blob: bd08c4f3007fc7f5297326a113231d5267acb573 [file] [log] [blame]
from test.support import verbose, is_android
from test.support.warnings_helper import check_warnings
import unittest
import locale
import sys
import codecs
class BaseLocalizedTest(unittest.TestCase):
#
# Base class for tests using a real locale
#
@classmethod
def setUpClass(cls):
if sys.platform == 'darwin':
import os
tlocs = ("en_US.UTF-8", "en_US.ISO8859-1", "en_US")
if int(os.uname().release.split('.')[0]) < 10:
# The locale test work fine on OSX 10.6, I (ronaldoussoren)
# haven't had time yet to verify if tests work on OSX 10.5
# (10.4 is known to be bad)
raise unittest.SkipTest("Locale support on MacOSX is minimal")
elif sys.platform.startswith("win"):
tlocs = ("En", "English")
else:
tlocs = ("en_US.UTF-8", "en_US.ISO8859-1",
"en_US.US-ASCII", "en_US")
try:
oldlocale = locale.setlocale(locale.LC_NUMERIC)
for tloc in tlocs:
try:
locale.setlocale(locale.LC_NUMERIC, tloc)
except locale.Error:
continue
break
else:
raise unittest.SkipTest("Test locale not supported "
"(tried %s)" % (', '.join(tlocs)))
cls.enUS_locale = tloc
finally:
locale.setlocale(locale.LC_NUMERIC, oldlocale)
def setUp(self):
oldlocale = locale.setlocale(self.locale_type)
self.addCleanup(locale.setlocale, self.locale_type, oldlocale)
locale.setlocale(self.locale_type, self.enUS_locale)
if verbose:
print("testing with %r..." % self.enUS_locale, end=' ', flush=True)
class BaseCookedTest(unittest.TestCase):
#
# Base class for tests using cooked localeconv() values
#
def setUp(self):
locale._override_localeconv = self.cooked_values
def tearDown(self):
locale._override_localeconv = {}
class CCookedTest(BaseCookedTest):
# A cooked "C" locale
cooked_values = {
'currency_symbol': '',
'decimal_point': '.',
'frac_digits': 127,
'grouping': [],
'int_curr_symbol': '',
'int_frac_digits': 127,
'mon_decimal_point': '',
'mon_grouping': [],
'mon_thousands_sep': '',
'n_cs_precedes': 127,
'n_sep_by_space': 127,
'n_sign_posn': 127,
'negative_sign': '',
'p_cs_precedes': 127,
'p_sep_by_space': 127,
'p_sign_posn': 127,
'positive_sign': '',
'thousands_sep': ''
}
class EnUSCookedTest(BaseCookedTest):
# A cooked "en_US" locale
cooked_values = {
'currency_symbol': '$',
'decimal_point': '.',
'frac_digits': 2,
'grouping': [3, 3, 0],
'int_curr_symbol': 'USD ',
'int_frac_digits': 2,
'mon_decimal_point': '.',
'mon_grouping': [3, 3, 0],
'mon_thousands_sep': ',',
'n_cs_precedes': 1,
'n_sep_by_space': 0,
'n_sign_posn': 1,
'negative_sign': '-',
'p_cs_precedes': 1,
'p_sep_by_space': 0,
'p_sign_posn': 1,
'positive_sign': '',
'thousands_sep': ','
}
class FrFRCookedTest(BaseCookedTest):
