
|
("\uC138\uACC4\uC758\uBAA8\uB4E0\uC0AC\uB78C\uB4E4\uC774"
|
|
"\uD55C\uAD6D\uC5B4\uB97C\uC774\uD574\uD55C\uB2E4\uBA74"
|
|
"\uC5BC\uB9C8\uB098\uC88B\uC744\uAE4C",
|
|
b"989aomsvi5e83db1d2a355cv1e0vak1dwrv93d5xbh15a0dt30a5j"
|
|
b"psd879ccm6fea98c"),
|
|
|
|
# (I) Russian (Cyrillic):
|
|
("\u043F\u043E\u0447\u0435\u043C\u0443\u0436\u0435\u043E"
|
|
"\u043D\u0438\u043D\u0435\u0433\u043E\u0432\u043E\u0440"
|
|
"\u044F\u0442\u043F\u043E\u0440\u0443\u0441\u0441\u043A"
|
|
"\u0438",
|
|
b"b1abfaaepdrnnbgefbaDotcwatmq2g4l"),
|
|
|
|
# (J) Spanish: Porqu<eacute>nopuedensimplementehablarenEspa<ntilde>ol
|
|
("\u0050\u006F\u0072\u0071\u0075\u00E9\u006E\u006F\u0070"
|
|
"\u0075\u0065\u0064\u0065\u006E\u0073\u0069\u006D\u0070"
|
|
"\u006C\u0065\u006D\u0065\u006E\u0074\u0065\u0068\u0061"
|
|
"\u0062\u006C\u0061\u0072\u0065\u006E\u0045\u0073\u0070"
|
|
"\u0061\u00F1\u006F\u006C",
|
|
b"PorqunopuedensimplementehablarenEspaol-fmd56a"),
|
|
|
|
# (K) Vietnamese:
|
|
# T<adotbelow>isaoh<odotbelow>kh<ocirc>ngth<ecirchookabove>ch\
|
|
# <ihookabove>n<oacute>iti<ecircacute>ngVi<ecircdotbelow>t
|
|
("\u0054\u1EA1\u0069\u0073\u0061\u006F\u0068\u1ECD\u006B"
|
|
"\u0068\u00F4\u006E\u0067\u0074\u0068\u1EC3\u0063\u0068"
|
|
"\u1EC9\u006E\u00F3\u0069\u0074\u0069\u1EBF\u006E\u0067"
|
|
"\u0056\u0069\u1EC7\u0074",
|
|
b"TisaohkhngthchnitingVit-kjcr8268qyxafd2f1b9g"),
|
|
|
|
#(L) 3<nen>B<gumi><kinpachi><sensei>
|
|
("\u0033\u5E74\u0042\u7D44\u91D1\u516B\u5148\u751F",
|
|
b"3B-ww4c5e180e575a65lsy2b"),
|
|
|
|
# (M) <amuro><namie>-with-SUPER-MONKEYS
|
|
("\u5B89\u5BA4\u5948\u7F8E\u6075\u002D\u0077\u0069\u0074"
|
|
"\u0068\u002D\u0053\u0055\u0050\u0045\u0052\u002D\u004D"
|
|
"\u004F\u004E\u004B\u0045\u0059\u0053",
|
|
b"-with-SUPER-MONKEYS-pc58ag80a8qai00g7n9n"),
|
|
|
|
# (N) Hello-Another-Way-<sorezore><no><basho>
|
|
("\u0048\u0065\u006C\u006C\u006F\u002D\u0041\u006E\u006F"
|
|
"\u0074\u0068\u0065\u0072\u002D\u0057\u0061\u0079\u002D"
|
|
"\u305D\u308C\u305E\u308C\u306E\u5834\u6240",
|
|
b"Hello-Another-Way--fc4qua05auwb3674vfr0b"),
|
|
|
|
# (O) <hitotsu><yane><no><shita>2
|
|
("\u3072\u3068\u3064\u5C4B\u6839\u306E\u4E0B\u0032",
|
|
b"2-u9tlzr9756bt3uc0v"),
|
|
|
|
# (P) Maji<de>Koi<suru>5<byou><mae>
|
|
("\u004D\u0061\u006A\u0069\u3067\u004B\u006F\u0069\u3059"
|
|
"\u308B\u0035\u79D2\u524D",
|
|
b"MajiKoi5-783gue6qz075azm5e"),
|
|
|
|
# (Q) <pafii>de<runba>
|
|
("\u30D1\u30D5\u30A3\u30FC\u0064\u0065\u30EB\u30F3\u30D0",
|
|
b"de-jg4avhby1noc0d"),
|
|
|
|
# (R) <sono><supiido><de>
|
|
("\u305D\u306E\u30B9\u30D4\u30FC\u30C9\u3067",
|
|
b"d9juau41awczczp"),
|
|
|
|
# (S) -> $1.00 <-
|
|
("\u002D\u003E\u0020\u0024\u0031\u002E\u0030\u0030\u0020"
|
|
"\u003C\u002D",
|
|
b"-> $1.00 <--")
|
|
]
|
|
|
|
for i in punycode_testcases:
|
|
if len(i)!=2:
|
|
print(repr(i))
|
|
|
|
class PunycodeTest(unittest.TestCase):
|
|
def test_encode(self):
|
|
for uni, puny in punycode_testcases:
