mirror of
https://github.com/python/cpython.git
synced 2024-11-25 19:03:49 +08:00
89fad14944
svn+ssh://pythondev@svn.python.org/python/trunk ........ r78018 | georg.brandl | 2010-02-06 11:08:21 +0100 (Sa, 06 Feb 2010) | 1 line #7864: make deprecation notices a bit clearer. ........ r78035 | georg.brandl | 2010-02-06 23:44:17 +0100 (Sa, 06 Feb 2010) | 1 line Fix duplicate import. ........ r78036 | georg.brandl | 2010-02-06 23:49:47 +0100 (Sa, 06 Feb 2010) | 1 line Remove unused import. ........ r78037 | georg.brandl | 2010-02-06 23:59:15 +0100 (Sa, 06 Feb 2010) | 1 line No need to assign the results of expressions used only for side effects. ........ r78038 | georg.brandl | 2010-02-07 00:02:29 +0100 (So, 07 Feb 2010) | 1 line Add a missing import. ........ r78039 | georg.brandl | 2010-02-07 00:06:24 +0100 (So, 07 Feb 2010) | 1 line Add missing imports. ........ r78040 | georg.brandl | 2010-02-07 00:08:00 +0100 (So, 07 Feb 2010) | 1 line Fix a few UnboundLocalErrors in test_long. ........ r78042 | georg.brandl | 2010-02-07 00:12:12 +0100 (So, 07 Feb 2010) | 1 line Add missing import. ........ r78043 | georg.brandl | 2010-02-07 00:12:19 +0100 (So, 07 Feb 2010) | 1 line Remove duplicate test method. ........ r78046 | georg.brandl | 2010-02-07 00:18:00 +0100 (So, 07 Feb 2010) | 1 line Fix various missing import/unbound name errors. ........ r78048 | georg.brandl | 2010-02-07 00:23:45 +0100 (So, 07 Feb 2010) | 1 line We heard you like test failures so we put unbound locals in your test so that you can fail while you fail. ........ r78049 | georg.brandl | 2010-02-07 00:33:33 +0100 (So, 07 Feb 2010) | 1 line Fix import/access for some identifiers. _TestSharedCTypes does not seem to be executed? ........ r78050 | georg.brandl | 2010-02-07 00:34:10 +0100 (So, 07 Feb 2010) | 1 line Fix more unbound locals in code paths that do not seem to be used. ........ r78051 | georg.brandl | 2010-02-07 00:53:52 +0100 (So, 07 Feb 2010) | 1 line Add missing import when running these tests standalone. ........ r78052 | georg.brandl | 2010-02-07 00:54:04 +0100 (So, 07 Feb 2010) | 1 line Add missing import when running these tests standalone. ........ r78054 | georg.brandl | 2010-02-07 00:58:25 +0100 (So, 07 Feb 2010) | 1 line Add missing import. ........ r78059 | georg.brandl | 2010-02-07 12:34:15 +0100 (So, 07 Feb 2010) | 1 line Use "regexp" consistently. ........ r78075 | georg.brandl | 2010-02-07 13:16:12 +0100 (So, 07 Feb 2010) | 1 line Fix another duplicated test method. ........ r78076 | georg.brandl | 2010-02-07 13:19:43 +0100 (So, 07 Feb 2010) | 1 line Fix wrong usage of "except X, Y:". ........ r78077 | georg.brandl | 2010-02-07 13:25:50 +0100 (So, 07 Feb 2010) | 1 line Fix two redefined test methods. ........ r78078 | georg.brandl | 2010-02-07 13:27:06 +0100 (So, 07 Feb 2010) | 1 line Fix a redefined test method. ........ r78079 | georg.brandl | 2010-02-07 13:34:26 +0100 (So, 07 Feb 2010) | 1 line Add a minimal test for fnmatchcase(). ........ r78080 | georg.brandl | 2010-02-07 13:55:12 +0100 (So, 07 Feb 2010) | 1 line Remove duplicate test method. ........
948 lines
36 KiB
Python
948 lines
36 KiB
Python
# -*- coding: utf-8 -*-
|
|
# Copyright (C) 2001,2002 Python Software Foundation
|
|
# csv package unit tests
|
|
|
|
import io
|
|
import sys
|
|
import os
|
|
import unittest
|
|
from io import StringIO
|
|
from tempfile import TemporaryFile
|
|
import csv
|
|
import gc
|
|
from test import support
|
|
|
|
class Test_Csv(unittest.TestCase):
|
|
"""
|
|
Test the underlying C csv parser in ways that are not appropriate
|
|
from the high level interface. Further tests of this nature are done
|
|
in TestDialectRegistry.
