| """ Test script for the Unicode implementation. |
| |
| Written by Bill Tutt. |
| Modified for Python 2.0 by Fredrik Lundh (fredrik@pythonware.com) |
| |
| (c) Copyright CNRI, All Rights Reserved. NO WARRANTY. |
| |
| """#" |
| |
| import unittest |
| |
| from test import support |
| |
| class UnicodeNamesTest(unittest.TestCase): |
| |
| def checkletter(self, name, code): |
| # Helper that put all \N escapes inside eval'd raw strings, |
| # to make sure this script runs even if the compiler |
| # chokes on \N escapes |
| res = eval(r'"\N{%s}"' % name) |
| self.assertEqual(res, code) |
| return res |
| |
| def test_general(self): |
| # General and case insensitivity test: |
| chars = [ |
| "LATIN CAPITAL LETTER T", |
| "LATIN SMALL LETTER H", |
| "LATIN SMALL LETTER E", |
| "SPACE", |
| "LATIN SMALL LETTER R", |
| "LATIN CAPITAL LETTER E", |
| "LATIN SMALL LETTER D", |
| "SPACE", |
| "LATIN SMALL LETTER f", |
| "LATIN CAPITAL LeTtEr o", |
| "LATIN SMaLl LETTER x", |
| "SPACE", |
| "LATIN SMALL LETTER A", |
| "LATIN SMALL LETTER T", |
| "LATIN SMALL LETTER E", |
| "SPACE", |
| "LATIN SMALL LETTER T", |
| "LATIN SMALL LETTER H", |
| "LATIN SMALL LETTER E", |
| "SpAcE", |
| "LATIN SMALL LETTER S", |
| "LATIN SMALL LETTER H", |
| "LATIN small LETTER e", |
| "LATIN small LETTER e", |
| "LATIN SMALL LETTER P", |
| "FULL STOP" |
| ] |
| string = "The rEd fOx ate the sheep." |
| |
| self.assertEqual( |
| "".join([self.checkletter(*args) for args in zip(chars, string)]), |
| string |
| ) |
| |
| def test_ascii_letters(self): |
| import unicodedata |
| |
| for char in "".join(map(chr, range(ord("a"), ord("z")))): |
| name = "LATIN SMALL LETTER %s" % char.upper() |
| code = unicodedata.lookup(name) |
| self.assertEqual(unicodedata.name(code), name) |
| |
| def test_hangul_syllables(self): |
| self.checkletter("HANGUL SYLLABLE GA", "\uac00") |
| self.checkletter("HANGUL SYLLABLE GGWEOSS", "\uafe8") |
| self.checkletter("HANGUL SYLLABLE DOLS", "\ub3d0") |
| self.checkletter("HANGUL SYLLABLE RYAN", "\ub7b8") |
| self.checkletter("HANGUL SYLLABLE MWIK", "\ubba0") |
| self.checkletter("HANGUL SYLLABLE BBWAEM", "\ubf88") |
| self.checkletter("HANGUL SYLLABLE SSEOL", "\uc370") |
| self.checkletter("HANGUL SYLLABLE YI", "\uc758") |
| self.checkletter("HANGUL SYLLABLE JJYOSS", "\ucb40") |
| self.checkletter("HANGUL SYLLABLE KYEOLS", "\ucf28") |
| self.checkletter("HANGUL SYLLABLE PAN", "\ud310") |
| self.checkletter("HANGUL SYLLABLE HWEOK", "\ud6f8") |
| self.checkletter("HANGUL SYLLABLE HIH", "\ud7a3") |
| |
| import unicodedata |
| self.assertRaises(ValueError, unicodedata.name, "\ud7a4") |
| |
| def test_cjk_unified_ideographs(self): |
| self.checkletter("CJK UNIFIED IDEOGRAPH-3400", "\u3400") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-4DB5", "\u4db5") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-4E00", "\u4e00") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-9FCB", "\u9fCB") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-20000", "\U00020000") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-2A6D6", "\U0002a6d6") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-2A700", "\U0002A700") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-2B734", "\U0002B734") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-2B740", "\U0002B740") |
| self.checkletter("CJK UNIFIED IDEOGRAPH-2B81D", "\U0002B81D") |
| |
| def test_bmp_characters(self): |
| import unicodedata |
| count = 0 |
| for code in range(0x10000): |
| char = chr(code) |
| name = unicodedata.name(char, None) |
| if name is not None: |
| self.assertEqual(unicodedata.lookup(name), char) |
| count += 1 |
| |
| def test_misc_symbols(self): |
| self.checkletter("PILCROW SIGN", "\u00b6") |
| self.checkletter("REPLACEMENT CHARACTER", "\uFFFD") |
| self.checkletter("HALFWIDTH KATAKANA SEMI-VOICED SOUND MARK", "\uFF9F") |
| self.checkletter("FULLWIDTH LATIN SMALL LETTER A", "\uFF41") |
| |
| def test_errors(self): |
| import unicodedata |
| self.assertRaises(TypeError, unicodedata.name) |
| self.assertRaises(TypeError, unicodedata.name, 'xx') |
| self.assertRaises(TypeError, unicodedata.lookup) |
| self.assertRaises(KeyError, unicodedata.lookup, 'unknown') |
| |
| def test_strict_error_handling(self): |
| # bogus character name |
| self.assertRaises( |
| UnicodeError, |
| str, b"\\N{blah}", 'unicode-escape', 'strict' |
| ) |
| # long bogus character name |
| self.assertRaises( |
| UnicodeError, |
| str, bytes("\\N{%s}" % ("x" * 100000), "ascii"), 'unicode-escape', 'strict' |
| ) |
| # missing closing brace |
| self.assertRaises( |
| UnicodeError, |
| str, b"\\N{SPACE", 'unicode-escape', 'strict' |
| ) |
| # missing opening brace |
| self.assertRaises( |
| UnicodeError, |
| str, b"\\NSPACE", 'unicode-escape', 'strict' |
| ) |
| |
| def test_main(): |
| support.run_unittest(UnicodeNamesTest) |
| |
| if __name__ == "__main__": |
| test_main() |