|
|
# Need to convert both strings to lower case, since
|
|
# some of the extended encodings use upper case, but our
|
|
# code produces only lower case. Converting just puny to
|
|
# lower is also insufficient, since some of the input characters
|
|
# are upper case.
|
|
self.assertEquals(
|
|
str(uni.encode("punycode"), "ascii").lower(),
|
|
str(puny, "ascii").lower()
|
|
)
|
|
|
|
def test_decode(self):
|
|
for uni, puny in punycode_testcases:
|
|
self.assertEquals(uni, puny.decode("punycode"))
|
|
self.assertEquals(uni, puny.decode("ascii").decode("punycode"))
|
|
|
|
class UnicodeInternalTest(unittest.TestCase):
|
|
def test_bug1251300(self):
|
|
# Decoding with unicode_internal used to not correctly handle "code
|
|
# points" above 0x10ffff on UCS-4 builds.
|
|
if sys.maxunicode > 0xffff:
|
|
ok = [
|
|
(b"\x00\x10\xff\xff", "\U0010ffff"),
|
|
(b"\x00\x00\x01\x01", "\U00000101"),
|
|
(b"", ""),
|
|
]
|
|
not_ok = [
|
|
b"\x7f\xff\xff\xff",
|
|
b"\x80\x00\x00\x00",
|
|
b"\x81\x00\x00\x00",
|
|
b"\x00",
|
|
b"\x00\x00\x00\x00\x00",
|
|
]
|
|
for internal, uni in ok:
|
|
if sys.byteorder == "little":
|
|
internal = bytes(reversed(internal))
|
|
self.assertEquals(uni, internal.decode("unicode_internal"))
|
|
for internal in not_ok:
|
|
if sys.byteorder == "little":
|
|
internal = bytes(reversed(internal))
|
|
self.assertRaises(UnicodeDecodeError, internal.decode,
|
|
"unicode_internal")
|
|
|
|
def test_decode_error_attributes(self):
|
|
if sys.maxunicode > 0xffff:
|
|
try:
|
|
b"\x00\x00\x00\x00\x00\x11\x11\x00".decode("unicode_internal")
|
|
except UnicodeDecodeError as ex:
|
|
self.assertEquals("unicode_internal", ex.encoding)
|
|
self.assertEquals(b"\x00\x00\x00\x00\x00\x11\x11\x00", ex.object)
|
|
self.assertEquals(4, ex.start)
|
|
self.assertEquals(8, ex.end)
|
|
else:
|
|
self.fail()
|
|
|
|
def test_decode_callback(self):
|
|
if sys.maxunicode > 0xffff:
|
|
codecs.register_error("UnicodeInternalTest", codecs.ignore_errors)
|
|
decoder = codecs.getdecoder("unicode_internal")
|
|
ab = "ab".encode("unicode_internal")
|
|
ignored = decoder(bytes("%s\x22\x22\x22\x22%s" % (ab[:4], ab[4:])),
|
|
"UnicodeInternalTest")
|
|
self.assertEquals(("ab", 12), ignored)
|
|
|
|
# From http://www.gnu.org/software/libidn/draft-josefsson-idn-test-vectors.html
|
|
nameprep_tests = [
|
|
# 3.1 Map to nothing.
|
|
(b'foo\xc2\xad\xcd\x8f\xe1\xa0\x86\xe1\xa0\x8bbar'
|
|
b'\xe2\x80\x8b\xe2\x81\xa0baz\xef\xb8\x80\xef\xb8\x88\xef'