# A cooked "fr_FR" locale with a space character as decimal separator
# and a non-ASCII currency symbol.
cooked_values = {
'currency_symbol': '\u20ac',
'decimal_point': ',',
'frac_digits': 2,
'grouping': [3, 3, 0],
'int_curr_symbol': 'EUR ',
'int_frac_digits': 2,
'mon_decimal_point': ',',
'mon_grouping': [3, 3, 0],
'mon_thousands_sep': ' ',
'n_cs_precedes': 0,
'n_sep_by_space': 1,
'n_sign_posn': 1,
'negative_sign': '-',
'p_cs_precedes': 0,
'p_sep_by_space': 1,
'p_sign_posn': 1,
'positive_sign': '',
'thousands_sep': ' '
}
class BaseFormattingTest(object):
#
# Utility functions for formatting tests
#
def _test_formatfunc(self, format, value, out, func, **format_opts):
self.assertEqual(
func(format, value, **format_opts), out)
def _test_format(self, format, value, out, **format_opts):
with check_warnings(('', DeprecationWarning)):
self._test_formatfunc(format, value, out,
func=locale.format, **format_opts)
def _test_format_string(self, format, value, out, **format_opts):
self._test_formatfunc(format, value, out,
func=locale.format_string, **format_opts)
def _test_currency(self, value, out, **format_opts):
self.assertEqual(locale.currency(value, **format_opts), out)
class EnUSNumberFormatting(BaseFormattingTest):
# XXX there is a grouping + padding bug when the thousands separator
# is empty but the grouping array contains values (e.g. Solaris 10)
def setUp(self):
self.sep = locale.localeconv()['thousands_sep']
def test_grouping(self):
self._test_format("%f", 1024, grouping=1, out='1%s024.000000' % self.sep)
self._test_format("%f", 102, grouping=1, out='102.000000')
self._test_format("%f", -42, grouping=1, out='-42.000000')
self._test_format("%+f", -42, grouping=1, out='-42.000000')
def test_grouping_and_padding(self):
self._test_format("%20.f", -42, grouping=1, out='-42'.rjust(20))
if self.sep:
self._test_format("%+10.f", -4200, grouping=1,
out=('-4%s200' % self.sep).rjust(10))
self._test_format("%-10.f", -4200, grouping=1,
out=('-4%s200' % self.sep).ljust(10))
def test_integer_grouping(self):
self._test_format("%d", 4200, grouping=True, out='4%s200' % self.sep)
self._test_format("%+d", 4200, grouping=True, out='+4%s200' % self.sep)
self._test_format("%+d", -4200, grouping=True, out='-4%s200' % self.sep)
def test_integer_grouping_and_padding(self):
self._test_format("%10d", 4200, grouping=True,
out=('4%s200' % self.sep).rjust(10))
self._test_format("%-10d", -4200, grouping=True,
out=('-4%s200' % self.sep).ljust(10))
def test_simple(self):
self._test_format("%f", 1024, grouping=0, out='1024.000000')
self._test_format("%f", 102, grouping=0, out='102.000000')
self._test_format("%f", -42, grouping=0, out='-42.000000')
self._test_format("%+f", -42, grouping=0, out='-42.000000')
def test_padding(self):
self._test_format("%20.f", -42, grouping=0, out='-42'.rjust(20))
self._test_format("%+10.f", -4200, grouping=0, out='-4200'.rjust(10))
self._test_format("%-10.f", 4200, grouping=0, out='4200'.ljust(10))
def test_format_deprecation(self):
with self.assertWarns(DeprecationWarning):
locale.format("%-10.f", 4200, grouping=True)
def test_complex_formatting(self):
# Spaces in formatting string
self._test_format_string("One million is %i", 1000000, grouping=1,
out='One million is 1%s000%s000' % (self.sep, self.sep))
self._test_format_string("One million is %i", 1000000, grouping=1,
out='One million is 1%s000%s000' % (self.sep, self.sep))
# Dots in formatting string
self._test_format_string(".%f.", 1000.0, out='.1000.000000.')
# Padding
if self.sep:
self._test_format_string("--> %10.2f", 4200, grouping=1,
out='--> ' + ('4%s200.00' % self.sep).rjust(10))
# Asterisk formats
self._test_format_string("%10.*f", (2, 1000), grouping=0,
out='1000.00'.rjust(10))
if self.sep:
self._test_format_string("%*.*f", (10, 2, 1000), grouping=1,
out=('1%s000.00' % self.sep).rjust(10))
# Test more-in-one
if self.sep:
self._test_format_string("int %i float %.2f str %s",
(1000, 1000.0, 'str'), grouping=1,
out='int 1%s000 float 1%s000.00 str str' %
(self.sep, self.sep))
class TestFormatPatternArg(unittest.TestCase):
# Test handling of pattern argument of format
def test_onlyOnePattern(self):
with check_warnings(('', DeprecationWarning)):