|
|
"""
|
|
def _test_arg_valid(self, ctor, arg):
|
|
self.assertRaises(TypeError, ctor)
|
|
self.assertRaises(TypeError, ctor, None)
|
|
self.assertRaises(TypeError, ctor, arg, bad_attr = 0)
|
|
self.assertRaises(TypeError, ctor, arg, delimiter = 0)
|
|
self.assertRaises(TypeError, ctor, arg, delimiter = 'XX')
|
|
self.assertRaises(csv.Error, ctor, arg, 'foo')
|
|
self.assertRaises(TypeError, ctor, arg, delimiter=None)
|
|
self.assertRaises(TypeError, ctor, arg, delimiter=1)
|
|
self.assertRaises(TypeError, ctor, arg, quotechar=1)
|
|
self.assertRaises(TypeError, ctor, arg, lineterminator=None)
|
|
self.assertRaises(TypeError, ctor, arg, lineterminator=1)
|
|
self.assertRaises(TypeError, ctor, arg, quoting=None)
|
|
self.assertRaises(TypeError, ctor, arg,
|
|
quoting=csv.QUOTE_ALL, quotechar='')
|
|
self.assertRaises(TypeError, ctor, arg,
|
|
quoting=csv.QUOTE_ALL, quotechar=None)
|
|
|
|
def test_reader_arg_valid(self):
|
|
self._test_arg_valid(csv.reader, [])
|
|
|
|
def test_writer_arg_valid(self):
|
|
self._test_arg_valid(csv.writer, StringIO())
|
|
|
|
def _test_default_attrs(self, ctor, *args):
|
|
obj = ctor(*args)
|
|
# Check defaults
|
|
self.assertEqual(obj.dialect.delimiter, ',')
|
|
self.assertEqual(obj.dialect.doublequote, True)
|
|
self.assertEqual(obj.dialect.escapechar, None)
|
|
self.assertEqual(obj.dialect.lineterminator, "\r\n")
|
|
self.assertEqual(obj.dialect.quotechar, '"')
|
|
self.assertEqual(obj.dialect.quoting, csv.QUOTE_MINIMAL)
|
|
self.assertEqual(obj.dialect.skipinitialspace, False)
|
|
self.assertEqual(obj.dialect.strict, False)
|
|
# Try deleting or changing attributes (they are read-only)
|
|
self.assertRaises(AttributeError, delattr, obj.dialect, 'delimiter')
|
|
self.assertRaises(AttributeError, setattr, obj.dialect, 'delimiter', ':')
|
|
self.assertRaises(AttributeError, delattr, obj.dialect, 'quoting')
|
|
self.assertRaises(AttributeError, setattr, obj.dialect,
|
|
'quoting', None)
|
|
|
|
def test_reader_attrs(self):
|
|
self._test_default_attrs(csv.reader, [])
|
|
|
|
def test_writer_attrs(self):
|
|
self._test_default_attrs(csv.writer, StringIO())
|
|
|
|
def _test_kw_attrs(self, ctor, *args):
|
|
# Now try with alternate options
|
|
kwargs = dict(delimiter=':', doublequote=False, escapechar='\\',
|
|
lineterminator='\r', quotechar='*',
|
|
quoting=csv.QUOTE_NONE, skipinitialspace=True,
|
|
strict=True)
|
|
obj = ctor(*args, **kwargs)
|
|
self.assertEqual(obj.dialect.delimiter, ':')
|
|
self.assertEqual(obj.dialect.doublequote, False)
|
|
self.assertEqual(obj.dialect.escapechar, '\\')
|
|
self.assertEqual(obj.dialect.lineterminator, "\r")
|
|
self.assertEqual(obj.dialect.quotechar, '*')
|
|
self.assertEqual(obj.dialect.quoting, csv.QUOTE_NONE)
|
|
self.assertEqual(obj.dialect.skipinitialspace, True)
|
|
self.assertEqual(obj.dialect.strict, True)
|
|
|
|
def test_reader_kw_attrs(self):
|
|
self._test_kw_attrs(csv.reader, [])
|
|
|
|
def test_writer_kw_attrs(self):
|
|
self._test_kw_attrs(csv.writer, StringIO())
|
|
|
|
def _test_dialect_attrs(self, ctor, *args):
|
|
# Now try with dialect-derived options
|
|
class dialect:
|
|
delimiter='-'
|
|
doublequote=False
|
|
escapechar='^'
|
|
lineterminator='$'
|
|
quotechar='#'
|
|
quoting=csv.QUOTE_ALL
|
|
skipinitialspace=True
|
|
strict=False
|
|
args = args + (dialect,)
|
|
obj = ctor(*args)
|
|
self.assertEqual(obj.dialect.delimiter, '-')
|
|
self.assertEqual(obj.dialect.doublequote, False)
|
|
self.assertEqual(obj.dialect.escapechar, '^')
|
|
self.assertEqual(obj.dialect.lineterminator, "$")
|
|
self.assertEqual(obj.dialect.quotechar, '#')
|
|
self.assertEqual(obj.dialect.quoting, csv.QUOTE_ALL)
|
|
self.assertEqual(obj.dialect.skipinitialspace, True)
|
|
self.assertEqual(obj.dialect.strict, False)
|
|
|
|
def test_reader_dialect_attrs(self):
|
|
self._test_dialect_attrs(csv.reader, [])
|
|
|
|
def test_writer_dialect_attrs(self):
|
|
self._test_dialect_attrs(csv.writer, StringIO())
|
|
|
|
|
|
def _write_test(self, fields, expect, **kwargs):
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj, **kwargs)
|
|
writer.writerow(fields)
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(),
|
|
expect + writer.dialect.lineterminator)
|
|
|
|
def test_write_arg_valid(self):
|
|
self.assertRaises(csv.Error, self._write_test, None, '')
|
|
self._write_test((), '')
|
|
self._write_test([None], '""')
|
|
self.assertRaises(csv.Error, self._write_test,
