from test import support import unittest import urllib.parse RFC1808_BASE = "http://a/b/c/d;p?q#f" RFC2396_BASE = "http://a/b/c/d;p?q" RFC3986_BASE = 'http://a/b/c/d;p?q' SIMPLE_BASE = 'http://a/b/c/d' # A list of test cases. Each test case is a two-tuple that contains # a string with the query and a dictionary with the expected result. parse_qsl_test_cases = [ ("", []), ("&", []), ("&&", []), ("=", [('', '')]), ("=a", [('', 'a')]), ("a", [('a', '')]), ("a=", [('a', '')]), ("a=", [('a', '')]), ("&a=b", [('a', 'b')]), ("a=a+b&b=b+c", [('a', 'a b'), ('b', 'b c')]), ("a=1&a=2", [('a', '1'), ('a', '2')]), (b"", []), (b"&", []), (b"&&", []), (b"=", [(b'', b'')]), (b"=a", [(b'', b'a')]), (b"a", [(b'a', b'')]), (b"a=", [(b'a', b'')]), (b"a=", [(b'a', b'')]), (b"&a=b", [(b'a', b'b')]), (b"a=a+b&b=b+c", [(b'a', b'a b'), (b'b', b'b c')]), (b"a=1&a=2", [(b'a', b'1'), (b'a', b'2')]), ] class UrlParseTestCase(unittest.TestCase): def checkRoundtrips(self, url, parsed, split): result = urllib.parse.urlparse(url) self.assertEqual(result, parsed) t = (result.scheme, result.netloc, result.path, result.params, result.query, result.fragment) self.assertEqual(t, parsed) # put it back together and it should be the same result2 = urllib.parse.urlunparse(result) self.assertEqual(result2, url) self.assertEqual(result2, result.geturl()) # the result of geturl() is a fixpoint; we can always parse it # again to get the same result: result3 = urllib.parse.urlparse(result.geturl()) self.assertEqual(result3.geturl(), result.geturl()) self.assertEqual(result3, result) self.assertEqual(result3.scheme, result.scheme) self.assertEqual(result3.netloc, result.netloc) self.assertEqual(result3.path, result.path) self.assertEqual(result3.params, result.params) self.assertEqual(result3.query, result.query) self.assertEqual(result3.fragment, result.fragment) self.assertEqual(result3.username, result.username) self.assertEqual(result3.password, result.password) self.assertEqual(result3.hostname, result.hostname) self.assertEqual(result3.port, result.port) # check the roundtrip using urlsplit() as well result = urllib.parse.urlsplit(url) self.assertEqual(result, split) t = (result.scheme, result.netloc, result.path, result.query, result.fragment) self.assertEqual(t, split) result2 = urllib.parse.urlunsplit(result) self.assertEqual(result2, url) self.assertEqual(result2, result.geturl()) # check the fixpoint property of re-parsing the result of geturl() result3 = urllib.parse.urlsplit(result.geturl()) self.assertEqual(result3.geturl(), result.geturl()) self.assertEqual(result3, result) self.assertEqual(result3.scheme, result.scheme) self.assertEqual(result3.netloc, result.netloc) self.assertEqual(result3.path, result.path) self.assertEqual(result3.query, result.query) self.assertEqual(result3.fragment, result.fragment) self.assertEqual(result3.username, result.username) self.assertEqual(result3.password, result.password) self.assertEqual(result3.hostname, result.hostname) self.assertEqual(result3.port, result.port) def test_qsl(self): for orig, expect in parse_qsl_test_cases: result = urllib.parse.parse_qsl(orig, keep_blank_values=True) self.assertEqual(result, expect, "Error parsing %r" % orig) expect_without_blanks = [v for v in expect if len(v[1])] result = urllib.parse.parse_qsl(orig, keep_blank_values=False) self.assertEqual(result, expect_without_blanks, "Error parsing %r" % orig) def test_roundtrips(self): str_cases = [ ('file:///tmp/junk.txt', ('file', '', '/tmp/junk.txt', '', '', ''), ('file', '', '/tmp/junk.txt', '', '')), ('imap://mail.python.org/mbox1', ('imap', 'mail.python.org', '/mbox1', '', '', ''), ('imap', 'mail.python.org', '/mbox1', '', '')), ('mms://wms.sys.hinet.net/cts/Drama/09006251100.asf', ('mms', 'wms.sys.hinet.net', '/cts/Drama/09006251100.asf', '', '', ''), ('mms', 'wms.sys.hinet.net', '/cts/Drama/09006251100.asf', '', '')), ('nfs://server/path/to/file.txt', ('nfs', 'server', '/path/to/file.txt', '', '', ''), ('nfs', 'server', '/path/to/file.txt', '', '')), ('svn+ssh://svn.zope.org/repos/main/ZConfig/trunk/', ('svn+ssh', 'svn.zope.org', '/repos/main/ZConfig/trunk/', '', '', ''), ('svn+ssh', 'svn.zope.org', '/repos/main/ZConfig/trunk/', '', '')), ('git+ssh://git@github.com/user/project.git', ('git+ssh', 'git@github.com','/user/project.git', '','',''), ('git+ssh', 'git@github.com','/user/project.git', '', '')), ] def _encode(t): return (t[0].encode('ascii'), tuple(x.encode('ascii') for x in t[1]), tuple(x.encode('ascii') for x in t[2])) bytes_cases = [_encode(x) for x in str_cases] for