diff options
author | Zachary Ware <zachary.ware@gmail.com> | 2016-09-09 19:55:37 (GMT) |
---|---|---|
committer | Zachary Ware <zachary.ware@gmail.com> | 2016-09-09 19:55:37 (GMT) |
commit | 724f6a67f241f39394b15c419570920ebdfefacb (patch) | |
tree | dc795f9edbd8de67b4ad14fb4ed2ec45f9481ae2 /Lib/test/test_unicode_file_functions.py | |
parent | a0154c0f0ee38b1b8a05eac181a35c78ca1508f4 (diff) | |
download | cpython-724f6a67f241f39394b15c419570920ebdfefacb.zip cpython-724f6a67f241f39394b15c419570920ebdfefacb.tar.gz cpython-724f6a67f241f39394b15c419570920ebdfefacb.tar.bz2 |
Rename test_pep####.py files
Diffstat (limited to 'Lib/test/test_unicode_file_functions.py')
-rw-r--r-- | Lib/test/test_unicode_file_functions.py | 195 |
1 files changed, 195 insertions, 0 deletions
diff --git a/Lib/test/test_unicode_file_functions.py b/Lib/test/test_unicode_file_functions.py new file mode 100644 index 0000000..98c716b --- /dev/null +++ b/Lib/test/test_unicode_file_functions.py @@ -0,0 +1,195 @@ +# Test the Unicode versions of normal file functions +# open, os.open, os.stat. os.listdir, os.rename, os.remove, os.mkdir, os.chdir, os.rmdir +import os +import sys +import unittest +import warnings +from unicodedata import normalize +from test import support + +filenames = [ + '1_abc', + '2_ascii', + '3_Gr\xfc\xdf-Gott', + '4_\u0393\u03b5\u03b9\u03ac-\u03c3\u03b1\u03c2', + '5_\u0417\u0434\u0440\u0430\u0432\u0441\u0442\u0432\u0443\u0439\u0442\u0435', + '6_\u306b\u307d\u3093', + '7_\u05d4\u05e9\u05e7\u05e6\u05e5\u05e1', + '8_\u66e8\u66e9\u66eb', + '9_\u66e8\u05e9\u3093\u0434\u0393\xdf', + # Specific code points: fn, NFC(fn) and NFKC(fn) all differents + '10_\u1fee\u1ffd', + ] + +# Mac OS X decomposes Unicode names, using Normal Form D. +# http://developer.apple.com/mac/library/qa/qa2001/qa1173.html +# "However, most volume formats do not follow the exact specification for +# these normal forms. For example, HFS Plus uses a variant of Normal Form D +# in which U+2000 through U+2FFF, U+F900 through U+FAFF, and U+2F800 through +# U+2FAFF are not decomposed." +if sys.platform != 'darwin': + filenames.extend([ + # Specific code points: NFC(fn), NFD(fn), NFKC(fn) and NFKD(fn) all differents + '11_\u0385\u03d3\u03d4', + '12_\u00a8\u0301\u03d2\u0301\u03d2\u0308', # == NFD('\u0385\u03d3\u03d4') + '13_\u0020\u0308\u0301\u038e\u03ab', # == NFKC('\u0385\u03d3\u03d4') + '14_\u1e9b\u1fc1\u1fcd\u1fce\u1fcf\u1fdd\u1fde\u1fdf\u1fed', + + # Specific code points: fn, NFC(fn) and NFKC(fn) all differents + '15_\u1fee\u1ffd\ufad1', + '16_\u2000\u2000\u2000A', + '17_\u2001\u2001\u2001A', + '18_\u2003\u2003\u2003A', # == NFC('\u2001\u2001\u2001A') + '19_\u0020\u0020\u0020A', # '\u0020' == ' ' == NFKC('\u2000') == + # NFKC('\u2001') == NFKC('\u2003') + ]) + + +# Is it Unicode-friendly? +if not os.path.supports_unicode_filenames: + fsencoding = sys.getfilesystemencoding() + try: + for name in filenames: + name.encode(fsencoding) + except UnicodeEncodeError: + raise unittest.SkipTest("only NT+ and systems with " + "Unicode-friendly filesystem encoding") + + +class UnicodeFileTests(unittest.TestCase): + files = set(filenames) + normal_form = None + + def setUp(self): + try: + os.mkdir(support.TESTFN) + except FileExistsError: + pass + self.addCleanup(support.rmtree, support.TESTFN) + + files = set() + for name in self.files: + name = os.path.join(support.TESTFN, self.norm(name)) + with open(name, 'wb') as f: + f.write((name+'\n').encode("utf-8")) + os.stat(name) + files.add(name) + self.files = files + + def norm(self, s): + if self.normal_form: + return normalize(self.normal_form, s) + return s + + def _apply_failure(self, fn, filename, + expected_exception=FileNotFoundError, + check_filename=True): + with self.assertRaises(expected_exception) as c: + fn(filename) + exc_filename = c.exception.filename + if check_filename: + self.assertEqual(exc_filename, filename, "Function '%s(%a) failed " + "with bad filename in the exception: %a" % + (fn.