|
|
[None], None, quoting = csv.QUOTE_NONE)
|
|
# Check that exceptions are passed up the chain
|
|
class BadList:
|
|
def __len__(self):
|
|
return 10;
|
|
def __getitem__(self, i):
|
|
if i > 2:
|
|
raise IOError
|
|
self.assertRaises(IOError, self._write_test, BadList(), '')
|
|
class BadItem:
|
|
def __str__(self):
|
|
raise IOError
|
|
self.assertRaises(IOError, self._write_test, [BadItem()], '')
|
|
|
|
def test_write_bigfield(self):
|
|
# This exercises the buffer realloc functionality
|
|
bigstring = 'X' * 50000
|
|
self._write_test([bigstring,bigstring], '%s,%s' % \
|
|
(bigstring, bigstring))
|
|
|
|
def test_write_quoting(self):
|
|
self._write_test(['a',1,'p,q'], 'a,1,"p,q"')
|
|
self.assertRaises(csv.Error,
|
|
self._write_test,
|
|
['a',1,'p,q'], 'a,1,p,q',
|
|
quoting = csv.QUOTE_NONE)
|
|
self._write_test(['a',1,'p,q'], 'a,1,"p,q"',
|
|
quoting = csv.QUOTE_MINIMAL)
|
|
self._write_test(['a',1,'p,q'], '"a",1,"p,q"',
|
|
quoting = csv.QUOTE_NONNUMERIC)
|
|
self._write_test(['a',1,'p,q'], '"a","1","p,q"',
|
|
quoting = csv.QUOTE_ALL)
|
|
self._write_test(['a\nb',1], '"a\nb","1"',
|
|
quoting = csv.QUOTE_ALL)
|
|
|
|
def test_write_escape(self):
|
|
self._write_test(['a',1,'p,q'], 'a,1,"p,q"',
|
|
escapechar='\\')
|
|
self.assertRaises(csv.Error,
|
|
self._write_test,
|
|
['a',1,'p,"q"'], 'a,1,"p,\\"q\\""',
|
|
escapechar=None, doublequote=False)
|
|
self._write_test(['a',1,'p,"q"'], 'a,1,"p,\\"q\\""',
|
|
escapechar='\\', doublequote = False)
|
|
self._write_test(['"'], '""""',
|
|
escapechar='\\', quoting = csv.QUOTE_MINIMAL)
|
|
self._write_test(['"'], '\\"',
|
|
escapechar='\\', quoting = csv.QUOTE_MINIMAL,
|
|
doublequote = False)
|
|
self._write_test(['"'], '\\"',
|
|
escapechar='\\', quoting = csv.QUOTE_NONE)
|
|
self._write_test(['a',1,'p,q'], 'a,1,p\\,q',
|
|
escapechar='\\', quoting = csv.QUOTE_NONE)
|
|
|
|
def test_writerows(self):
|
|
class BrokenFile:
|
|
def write(self, buf):
|
|
raise IOError
|
|
writer = csv.writer(BrokenFile())
|
|
self.assertRaises(IOError, writer.writerows, [['a']])
|
|
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj)
|
|
self.assertRaises(TypeError, writer.writerows, None)
|
|
writer.writerows([['a','b'],['c','d']])
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), "a,b\r\nc,d\r\n")
|
|
|
|
def _read_test(self, input, expect, **kwargs):
|
|
reader = csv.reader(input, **kwargs)
|
|
result = list(reader)
|
|
self.assertEqual(result, expect)
|
|
|
|
def test_read_oddinputs(self):
|
|
self._read_test([], [])
|
|
self._read_test([''], [[]])
|
|
self.assertRaises(csv.Error, self._read_test,
|
|
['"ab"c'], None, strict = 1)
|
|
# cannot handle null bytes for the moment
|
|
self.assertRaises(csv.Error, self._read_test,
|
|
['ab\0c'], None, strict = 1)
|
|
self._read_test(['"ab"c'], [['abc']], doublequote = 0)
|
|
|
|
self.assertRaises(csv.Error, self._read_test,
|
|
[b'ab\0c'], None)
|
|
|
|
|
|
def test_read_eol(self):
|
|
self._read_test(['a,b'], [['a','b']])
|
|
self._read_test(['a,b\n'], [['a','b']])
|
|
self._read_test(['a,b\r\n'], [['a','b']])
|
|
self._read_test(['a,b\r'], [['a','b']])
|
|
self.assertRaises(csv.Error, self._read_test, ['a,b\rc,d'], [])
|
|
self.assertRaises(csv.Error, self._read_test, ['a,b\nc,d'], [])
|
|
self.assertRaises(csv.Error, self._read_test, ['a,b\r\nc,d'], [])
|
|
|
|
def test_read_escape(self):
|
|
self._read_test(['a,\\b,c'], [['a', 'b', 'c']], escapechar='\\')
|
|
self._read_test(['a,b\\,c'], [['a', 'b,c']], escapechar='\\')
|
|
self._read_test(['a,"b\\,c"'], [['a', 'b,c']], escapechar='\\')
|
|
self._read_test(['a,"b,\\c"'], [['a', 'b,c']], escapechar='\\')
|
|
self._read_test(['a,"b,c\\""'], [['a', 'b,c"']], escapechar='\\')
|
|
self._read_test(['a,"b,c"\\'], [['a', 'b,c\\']], escapechar='\\')
|
|
|
|
def test_read_quoting(self):
|
|
self._read_test(['1,",3,",5'], [['1', ',3,', '5']])
|
|
self._read_test(['1,",3,",5'], [['1', '"', '3', '"', '5']],
|
|
quotechar=None, escapechar='\\')
|
|
self._read_test(['1,",3,",5'], [['1', '"', '3', '"', '5']],
|
|
quoting=csv.QUOTE_NONE, escapechar='\\')
|
|
# will this fail where locale uses comma for decimals?
|
|
self._read_test([',3,"5",7.3, 9'], [['', 3, '5', 7.3, 9]],
|
|
quoting=csv.QUOTE_NONNUMERIC)
|
|
self._read_test(['"a\nb", 7'], [['a\nb', ' 7']])
|
|
self.assertRaises(ValueError, self._read_test,
|
|
['abc,3'], [[]],
|
|
quoting=csv.QUOTE_NONNUMERIC)
|
|
|
|
def test_read_bigfield(self):