url, parsed, split in str_cases + bytes_cases: self.checkRoundtrips(url, parsed, split) def test_http_roundtrips(self): # urllib.parse.urlsplit treats 'http:' as an optimized special case, # so we test both 'http:' and 'https:' in all the following. # Three cheers for white box knowledge! str_cases = [ ('://www.python.org', ('www.python.org', '', '', '', ''), ('www.python.org', '', '', '')), ('://www.python.org#abc', ('www.python.org', '', '', '', 'abc'), ('www.python.org', '', '', 'abc')), ('://www.python.org?q=abc', ('www.python.org', '', '', 'q=abc', ''), ('www.python.org', '', 'q=abc', '')), ('://www.python.org/#abc', ('www.python.org', '/', '', '', 'abc'), ('www.python.org', '/', '', 'abc')), ('://a/b/c/d;p?q#f', ('a', '/b/c/d', 'p', 'q', 'f'), ('a', '/b/c/d;p', 'q', 'f')), ] def _encode(t): return (t[0].encode('ascii'), tuple(x.encode('ascii') for x in t[1]), tuple(x.encode('ascii') for x in t[2])) bytes_cases = [_encode(x) for x in str_cases] str_schemes = ('http', 'https') bytes_schemes = (b'http', b'https') str_tests = str_schemes, str_cases bytes_tests = bytes_schemes, bytes_cases for schemes, test_cases in (str_tests, bytes_tests): for scheme in schemes: for url, parsed, split in test_cases: url = scheme + url parsed = (scheme,) + parsed split = (scheme,) + split self.checkRoundtrips(url, parsed, split) def checkJoin(self, base, relurl, expected): str_components = (base, relurl, expected) self.assertEqual(urllib.parse.urljoin(base, relurl), expected) bytes_components = baseb, relurlb, expectedb = [ x.encode('ascii') for x in str_components] self.assertEqual(urllib.parse.urljoin(baseb, relurlb), expectedb) def test_unparse_parse(self): str_cases = ['Python', './Python','x-newscheme://foo.com/stuff','x://y','x:/y','x:/','/',] bytes_cases = [x.encode('ascii') for x in str_cases] for u in str_cases + bytes_cases: self.assertEqual(urllib.parse.urlunsplit(urllib.parse.urlsplit(u)), u) self.assertEqual(urllib.parse.urlunparse(urllib.parse.urlparse(u)), u) def test_RFC1808(self): # "normal" cases from RFC 1808: self.checkJoin(RFC1808_BASE, 'g:h', 'g:h') self.checkJoin(RFC1808_BASE, 'g', 'http://a/b/c/g') self.checkJoin(RFC1808_BASE, './g', 'http://a/b/c/g') self.checkJoin(RFC1808_BASE, 'g/', 'http://a/b/c/g/') self.checkJoin(RFC1808_BASE, '/g', 'http://a/g') self.checkJoin(RFC1808_BASE, '//g', 'http://g') self.checkJoin(RFC1808_BASE, 'g?y', 'http://a/b/c/g?y') self.checkJoin(RFC1808_BASE, 'g?y/./x', 'http://a/b/c/g?y/./x') self.checkJoin(RFC1808_BASE, '#s', 'http://a/b/c/d;p?q#s') self.checkJoin(RFC1808_BASE, 'g#s', 'http://a/b/c/g#s') self.checkJoin(RFC1808_BASE, 'g#s/./x', 'http://a/b/c/g#s/./x') self.checkJoin(RFC1808_BASE, 'g?y#s', 'http://a/b/c/g?y#s') self.checkJoin(RFC1808_BASE, 'g;x', 'http://a/b/c/g;x') self.checkJoin(RFC1808_BASE, 'g;x?y#s', 'http://a/b/c/g;x?y#s') self.checkJoin(RFC1808_BASE, '.', 'http://a/b/c/') self.checkJoin(RFC1808_BASE, './', 'http://a/b/c/') self.checkJoin(RFC1808_BASE, '..', 'http://a/b/') self.checkJoin(RFC1808_BASE, '../', 'http://a/b/') self.checkJoin(RFC1808_BASE, '../g', 'http://a/b/g') self.checkJoin(RFC1808_BASE, '../..', 'http://a/') self.checkJoin(RFC1808_BASE, '../../', 'http://a/') self.checkJoin(RFC1808_BASE, '../../g', 'http://a/g') # "abnormal" cases from RFC 1808: self.checkJoin(RFC1808_BASE, '', 'http://a/b/c/d;p?q#f') self.checkJoin(RFC1808_BASE, '../../../g', 'http://a/../g') self.checkJoin(RFC1808_BASE, '../../../../g', 'http://a/../../g') self.checkJoin(RFC1808_BASE, '/./g', 'http://a/./g') self.checkJoin(RFC1808_BASE, '/../g', 'http://a/../g') self.checkJoin(RFC1808_BASE, 'g.', 'http://a/b/c/g.') self.checkJoin(RFC1808_BASE, '.g', 'http://a/b/c/.g') self.checkJoin(RFC1808_BASE, 'g..', 'http://a/b/c/g..') self.checkJoin(RFC1808_BASE, '..g', 'http://a/b/c/..g') self.checkJoin(RFC1808_BASE, './../g', 'http://a/b/g') self.checkJoin(RFC1808_BASE, './g/.', 'http://a/b/c/g/') self.checkJoin(RFC1808_BASE, 'g/./h', 'http://a/b/c/g/h') self.checkJoin(RFC1808_BASE, 'g/../h', 'http://a/b/c/h') # RFC 1808 and RFC 1630 disagree on these (according to RFC 1808), # so we'll not actually run these tests (which expect 1808 behavior). #self.checkJoin(RFC1808_BASE, 'http:g', 'http:g') #self.checkJoin(RFC1808_BASE, 'http:', 'http:') def test_RFC2368(self): # Issue 11467: path that starts with a number is not parsed correctly self.assertEqual(urllib.parse.urlparse('mailto:1337@example.org'), ('mailto', '', '1337@example.org', '', '', '')) def test_RFC2396(self): # cases from RFC 2396 self.checkJoin(RFC2396_BASE, 'g:h', 