mirror of
https://github.com/python/cpython.git
synced 2024-12-24 09:15:04 +08:00
a27c5bd2bd
svn+ssh://pythondev@svn.python.org/python/trunk ........ r72979 | philip.jenvey | 2009-05-27 22:58:44 -0700 (Wed, 27 May 2009) | 2 lines explicitly close files ........
129 lines
3.7 KiB
Python
129 lines
3.7 KiB
Python
# Tests universal newline support for both reading and parsing files.
|
|
import io
|
|
import _pyio as pyio
|
|
import unittest
|
|
import os
|
|
import sys
|
|
from test import support
|
|
|
|
if not hasattr(sys.stdin, 'newlines'):
|
|
raise unittest.SkipTest(
|
|
"This Python does not have universal newline support")
|
|
|
|
FATX = 'x' * (2**14)
|
|
|
|
DATA_TEMPLATE = [
|
|
"line1=1",
|
|
"line2='this is a very long line designed to go past any default " +
|
|
"buffer limits that exist in io.py but we also want to test " +
|
|
"the uncommon case, naturally.'",
|
|
"def line3():pass",
|
|
"line4 = '%s'" % FATX,
|
|
]
|
|
|
|
DATA_LF = "\n".join(DATA_TEMPLATE) + "\n"
|
|
DATA_CR = "\r".join(DATA_TEMPLATE) + "\r"
|
|
DATA_CRLF = "\r\n".join(DATA_TEMPLATE) + "\r\n"
|
|
|
|
# Note that DATA_MIXED also tests the ability to recognize a lone \r
|
|
# before end-of-file.
|
|
DATA_MIXED = "\n".join(DATA_TEMPLATE) + "\r"
|
|
DATA_SPLIT = [x + "\n" for x in DATA_TEMPLATE]
|
|
|
|
class TestGenericUnivNewlines(unittest.TestCase):
|
|
# use a class variable DATA to define the data to write to the file
|
|
# and a class variable NEWLINE to set the expected newlines value
|
|
READMODE = 'r'
|
|
WRITEMODE = 'wb'
|
|
|
|
def setUp(self):
|
|
data = self.DATA
|
|
if "b" in self.WRITEMODE:
|
|
data = data.encode("ascii")
|
|
with self.open(support.TESTFN, self.WRITEMODE) as fp:
|
|
fp.write(data)
|
|
|
|
def tearDown(self):
|
|
try:
|
|
os.unlink(support.TESTFN)
|
|
except:
|
|
pass
|
|
|
|
def test_read(self):
|
|
with self.open(support.TESTFN, self.READMODE) as fp:
|
|
data = fp.read()
|
|
self.assertEqual(data, DATA_LF)
|
|
self.assertEqual(repr(fp.newlines), repr(self.NEWLINE))
|
|
|
|
def test_readlines(self):
|
|
with self.open(support.TESTFN, self.READMODE) as fp:
|
|
data = fp.readlines()
|
|
self.assertEqual(data, DATA_SPLIT)
|
|
self.assertEqual(repr(fp.newlines), repr(self.NEWLINE))
|
|
|
|
def test_readline(self):
|
|
with self.open(support.TESTFN, self.READMODE) as fp:
|
|
data = []
|
|
d = fp.readline()
|
|
while d:
|
|
data.append(d)
|
|
d = fp.readline()
|
|
self.assertEqual(data, DATA_SPLIT)
|
|
self.assertEqual(repr(fp.newlines), repr(self.NEWLINE))
|
|
|
|
def test_seek(self):
|
|
with self.open(support.TESTFN, self.READMODE) as fp:
|
|
fp.readline()
|
|
pos = fp.tell()
|
|
data = fp.readlines()
|
|
self.assertEqual(data, DATA_SPLIT[1:])
|
|
fp.seek(pos)
|
|
data = fp.readlines()
|
|
self.assertEqual(data, DATA_SPLIT[1:])
|
|
|
|
|
|
class TestCRNewlines(TestGenericUnivNewlines):
|
|
NEWLINE = '\r'
|
|
DATA = DATA_CR
|
|
|
|
class TestLFNewlines(TestGenericUnivNewlines):
|
|
NEWLINE = '\n'
|
|
DATA = DATA_LF
|
|
|
|
class TestCRLFNewlines(TestGenericUnivNewlines):
|
|
NEWLINE = '\r\n'
|
|
DATA = DATA_CRLF
|
|
|
|
def test_tell(self):
|
|
with self.open(support.TESTFN, self.READMODE) as fp:
|
|
self.assertEqual(repr(fp.newlines), repr(None))
|
|
data = fp.readline()
|
|
pos = fp.tell()
|
|
self.assertEqual(repr(fp.newlines), repr(self.NEWLINE))
|
|
|
|
class TestMixedNewlines(TestGenericUnivNewlines):
|
|
NEWLINE = ('\r', '\n')
|
|
DATA = DATA_MIXED
|
|
|
|
|
|
def test_main():
|
|
base_tests = (TestCRNewlines,
|
|
TestLFNewlines,
|
|
TestCRLFNewlines,
|
|
TestMixedNewlines)
|
|
tests = []
|
|
# Test the C and Python implementations.
|
|
for test in base_tests:
|
|
class CTest(test):
|
|
open = io.open
|
|
CTest.__name__ = "C" + test.__name__
|
|
class PyTest(test):
|
|
open = staticmethod(pyio.open)
|
|
PyTest.__name__ = "Py" + test.__name__
|
|
tests.append(CTest)
|
|
tests.append(PyTest)
|
|
support.run_unittest(*tests)
|
|
|
|
if __name__ == '__main__':
|
|
test_main()
|