|
|
# This exercises the buffer realloc functionality and field size
|
|
# limits.
|
|
limit = csv.field_size_limit()
|
|
try:
|
|
size = 50000
|
|
bigstring = 'X' * size
|
|
bigline = '%s,%s' % (bigstring, bigstring)
|
|
self._read_test([bigline], [[bigstring, bigstring]])
|
|
csv.field_size_limit(size)
|
|
self._read_test([bigline], [[bigstring, bigstring]])
|
|
self.assertEqual(csv.field_size_limit(), size)
|
|
csv.field_size_limit(size-1)
|
|
self.assertRaises(csv.Error, self._read_test, [bigline], [])
|
|
self.assertRaises(TypeError, csv.field_size_limit, None)
|
|
self.assertRaises(TypeError, csv.field_size_limit, 1, None)
|
|
finally:
|
|
csv.field_size_limit(limit)
|
|
|
|
def test_read_linenum(self):
|
|
r = csv.reader(['line,1', 'line,2', 'line,3'])
|
|
self.assertEqual(r.line_num, 0)
|
|
next(r)
|
|
self.assertEqual(r.line_num, 1)
|
|
next(r)
|
|
self.assertEqual(r.line_num, 2)
|
|
next(r)
|
|
self.assertEqual(r.line_num, 3)
|
|
self.assertRaises(StopIteration, next, r)
|
|
self.assertEqual(r.line_num, 3)
|
|
|
|
def test_roundtrip_quoteed_newlines(self):
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj)
|
|
self.assertRaises(TypeError, writer.writerows, None)
|
|
rows = [['a\nb','b'],['c','x\r\nd']]
|
|
writer.writerows(rows)
|
|
fileobj.seek(0)
|
|
for i, row in enumerate(csv.reader(fileobj)):
|
|
self.assertEqual(row, rows[i])
|
|
|
|
class TestDialectRegistry(unittest.TestCase):
|
|
def test_registry_badargs(self):
|
|
self.assertRaises(TypeError, csv.list_dialects, None)
|
|
self.assertRaises(TypeError, csv.get_dialect)
|
|
self.assertRaises(csv.Error, csv.get_dialect, None)
|
|
self.assertRaises(csv.Error, csv.get_dialect, "nonesuch")
|
|
self.assertRaises(TypeError, csv.unregister_dialect)
|
|
self.assertRaises(csv.Error, csv.unregister_dialect, None)
|
|
self.assertRaises(csv.Error, csv.unregister_dialect, "nonesuch")
|
|
self.assertRaises(TypeError, csv.register_dialect, None)
|
|
self.assertRaises(TypeError, csv.register_dialect, None, None)
|
|
self.assertRaises(TypeError, csv.register_dialect, "nonesuch", 0, 0)
|
|
self.assertRaises(TypeError, csv.register_dialect, "nonesuch",
|
|
badargument=None)
|
|
self.assertRaises(TypeError, csv.register_dialect, "nonesuch",
|
|
quoting=None)
|
|
self.assertRaises(TypeError, csv.register_dialect, [])
|
|
|
|
def test_registry(self):
|
|
class myexceltsv(csv.excel):
|
|
delimiter = "\t"
|
|
name = "myexceltsv"
|
|
expected_dialects = csv.list_dialects() + [name]
|
|
expected_dialects.sort()
|
|
csv.register_dialect(name, myexceltsv)
|
|
try:
|
|
self.assertTrue(csv.get_dialect(name).delimiter, '\t')
|
|
got_dialects = csv.list_dialects()
|
|
got_dialects.sort()
|
|
self.assertEqual(expected_dialects, got_dialects)
|
|
finally:
|
|
csv.unregister_dialect(name)
|
|
|
|
def test_register_kwargs(self):
|
|
name = 'fedcba'
|
|
csv.register_dialect(name, delimiter=';')
|
|
try:
|
|
self.assertTrue(csv.get_dialect(name).delimiter, '\t')
|
|
self.assertTrue(list(csv.reader('X;Y;Z', name)), ['X', 'Y', 'Z'])
|
|
finally:
|
|
csv.unregister_dialect(name)
|
|
|
|
def test_incomplete_dialect(self):
|
|
class myexceltsv(csv.Dialect):
|
|
delimiter = "\t"
|
|
self.assertRaises(csv.Error, myexceltsv)
|
|
|
|
def test_space_dialect(self):
|
|
class space(csv.excel):
|
|
delimiter = " "
|
|
quoting = csv.QUOTE_NONE
|
|
escapechar = "\\"
|
|
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("abc def\nc1ccccc1 benzene\n")
|
|
fileobj.seek(0)
|
|
reader = csv.reader(fileobj, dialect=space())
|
|
self.assertEqual(next(reader), ["abc", "def"])
|
|
self.assertEqual(next(reader), ["c1ccccc1", "benzene"])
|
|
|
|
def compare_dialect_123(self, expected, *writeargs, **kwwriteargs):
|
|
|
|
with TemporaryFile("w+", newline='', encoding="utf-8") as fileobj:
|
|
|
|
writer = csv.writer(fileobj, *writeargs, **kwwriteargs)
|
|
writer.writerow([1,2,3])
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected)
|
|
|
|
def test_dialect_apply(self):
|
|
class testA(csv.excel):
|
|
delimiter = "\t"
|
|
class testB(csv.excel):
|
|
delimiter = ":"
|
|
class testC(csv.excel):
|
|
delimiter = "|"
|
|
class testUni(csv.excel):
|
|
delimiter = "\u039B"
|
|
|
|
csv.register_dialect('testC', testC)
|
|
try:
|
|
self.compare_dialect_123("1,2,3\r\n")
|
|
self.compare_dialect_123("1\t2\t3\r\n", testA)
|
|
self.compare_dialect_123("1:2:3\r\n", dialect=testB())
|
|
self.compare_dialect_123("1|2|3\r\n", dialect='testC')
|
|
self.compare_dialect_123("1;2;3\r\n", dialect=testA,
|
|
delimiter=';')
|
|
self.compare_dialect_123("1\u039B2\u039B3\r\n",
|
|
dialect=testUni)
|
|
|
|
finally:
|
|
csv.unregister_dialect('testC')
|
|
|
|
def test_bad_dialect(self):
|
|
# Unknown parameter
|
|
self.assertRaises(TypeError, csv.reader, [], bad_attr = 0)
|
|
# Bad values
|
|
self.assertRaises(TypeError, csv.reader, [], delimiter = None)
|
|
self.assertRaises(TypeError, csv.reader, [], quoting = -1)
|
|
self.assertRaises(TypeError, csv.reader, [], quoting = 100)
|
|
|
|
class TestCsvBase(unittest.TestCase):
|
|
def readerAssertEqual(self, input, expected_result):
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
fileobj.write(input)
|
|
fileobj.seek(0)
|
|
reader = csv.reader(fileobj, dialect = self.dialect)
|
|
fields = list(reader)
|
|