# Issue 2522: accept exactly one % pattern, and no extra chars.
self.assertRaises(ValueError, locale.format, "%f\n", 'foo')
self.assertRaises(ValueError, locale.format, "%f\r", 'foo')
self.assertRaises(ValueError, locale.format, "%f\r\n", 'foo')
self.assertRaises(ValueError, locale.format, " %f", 'foo')
self.assertRaises(ValueError, locale.format, "%fg", 'foo')
self.assertRaises(ValueError, locale.format, "%^g", 'foo')
self.assertRaises(ValueError, locale.format, "%f%%", 'foo')
class TestLocaleFormatString(unittest.TestCase):
"""General tests on locale.format_string"""
def test_percent_escape(self):
self.assertEqual(locale.format_string('%f%%', 1.0), '%f%%' % 1.0)
self.assertEqual(locale.format_string('%d %f%%d', (1, 1.0)),
'%d %f%%d' % (1, 1.0))
self.assertEqual(locale.format_string('%(foo)s %%d', {'foo': 'bar'}),
('%(foo)s %%d' % {'foo': 'bar'}))
def test_mapping(self):
self.assertEqual(locale.format_string('%(foo)s bing.', {'foo': 'bar'}),
('%(foo)s bing.' % {'foo': 'bar'}))
self.assertEqual(locale.format_string('%(foo)s', {'foo': 'bar'}),
('%(foo)s' % {'foo': 'bar'}))
class TestNumberFormatting(BaseLocalizedTest, EnUSNumberFormatting):
# Test number formatting with a real English locale.
locale_type = locale.LC_NUMERIC
def setUp(self):
BaseLocalizedTest.setUp(self)
EnUSNumberFormatting.setUp(self)
class TestEnUSNumberFormatting(EnUSCookedTest, EnUSNumberFormatting):
# Test number formatting with a cooked "en_US" locale.
def setUp(self):
EnUSCookedTest.setUp(self)
EnUSNumberFormatting.setUp(self)
def test_currency(self):
self._test_currency(50000, "$50000.00")
self._test_currency(50000, "$50,000.00", grouping=True)
self._test_currency(50000, "USD 50,000.00",
grouping=True, international=True)
class TestCNumberFormatting(CCookedTest, BaseFormattingTest):
# Test number formatting with a cooked "C" locale.
def test_grouping(self):
self._test_format("%.2f", 12345.67, grouping=True, out='12345.67')
def test_grouping_and_padding(self):
self._test_format("%9.2f", 12345.67, grouping=True, out=' 12345.67')
class TestFrFRNumberFormatting(FrFRCookedTest, BaseFormattingTest):
# Test number formatting with a cooked "fr_FR" locale.
def test_decimal_point(self):
self._test_format("%.2f", 12345.67, out='12345,67')
def test_grouping(self):
self._test_format("%.2f", 345.67, grouping=True, out='345,67')
self._test_format("%.2f", 12345.67, grouping=True, out='12 345,67')
def test_grouping_and_padding(self):
self._test_format("%6.2f", 345.67, grouping=True, out='345,67')
self._test_format("%7.2f", 345.67, grouping=True, out=' 345,67')
self._test_format("%8.2f", 12345.67, grouping=True, out='12 345,67')
self._test_format("%9.2f", 12345.67, grouping=True, out='12 345,67')
self._test_format("%10.2f", 12345.67, grouping=True, out=' 12 345,67')
self._test_format("%-6.2f", 345.67, grouping=True, out='345,67')
self._test_format("%-7.2f", 345.67, grouping=True, out='345,67 ')
self._test_format("%-8.2f", 12345.67, grouping=True, out='12 345,67')
self._test_format("%-9.2f", 12345.67, grouping=True, out='12 345,67')
self._test_format("%-10.2f", 12345.67, grouping=True, out='12 345,67 ')
def test_integer_grouping(self):
self._test_format("%d", 200, grouping=True, out='200')