'g:h') self.checkJoin(RFC2396_BASE, 'g', 'http://a/b/c/g') self.checkJoin(RFC2396_BASE, './g', 'http://a/b/c/g') self.checkJoin(RFC2396_BASE, 'g/', 'http://a/b/c/g/') self.checkJoin(RFC2396_BASE, '/g', 'http://a/g') self.checkJoin(RFC2396_BASE, '//g', 'http://g') self.checkJoin(RFC2396_BASE, 'g?y', 'http://a/b/c/g?y') self.checkJoin(RFC2396_BASE, '#s', 'http://a/b/c/d;p?q#s') self.checkJoin(RFC2396_BASE, 'g#s', 'http://a/b/c/g#s') self.checkJoin(RFC2396_BASE, 'g?y#s', 'http://a/b/c/g?y#s') self.checkJoin(RFC2396_BASE, 'g;x', 'http://a/b/c/g;x') self.checkJoin(RFC2396_BASE, 'g;x?y#s', 'http://a/b/c/g;x?y#s') self.checkJoin(RFC2396_BASE, '.', 'http://a/b/c/') self.checkJoin(RFC2396_BASE, './', 'http://a/b/c/') self.checkJoin(RFC2396_BASE, '..', 'http://a/b/') self.checkJoin(RFC2396_BASE, '../', 'http://a/b/') self.checkJoin(RFC2396_BASE, '../g', 'http://a/b/g') self.checkJoin(RFC2396_BASE, '../..', 'http://a/') self.checkJoin(RFC2396_BASE, '../../', 'http://a/') self.checkJoin(RFC2396_BASE, '../../g', 'http://a/g') self.checkJoin(RFC2396_BASE, '', RFC2396_BASE) self.checkJoin(RFC2396_BASE, '../../../g', 'http://a/../g') self.checkJoin(RFC2396_BASE, '../../../../g', 'http://a/../../g') self.checkJoin(RFC2396_BASE, '/./g', 'http://a/./g') self.checkJoin(RFC2396_BASE, '/../g', 'http://a/../g') self.checkJoin(RFC2396_BASE, 'g.', 'http://a/b/c/g.') self.checkJoin(RFC2396_BASE, '.g', 'http://a/b/c/.g') self.checkJoin(RFC2396_BASE, 'g..', 'http://a/b/c/g..') self.checkJoin(RFC2396_BASE, '..g', 'http://a/b/c/..g') self.checkJoin(RFC2396_BASE, './../g', 'http://a/b/g') self.checkJoin(RFC2396_BASE, './g/.', 'http://a/b/c/g/') self.checkJoin(RFC2396_BASE, 'g/./h', 'http://a/b/c/g/h') self.checkJoin(RFC2396_BASE, 'g/../h', 'http://a/b/c/h') self.checkJoin(RFC2396_BASE, 'g;x=1/./y', 'http://a/b/c/g;x=1/y') self.checkJoin(RFC2396_BASE, 'g;x=1/../y', 'http://a/b/c/y') self.checkJoin(RFC2396_BASE, 'g?y/./x', 'http://a/b/c/g?y/./x') self.checkJoin(RFC2396_BASE, 'g?y/../x', 'http://a/b/c/g?y/../x') self.checkJoin(RFC2396_BASE, 'g#s/./x', 'http://a/b/c/g#s/./x') self.checkJoin(RFC2396_BASE, 'g#s/../x', 'http://a/b/c/g#s/../x') def test_RFC3986(self): # Test cases from RFC3986 self.checkJoin(RFC3986_BASE, '?y','http://a/b/c/d;p?y') self.checkJoin(RFC2396_BASE, ';x', 'http://a/b/c/;x') self.checkJoin(RFC3986_BASE, 'g:h','g:h') self.checkJoin(RFC3986_BASE, 'g','http://a/b/c/g') self.checkJoin(RFC3986_BASE, './g','http://a/b/c/g') self.checkJoin(RFC3986_BASE, 'g/','http://a/b/c/g/') self.checkJoin(RFC3986_BASE, '/g','http://a/g') self.checkJoin(RFC3986_BASE, '//g','http://g') self.checkJoin(RFC3986_BASE, '?y','http://a/b/c/d;p?y') self.checkJoin(RFC3986_BASE, 'g?y','http://a/b/c/g?y') self.checkJoin(RFC3986_BASE, '#s','http://a/b/c/d;p?q#s') self.checkJoin(RFC3986_BASE, 'g#s','http://a/b/c/g#s') self.checkJoin(RFC3986_BASE, 'g?y#s','http://a/b/c/g?y#s') self.checkJoin(RFC3986_BASE, ';x','http://a/b/c/;x') self.checkJoin(RFC3986_BASE, 'g;x','http://a/b/c/g;x') self.checkJoin(RFC3986_BASE, 'g;x?y#s','http://a/b/c/g;x?y#s') self.checkJoin(RFC3986_BASE, '','http://a/b/c/d;p?q') self.checkJoin(RFC3986_BASE, '.','http://a/b/c/') self.checkJoin(RFC3986_BASE, './','http://a/b/c/') self.checkJoin(RFC3986_BASE, '..','http://a/b/') self.checkJoin(RFC3986_BASE, '../','http://a/b/') self.checkJoin(RFC3986_BASE, '../g','http://a/b/g') self.checkJoin(RFC3986_BASE, '../..','http://a/') self.checkJoin(RFC3986_BASE, '../../','http://a/') self.checkJoin(RFC3986_BASE, '../../g','http://a/g') #Abnormal Examples # The 'abnormal scenarios' are incompatible with RFC2986 parsing # Tests are here for reference. #self.checkJoin(RFC3986_BASE, '../../../g','http://a/g') #self.checkJoin(RFC3986_BASE, '../../../../g','http://a/g') #self.checkJoin(RFC3986_BASE, '/./g','http://a/g') #self.checkJoin(RFC3986_BASE, '/../g','http://a/g') self.checkJoin(RFC3986_BASE, 'g.','http://a/b/c/g.') self.checkJoin(RFC3986_BASE, '.g','http://a/b/c/.g') self.checkJoin(RFC3986_BASE, 'g..','http://a/b/c/g..') self.checkJoin(RFC3986_BASE, '..g','http://a/b/c/..g') self.checkJoin(RFC3986_BASE, './../g','http://a/b/g') self.checkJoin(RFC3986_BASE, './g/.','http://a/b/c/g/') self.checkJoin(RFC3986_BASE, 'g/./h','http://a/b/c/g/h') self.checkJoin(RFC3986_BASE, 'g/../h','http://a/b/c/h') self.checkJoin(RFC3986_BASE, 'g;x=1/./y','http://a/b/c/g;x=1/y') self.checkJoin(RFC3986_BASE, 'g;x=1/../y','http://a/b/c/y') self.checkJoin(RFC3986_BASE, 'g?y/./x','http://a/b/c/g?y/./x') self.checkJoin(RFC3986_BASE, 'g?y/../x','http://a/b/c/g?y/../x') self.checkJoin(RFC3986_BASE, 