|
|
b'\xb8\x8f\xef\xbb\xbf',
|
|
b'foobarbaz'),
|
|
# 3.2 Case folding ASCII U+0043 U+0041 U+0046 U+0045.
|
|
(b'CAFE',
|
|
b'cafe'),
|
|
# 3.3 Case folding 8bit U+00DF (german sharp s).
|
|
# The original test case is bogus; it says \xc3\xdf
|
|
(b'\xc3\x9f',
|
|
b'ss'),
|
|
# 3.4 Case folding U+0130 (turkish capital I with dot).
|
|
(b'\xc4\xb0',
|
|
b'i\xcc\x87'),
|
|
# 3.5 Case folding multibyte U+0143 U+037A.
|
|
(b'\xc5\x83\xcd\xba',
|
|
b'\xc5\x84 \xce\xb9'),
|
|
# 3.6 Case folding U+2121 U+33C6 U+1D7BB.
|
|
# XXX: skip this as it fails in UCS-2 mode
|
|
#('\xe2\x84\xa1\xe3\x8f\x86\xf0\x9d\x9e\xbb',
|
|
# 'telc\xe2\x88\x95kg\xcf\x83'),
|
|
(None, None),
|
|
# 3.7 Normalization of U+006a U+030c U+00A0 U+00AA.
|
|
(b'j\xcc\x8c\xc2\xa0\xc2\xaa',
|
|
b'\xc7\xb0 a'),
|
|
# 3.8 Case folding U+1FB7 and normalization.
|
|
(b'\xe1\xbe\xb7',
|
|
b'\xe1\xbe\xb6\xce\xb9'),
|
|
# 3.9 Self-reverting case folding U+01F0 and normalization.
|
|
# The original test case is bogus, it says `\xc7\xf0'
|
|
(b'\xc7\xb0',
|
|
b'\xc7\xb0'),
|
|
# 3.10 Self-reverting case folding U+0390 and normalization.
|
|
(b'\xce\x90',
|
|
b'\xce\x90'),
|
|
# 3.11 Self-reverting case folding U+03B0 and normalization.
|
|
(b'\xce\xb0',
|
|
b'\xce\xb0'),
|
|
# 3.12 Self-reverting case folding U+1E96 and normalization.
|
|
(b'\xe1\xba\x96',
|
|
b'\xe1\xba\x96'),
|
|
# 3.13 Self-reverting case folding U+1F56 and normalization.
|
|
(b'\xe1\xbd\x96',
|
|
b'\xe1\xbd\x96'),
|
|
# 3.14 ASCII space character U+0020.
|
|
(b' ',
|
|
b' '),
|
|
# 3.15 Non-ASCII 8bit space character U+00A0.
|
|
(b'\xc2\xa0',
|
|
b' '),
|
|
# 3.16 Non-ASCII multibyte space character U+1680.
|
|
(b'\xe1\x9a\x80',
|
|
None),
|
|
# 3.17 Non-ASCII multibyte space character U+2000.
|
|
(b'\xe2\x80\x80',
|
|
b' '),
|
|
# 3.18 Zero Width Space U+200b.
|
|
(b'\xe2\x80\x8b',
|
|
b''),
|
|
# 3.19 Non-ASCII multibyte space character U+3000.
|
|
(b'\xe3\x80\x80',
|
|
b' '),
|
|
# 3.20 ASCII control characters U+0010 U+007F.
|
|
(b'\x10\x7f',
|
|
b'\x10\x7f'),
|
|
# 3.21 Non-ASCII 8bit control character U+0085.
|
|
(b'\xc2\x85',
|
|
None),
|
|
# 3.22 Non-ASCII multibyte control character U+180E.
|
|
(b'\xe1\xa0\x8e',
|
|
None),
|
|
# 3.23 Zero Width No-Break Space U+FEFF.
|
|
(b'\xef\xbb\xbf',
|
|
b''),
|
|
# 3.24 Non-ASCII control character U+1D175.
|
|
(b'\xf0\x9d\x85\xb5',
|
|
None),
|
|
# 3.25 Plane 0 private use character U+F123.
|
|
(b'\xef\x84\xa3',
|
|
None),
|
|
# 3.26 Plane 15 private use character U+F1234.
|
|
(b'\xf3\xb1\x88\xb4',
|
|
None),
|
|
# 3.27 Plane 16 private use character U+10F234.
|
|
(b'\xf4\x8f\x88\xb4',
|
|
None),
|
|
# 3.28 Non-character code point U+8FFFE.
|
|
(b'\xf2\x8f\xbf\xbe',
|
|
None),
|
|
# 3.29 Non-character code point U+10FFFF.
|
|
(b'\xf4\x8f\xbf\xbf',
|
|
None),
|
|
# 3.30 Surrogate code U+DF42.
|
|
(b'\xed\xbd\x82',
|
|
None),
|
|
# 3.31 Non-plain text character U+FFFD.
|
|
(b'\xef\xbf\xbd',
|
|
None),
|
|
# 3.32 Ideographic description character U+2FF5.
|
|
(b'\xe2\xbf\xb5',
|
|
None),
|
|
# 3.33 Display property character U+0341.
|
|
(b'\xcd\x81',
|
|
b'\xcc\x81'),
|
|
# 3.34 Left-to-right mark U+200E.
|
|
(b'\xe2\x80\x8e',
|
|
None),
|
|
