Staging
v0.5.1
https://github.com/python/cpython
Raw File
Tip revision: 11591c3dda25a4fd27d2ae012972f74d3b8fe578 authored by Benjamin Peterson on 06 June 2010, 00:54:29 UTC
fix sphinx warning with an extra space
Tip revision: 11591c3
test_unicodedata.py
""" Test script for the unicodedata module.

    Written by Marc-Andre Lemburg (mal@lemburg.com).

    (c) Copyright CNRI, All Rights Reserved. NO WARRANTY.

"""

import sys
import unittest
import hashlib
import subprocess
import test.test_support

encoding = 'utf-8'


### Run tests

class UnicodeMethodsTest(unittest.TestCase):

    # update this, if the database changes
    expectedchecksum = '4504dffd035baea02c5b9de82bebc3d65e0e0baf'

    def test_method_checksum(self):
        h = hashlib.sha1()
        for i in range(0x10000):
            char = unichr(i)
            data = [
                # Predicates (single char)
                u"01"[char.isalnum()],
                u"01"[char.isalpha()],
                u"01"[char.isdecimal()],
                u"01"[char.isdigit()],
                u"01"[char.islower()],
                u"01"[char.isnumeric()],
                u"01"[char.isspace()],
                u"01"[char.istitle()],
                u"01"[char.isupper()],

                # Predicates (multiple chars)
                u"01"[(char + u'abc').isalnum()],
                u"01"[(char + u'abc').isalpha()],
                u"01"[(char + u'123').isdecimal()],
                u"01"[(char + u'123').isdigit()],
                u"01"[(char + u'abc').islower()],
                u"01"[(char + u'123').isnumeric()],
                u"01"[(char + u' \t').isspace()],
                u"01"[(char + u'abc').istitle()],
                u"01"[(char + u'ABC').isupper()],

                # Mappings (single char)
                char.lower(),
                char.upper(),
                char.title(),

                # Mappings (multiple chars)
                (char + u'abc').lower(),
                (char + u'ABC').upper(),
                (char + u'abc').title(),
                (char + u'ABC').title(),

                ]
            h.update(u''.join(data).encode(encoding))
        result = h.hexdigest()
        self.assertEqual(result, self.expectedchecksum)

class UnicodeDatabaseTest(unittest.TestCase):

    def setUp(self):
        # In case unicodedata is not available, this will raise an ImportError,
        # but the other test cases will still be run
        import unicodedata
        self.db = unicodedata

    def tearDown(self):
        del self.db

class UnicodeFunctionsTest(UnicodeDatabaseTest):

    # update this, if the database changes
    expectedchecksum = '6ccf1b1a36460d2694f9b0b0f0324942fe70ede6'

    def test_function_checksum(self):
        data = []
        h = hashlib.sha1()

        for i in range(0x10000):
            char = unichr(i)
            data = [
                # Properties
                str(self.db.digit(char, -1)),
                str(self.db.numeric(char, -1)),
                str(self.db.decimal(char, -1)),
                self.db.category(char),
                self.db.bidirectional(char),
                self.db.decomposition(char),
                str(self.db.mirrored(char)),
                str(self.db.combining(char)),
            ]
            h.update(''.join(data))
        result = h.hexdigest()
        self.assertEqual(result, self.expectedchecksum)

    def test_digit(self):
        self.assertEqual(self.db.digit(u'A', None), None)
        self.assertEqual(self.db.digit(u'9'), 9)
        self.assertEqual(self.db.digit(u'\u215b', None), None)
        self.assertEqual(self.db.digit(u'\u2468'), 9)
        self.assertEqual(self.db.digit(u'\U00020000', None), None)

        self.assertRaises(TypeError, self.db.digit)
        self.assertRaises(TypeError, self.db.digit, u'xx')
        self.assertRaises(ValueError, self.db.digit, u'x')

    def test_numeric(self):
        self.assertEqual(self.db.numeric(u'A',None), None)
        self.assertEqual(self.db.numeric(u'9'), 9)
        self.assertEqual(self.db.numeric(u'\u215b'), 0.125)
        self.assertEqual(self.db.numeric(u'\u2468'), 9.0)
        self.assertEqual(self.db.numeric(u'\ua627'), 7.0)
        self.assertEqual(self.db.numeric(u'\U00020000', None), None)