'g#s/./x','http://a/b/c/g#s/./x') self.checkJoin(RFC3986_BASE, 'g#s/../x','http://a/b/c/g#s/../x') #self.checkJoin(RFC3986_BASE, 'http:g','http:g') # strict parser self.checkJoin(RFC3986_BASE, 'http:g','http://a/b/c/g') #relaxed parser # Test for issue9721 self.checkJoin('http://a/b/c/de', ';x','http://a/b/c/;x') def test_urljoins(self): self.checkJoin(SIMPLE_BASE, 'g:h','g:h') self.checkJoin(SIMPLE_BASE, 'http:g','http://a/b/c/g') self.checkJoin(SIMPLE_BASE, 'http:','http://a/b/c/d') self.checkJoin(SIMPLE_BASE, 'g','http://a/b/c/g') self.checkJoin(SIMPLE_BASE, './g','http://a/b/c/g') self.checkJoin(SIMPLE_BASE, 'g/','http://a/b/c/g/') self.checkJoin(SIMPLE_BASE, '/g','http://a/g') self.checkJoin(SIMPLE_BASE, '//g','http://g') self.checkJoin(SIMPLE_BASE, '?y','http://a/b/c/d?y') self.checkJoin(SIMPLE_BASE, 'g?y','http://a/b/c/g?y') self.checkJoin(SIMPLE_BASE, 'g?y/./x','http://a/b/c/g?y/./x') self.checkJoin(SIMPLE_BASE, '.','http://a/b/c/') self.checkJoin(SIMPLE_BASE, './','http://a/b/c/') self.checkJoin(SIMPLE_BASE, '..','http://a/b/') self.checkJoin(SIMPLE_BASE, '../','http://a/b/') self.checkJoin(SIMPLE_BASE, '../g','http://a/b/g') self.checkJoin(SIMPLE_BASE, '../..','http://a/') self.checkJoin(SIMPLE_BASE, '../../g','http://a/g') self.checkJoin(SIMPLE_BASE, '../../../g','http://a/../g') self.checkJoin(SIMPLE_BASE, './../g','http://a/b/g') self.checkJoin(SIMPLE_BASE, './g/.','http://a/b/c/g/') self.checkJoin(SIMPLE_BASE, '/./g','http://a/./g') self.checkJoin(SIMPLE_BASE, 'g/./h','http://a/b/c/g/h') self.checkJoin(SIMPLE_BASE, 'g/../h','http://a/b/c/h') self.checkJoin(SIMPLE_BASE, 'http:g','http://a/b/c/g') self.checkJoin(SIMPLE_BASE, 'http:','http://a/b/c/d') self.checkJoin(SIMPLE_BASE, 'http:?y','http://a/b/c/d?y') self.checkJoin(SIMPLE_BASE, 'http:g?y','http://a/b/c/g?y') self.checkJoin(SIMPLE_BASE, 'http:g?y/./x','http://a/b/c/g?y/./x') self.checkJoin('http:///', '..','http:///') self.checkJoin('', 'http://a/b/c/g?y/./x','http://a/b/c/g?y/./x') self.checkJoin('', 'http://a/./g', 'http://a/./g') self.checkJoin('svn://pathtorepo/dir1', 'dir2', 'svn://pathtorepo/dir2') self.checkJoin('svn+ssh://pathtorepo/dir1', 'dir2', 'svn+ssh://pathtorepo/dir2') def test_RFC2732(self): str_cases = [ ('http://Test.python.org:5432/foo/', 'test.python.org', 5432), ('http://12.34.56.78:5432/foo/', '12.34.56.78', 5432), ('http://[::1]:5432/foo/', '::1', 5432), ('http://[dead:beef::1]:5432/foo/', 'dead:beef::1', 5432), ('http://[dead:beef::]:5432/foo/', 'dead:beef::', 5432), ('http://[dead:beef:cafe:5417:affe:8FA3:deaf:feed]:5432/foo/', 'dead:beef:cafe:5417:affe:8fa3:deaf:feed', 5432), ('http://[::12.34.56.78]:5432/foo/', '::12.34.56.78', 5432), ('http://[::ffff:12.34.56.78]:5432/foo/', '::ffff:12.34.56.78', 5432), ('http://Test.python.org/foo/', 'test.python.org', None), ('http://12.34.56.78/foo/', '12.34.56.78', None), ('http://[::1]/foo/', '::1', None), ('http://[dead:beef::1]/foo/', 'dead:beef::1', None), ('http://[dead:beef::]/foo/', 'dead:beef::', None), ('http://[dead:beef:cafe:5417:affe:8FA3:deaf:feed]/foo/', 'dead:beef:cafe:5417:affe:8fa3:deaf:feed', None), ('http://[::12.34.56.78]/foo/', '::12.34.56.78', None), ('http://[::ffff:12.34.56.78]/foo/', '::ffff:12.34.56.78', None), ('http://Test.python.org:/foo/', 'test.python.org', None), ('http://12.34.56.78:/foo/', '12.34.56.78', None), ('http://[::1]:/foo/', '::1', None), ('http://[dead:beef::1]:/foo/', 'dead:beef::1', None), ('http://[dead:beef::]:/foo/', 'dead:beef::', None), ('http://[dead:beef:cafe:5417:affe:8FA3:deaf:feed]:/foo/', 'dead:beef:cafe:5417:affe:8fa3:deaf:feed', None), ('http://[::12.34.56.78]:/foo/', '::12.34.56.78', None), ('http://[::ffff:12.34.56.78]:/foo/', '::ffff:12.34.56.78', None), ] def _encode(t): return t[0].encode('ascii'), t[1].encode('ascii'), t[2] bytes_cases = [_encode(x) for x in str_cases] for url, hostname, port in str_cases + bytes_cases: urlparsed = urllib.parse.urlparse(url) self.assertEqual((urlparsed.hostname, urlparsed.port) , (hostname, port)) str_cases = [ 'http://::12.34.56.78]/', 'http://[::1/foo/', 'ftp://[::1/foo/bad]/bad', 'http://[::1/foo/bad]/bad', 'http://[::ffff:12.34.56.78'] bytes_cases = [x.encode('ascii') for x in str_cases] for invalid_url in str_cases + bytes_cases: self.assertRaises(ValueError, urllib.parse.urlparse, invalid_url) def test_urldefrag(self): str_cases = [ ('http://python.org#frag', 'http://python.org', 'frag'), ('http://python.org', 'http://python.org', ''), ('http://python.org/#frag', 'http://python.org/', 'frag'), ('http://python.org/', 