self.assertEqual(fields, expected_result)
|
|
|
|
def writerAssertEqual(self, input, expected_result):
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj, dialect = self.dialect)
|
|
writer.writerows(input)
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected_result)
|
|
|
|
class TestDialectExcel(TestCsvBase):
|
|
dialect = 'excel'
|
|
|
|
def test_single(self):
|
|
self.readerAssertEqual('abc', [['abc']])
|
|
|
|
def test_simple(self):
|
|
self.readerAssertEqual('1,2,3,4,5', [['1','2','3','4','5']])
|
|
|
|
def test_blankline(self):
|
|
self.readerAssertEqual('', [])
|
|
|
|
def test_empty_fields(self):
|
|
self.readerAssertEqual(',', [['', '']])
|
|
|
|
def test_singlequoted(self):
|
|
self.readerAssertEqual('""', [['']])
|
|
|
|
def test_singlequoted_left_empty(self):
|
|
self.readerAssertEqual('"",', [['','']])
|
|
|
|
def test_singlequoted_right_empty(self):
|
|
self.readerAssertEqual(',""', [['','']])
|
|
|
|
def test_single_quoted_quote(self):
|
|
self.readerAssertEqual('""""', [['"']])
|
|
|
|
def test_quoted_quotes(self):
|
|
self.readerAssertEqual('""""""', [['""']])
|
|
|
|
def test_inline_quote(self):
|
|
self.readerAssertEqual('a""b', [['a""b']])
|
|
|
|
def test_inline_quotes(self):
|
|
self.readerAssertEqual('a"b"c', [['a"b"c']])
|
|
|
|
def test_quotes_and_more(self):
|
|
# Excel would never write a field containing '"a"b', but when
|
|
# reading one, it will return 'ab'.
|
|
self.readerAssertEqual('"a"b', [['ab']])
|
|
|
|
def test_lone_quote(self):
|
|
self.readerAssertEqual('a"b', [['a"b']])
|
|
|
|
def test_quote_and_quote(self):
|
|
# Excel would never write a field containing '"a" "b"', but when
|
|
# reading one, it will return 'a "b"'.
|
|
self.readerAssertEqual('"a" "b"', [['a "b"']])
|
|
|
|
def test_space_and_quote(self):
|
|
self.readerAssertEqual(' "a"', [[' "a"']])
|
|
|
|
def test_quoted(self):
|
|
self.readerAssertEqual('1,2,3,"I think, therefore I am",5,6',
|
|
[['1', '2', '3',
|
|
'I think, therefore I am',
|
|
'5', '6']])
|
|
|
|
def test_quoted_quote(self):
|
|
self.readerAssertEqual('1,2,3,"""I see,"" said the blind man","as he picked up his hammer and saw"',
|
|
[['1', '2', '3',
|
|
'"I see," said the blind man',
|
|
'as he picked up his hammer and saw']])
|
|
|
|
def test_quoted_nl(self):
|
|
input = '''\
|
|
1,2,3,"""I see,""
|
|
said the blind man","as he picked up his
|
|
hammer and saw"
|
|
9,8,7,6'''
|
|
self.readerAssertEqual(input,
|
|
[['1', '2', '3',
|
|
'"I see,"\nsaid the blind man',
|
|
'as he picked up his\nhammer and saw'],
|
|
['9','8','7','6']])
|
|
|
|
def test_dubious_quote(self):
|
|
self.readerAssertEqual('12,12,1",', [['12', '12', '1"', '']])
|
|
|
|
def test_null(self):
|
|
self.writerAssertEqual([], '')
|
|
|
|
def test_single_writer(self):
|
|
self.writerAssertEqual([['abc']], 'abc\r\n')
|
|
|
|
def test_simple_writer(self):
|
|
self.writerAssertEqual([[1, 2, 'abc', 3, 4]], '1,2,abc,3,4\r\n')
|
|
|
|
def test_quotes(self):
|
|
self.writerAssertEqual([[1, 2, 'a"bc"', 3, 4]], '1,2,"a""bc""",3,4\r\n')
|
|
|
|
def test_quote_fieldsep(self):
|
|
self.writerAssertEqual([['abc,def']], '"abc,def"\r\n')
|
|
|
|
def test_newlines(self):
|
|
self.writerAssertEqual([[1, 2, 'a\nbc', 3, 4]], '1,2,"a\nbc",3,4\r\n')
|
|
|
|
class EscapedExcel(csv.excel):
|
|
quoting = csv.QUOTE_NONE
|
|
escapechar = '\\'
|
|
|
|
class TestEscapedExcel(TestCsvBase):
|
|
dialect = EscapedExcel()
|
|
|
|
def test_escape_fieldsep(self):
|
|
self.writerAssertEqual([['abc,def']], 'abc\\,def\r\n')
|
|
|
|
def test_read_escape_fieldsep(self):
|
|
self.readerAssertEqual('abc\\,def\r\n', [['abc,def']])
|
|
|
|
class QuotedEscapedExcel(csv.excel):
|
|
quoting = csv.QUOTE_NONNUMERIC
|
|
escapechar = '\\'
|
|
|
|
class TestQuotedEscapedExcel(TestCsvBase):
|
|
dialect = QuotedEscapedExcel()
|
|
|
|
def test_write_escape_fieldsep(self):
|
|
self.writerAssertEqual([['abc,def']], '"abc,def"\r\n')
|
|
|
|
def test_read_escape_fieldsep(self):
|
|
self.readerAssertEqual('"abc\\,def"\r\n', [['abc,def']])
|
|
|
|
class TestDictFields(unittest.TestCase):
|
|
### "long" means the row is longer than the number of fieldnames
|
|
### "short" means there are fewer elements in the row than fieldnames
|
|
def test_write_simple_dict(self):
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.DictWriter(fileobj, fieldnames = ["f1", "f2", "f3"])
|
|
writer.writeheader()
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.readline(), "f1,f2,f3\r\n")
|
|
writer.writerow({"f1": 10, "f3": "abc"})
|
|
fileobj.seek(0)
|
|
fileobj.readline() # header
|
|
self.assertEqual(fileobj.read(), "10,,abc\r\n")
|
|
|
|
def test_write_no_fields(self):
|
|
fileobj = StringIO()
|
|
self.assertRaises(TypeError, csv.DictWriter, fileobj)
|
|
|
|
def test_read_dict_fields(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("1,2,abc\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj,
|
|
fieldnames=["f1", "f2", "f3"])
|
|
self.assertEqual(next(reader), {"f1": '1', "f2": '2', "f3": 'abc'})
|
|
|
|
def test_read_dict_no_fieldnames(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("f1,f2,f3\r\n1,2,abc\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj)
|
|
self.assertEqual(next(reader), {"f1": '1', "f2": '2', "f3": 'abc'})
|
|
self.assertEqual(reader.fieldnames, ["f1", "f2", "f3"])