# 3.35 Deprecated U+202A.
|
|
(b'\xe2\x80\xaa',
|
|
None),
|
|
# 3.36 Language tagging character U+E0001.
|
|
(b'\xf3\xa0\x80\x81',
|
|
None),
|
|
# 3.37 Language tagging character U+E0042.
|
|
(b'\xf3\xa0\x81\x82',
|
|
None),
|
|
# 3.38 Bidi: RandALCat character U+05BE and LCat characters.
|
|
(b'foo\xd6\xbebar',
|
|
None),
|
|
# 3.39 Bidi: RandALCat character U+FD50 and LCat characters.
|
|
(b'foo\xef\xb5\x90bar',
|
|
None),
|
|
# 3.40 Bidi: RandALCat character U+FB38 and LCat characters.
|
|
(b'foo\xef\xb9\xb6bar',
|
|
b'foo \xd9\x8ebar'),
|
|
# 3.41 Bidi: RandALCat without trailing RandALCat U+0627 U+0031.
|
|
(b'\xd8\xa71',
|
|
None),
|
|
# 3.42 Bidi: RandALCat character U+0627 U+0031 U+0628.
|
|
(b'\xd8\xa71\xd8\xa8',
|
|
b'\xd8\xa71\xd8\xa8'),
|
|
# 3.43 Unassigned code point U+E0002.
|
|
# Skip this test as we allow unassigned
|
|
#(b'\xf3\xa0\x80\x82',
|
|
# None),
|
|
(None, None),
|
|
# 3.44 Larger test (shrinking).
|
|
# Original test case reads \xc3\xdf
|
|
(b'X\xc2\xad\xc3\x9f\xc4\xb0\xe2\x84\xa1j\xcc\x8c\xc2\xa0\xc2'
|
|
b'\xaa\xce\xb0\xe2\x80\x80',
|
|
b'xssi\xcc\x87tel\xc7\xb0 a\xce\xb0 '),
|
|
# 3.45 Larger test (expanding).
|
|
# Original test case reads \xc3\x9f
|
|
(b'X\xc3\x9f\xe3\x8c\x96\xc4\xb0\xe2\x84\xa1\xe2\x92\x9f\xe3\x8c'
|
|
b'\x80',
|
|
b'xss\xe3\x82\xad\xe3\x83\xad\xe3\x83\xa1\xe3\x83\xbc\xe3'
|
|
b'\x83\x88\xe3\x83\xabi\xcc\x87tel\x28d\x29\xe3\x82'
|
|
b'\xa2\xe3\x83\x91\xe3\x83\xbc\xe3\x83\x88')
|
|
]
|
|
|
|
|
|
class NameprepTest(unittest.TestCase):
|
|
def test_nameprep(self):
|
|
from encodings.idna import nameprep
|
|
for pos, (orig, prepped) in enumerate(nameprep_tests):
|
|
if orig is None:
|
|
# Skipped
|
|
continue
|
|
# The Unicode strings are given in UTF-8
|
|
orig = str(orig, "utf-8")
|
|
if prepped is None:
|
|
# Input contains prohibited characters
|
|
self.assertRaises(UnicodeError, nameprep, orig)
|
|
else:
|
|
prepped = str(prepped, "utf-8")
|
|
try:
|
|
self.assertEquals(nameprep(orig), prepped)
|
|
except Exception as e:
|
|
raise test_support.TestFailed("Test 3.%d: %s" % (pos+1, str(e)))
|
|
|
|
class IDNACodecTest(unittest.TestCase):
|
|
def test_builtin_decode(self):
|
|
self.assertEquals(str(b"python.org", "idna"), "python.org")
|
|
self.assertEquals(str(b"python.org.", "idna"), "python.org.")
|
|
self.assertEquals(str(b"xn--pythn-mua.org", "idna"), "pyth\xf6n.org")
|
|
self.assertEquals(str(b"xn--pythn-mua.org.", "idna"), "pyth\xf6n.org.")
|
|
|
|
def test_builtin_encode(self):
|
|
self.assertEquals("python.org".encode("idna"), b"python.org")
|
|
self.assertEquals("python.org.".encode("idna"), b"python.org.")
|
|
self.assertEquals("pyth\xf6n.org".encode("idna"), b"xn--pythn-mua.org")
|
|
self.assertEquals("pyth\xf6n.org.".encode("idna"), b"xn--pythn-mua.org.")
|
|
|
|
def test_stream(self):
|
|
r = codecs.getreader("idna")(io.BytesIO(b"abc"))
|
|
r.read(3)
|
|
self.assertEquals(r.read(), "")
|
|
|
|
def test_incremental_decode(self):
|
|
self.assertEquals(
|
|
"".join(codecs.iterdecode((bytes([c]) for c in b"python.org"), "idna")),
|
|
"python.org"
|
|
)
|
|
self.assertEquals(
|
|
"".join(codecs.iterdecode((bytes([c]) for c in b"python.org."), "idna")),
|
|
"python.org."