self._test_format("%d", 4200, grouping=True, out='4 200')
def test_integer_grouping_and_padding(self):
self._test_format("%4d", 4200, grouping=True, out='4 200')
self._test_format("%5d", 4200, grouping=True, out='4 200')
self._test_format("%10d", 4200, grouping=True, out='4 200'.rjust(10))
self._test_format("%-4d", 4200, grouping=True, out='4 200')
self._test_format("%-5d", 4200, grouping=True, out='4 200')
self._test_format("%-10d", 4200, grouping=True, out='4 200'.ljust(10))
def test_currency(self):
euro = '\u20ac'
self._test_currency(50000, "50000,00 " + euro)
self._test_currency(50000, "50 000,00 " + euro, grouping=True)
self._test_currency(50000, "50 000,00 EUR",
grouping=True, international=True)
class TestCollation(unittest.TestCase):
# Test string collation functions
def test_strcoll(self):
self.assertLess(locale.strcoll('a', 'b'), 0)
self.assertEqual(locale.strcoll('a', 'a'), 0)
self.assertGreater(locale.strcoll('b', 'a'), 0)
# embedded null character
self.assertRaises(ValueError, locale.strcoll, 'a\0', 'a')
self.assertRaises(ValueError, locale.strcoll, 'a', 'a\0')
def test_strxfrm(self):
self.assertLess(locale.strxfrm('a'), locale.strxfrm('b'))
# embedded null character
self.assertRaises(ValueError, locale.strxfrm, 'a\0')
class TestEnUSCollation(BaseLocalizedTest, TestCollation):
# Test string collation functions with a real English locale
locale_type = locale.LC_ALL
def setUp(self):
enc = codecs.lookup(locale.getpreferredencoding(False) or 'ascii').name
if enc not in ('utf-8', 'iso8859-1', 'cp1252'):
raise unittest.SkipTest('encoding not suitable')
if enc != 'iso8859-1' and (sys.platform == 'darwin' or is_android or
sys.platform.startswith('freebsd')):
raise unittest.SkipTest('wcscoll/wcsxfrm have known bugs')
BaseLocalizedTest.setUp(self)
@unittest.skipIf(sys.platform.startswith('aix'),
'bpo-29972: broken test on AIX')
def test_strcoll_with_diacritic(self):
self.assertLess(locale.strcoll('à', 'b'), 0)
@unittest.skipIf(sys.platform.startswith('aix'),
'bpo-29972: broken test on AIX')
def test_strxfrm_with_diacritic(self):
self.assertLess(locale.strxfrm('à'), locale.strxfrm('b'))
class NormalizeTest(unittest.TestCase):
def check(self, localename, expected):
self.assertEqual(locale.normalize(localename), expected, msg=localename)
def test_locale_alias(self):
for localename, alias in locale.locale_alias.items():
with self.subTest(locale=(localename, alias)):
self.check(localename, alias)
def test_empty(self):
self.check('', '')
def test_c(self):
self.check('c', 'C')
self.check('posix', 'C')
def test_english(self):
self.check('en', 'en_US.ISO8859-1')
self.check('EN', 'en_US.ISO8859-1')
self.check('en.iso88591', 'en_US.ISO8859-1')
self.check('en_US', 'en_US.ISO8859-1')
self.check('en_us', 'en_US.ISO8859-1')
self.check('en_GB', 'en_GB.ISO8859-1')
self.check('en_US.UTF-8', 'en_US.UTF-8')
self.check('en_US.utf8', 'en_US.UTF-8')
self.check('en_US:UTF-8', 'en_US.UTF-8')
self.check('en_US.ISO8859-1', 'en_US.ISO8859-1')
self.check('en_US.US-ASCII', 'en_US.ISO8859-1')
self.check('en_US.88591', 'en_US.ISO8859-1')
self.check('en_US.885915', 'en_US.ISO8859-15')
self.check('english', 'en_EN.ISO8859-1')
