from youtube_dl.utils import get_filesystem_encoding
from youtube_dl.compat import (
compat_getenv,
+ compat_etree_fromstring,
compat_expanduser,
+ compat_shlex_split,
+ compat_str,
+ compat_urllib_parse_unquote,
+ compat_urllib_parse_unquote_plus,
)
dir(youtube_dl.compat))) - set(['unicode_literals'])
self.assertEqual(all_names, sorted(present_names))
+ def test_compat_urllib_parse_unquote(self):
+ self.assertEqual(compat_urllib_parse_unquote('abc%20def'), 'abc def')
+ self.assertEqual(compat_urllib_parse_unquote('%7e/abc+def'), '~/abc+def')
+ self.assertEqual(compat_urllib_parse_unquote(''), '')
+ self.assertEqual(compat_urllib_parse_unquote('%'), '%')
+ self.assertEqual(compat_urllib_parse_unquote('%%'), '%%')
+ self.assertEqual(compat_urllib_parse_unquote('%%%'), '%%%')
+ self.assertEqual(compat_urllib_parse_unquote('%2F'), '/')
+ self.assertEqual(compat_urllib_parse_unquote('%2f'), '/')
+ self.assertEqual(compat_urllib_parse_unquote('%E6%B4%A5%E6%B3%A2'), '津波')
+ self.assertEqual(
+ compat_urllib_parse_unquote('''<meta property="og:description" content="%E2%96%81%E2%96%82%E2%96%83%E2%96%84%25%E2%96%85%E2%96%86%E2%96%87%E2%96%88" />
+%<a href="https://ar.wikipedia.org/wiki/%D8%AA%D8%B3%D9%88%D9%86%D8%A7%D9%85%D9%8A">%a'''),
+ '''<meta property="og:description" content="▁▂▃▄%▅▆▇█" />
+%<a href="https://ar.wikipedia.org/wiki/تسونامي">%a''')
+ self.assertEqual(
+ compat_urllib_parse_unquote('''%28%5E%E2%97%A3_%E2%97%A2%5E%29%E3%81%A3%EF%B8%BB%E3%83%87%E2%95%90%E4%B8%80 %E2%87%80 %E2%87%80 %E2%87%80 %E2%87%80 %E2%87%80 %E2%86%B6%I%Break%25Things%'''),
+ '''(^◣_◢^)っ︻デ═一 ⇀ ⇀ ⇀ ⇀ ⇀ ↶%I%Break%Things%''')
+
+ def test_compat_urllib_parse_unquote_plus(self):
+ self.assertEqual(compat_urllib_parse_unquote_plus('abc%20def'), 'abc def')
+ self.assertEqual(compat_urllib_parse_unquote_plus('%7e/abc+def'), '~/abc def')
+
+ def test_compat_shlex_split(self):
+ self.assertEqual(compat_shlex_split('-option "one two"'), ['-option', 'one two'])
+
+ def test_compat_etree_fromstring(self):
+ xml = '''
+ <root foo="bar" spam="中文">
+ <normal>foo</normal>
+ <chinese>中文</chinese>
+ <foo><bar>spam</bar></foo>
+ </root>
+ '''
+ doc = compat_etree_fromstring(xml.encode('utf-8'))
+ self.assertTrue(isinstance(doc.attrib['foo'], compat_str))
+ self.assertTrue(isinstance(doc.attrib['spam'], compat_str))
+ self.assertTrue(isinstance(doc.find('normal').text, compat_str))
+ self.assertTrue(isinstance(doc.find('chinese').text, compat_str))
+ self.assertTrue(isinstance(doc.find('foo/bar').text, compat_str))
+
if __name__ == '__main__':
unittest.main()