|
|
)
|
|
self.assertEquals(
|
|
"".join(codecs.iterdecode((bytes([c]) for c in b"xn--pythn-mua.org."), "idna")),
|
|
"pyth\xf6n.org."
|
|
)
|
|
self.assertEquals(
|
|
"".join(codecs.iterdecode((bytes([c]) for c in b"xn--pythn-mua.org."), "idna")),
|
|
"pyth\xf6n.org."
|
|
)
|
|
|
|
decoder = codecs.getincrementaldecoder("idna")()
|
|
self.assertEquals(decoder.decode(b"xn--xam", ), "")
|
|
self.assertEquals(decoder.decode(b"ple-9ta.o", ), "\xe4xample.")
|
|
self.assertEquals(decoder.decode(b"rg"), "")
|
|
self.assertEquals(decoder.decode(b"", True), "org")
|
|
|
|
decoder.reset()
|
|
self.assertEquals(decoder.decode(b"xn--xam", ), "")
|
|
self.assertEquals(decoder.decode(b"ple-9ta.o", ), "\xe4xample.")
|
|
self.assertEquals(decoder.decode(b"rg."), "org.")
|
|
self.assertEquals(decoder.decode(b"", True), "")
|
|
|
|
def test_incremental_encode(self):
|
|
self.assertEquals(
|
|
b"".join(codecs.iterencode("python.org", "idna")),
|
|
b"python.org"
|
|
)
|
|
self.assertEquals(
|
|
b"".join(codecs.iterencode("python.org.", "idna")),
|
|
b"python.org."
|
|
)
|
|
self.assertEquals(
|
|
b"".join(codecs.iterencode("pyth\xf6n.org.", "idna")),
|
|
b"xn--pythn-mua.org."
|
|
)
|
|
self.assertEquals(
|
|
b"".join(codecs.iterencode("pyth\xf6n.org.", "idna")),
|
|
b"xn--pythn-mua.org."
|
|
)
|
|
|
|
encoder = codecs.getincrementalencoder("idna")()
|
|
self.assertEquals(encoder.encode("\xe4x"), b"")
|
|
self.assertEquals(encoder.encode("ample.org"), b"xn--xample-9ta.")
|
|
self.assertEquals(encoder.encode("", True), b"org")
|
|
|
|
encoder.reset()
|
|
self.assertEquals(encoder.encode("\xe4x"), b"")
|
|
self.assertEquals(encoder.encode("ample.org."), b"xn--xample-9ta.org.")
|
|
self.assertEquals(encoder.encode("", True), b"")
|
|
|
|
class CodecsModuleTest(unittest.TestCase):
|
|
|
|
def test_decode(self):
|
|
self.assertEquals(codecs.decode(b'\xe4\xf6\xfc', 'latin-1'),
|
|
'\xe4\xf6\xfc')
|
|
self.assertRaises(TypeError, codecs.decode)
|
|
self.assertEquals(codecs.decode(b'abc'), 'abc')
|
|
self.assertRaises(UnicodeDecodeError, codecs.decode, b'\xff', 'ascii')
|
|
|
|
def test_encode(self):
|
|
self.assertEquals(codecs.encode('\xe4\xf6\xfc', 'latin-1'),
|
|
b'\xe4\xf6\xfc')
|
|
self.assertRaises(TypeError, codecs.encode)
|
|
self.assertRaises(LookupError, codecs.encode, "foo", "__spam__")
|
|
self.assertEquals(codecs.encode('abc'), b'abc')
|
|
self.assertRaises(UnicodeEncodeError, codecs.encode, '\xffff', 'ascii')
|
|
|
|
def test_register(self):
|
|
self.assertRaises(TypeError, codecs.register)
|
|
self.assertRaises(TypeError, codecs.register, 42)
|
|
|
|
def test_lookup(self):
|
|
self.assertRaises(TypeError, codecs.lookup)
|
|
self.assertRaises(LookupError, codecs.lookup, "__spam__")
|
|
self.assertRaises(LookupError, codecs.lookup, " ")
|
|
|
|
def test_getencoder(self):
|
|
self.assertRaises(TypeError, codecs.getencoder)
|
|
self.assertRaises(LookupError, codecs.getencoder, "__spam__")
|
|
|
|
def test_getdecoder(self):
|
|
self.assertRaises(TypeError, codecs.getdecoder)
|
|
self.assertRaises(LookupError, codecs.getdecoder, "__spam__")
|
|
|
|
def test_getreader(self):
|
|
self.assertRaises(TypeError, codecs.getreader)
|
|
self.assertRaises(LookupError, codecs.getreader, "__spam__")
|
|
|
|
def test_getwriter(self):
|
|
self.assertRaises(TypeError, codecs.getwriter)
|
|