'http://python.org/', ''), ('http://python.org/?q#frag', 'http://python.org/?q', 'frag'), ('http://python.org/?q', 'http://python.org/?q', ''), ('http://python.org/p#frag', 'http://python.org/p', 'frag'), ('http://python.org/p?q', 'http://python.org/p?q', ''), (RFC1808_BASE, 'http://a/b/c/d;p?q', 'f'), (RFC2396_BASE, 'http://a/b/c/d;p?q', ''), ] def _encode(t): return type(t)(x.encode('ascii') for x in t) bytes_cases = [_encode(x) for x in str_cases] for url, defrag, frag in str_cases + bytes_cases: result = urllib.parse.urldefrag(url) self.assertEqual(result.geturl(), url) self.assertEqual(result, (defrag, frag)) self.assertEqual(result.url, defrag) self.assertEqual(result.fragment, frag) def test_urlsplit_attributes(self): url = "HTTP://WWW.PYTHON.ORG/doc/#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.scheme, "http") self.assertEqual(p.netloc, "WWW.PYTHON.ORG") self.assertEqual(p.path, "/doc/") self.assertEqual(p.query, "") self.assertEqual(p.fragment, "frag") self.assertEqual(p.username, None) self.assertEqual(p.password, None) self.assertEqual(p.hostname, "www.python.org") self.assertEqual(p.port, None) # geturl() won't return exactly the original URL in this case # since the scheme is always case-normalized # We handle this by ignoring the first 4 characters of the URL self.assertEqual(p.geturl()[4:], url[4:]) url = "http://User:Pass@www.python.org:080/doc/?query=yes#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.scheme, "http") self.assertEqual(p.netloc, "User:Pass@www.python.org:080") self.assertEqual(p.path, "/doc/") self.assertEqual(p.query, "query=yes") self.assertEqual(p.fragment, "frag") self.assertEqual(p.username, "User") self.assertEqual(p.password, "Pass") self.assertEqual(p.hostname, "www.python.org") self.assertEqual(p.port, 80) self.assertEqual(p.geturl(), url) # Addressing issue1698, which suggests Username can contain # "@" characters. Though not RFC compliant, many ftp sites allow # and request email addresses as usernames. url = "http://User@example.com:Pass@www.python.org:080/doc/?query=yes#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.scheme, "http") self.assertEqual(p.netloc, "User@example.com:Pass@www.python.org:080") self.assertEqual(p.path, "/doc/") self.assertEqual(p.query, "query=yes") self.assertEqual(p.fragment, "frag") self.assertEqual(p.username, "User@example.com") self.assertEqual(p.password, "Pass") self.assertEqual(p.hostname, "www.python.org") self.assertEqual(p.port, 80) self.assertEqual(p.geturl(), url) # And check them all again, only with bytes this time url = b"HTTP://WWW.PYTHON.ORG/doc/#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.scheme, b"http") self.assertEqual(p.netloc, b"WWW.PYTHON.ORG") self.assertEqual(p.path, b"/doc/") self.assertEqual(p.query, b"") self.assertEqual(p.fragment, b"frag") self.assertEqual(p.username, None) self.assertEqual(p.password, None) self.assertEqual(p.hostname, b"www.python.org") self.assertEqual(p.port, None) self.assertEqual(p.geturl()[4:], url[4:]) url = b"http://User:Pass@www.python.org:080/doc/?query=yes#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.scheme, b"http") self.assertEqual(p.netloc, b"User:Pass@www.python.org:080") self.assertEqual(p.path, b"/doc/") self.assertEqual(p.query, b"query=yes") self.assertEqual(p.fragment, b"frag") self.assertEqual(p.username, b"User") self.assertEqual(p.password, b"Pass") self.assertEqual(p.hostname, b"www.python.org") self.assertEqual(p.port, 80) self.assertEqual(p.geturl(), url) url = b"http://User@example.com:Pass@www.python.org:080/doc/?query=yes#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.scheme, b"http") self.assertEqual(p.netloc, b"User@example.com:Pass@www.python.org:080") self.assertEqual(p.path, b"/doc/") self.assertEqual(p.query, b"query=yes") self.assertEqual(p.fragment, b"frag") self.assertEqual(p.username, b"User@example.com") self.assertEqual(p.password, b"Pass") self.assertEqual(p.hostname, b"www.python.org") self.assertEqual(p.port, 80) self.assertEqual(p.geturl(), url) # Verify an illegal port is returned as None url = b"HTTP://WWW.PYTHON.ORG:65536/doc/#frag" p = urllib.parse.urlsplit(url) self.assertEqual(p.port, None) def test_attributes_bad_port(self): """Check handling of non-integer ports.""" p = urllib.parse.urlsplit("http://www.example.net:foo") self.assertEqual(p.netloc, "www.example.net:foo") self.assertRaises(ValueError, lambda: p.port) p = urllib.parse.urlparse("http://www.example.net:foo") self.assertEqual(p.netloc, "www.example.net:foo") self.assertRaises(ValueError, lambda: p.port) # Once again, repeat ourselves to test bytes p = urllib.parse.urlsplit(b"http://www.example.net:foo") self.assertEqual(p.netloc, b"www.example.net:foo") self.assertRaises(ValueError, lambda: p.port) p = urllib.parse.urlparse(b"http://www.example.net:foo") self.assertEqual(p.netloc, b"www.example.net:foo") self.assertRaises(ValueError, lambda: p.port) def test_attributes_without_netloc(self): # This example is straight from RFC 3261. It looks like it # should allow the username, hostname, and port to be filled # in, but doesn't. Since it's a URI and doesn't use the # scheme://netloc syntax, the netloc and related attributes # should be left empty. uri = "sip:alice@atlanta.com;maddr=239.255.255.1;ttl=15" p = urllib.parse.urlsplit(uri) self.assertEqual(p.netloc, "") self.assertEqual(p.username, None) self.assertEqual(p.password, None) self.assertEqual(p.hostname, None) self.assertEqual(p.port, None) self.assertEqual(p.geturl(), uri) p = urllib.parse.urlparse(uri) self.assertEqual(p.netloc, "") self.assertEqual(p.username, None) self.assertEqual(p.password, None) self.assertEqual(p.hostname, None) self.assertEqual(p.port, None) self.assertEqual(p.geturl(), uri) # You guessed it, repeating the test with bytes input uri = b"sip:alice@atlanta.com;maddr=239.255.255.1;ttl=15" p = urllib.parse.urlsplit(uri) self.assertEqual(p.netloc, b"") self.assertEqual(p.username, None) self.assertEqual(p.password, None) self.assertEqual(p.hostname, None) self.assertEqual(p.port, None) self.assertEqual(p.geturl(), uri) p = urllib.parse.urlparse(uri) self.assertEqual(p.netloc, b"") self.assertEqual(p.username, None) self.assertEqual(p.password, None) self.assertEqual(p.hostname, None) self.assertEqual(p.port, None) self.assertEqual(p.geturl(), uri) def test_noslash(self): # Issue 1637: http://foo.com?query is legal self.assertEqual(urllib.parse.urlparse("http://example.com?blahblah=/foo"), ('http', 'example.com', '', '', 'blahblah=/foo', '')) self.assertEqual(urllib.parse.urlparse(b"http://example.com?blahblah=/foo"), (b'http', b'example.com', b'', b'', b'blahblah=/foo', b'')) def test_withoutscheme(self): # Test urlparse without scheme # Issue 754016: urlparse goes wrong with IP:port without scheme # RFC 1808 specifies that netloc should start with //, urlparse expects # the same, otherwise it classifies the portion of url as path. self.assertEqual(urllib.parse.urlparse("path"), ('','','path','','','')) self.assertEqual(urllib.parse.urlparse("//www.python.org:80"), ('','www.python.org:80','','','','')) self.assertEqual(urllib.parse.urlparse("http://www.python.org:80"), ('http','www.python.org:80','','','','')) # Repeat for bytes input self.assertEqual(urllib.parse.urlparse(b"path"), (b'',b'',b'path',b'',b'',b'')) self.assertEqual(urllib.parse.urlparse(b"//www.python.org:80"), (b'',b'www.python.org:80',b'',b'',b'',b'')) self.assertEqual(urllib.parse.urlparse(b"http://www.python.org:80"), (b'http',b'www.python.org:80',b'',b'',b'',b'')) def test_portseparator(self): # Issue 754016 makes changes for port separator ':' from scheme separator self.assertEqual(urllib.parse.urlparse("path:80"), ('','','path:80','','','')) self.assertEqual(urllib.parse.urlparse("http:"),('http','','','','','')) self.assertEqual(urllib.parse.urlparse("https:"),('https','','','','','')) self.assertEqual(urllib.parse.urlparse("http://www.python.org:80"), ('http','www.python.org:80','','','','')) # As usual, need to check bytes input as well self.assertEqual(urllib.parse.urlparse(b"path:80"), (b'',b'',b'path:80',b'',b'',b'')) self.assertEqual(urllib.parse.urlparse(b"http:"),(b'http',b'',b'',b'',b'',b'')) self.assertEqual(urllib.parse.urlparse(b"https:"),(b'https',b'',b'',b'',b'',b'')) self.assertEqual(urllib.parse.urlparse(b"http://www.python.org:80"), (b'http',b'www.python.org:80',b'',b'',b'',b'')) def test_usingsys(self): # Issue 3314: sys module is used in the error self.assertRaises(TypeError, urllib.parse.urlencode, "foo") def test_anyscheme(self): # Issue 7904: s3://foo.com/stuff has netloc "foo.com". self.assertEqual(urllib.parse.urlparse("s3://foo.com/stuff"), ('s3', 'foo.com', '/stuff', '', '', '')) self.assertEqual(urllib.parse.urlparse("x-newscheme://foo.com/stuff"), ('x-newscheme', 'foo.com', '/stuff', '', '', '')) self.assertEqual(urllib.parse.urlparse("x-newscheme://foo.com/stuff?query#fragment"), ('x-newscheme', 'foo.com', '/stuff', '', 'query', 'fragment')) self.assertEqual(urllib.parse.urlparse("x-newscheme://foo.com/stuff?query"), ('x-newscheme', 'foo.com', '/stuff', '', 'query', '')) # And