self.check('english_uk.ascii', 'en_GB.ISO8859-1')
def test_hyphenated_encoding(self):
self.check('az_AZ.iso88599e', 'az_AZ.ISO8859-9E')
self.check('az_AZ.ISO8859-9E', 'az_AZ.ISO8859-9E')
self.check('tt_RU.koi8c', 'tt_RU.KOI8-C')
self.check('tt_RU.KOI8-C', 'tt_RU.KOI8-C')
self.check('lo_LA.cp1133', 'lo_LA.IBM-CP1133')
self.check('lo_LA.ibmcp1133', 'lo_LA.IBM-CP1133')
self.check('lo_LA.IBM-CP1133', 'lo_LA.IBM-CP1133')
self.check('uk_ua.microsoftcp1251', 'uk_UA.CP1251')
self.check('uk_ua.microsoft-cp1251', 'uk_UA.CP1251')
self.check('ka_ge.georgianacademy', 'ka_GE.GEORGIAN-ACADEMY')
self.check('ka_GE.GEORGIAN-ACADEMY', 'ka_GE.GEORGIAN-ACADEMY')
self.check('cs_CZ.iso88592', 'cs_CZ.ISO8859-2')
self.check('cs_CZ.ISO8859-2', 'cs_CZ.ISO8859-2')
def test_euro_modifier(self):
self.check('de_DE@euro', 'de_DE.ISO8859-15')
self.check('en_US.ISO8859-15@euro', 'en_US.ISO8859-15')
self.check('de_DE.utf8@euro', 'de_DE.UTF-8')
def test_latin_modifier(self):
self.check('be_BY.UTF-8@latin', 'be_BY.UTF-8@latin')
self.check('sr_RS.UTF-8@latin', 'sr_RS.UTF-8@latin')
self.check('sr_RS.UTF-8@latn', 'sr_RS.UTF-8@latin')
def test_valencia_modifier(self):
self.check('ca_ES.UTF-8@valencia', 'ca_ES.UTF-8@valencia')
self.check('ca_ES@valencia', 'ca_ES.UTF-8@valencia')
self.check('ca@valencia', 'ca_ES.ISO8859-1@valencia')
def test_devanagari_modifier(self):
self.check('ks_IN.UTF-8@devanagari', 'ks_IN.UTF-8@devanagari')
self.check('ks_IN@devanagari', 'ks_IN.UTF-8@devanagari')
self.check('ks@devanagari', 'ks_IN.UTF-8@devanagari')
self.check('ks_IN.UTF-8', 'ks_IN.UTF-8')
self.check('ks_IN', 'ks_IN.UTF-8')
self.check('ks', 'ks_IN.UTF-8')
self.check('sd_IN.UTF-8@devanagari', 'sd_IN.UTF-8@devanagari')
self.check('sd_IN@devanagari', 'sd_IN.UTF-8@devanagari')
self.check('sd@devanagari', 'sd_IN.UTF-8@devanagari')
self.check('sd_IN.UTF-8', 'sd_IN.UTF-8')
self.check('sd_IN', 'sd_IN.UTF-8')
self.check('sd', 'sd_IN.UTF-8')
def test_euc_encoding(self):
self.check('ja_jp.euc', 'ja_JP.eucJP')
self.check('ja_jp.eucjp', 'ja_JP.eucJP')
self.check('ko_kr.euc', 'ko_KR.eucKR')
self.check('ko_kr.euckr', 'ko_KR.eucKR')
self.check('zh_cn.euc', 'zh_CN.eucCN')
self.check('zh_tw.euc', 'zh_TW.eucTW')
self.check('zh_tw.euctw', 'zh_TW.eucTW')
def test_japanese(self):
self.check('ja', 'ja_JP.eucJP')
self.check('ja.jis', 'ja_JP.JIS7')
self.check('ja.sjis', 'ja_JP.SJIS')
self.check('ja_jp', 'ja_JP.eucJP')
self.check('ja_jp.ajec', 'ja_JP.eucJP')
self.check('ja_jp.euc', 'ja_JP.eucJP')
self.check('ja_jp.eucjp', 'ja_JP.eucJP')
self.check('ja_jp.iso-2022-jp', 'ja_JP.JIS7')
self.check('ja_jp.iso2022jp', 'ja_JP.JIS7')
self.check('ja_jp.jis', 'ja_JP.JIS7')
self.check('ja_jp.jis7', 'ja_JP.JIS7')
self.check('ja_jp.mscode', 'ja_JP.SJIS')
self.check('ja_jp.pck', 'ja_JP.SJIS')
self.check('ja_jp.sjis', 'ja_JP.SJIS')
self.check('ja_jp.ujis', 'ja_JP.eucJP')
self.check('ja_jp.utf8', 'ja_JP.UTF-8')
self.check('japan', 'ja_JP.eucJP')
self.check('japanese', 'ja_JP.eucJP')
self.check('japanese-euc', 'ja_JP.eucJP')
self.check('japanese.euc', 'ja_JP.eucJP')
self.check('japanese.sjis', 'ja_JP.SJIS')
self.check('jp_jp', 'ja_JP.eucJP')
class TestMiscellaneous(unittest.TestCase):
def test_defaults_UTF8(self):