        self.assertRaises(TypeError, self.db.numeric)
        self.assertRaises(TypeError, self.db.numeric, u'xx')
        self.assertRaises(ValueError, self.db.numeric, u'x')

    def test_decimal(self):
        self.assertEqual(self.db.decimal(u'A',None), None)
        self.assertEqual(self.db.decimal(u'9'), 9)
        self.assertEqual(self.db.decimal(u'\u215b', None), None)
        self.assertEqual(self.db.decimal(u'\u2468', None), None)
        self.assertEqual(self.db.decimal(u'\U00020000', None), None)

        self.assertRaises(TypeError, self.db.decimal)
        self.assertRaises(TypeError, self.db.decimal, u'xx')
        self.assertRaises(ValueError, self.db.decimal, u'x')

    def test_category(self):
        self.assertEqual(self.db.category(u'\uFFFE'), 'Cn')
        self.assertEqual(self.db.category(u'a'), 'Ll')
        self.assertEqual(self.db.category(u'A'), 'Lu')
        self.assertEqual(self.db.category(u'\U00020000'), 'Lo')

        self.assertRaises(TypeError, self.db.category)
        self.assertRaises(TypeError, self.db.category, u'xx')

    def test_bidirectional(self):
        self.assertEqual(self.db.bidirectional(u'\uFFFE'), '')
        self.assertEqual(self.db.bidirectional(u' '), 'WS')
        self.assertEqual(self.db.bidirectional(u'A'), 'L')
        self.assertEqual(self.db.bidirectional(u'\U00020000'), 'L')

        self.assertRaises(TypeError, self.db.bidirectional)
        self.assertRaises(TypeError, self.db.bidirectional, u'xx')

    def test_decomposition(self):
        self.assertEqual(self.db.decomposition(u'\uFFFE'),'')
        self.assertEqual(self.db.decomposition(u'\u00bc'), '<fraction> 0031 2044 0034')

        self.assertRaises(TypeError, self.db.decomposition)
        self.assertRaises(TypeError, self.db.decomposition, u'xx')

    def test_mirrored(self):
        self.assertEqual(self.db.mirrored(u'\uFFFE'), 0)
        self.assertEqual(self.db.mirrored(u'a'), 0)
        self.assertEqual(self.db.mirrored(u'\u2201'), 1)
        self.assertEqual(self.db.mirrored(u'\U00020000'), 0)

        self.assertRaises(TypeError, self.db.mirrored)
        self.assertRaises(TypeError, self.db.mirrored, u'xx')

    def test_combining(self):
        self.assertEqual(self.db.combining(u'\uFFFE'), 0)
        self.assertEqual(self.db.combining(u'a'), 0)
        self.assertEqual(self.db.combining(u'\u20e1'), 230)
        self.assertEqual(self.db.combining(u'\U00020000'), 0)

        self.assertRaises(TypeError, self.db.combining)
        self.assertRaises(TypeError, self.db.combining, u'xx')

    def test_normalize(self):
        self.assertRaises(TypeError, self.db.normalize)
        self.assertRaises(ValueError, self.db.normalize, 'unknown', u'xx')
        self.assertEqual(self.db.normalize('NFKC', u''), u'')
        # The rest can be found in test_normalization.py
        # which requires an external file.

    def test_pr29(self):
        # http://www.unicode.org/review/pr-29.html
        for text in (u"\u0b47\u0300\u0b3e", u"\u1100\u0300\u1161"):
            self.assertEqual(self.db.normalize('NFC', text), text)

    def test_east_asian_width(self):
        eaw = self.db.east_asian_width
        self.assertRaises(TypeError, eaw, 'a')
        self.assertRaises(TypeError, eaw, u'')
        self.assertRaises(TypeError, eaw, u'ra')
        self.assertEqual(eaw(u'\x1e'), 'N')
        self.assertEqual(eaw(u'\x20'), 'Na')
        self.assertEqual(eaw(u'\uC894'), 'W')
        self.assertEqual(eaw(u'\uFF66'), 'H')
        self.assertEqual(eaw(u'\uFF1F'), 'F')
        self.assertEqual(eaw(u'\u2010'), 'A')
        self.assertEqual(eaw(u'\U00020000'), 'W')

class UnicodeMiscTest(UnicodeDatabaseTest):

    def test_failed_import_during_compiling(self):
        # Issue 4367
        # Decoding \N escapes requires the unicodedata module. If it can't be
        # imported, we shouldn't segfault.