|
|
|
|
# Two test cases to make sure existing ways of implicitly setting
|
|
# fieldnames continue to work. Both arise from discussion in issue3436.
|
|
def test_read_dict_fieldnames_from_file(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("f1,f2,f3\r\n1,2,abc\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj,
|
|
fieldnames=next(csv.reader(fileobj)))
|
|
self.assertEqual(reader.fieldnames, ["f1", "f2", "f3"])
|
|
self.assertEqual(next(reader), {"f1": '1', "f2": '2', "f3": 'abc'})
|
|
|
|
def test_read_dict_fieldnames_chain(self):
|
|
import itertools
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("f1,f2,f3\r\n1,2,abc\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj)
|
|
first = next(reader)
|
|
for row in itertools.chain([first], reader):
|
|
self.assertEqual(reader.fieldnames, ["f1", "f2", "f3"])
|
|
self.assertEqual(row, {"f1": '1', "f2": '2', "f3": 'abc'})
|
|
|
|
def test_read_long(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("1,2,abc,4,5,6\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj,
|
|
fieldnames=["f1", "f2"])
|
|
self.assertEqual(next(reader), {"f1": '1', "f2": '2',
|
|
None: ["abc", "4", "5", "6"]})
|
|
|
|
def test_read_long_with_rest(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("1,2,abc,4,5,6\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj,
|
|
fieldnames=["f1", "f2"], restkey="_rest")
|
|
self.assertEqual(next(reader), {"f1": '1', "f2": '2',
|
|
"_rest": ["abc", "4", "5", "6"]})
|
|
|
|
def test_read_long_with_rest_no_fieldnames(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("f1,f2\r\n1,2,abc,4,5,6\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj, restkey="_rest")
|
|
self.assertEqual(reader.fieldnames, ["f1", "f2"])
|
|
self.assertEqual(next(reader), {"f1": '1', "f2": '2',
|
|
"_rest": ["abc", "4", "5", "6"]})
|
|
|
|
def test_read_short(self):
|
|
with TemporaryFile("w+") as fileobj:
|
|
fileobj.write("1,2,abc,4,5,6\r\n1,2,abc\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.DictReader(fileobj,
|
|
fieldnames="1 2 3 4 5 6".split(),
|
|
restval="DEFAULT")
|
|
self.assertEqual(next(reader), {"1": '1', "2": '2', "3": 'abc',
|
|
"4": '4', "5": '5', "6": '6'})
|
|
self.assertEqual(next(reader), {"1": '1', "2": '2', "3": 'abc',
|
|
"4": 'DEFAULT', "5": 'DEFAULT',
|
|
"6": 'DEFAULT'})
|
|
|
|
def test_read_multi(self):
|
|
sample = [
|
|
'2147483648,43.0e12,17,abc,def\r\n',
|
|
'147483648,43.0e2,17,abc,def\r\n',
|
|
'47483648,43.0,170,abc,def\r\n'
|
|
]
|
|
|
|
reader = csv.DictReader(sample,
|
|
fieldnames="i1 float i2 s1 s2".split())
|
|
self.assertEqual(next(reader), {"i1": '2147483648',
|
|
"float": '43.0e12',
|
|
"i2": '17',
|
|
"s1": 'abc',
|
|
"s2": 'def'})
|
|
|
|
def test_read_with_blanks(self):
|
|
reader = csv.DictReader(["1,2,abc,4,5,6\r\n","\r\n",
|
|
"1,2,abc,4,5,6\r\n"],
|
|
fieldnames="1 2 3 4 5 6".split())
|
|
self.assertEqual(next(reader), {"1": '1', "2": '2', "3": 'abc',
|
|
"4": '4', "5": '5', "6": '6'})
|
|
self.assertEqual(next(reader), {"1": '1', "2": '2', "3": 'abc',
|
|
"4": '4', "5": '5', "6": '6'})
|
|
|
|
def test_read_semi_sep(self):
|
|
reader = csv.DictReader(["1;2;abc;4;5;6\r\n"],
|
|
fieldnames="1 2 3 4 5 6".split(),
|
|
delimiter=';')
|
|
self.assertEqual(next(reader), {"1": '1', "2": '2', "3": 'abc',
|
|
"4": '4', "5": '5', "6": '6'})
|
|
|
|
class TestArrayWrites(unittest.TestCase):
|
|
def test_int_write(self):
|
|
import array
|
|
contents = [(20-i) for i in range(20)]
|
|
a = array.array('i', contents)
|
|
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj, dialect="excel")
|
|
writer.writerow(a)
|
|
expected = ",".join([str(i) for i in a])+"\r\n"
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected)
|
|
|
|
def test_double_write(self):
|
|
import array
|
|
contents = [(20-i)*0.1 for i in range(20)]
|
|
a = array.array('d', contents)
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj, dialect="excel")
|
|
writer.writerow(a)
|
|
expected = ",".join([str(i) for i in a])+"\r\n"
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected)
|
|
|
|
def test_float_write(self):
|
|
import array
|
|
contents = [(20-i)*0.1 for i in range(20)]
|
|
a = array.array('f', contents)