self.assertRaises(LookupError, codecs.getwriter, "__spam__")
|
|
|
|
class StreamReaderTest(unittest.TestCase):
|
|
|
|
def setUp(self):
|
|
self.reader = codecs.getreader('utf-8')
|
|
self.stream = io.BytesIO(b'\xed\x95\x9c\n\xea\xb8\x80')
|
|
|
|
def test_readlines(self):
|
|
f = self.reader(self.stream)
|
|
self.assertEquals(f.readlines(), ['\ud55c\n', '\uae00'])
|
|
|
|
class EncodedFileTest(unittest.TestCase):
|
|
|
|
def test_basic(self):
|
|
f = io.BytesIO(b'\xed\x95\x9c\n\xea\xb8\x80')
|
|
ef = codecs.EncodedFile(f, 'utf-16-le', 'utf-8')
|
|
self.assertEquals(ef.read(), b'\\\xd5\n\x00\x00\xae')
|
|
|
|
f = io.BytesIO()
|
|
ef = codecs.EncodedFile(f, 'utf-8', 'latin1')
|
|
ef.write(b'\xc3\xbc')
|
|
self.assertEquals(f.getvalue(), b'\xfc')
|
|
|
|
all_unicode_encodings = [
|
|
"ascii",
|
|
"big5",
|
|
"big5hkscs",
|
|
"charmap",
|
|
"cp037",
|
|
"cp1006",
|
|
"cp1026",
|
|
"cp1140",
|
|
"cp1250",
|
|
"cp1251",
|
|
"cp1252",
|
|
"cp1253",
|
|
"cp1254",
|
|
"cp1255",
|
|
"cp1256",
|
|
"cp1257",
|
|
"cp1258",
|
|
"cp424",
|
|
"cp437",
|
|
"cp500",
|
|
"cp737",
|
|
"cp775",
|
|
"cp850",
|
|
"cp852",
|
|
"cp855",
|
|
"cp856",
|
|
"cp857",
|
|
"cp860",
|
|
"cp861",
|
|
"cp862",
|
|
"cp863",
|
|
"cp864",
|
|
"cp865",
|
|
"cp866",
|
|
"cp869",
|
|
"cp874",
|
|
"cp875",
|
|
"cp932",
|
|
"cp949",
|
|
"cp950",
|
|
"euc_jis_2004",
|
|
"euc_jisx0213",
|
|
"euc_jp",
|
|
"euc_kr",
|
|
"gb18030",
|
|
"gb2312",
|
|
"gbk",
|
|
"hp_roman8",
|
|
"hz",
|
|
"idna",
|
|
"iso2022_jp",
|
|
"iso2022_jp_1",
|
|
"iso2022_jp_2",
|
|
"iso2022_jp_2004",
|
|
"iso2022_jp_3",
|
|
"iso2022_jp_ext",
|
|
"iso2022_kr",
|
|
"iso8859_1",
|
|
"iso8859_10",
|
|
"iso8859_11",
|
|
"iso8859_13",
|
|
"iso8859_14",
|
|
"iso8859_15",
|
|
"iso8859_16",
|
|
"iso8859_2",
|
|
"iso8859_3",
|
|
"iso8859_4",
|
|
"iso8859_5",
|
|
"iso8859_6",
|
|
"iso8859_7",
|
|
"iso8859_8",
|
|
"iso8859_9",
|
|
"johab",
|
|
"koi8_r",
|
|
"koi8_u",
|
|
"latin_1",
|
|
"mac_cyrillic",
|
|
"mac_greek",
|
|
"mac_iceland",
|
|
"mac_latin2",
|
|
"mac_roman",
|
|
"mac_turkish",
|
|
"palmos",
|
|
"ptcp154",
|
|
"punycode",
|
|
"raw_unicode_escape",
|
|
"shift_jis",
|
|
"shift_jis_2004",
|
|
"shift_jisx0213",
|
|
"tis_620",
|
|
"unicode_escape",
|
|
"unicode_internal",
|
|
"utf_16",
|
|
"utf_16_be",
|
|
"utf_16_le",
|
|
"utf_7",
|
|
"utf_8",
|
|
]
|
|
|
|
if hasattr(codecs, "mbcs_encode"):
|
|
all_unicode_encodings.append("mbcs")
|
|
|
|
# The following encoding is not tested, because it's not supposed
|
|
# to work:
|
|
# "undefined"
|
|
|
|
# The following encodings don't work in stateful mode
|
|
broken_unicode_with_streams = [
|
|
"punycode",
|
|
"unicode_internal"
|
|
]
|
|
broken_incremental_coders = broken_unicode_with_streams + [
|
|
"idna",
|
|
]
|
|
|
|
# The following encodings only support "strict" mode
|
|
only_strict_mode = [
|
|
"idna",
|
|
]
|
|
|
|
class BasicUnicodeTest(unittest.TestCase, MixInCheckStateHandling):
|
|
def test_basics(self):
|
|
s = "abc123" # all codecs should be able to encode these
|
|
for encoding in all_unicode_encodings:
|
|
name = codecs.lookup(encoding).name
|
|
if encoding.endswith("_codec"):
|
|
name += "_codec"
|
|
elif encoding == "latin_1":
|
|
name = "latin_1"
|
|
self.assertEqual(encoding.replace("_", "-"), name.replace("_", "-"))
|
|
(b, size) = codecs.getencoder(encoding)(s)
|
|
if encoding != "unicode_internal":