        # This program should raise a SyntaxError in the eval.
        code = "import sys;" \
            "sys.modules['unicodedata'] = None;" \
            """eval("u'\N{SOFT HYPHEN}'")"""
        args = [sys.executable, "-c", code]
        # We use a subprocess because the unicodedata module may already have
        # been loaded in this process.
        popen = subprocess.Popen(args, stderr=subprocess.PIPE)
        popen.wait()
        self.assertEqual(popen.returncode, 1)
        error = "SyntaxError: (unicode error) \N escapes not supported " \
            "(can't load unicodedata module)"
        self.assertIn(error, popen.stderr.read())

    def test_decimal_numeric_consistent(self):
        # Test that decimal and numeric are consistent,
        # i.e. if a character has a decimal value,
        # its numeric value should be the same.
        count = 0
        for i in xrange(0x10000):
            c = unichr(i)
            dec = self.db.decimal(c, -1)
            if dec != -1:
                self.assertEqual(dec, self.db.numeric(c))
                count += 1
        self.assertTrue(count >= 10) # should have tested at least the ASCII digits

    def test_digit_numeric_consistent(self):
        # Test that digit and numeric are consistent,
        # i.e. if a character has a digit value,
        # its numeric value should be the same.
        count = 0
        for i in xrange(0x10000):
            c = unichr(i)
            dec = self.db.digit(c, -1)
            if dec != -1:
                self.assertEqual(dec, self.db.numeric(c))
                count += 1
        self.assertTrue(count >= 10) # should have tested at least the ASCII digits

    def test_bug_1704793(self):
        self.assertEquals(self.db.lookup("GOTHIC LETTER FAIHU"), u'\U00010346')

    def test_ucd_510(self):
        import unicodedata
        # In UCD 5.1.0, a mirrored property changed wrt. UCD 3.2.0
        self.assertTrue(unicodedata.mirrored(u"\u0f3a"))
        self.assertTrue(not unicodedata.ucd_3_2_0.mirrored(u"\u0f3a"))
        # Also, we now have two ways of representing
        # the upper-case mapping: as delta, or as absolute value
        self.assertTrue(u"a".upper()==u'A')
        self.assertTrue(u"\u1d79".upper()==u'\ua77d')
        self.assertTrue(u".".upper()==u".")

    def test_bug_5828(self):
        self.assertEqual(u"\u1d79".lower(), u"\u1d79")
        # Only U+0000 should have U+0000 as its upper/lower/titlecase variant
        self.assertEqual(
            [
                c for c in range(sys.maxunicode+1)
                if u"\x00" in unichr(c).lower()+unichr(c).upper()+unichr(c).title()
            ],
            [0]
        )

    def test_bug_4971(self):
        # LETTER DZ WITH CARON: DZ, Dz, dz
        self.assertEqual(u"\u01c4".title(), u"\u01c5")
        self.assertEqual(u"\u01c5".title(), u"\u01c5")
        self.assertEqual(u"\u01c6".title(), u"\u01c5")

    def test_linebreak_7643(self):
        for i in range(0x10000):
            lines = (unichr(i) + u'A').splitlines()
            if i in (0x0a, 0x0b, 0x0c, 0x0d, 0x85,
                     0x1c, 0x1d, 0x1e, 0x2028, 0x2029):
                self.assertEqual(len(lines), 2,
                                 r"\u%.4x should be a linebreak" % i)
            else:
                self.assertEqual(len(lines), 1,
                                 r"\u%.4x should not be a linebreak" % i)

def test_main():
    test.test_support.run_unittest(
        UnicodeMiscTest,
        UnicodeMethodsTest,
        UnicodeFunctionsTest
    )

if __name__ == "__main__":
    test_main()
back to top