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj, dialect="excel")
|
|
writer.writerow(a)
|
|
expected = ",".join([str(i) for i in a])+"\r\n"
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected)
|
|
|
|
def test_char_write(self):
|
|
import array, string
|
|
a = array.array('u', string.ascii_letters)
|
|
|
|
with TemporaryFile("w+", newline='') as fileobj:
|
|
writer = csv.writer(fileobj, dialect="excel")
|
|
writer.writerow(a)
|
|
expected = ",".join(a)+"\r\n"
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected)
|
|
|
|
class TestDialectValidity(unittest.TestCase):
|
|
def test_quoting(self):
|
|
class mydialect(csv.Dialect):
|
|
delimiter = ";"
|
|
escapechar = '\\'
|
|
doublequote = False
|
|
skipinitialspace = True
|
|
lineterminator = '\r\n'
|
|
quoting = csv.QUOTE_NONE
|
|
d = mydialect()
|
|
|
|
mydialect.quoting = None
|
|
self.assertRaises(csv.Error, mydialect)
|
|
|
|
mydialect.doublequote = True
|
|
mydialect.quoting = csv.QUOTE_ALL
|
|
mydialect.quotechar = '"'
|
|
d = mydialect()
|
|
|
|
mydialect.quotechar = "''"
|
|
self.assertRaises(csv.Error, mydialect)
|
|
|
|
mydialect.quotechar = 4
|
|
self.assertRaises(csv.Error, mydialect)
|
|
|
|
def test_delimiter(self):
|
|
class mydialect(csv.Dialect):
|
|
delimiter = ";"
|
|
escapechar = '\\'
|
|
doublequote = False
|
|
skipinitialspace = True
|
|
lineterminator = '\r\n'
|
|
quoting = csv.QUOTE_NONE
|
|
d = mydialect()
|
|
|
|
mydialect.delimiter = ":::"
|
|
self.assertRaises(csv.Error, mydialect)
|
|
|
|
mydialect.delimiter = 4
|
|
self.assertRaises(csv.Error, mydialect)
|
|
|
|
def test_lineterminator(self):
|
|
class mydialect(csv.Dialect):
|
|
delimiter = ";"
|
|
escapechar = '\\'
|
|
doublequote = False
|
|
skipinitialspace = True
|
|
lineterminator = '\r\n'
|
|
quoting = csv.QUOTE_NONE
|
|
d = mydialect()
|
|
|
|
mydialect.lineterminator = ":::"
|
|
d = mydialect()
|
|
|
|
mydialect.lineterminator = 4
|
|
self.assertRaises(csv.Error, mydialect)
|
|
|
|
|
|
class TestSniffer(unittest.TestCase):
|
|
sample1 = """\
|
|
Harry's, Arlington Heights, IL, 2/1/03, Kimi Hayes
|
|
Shark City, Glendale Heights, IL, 12/28/02, Prezence
|
|
Tommy's Place, Blue Island, IL, 12/28/02, Blue Sunday/White Crow
|
|
Stonecutters Seafood and Chop House, Lemont, IL, 12/19/02, Week Back
|
|
"""
|
|
sample2 = """\
|
|
'Harry''s':'Arlington Heights':'IL':'2/1/03':'Kimi Hayes'
|
|
'Shark City':'Glendale Heights':'IL':'12/28/02':'Prezence'
|
|
'Tommy''s Place':'Blue Island':'IL':'12/28/02':'Blue Sunday/White Crow'
|
|
'Stonecutters ''Seafood'' and Chop House':'Lemont':'IL':'12/19/02':'Week Back'
|
|
"""
|
|
header = '''\
|
|
"venue","city","state","date","performers"
|
|
'''
|
|
sample3 = '''\
|
|
05/05/03?05/05/03?05/05/03?05/05/03?05/05/03?05/05/03
|
|
05/05/03?05/05/03?05/05/03?05/05/03?05/05/03?05/05/03
|
|
05/05/03?05/05/03?05/05/03?05/05/03?05/05/03?05/05/03
|
|
'''
|
|
|
|
sample4 = '''\
|
|
2147483648;43.0e12;17;abc;def
|
|
147483648;43.0e2;17;abc;def
|
|
47483648;43.0;170;abc;def
|
|
'''
|
|
|
|
sample5 = "aaa\tbbb\r\nAAA\t\r\nBBB\t\r\n"
|
|
sample6 = "a|b|c\r\nd|e|f\r\n"
|
|
sample7 = "'a'|'b'|'c'\r\n'd'|e|f\r\n"
|
|
|
|
def test_has_header(self):
|
|
sniffer = csv.Sniffer()
|
|
self.assertEqual(sniffer.has_header(self.sample1), False)
|
|
self.assertEqual(sniffer.has_header(self.header+self.sample1), True)
|
|
|
|
def test_sniff(self):
|
|
sniffer = csv.Sniffer()
|
|
dialect = sniffer.sniff(self.sample1)
|
|
self.assertEqual(dialect.delimiter, ",")
|
|
self.assertEqual(dialect.quotechar, '"')
|
|
self.assertEqual(dialect.skipinitialspace, True)
|
|
|
|
dialect = sniffer.sniff(self.sample2)
|
|
self.assertEqual(dialect.delimiter, ":")
|
|
self.assertEqual(dialect.quotechar, "'")
|
|
self.assertEqual(dialect.skipinitialspace, False)
|
|
|
|
def test_delimiters(self):
|
|
sniffer = csv.Sniffer()
|
|
dialect = sniffer.sniff(self.sample3)
|
|
# given that all three lines in sample3 are equal,
|
|
# I think that any character could have been 'guessed' as the
|
|
# delimiter, depending on dictionary order
|
|
self.assertIn(dialect.delimiter, self.sample3)
|
|
dialect = sniffer.sniff(self.sample3, delimiters="?,")
|
|
self.assertEqual(dialect.delimiter, "?")