|
|
self.assertEqual(size, len(s), "%r != %r (encoding=%r)" % (size, len(s), encoding))
|
|
(chars, size) = codecs.getdecoder(encoding)(b)
|
|
self.assertEqual(chars, s, "%r != %r (encoding=%r)" % (chars, s, encoding))
|
|
|
|
if encoding not in broken_unicode_with_streams:
|
|
# check stream reader/writer
|
|
q = Queue(b"")
|
|
writer = codecs.getwriter(encoding)(q)
|
|
encodedresult = b""
|
|
for c in s:
|
|
writer.write(c)
|
|
encodedresult += q.read()
|
|
q = Queue(b"")
|
|
reader = codecs.getreader(encoding)(q)
|
|
decodedresult = ""
|
|
for c in encodedresult:
|
|
q.write(bytes([c]))
|
|
decodedresult += reader.read()
|
|
self.assertEqual(decodedresult, s, "%r != %r (encoding=%r)" % (decodedresult, s, encoding))
|
|
|
|
if encoding not in broken_incremental_coders:
|
|
# check incremental decoder/encoder (fetched via the Python
|
|
# and C API) and iterencode()/iterdecode()
|
|
try:
|
|
encoder = codecs.getincrementalencoder(encoding)()
|
|
cencoder = _testcapi.codec_incrementalencoder(encoding)
|
|
except LookupError: # no IncrementalEncoder
|
|
pass
|
|
else:
|
|
# check incremental decoder/encoder
|
|
encodedresult = b""
|
|
for c in s:
|
|
encodedresult += encoder.encode(c)
|
|
encodedresult += encoder.encode("", True)
|
|
decoder = codecs.getincrementaldecoder(encoding)()
|
|
decodedresult = ""
|
|
for c in encodedresult:
|
|
decodedresult += decoder.decode(bytes([c]))
|
|
decodedresult += decoder.decode(b"", True)
|
|
self.assertEqual(decodedresult, s, "%r != %r (encoding=%r)" % (decodedresult, s, encoding))
|
|
|
|
# check C API
|
|
encodedresult = b""
|
|
for c in s:
|
|
encodedresult += cencoder.encode(c)
|
|
encodedresult += cencoder.encode("", True)
|
|
cdecoder = _testcapi.codec_incrementaldecoder(encoding)
|
|
decodedresult = ""
|
|
for c in encodedresult:
|
|
decodedresult += cdecoder.decode(bytes([c]))
|
|
decodedresult += cdecoder.decode(b"", True)
|
|
self.assertEqual(decodedresult, s, "%r != %r (encoding=%r)" % (decodedresult, s, encoding))
|
|
|
|
# check iterencode()/iterdecode()
|
|
result = "".join(codecs.iterdecode(codecs.iterencode(s, encoding), encoding))
|
|
self.assertEqual(result, s, "%r != %r (encoding=%r)" % (result, s, encoding))
|
|
|
|
# check iterencode()/iterdecode() with empty string
|
|
result = "".join(codecs.iterdecode(codecs.iterencode("", encoding), encoding))
|
|
self.assertEqual(result, "")
|
|
|
|
if encoding not in only_strict_mode:
|
|
# check incremental decoder/encoder with errors argument
|
|
try:
|
|
encoder = codecs.getincrementalencoder(encoding)("ignore")
|
|
cencoder = _testcapi.codec_incrementalencoder(encoding, "ignore")
|
|
except LookupError: # no IncrementalEncoder
|
|
pass
|
|
else:
|
|
encodedresult = b"".join(encoder.encode(c) for c in s)
|
|
decoder = codecs.getincrementaldecoder(encoding)("ignore")
|
|
decodedresult = "".join(decoder.decode(bytes([c])) for c in encodedresult)
|
|
self.assertEqual(decodedresult, s, "%r != %r (encoding=%r)" % (decodedresult, s, encoding))
|
|
|
|
encodedresult = b"".join(cencoder.encode(c) for c in s)
|
|
cdecoder = _testcapi.codec_incrementaldecoder(encoding, "ignore")
|
|
decodedresult = "".join(cdecoder.decode(bytes([c])) for c in encodedresult)
|
|
self.assertEqual(decodedresult, s, "%r != %r (encoding=%r)" % (decodedresult, s, encoding))
|
|
|
|
def test_seek(self):
|