for bytes... self.assertEqual(urllib.parse.urlparse(b"s3://foo.com/stuff"), (b's3', b'foo.com', b'/stuff', b'', b'', b'')) self.assertEqual(urllib.parse.urlparse(b"x-newscheme://foo.com/stuff"), (b'x-newscheme', b'foo.com', b'/stuff', b'', b'', b'')) self.assertEqual(urllib.parse.urlparse(b"x-newscheme://foo.com/stuff?query#fragment"), (b'x-newscheme', b'foo.com', b'/stuff', b'', b'query', b'fragment')) self.assertEqual(urllib.parse.urlparse(b"x-newscheme://foo.com/stuff?query"), (b'x-newscheme', b'foo.com', b'/stuff', b'', b'query', b'')) def test_mixed_types_rejected(self): # Several functions that process either strings or ASCII encoded bytes # accept multiple arguments. Check they reject mixed type input with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlparse("www.python.org", b"http") with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlparse(b"www.python.org", "http") with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlsplit("www.python.org", b"http") with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlsplit(b"www.python.org", "http") with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlunparse(( b"http", "www.python.org","","","","")) with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlunparse(("http", b"www.python.org","","","","")) with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlunsplit((b"http", "www.python.org","","","")) with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urlunsplit(("http", b"www.python.org","","","")) with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urljoin("http://python.org", b"http://python.org") with self.assertRaisesRegex(TypeError, "Cannot mix str"): urllib.parse.urljoin(b"http://python.org", "http://python.org") def _check_result_type(self, str_type): num_args = len(str_type._fields) bytes_type = str_type._encoded_counterpart self.assertIs(bytes_type._decoded_counterpart, str_type) str_args = ('',) * num_args bytes_args = (b'',) * num_args str_result = str_type(*str_args) bytes_result = bytes_type(*bytes_args) encoding = 'ascii' errors = 'strict' self.assertEqual(str_result, str_args) self.assertEqual(bytes_result.decode(), str_args) self.assertEqual(bytes_result.decode(), str_result) self.assertEqual(bytes_result.decode(encoding), str_args) self.assertEqual(bytes_result.decode(encoding), str_result) self.assertEqual(bytes_result.decode(encoding, errors), str_args) self.assertEqual(bytes_result.decode(encoding, errors), str_result) self.assertEqual(bytes_result, bytes_args) self.assertEqual(str_result.encode(), bytes_args) self.assertEqual(str_result.encode(), bytes_result) self.assertEqual(str_result.encode(encoding), bytes_args) self.assertEqual(str_result.encode(encoding), bytes_result) self.assertEqual(str_result.encode(encoding, errors), bytes_args) self.assertEqual(str_result.encode(encoding, errors), bytes_result) def test_result_pairs(self): # Check encoding and decoding between result pairs result_types = [ urllib.parse.DefragResult, urllib.parse.SplitResult, urllib.parse.ParseResult, ] for result_type in result_types: self._check_result_type(result_type) def test_parse_qs_encoding(self): result = urllib.parse.parse_qs("key=\u0141%E9", encoding="latin-1") self.assertEqual(result, {'key': ['\u0141\xE9']}) result = urllib.parse.parse_qs("key=\u0141%C3%A9", encoding="utf-8") self.assertEqual(result, {'key': ['\u0141\xE9']}) result = urllib.parse.parse_qs("key=\u0141%C3%A9", encoding="ascii") self.assertEqual(result, {'key': ['\u0141\ufffd\ufffd']}) result = urllib.parse.parse_qs("key=\u0141%E9-", encoding="ascii") self.assertEqual(result, {'key': ['\u0141\ufffd-']}) result = urllib.parse.parse_qs("key=\u0141%E9-", encoding="ascii", errors="ignore") self.assertEqual(result, {'key': ['\u0141-']}) def test_parse_qsl_encoding(self): result = urllib.parse.parse_qsl("key=\u0141%E9", encoding="latin-1") self.assertEqual(result, [('key', '\u0141\xE9')]) result = urllib.parse.parse_qsl("key=\u0141%C3%A9", encoding="utf-8") self.assertEqual(result, [('key', '\u0141\xE9')]) result = urllib.parse.parse_qsl("key=\u0141%C3%A9", encoding="ascii") self.assertEqual(result, [('key', '\u0141\ufffd\ufffd')]) result = urllib.parse.parse_qsl("key=\u0141%E9-", encoding="ascii") self.assertEqual(result, [('key', '\u0141\ufffd-')]) result = urllib.parse.parse_qsl("key=\u0141%E9-", encoding="ascii", errors="ignore") self.assertEqual(result, [('key', '\u0141-')]) def test_splitport(self): splitport = urllib.parse.splitport