|
|
dialect = sniffer.sniff(self.sample3, delimiters="/,")
|
|
self.assertEqual(dialect.delimiter, "/")
|
|
dialect = sniffer.sniff(self.sample4)
|
|
self.assertEqual(dialect.delimiter, ";")
|
|
dialect = sniffer.sniff(self.sample5)
|
|
self.assertEqual(dialect.delimiter, "\t")
|
|
dialect = sniffer.sniff(self.sample6)
|
|
self.assertEqual(dialect.delimiter, "|")
|
|
dialect = sniffer.sniff(self.sample7)
|
|
self.assertEqual(dialect.delimiter, "|")
|
|
self.assertEqual(dialect.quotechar, "'")
|
|
|
|
def test_doublequote(self):
|
|
sniffer = csv.Sniffer()
|
|
dialect = sniffer.sniff(self.header)
|
|
self.assertFalse(dialect.doublequote)
|
|
dialect = sniffer.sniff(self.sample2)
|
|
self.assertTrue(dialect.doublequote)
|
|
|
|
if not hasattr(sys, "gettotalrefcount"):
|
|
if support.verbose: print("*** skipping leakage tests ***")
|
|
else:
|
|
class NUL:
|
|
def write(s, *args):
|
|
pass
|
|
writelines = write
|
|
|
|
class TestLeaks(unittest.TestCase):
|
|
def test_create_read(self):
|
|
delta = 0
|
|
lastrc = sys.gettotalrefcount()
|
|
for i in range(20):
|
|
gc.collect()
|
|
self.assertEqual(gc.garbage, [])
|
|
rc = sys.gettotalrefcount()
|
|
csv.reader(["a,b,c\r\n"])
|
|
csv.reader(["a,b,c\r\n"])
|
|
csv.reader(["a,b,c\r\n"])
|
|
delta = rc-lastrc
|
|
lastrc = rc
|
|
# if csv.reader() leaks, last delta should be 3 or more
|
|
self.assertEqual(delta < 3, True)
|
|
|
|
def test_create_write(self):
|
|
delta = 0
|
|
lastrc = sys.gettotalrefcount()
|
|
s = NUL()
|
|
for i in range(20):
|
|
gc.collect()
|
|
self.assertEqual(gc.garbage, [])
|
|
rc = sys.gettotalrefcount()
|
|
csv.writer(s)
|
|
csv.writer(s)
|
|
csv.writer(s)
|
|
delta = rc-lastrc
|
|
lastrc = rc
|
|
# if csv.writer() leaks, last delta should be 3 or more
|
|
self.assertEqual(delta < 3, True)
|
|
|
|
def test_read(self):
|
|
delta = 0
|
|
rows = ["a,b,c\r\n"]*5
|
|
lastrc = sys.gettotalrefcount()
|
|
for i in range(20):
|
|
gc.collect()
|
|
self.assertEqual(gc.garbage, [])
|
|
rc = sys.gettotalrefcount()
|
|
rdr = csv.reader(rows)
|
|
for row in rdr:
|
|
pass
|
|
delta = rc-lastrc
|
|
lastrc = rc
|
|
# if reader leaks during read, delta should be 5 or more
|
|
self.assertEqual(delta < 5, True)
|
|
|
|
def test_write(self):
|
|
delta = 0
|
|
rows = [[1,2,3]]*5
|
|
s = NUL()
|
|
lastrc = sys.gettotalrefcount()
|
|
for i in range(20):
|
|
gc.collect()
|
|
self.assertEqual(gc.garbage, [])
|
|
rc = sys.gettotalrefcount()
|
|
writer = csv.writer(s)
|
|
for row in rows:
|
|
writer.writerow(row)
|
|
delta = rc-lastrc
|
|
lastrc = rc
|
|
# if writer leaks during write, last delta should be 5 or more
|
|
self.assertEqual(delta < 5, True)
|
|
|
|
class TestUnicode(unittest.TestCase):
|
|
|
|
names = ["Martin von Löwis",
|
|
"Marc André Lemburg",
|
|
"Guido van Rossum",
|
|
"François Pinard"]
|
|
|
|
def test_unicode_read(self):
|
|
import io
|
|
with TemporaryFile("w+", newline='', encoding="utf-8") as fileobj:
|
|
fileobj.write(",".join(self.names) + "\r\n")
|
|
fileobj.seek(0)
|
|
reader = csv.reader(fileobj)
|
|
self.assertEqual(list(reader), [self.names])
|
|
|
|
|
|
def test_unicode_write(self):
|
|
import io
|
|
with TemporaryFile("w+", newline='', encoding="utf-8") as fileobj:
|
|
writer = csv.writer(fileobj)
|
|
writer.writerow(self.names)
|
|
expected = ",".join(self.names)+"\r\n"
|
|
fileobj.seek(0)
|
|
self.assertEqual(fileobj.read(), expected)
|
|
|
|
|
|
|
|
def test_main():
|
|
mod = sys.modules[__name__]
|
|
support.run_unittest(
|
|
*[getattr(mod, name) for name in dir(mod) if name.startswith('Test')]
|
|
)
|
|
|
|
if __name__ == '__main__':
|
|
test_main()
|