|
# all codecs should be able to encode these
|
|
s = "%s\n%s\n" % (100*"abc123", 100*"def456")
|
|
for encoding in all_unicode_encodings:
|
|
if encoding == "idna": # FIXME: See SF bug #1163178
|
|
continue
|
|
if encoding in broken_unicode_with_streams:
|
|
continue
|
|
reader = codecs.getreader(encoding)(io.BytesIO(s.encode(encoding)))
|
|
for t in range(5):
|
|
# Test that calling seek resets the internal codec state and buffers
|
|
reader.seek(0, 0)
|
|
data = reader.read()
|
|
self.assertEqual(s, data)
|
|
|
|
def test_bad_decode_args(self):
|
|
for encoding in all_unicode_encodings:
|
|
decoder = codecs.getdecoder(encoding)
|
|
self.assertRaises(TypeError, decoder)
|
|
if encoding not in ("idna", "punycode"):
|
|
self.assertRaises(TypeError, decoder, 42)
|
|
|
|
def test_bad_encode_args(self):
|
|
for encoding in all_unicode_encodings:
|
|
encoder = codecs.getencoder(encoding)
|
|
self.assertRaises(TypeError, encoder)
|
|
|
|
def test_encoding_map_type_initialized(self):
|
|
from encodings import cp1140
|
|
# This used to crash, we are only verifying there's no crash.
|
|
table_type = type(cp1140.encoding_table)
|
|
self.assertEqual(table_type, table_type)
|
|
|
|
def test_decoder_state(self):
|
|
# Check that getstate() and setstate() handle the state properly
|
|
u = "abc123"
|
|
for encoding in all_unicode_encodings:
|
|
if encoding not in broken_incremental_coders:
|
|
self.check_state_handling_decode(encoding, u, u.encode(encoding))
|
|
self.check_state_handling_encode(encoding, u, u.encode(encoding))
|
|
|
|
class CharmapTest(unittest.TestCase):
|
|
def test_decode_with_string_map(self):
|
|
self.assertEquals(
|
|
codecs.charmap_decode(b"\x00\x01\x02", "strict", "abc"),
|
|
("abc", 3)
|
|
)
|
|
|
|
self.assertEquals(
|
|
codecs.charmap_decode(b"\x00\x01\x02", "replace", "ab"),
|
|
("ab\ufffd", 3)
|
|
)
|
|
|
|
self.assertEquals(
|
|
codecs.charmap_decode(b"\x00\x01\x02", "replace", "ab\ufffe"),
|
|
("ab\ufffd", 3)
|
|
)
|
|
|
|
self.assertEquals(
|
|
codecs.charmap_decode(b"\x00\x01\x02", "ignore", "ab"),
|
|
("ab", 3)
|
|
)
|
|
|
|
self.assertEquals(
|
|
codecs.charmap_decode(b"\x00\x01\x02", "ignore", "ab\ufffe"),
|
|
("ab", 3)
|
|
)
|
|
|
|
allbytes = bytes(range(256))
|
|
self.assertEquals(
|
|
codecs.charmap_decode(allbytes, "ignore", ""),
|
|
("", len(allbytes))
|
|
)
|
|
|
|
class WithStmtTest(unittest.TestCase):
|
|
def test_encodedfile(self):
|
|
f = io.BytesIO(b"\xc3\xbc")
|
|
with codecs.EncodedFile(f, "latin-1", "utf-8") as ef:
|
|
self.assertEquals(ef.read(), b"\xfc")
|
|
|
|
def test_streamreaderwriter(self):
|
|
f = io.BytesIO(b"\xc3\xbc")
|
|
info = codecs.lookup("utf-8")
|
|
with codecs.StreamReaderWriter(f, info.streamreader,
|
|
info.streamwriter, 'strict') as srw:
|
|
self.assertEquals(srw.read(), "\xfc")
|
|
|
|
|
|
def test_main():
|
|
test_support.run_unittest(
|
|
UTF32Test,
|
|
UTF32LETest,
|
|
UTF32BETest,
|
|
UTF16Test,
|
|
UTF16LETest,
|
|
UTF16BETest,
|
|
UTF8Test,
|
|
UTF8SigTest,
|
|
UTF7Test,
|
|
UTF16ExTest,
|
|
ReadBufferTest,
|
|
CharBufferTest,
|
|
RecodingTest,
|
|
PunycodeTest,
|
|
UnicodeInternalTest,
|
|
NameprepTest,
|
|
IDNACodecTest,
|
|
CodecsModuleTest,
|
|
StreamReaderTest,
|
|
EncodedFileTest,
|
|
BasicUnicodeTest,
|
|
CharmapTest,
|
|
WithStmtTest,
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
test_main()
|