# Issue #18378: on (at least) macOS setting LC_CTYPE to "UTF-8" is
# valid. Futhermore LC_CTYPE=UTF is used by the UTF-8 locale coercing
# during interpreter startup (on macOS).
import _locale
import os
self.assertEqual(locale._parse_localename('UTF-8'), (None, 'UTF-8'))
if hasattr(_locale, '_getdefaultlocale'):
orig_getlocale = _locale._getdefaultlocale
del _locale._getdefaultlocale
else:
orig_getlocale = None
orig_env = {}
try:
for key in ('LC_ALL', 'LC_CTYPE', 'LANG', 'LANGUAGE'):
if key in os.environ:
orig_env[key] = os.environ[key]
del os.environ[key]
os.environ['LC_CTYPE'] = 'UTF-8'
self.assertEqual(locale.getdefaultlocale(), (None, 'UTF-8'))
finally:
for k in orig_env:
os.environ[k] = orig_env[k]
if 'LC_CTYPE' not in orig_env:
del os.environ['LC_CTYPE']
if orig_getlocale is not None:
_locale._getdefaultlocale = orig_getlocale
def test_getpreferredencoding(self):
# Invoke getpreferredencoding to make sure it does not cause exceptions.
enc = locale.getpreferredencoding()
if enc:
# If encoding non-empty, make sure it is valid
codecs.lookup(enc)
def test_strcoll_3303(self):
# test crasher from bug #3303
self.assertRaises(TypeError, locale.strcoll, "a", None)
self.assertRaises(TypeError, locale.strcoll, b"a", None)
def test_setlocale_category(self):
locale.setlocale(locale.LC_ALL)
locale.setlocale(locale.LC_TIME)
locale.setlocale(locale.LC_CTYPE)
locale.setlocale(locale.LC_COLLATE)
locale.setlocale(locale.LC_MONETARY)
locale.setlocale(locale.LC_NUMERIC)
# crasher from bug #7419
self.assertRaises(locale.Error, locale.setlocale, 12345)
def test_getsetlocale_issue1813(self):
# Issue #1813: setting and getting the locale under a Turkish locale
oldlocale = locale.setlocale(locale.LC_CTYPE)
self.addCleanup(locale.setlocale, locale.LC_CTYPE, oldlocale)
try:
locale.setlocale(locale.LC_CTYPE, 'tr_TR')
except locale.Error:
# Unsupported locale on this system
self.skipTest('test needs Turkish locale')
loc = locale.getlocale(locale.LC_CTYPE)
if verbose:
print('testing with %a' % (loc,), end=' ', flush=True)
locale.setlocale(locale.LC_CTYPE, loc)
self.assertEqual(loc, locale.getlocale(locale.LC_CTYPE))
def test_invalid_locale_format_in_localetuple(self):
with self.assertRaises(TypeError):
locale.setlocale(locale.LC_ALL, b'fi_FI')
def test_invalid_iterable_in_localetuple(self):
with self.assertRaises(TypeError):
locale.setlocale(locale.LC_ALL, (b'not', b'valid'))
class BaseDelocalizeTest(BaseLocalizedTest):
def _test_delocalize(self, value, out):
self.assertEqual(locale.delocalize(value), out)
def _test_atof(self, value, out):
self.assertEqual(locale.atof(value), out)
def _test_atoi(self, value, out):
self.assertEqual(locale.atoi(value), out)
class TestEnUSDelocalize(EnUSCookedTest, BaseDelocalizeTest):
def test_delocalize(self):
self._test_delocalize('50000.00', '50000.00')
self._test_delocalize('50,000.00', '50000.00')
def test_atof(self):
self._test_atof('50000.00', 50000.)
self._test_atof('50,000.00', 50000.)
def test_atoi(self):
self._test_atoi('50000', 50000)
self._test_atoi('50,000', 50000)
class TestCDelocalizeTest(CCookedTest, BaseDelocalizeTest):
def test_delocalize(self):
self._test_delocalize('50000.00', '50000.00')
def test_atof(self):
self._test_atof('50000.00', 50000.)
def test_atoi(self):
self._test_atoi('50000', 50000)
class TestfrFRDelocalizeTest(FrFRCookedTest, BaseDelocalizeTest):
def test_delocalize(self):
self._test_delocalize('50000,00', '50000.00')
self._test_delocalize('50 000,00', '50000.00')
def test_atof(self):
self._test_atof('50000,00', 50000.)
self._test_atof('50 000,00', 50000.)
def test_atoi(self):
self._test_atoi('50000', 50000)
self._test_atoi('50 000', 50000)
if __name__ == '__main__':
unittest.main()