self.assertEqual(splitport('parrot:88'), ('parrot', '88')) self.assertEqual(splitport('parrot'), ('parrot', None)) self.assertEqual(splitport('parrot:'), ('parrot', None)) self.assertEqual(splitport('127.0.0.1'), ('127.0.0.1', None)) self.assertEqual(splitport('parrot:cheese'), ('parrot:cheese', None)) def test_splitnport(self): splitnport = urllib.parse.splitnport self.assertEqual(splitnport('parrot:88'), ('parrot', 88)) self.assertEqual(splitnport('parrot'), ('parrot', -1)) self.assertEqual(splitnport('parrot', 55), ('parrot', 55)) self.assertEqual(splitnport('parrot:'), ('parrot', -1)) self.assertEqual(splitnport('parrot:', 55), ('parrot', 55)) self.assertEqual(splitnport('127.0.0.1'), ('127.0.0.1', -1)) self.assertEqual(splitnport('127.0.0.1', 55), ('127.0.0.1', 55)) self.assertEqual(splitnport('parrot:cheese'), ('parrot', None)) self.assertEqual(splitnport('parrot:cheese', 55), ('parrot', None)) def test_splitquery(self): # Normal cases are exercised by other tests; ensure that we also # catch cases with no port specified (testcase ensuring coverage) result = urllib.parse.splitquery('http://python.org/fake?foo=bar') self.assertEqual(result, ('http://python.org/fake', 'foo=bar')) result = urllib.parse.splitquery('http://python.org/fake?foo=bar?') self.assertEqual(result, ('http://python.org/fake?foo=bar', '')) result = urllib.parse.splitquery('http://python.org/fake') self.assertEqual(result, ('http://python.org/fake', None)) def test_splitvalue(self): # Normal cases are exercised by other tests; test pathological cases # with no key/value pairs. (testcase ensuring coverage) result = urllib.parse.splitvalue('foo=bar') self.assertEqual(result, ('foo', 'bar')) result = urllib.parse.splitvalue('foo=') self.assertEqual(result, ('foo', '')) result = urllib.parse.splitvalue('foobar') self.assertEqual(result, ('foobar', None)) def test_to_bytes(self): result = urllib.parse.to_bytes('http://www.python.org') self.assertEqual(result, 'http://www.python.org') self.assertRaises(UnicodeError, urllib.parse.to_bytes, 'http://www.python.org/medi\u00e6val') def test_urlencode_sequences(self): # Other tests incidentally urlencode things; test non-covered cases: # Sequence and object values. result = urllib.parse.urlencode({'a': [1, 2], 'b': (3, 4, 5)}, True) # we cannot rely on ordering here assert set(result.split('&')) == {'a=1', 'a=2', 'b=3', 'b=4', 'b=5'} class Trivial: def __str__(self): return 'trivial' result = urllib.parse.urlencode({'a': Trivial()}, True) self.assertEqual(result, 'a=trivial') def test_quote_from_bytes(self): self.assertRaises(TypeError, urllib.parse.quote_from_bytes, 'foo') result = urllib.parse.quote_from_bytes(b'archaeological arcana') self.assertEqual(result, 'archaeological%20arcana') result = urllib.parse.quote_from_bytes(b'') self.assertEqual(result, '') def test_unquote_to_bytes(self): result = urllib.parse.unquote_to_bytes('abc%20def') self.assertEqual(result, b'abc def') result = urllib.parse.unquote_to_bytes('') self.assertEqual(result, b'') def test_quote_errors(self): self.assertRaises(TypeError, urllib.parse.quote, b'foo', encoding='utf-8') self.assertRaises(TypeError, urllib.parse.quote, b'foo', errors='strict') def test_issue14072(self): p1 = urllib.parse.urlsplit('tel:+31-641044153') self.assertEqual(p1.scheme, 'tel') self.assertEqual(p1.path, '+31-641044153') p2 = urllib.parse.urlsplit('tel:+31641044153') self.assertEqual(p2.scheme, 'tel') self.assertEqual(p2.path, '+31641044153') # assert the behavior for urlparse p1 = urllib.parse.urlparse('tel:+31-641044153') self.assertEqual(p1.scheme, 'tel') self.assertEqual(p1.path, '+31-641044153') p2 = urllib.parse.urlparse('tel:+31641044153') self.assertEqual(p2.scheme, 'tel') self.assertEqual(p2.path, '+31641044153') def test_telurl_params(self): p1 = urllib.parse.urlparse('tel:123-4;phone-context=+1-650-516') self.assertEqual(p1.scheme, 'tel') self.assertEqual(p1.path, '123-4') self.assertEqual(p1.params, 'phone-context=+1-650-516') p1 = urllib.parse.urlparse('tel:+1-201-555-0123') self.assertEqual(p1.scheme, 'tel') self.assertEqual(p1.path, '+1-201-555-0123') self.assertEqual(p1.params, '') p1 = urllib.parse.urlparse('tel:7042;phone-context=example.com') self.assertEqual(p1.scheme, 'tel') self.assertEqual(p1.path, '7042') self.assertEqual(p1.params, 'phone-context=example.com') p1 = urllib.parse.urlparse('tel:863-1234;phone-context=+1-914-555') self.assertEqual(p1.scheme, 'tel') self.assertEqual(p1.path, '863-1234') self.assertEqual(p1.params, 'phone-context=+1-914-555') def test_main(): support.run_unittest(UrlParseTestCase) if __name__ == "__main__": test_main()