__name__, filename, exc_filename)) + + def test_failures(self): + # Pass non-existing Unicode filenames all over the place. + for name in self.files: + name = "not_" + name + self._apply_failure(open, name) + self._apply_failure(os.stat, name) + self._apply_failure(os.chdir, name) + self._apply_failure(os.rmdir, name) + self._apply_failure(os.remove, name) + self._apply_failure(os.listdir, name) + + if sys.platform == 'win32': + # Windows is lunatic. Issue #13366. + _listdir_failure = NotADirectoryError, FileNotFoundError + else: + _listdir_failure = NotADirectoryError + + def test_open(self): + for name in self.files: + f = open(name, 'wb') + f.write((name+'\n').encode("utf-8")) + f.close() + os.stat(name) + self._apply_failure(os.listdir, name, self._listdir_failure) + + # Skip the test on darwin, because darwin does normalize the filename to + # NFD (a variant of Unicode NFD form). Normalize the filename to NFC, NFKC, + # NFKD in Python is useless, because darwin will normalize it later and so + # open(), os.stat(), etc. don't raise any exception. + @unittest.skipIf(sys.platform == 'darwin', 'irrelevant test on Mac OS X') + def test_normalize(self): + files = set(self.files) + others = set() + for nf in set(['NFC', 'NFD', 'NFKC', 'NFKD']): + others |= set(normalize(nf, file) for file in files) + others -= files + for name in others: + self._apply_failure(open, name) + self._apply_failure(os.stat, name) + self._apply_failure(os.chdir, name) + self._apply_failure(os.rmdir, name) + self._apply_failure(os.remove, name) + self._apply_failure(os.listdir, name) + + # Skip the test on darwin, because darwin uses a normalization different + # than Python NFD normalization: filenames are different even if we use + # Python NFD normalization. + @unittest.skipIf(sys.platform == 'darwin', 'irrelevant test on Mac OS X') + def test_listdir(self): + sf0 = set(self.files) + with warnings.catch_warnings(): + warnings.simplefilter("ignore", DeprecationWarning) + f1 = os.listdir(support.TESTFN.encode(sys.getfilesystemencoding())) + f2 = os.listdir(support.TESTFN) + sf2 = set(os.path.join(support.TESTFN, f) for f in f2) + self.assertEqual(sf0, sf2, "%a != %a" % (sf0, sf2)) + self.assertEqual(len(f1), len(f2)) + + def test_rename(self): + for name in self.files: + os.rename(name, "tmp") + os.rename("tmp", name) + + def test_directory(self): + dirname = os.path.join(support.TESTFN, 'Gr\xfc\xdf-\u66e8\u66e9\u66eb') + filename = '\xdf-\u66e8\u66e9\u66eb' + with support.temp_cwd(dirname): + with open(filename, 'wb') as f: + f.write((filename + '\n').encode("utf-8")) + os.access(filename,os.R_OK) + os.remove(filename) + + +class UnicodeNFCFileTests(UnicodeFileTests): + normal_form = 'NFC' + + +class UnicodeNFDFileTests(UnicodeFileTests): + normal_form = 'NFD' + + +class UnicodeNFKCFileTests(UnicodeFileTests): + normal_form = 'NFKC' + + +class UnicodeNFKDFileTests(UnicodeFileTests): + normal_form = 'NFKD' + + +def test_main(): + support.run_unittest( + UnicodeFileTests, + UnicodeNFCFileTests, + UnicodeNFDFileTests, + UnicodeNFKCFileTests, + UnicodeNFKDFileTests, + ) + + +if __name__ == "__main__": + test_main() |