0x1949 Team - FAZEMRX - MANAGER
Edit File: test_url.py
# -*- coding: utf-8 -*- # Copyright (c) Twisted Matrix Laboratories. # See LICENSE for details. from __future__ import unicode_literals import sys import socket from typing import Any, Iterable, Optional, Text, Tuple, cast from .common import HyperlinkTestCase from .. import URL, URLParseError from .._url import inet_pton, SCHEME_PORT_MAP PY2 = sys.version_info[0] == 2 unicode = type("") BASIC_URL = "http://www.foo.com/a/nice/path/?zot=23&zut" # Examples from RFC 3986 section 5.4, Reference Resolution Examples relativeLinkBaseForRFC3986 = "http://a/b/c/d;p?q" relativeLinkTestsForRFC3986 = [ # "Normal" # ('g:h', 'g:h'), # can't click on a scheme-having url without an abs path ("g", "http://a/b/c/g"), ("./g", "http://a/b/c/g"), ("g/", "http://a/b/c/g/"), ("/g", "http://a/g"), ("//g", "http://g"), ("?y", "http://a/b/c/d;p?y"), ("g?y", "http://a/b/c/g?y"), ("#s", "http://a/b/c/d;p?q#s"), ("g#s", "http://a/b/c/g#s"), ("g?y#s", "http://a/b/c/g?y#s"), (";x", "http://a/b/c/;x"), ("g;x", "http://a/b/c/g;x"), ("g;x?y#s", "http://a/b/c/g;x?y#s"), ("", "http://a/b/c/d;p?q"), (".", "http://a/b/c/"), ("./", "http://a/b/c/"), ("..", "http://a/b/"), ("../", "http://a/b/"), ("../g", "http://a/b/g"), ("../..", "http://a/"), ("../../", "http://a/"), ("../../g", "http://a/g"), # Abnormal examples # ".." cannot be used to change the authority component of a URI. ("../../../g", "http://a/g"), ("../../../../g", "http://a/g"), # Only include "." and ".." when they are only part of a larger segment, # not by themselves. ("/./g", "http://a/g"), ("/../g", "http://a/g"), ("g.", "http://a/b/c/g."), (".g", "http://a/b/c/.g"), ("g..", "http://a/b/c/g.."), ("..g", "http://a/b/c/..g"), # Unnecessary or nonsensical forms of "." and "..". ("./../g", "http://a/b/g"), ("./g/.", "http://a/b/c/g/"), ("g/./h", "http://a/b/c/g/h"), ("g/../h", "http://a/b/c/h"), ("g;x=1/./y", "http://a/b/c/g;x=1/y"), ("g;x=1/../y", "http://a/b/c/y"), # Separating the reference's query and fragment components from the path. ("g?y/./x", "http://a/b/c/g?y/./x"), ("g?y/../x", "http://a/b/c/g?y/../x"), ("g#s/./x", "http://a/b/c/g#s/./x"), ("g#s/../x", "http://a/b/c/g#s/../x"), ] ROUNDTRIP_TESTS = ( "http://localhost", "http://localhost/", "http://127.0.0.1/", "http://[::127.0.0.1]/", "http://[::1]/", "http://localhost/foo", "http://localhost/foo/", "http://localhost/foo!!bar/", "http://localhost/foo%20bar/", "http://localhost/foo%2Fbar/", "http://localhost/foo?n", "http://localhost/foo?n=v", "http://localhost/foo?n=/a/b", "http://example.com/foo!@$bar?b!@z=123", "http://localhost/asd?a=asd%20sdf/345", "http://(%2525)/(%2525)?(%2525)&(%2525)=(%2525)#(%2525)", "http://(%C3%A9)/(%C3%A9)?(%C3%A9)&(%C3%A9)=(%C3%A9)#(%C3%A9)", "?sslrootcert=/Users/glyph/Downloads/rds-ca-2015-root.pem&sslmode=verify", # from boltons.urlutils' tests "http://googlewebsite.com/e-shops.aspx", "http://example.com:8080/search?q=123&business=Nothing%20Special", "http://hatnote.com:9000/?arg=1&arg=2&arg=3", "https://xn--bcher-kva.ch", "http://xn--ggbla1c4e.xn--ngbc5azd/", "http://tools.ietf.org/html/rfc3986#section-3.4", # 'http://wiki:pedia@hatnote.com', "ftp://ftp.rfc-editor.org/in-notes/tar/RFCs0001-0500.tar.gz", "http://[1080:0:0:0:8:800:200C:417A]/index.html", "ssh://192.0.2.16:2222/", "https://[::101.45.75.219]:80/?hi=bye", "ldap://[::192.9.5.5]/dc=example,dc=com??sub?(sn=Jensen)", "mailto:me@example.com?to=me@example.com&body=hi%20http://wikipedia.org", "news:alt.rec.motorcycle", "tel:+1-800-867-5309", "urn:oasis:member:A00024:x", ( "magnet:?xt=urn:btih:1a42b9e04e122b97a5254e3df77ab3c4b7da725f&dn=Puppy%" "20Linux%20precise-5.7.1.iso&tr=udp://tracker.openbittorrent.com:80&" "tr=udp://tracker.publicbt.com:80&tr=udp://tracker.istole.it:6969&" "tr=udp://tracker.ccc.de:80&tr=udp://open.demonii.com:1337" ), # percent-encoded delimiters in percent-encodable fields "https://%3A@example.com/", # colon in username "https://%40@example.com/", # at sign in username "https://%2f@example.com/", # slash in username "https://a:%3a@example.com/", # colon in password "https://a:%40@example.com/", # at sign in password "https://a:%2f@example.com/", # slash in password "https://a:%3f@example.com/", # question mark in password "https://example.com/%2F/", # slash in path "https://example.com/%3F/", # question mark in path "https://example.com/%23/", # hash in path "https://example.com/?%23=b", # hash in query param name "https://example.com/?%3D=b", # equals in query param name "https://example.com/?%26=b", # ampersand in query param name "https://example.com/?a=%23", # hash in query param value "https://example.com/?a=%26", # ampersand in query param value "https://example.com/?a=%3D", # equals in query param value "https://example.com/?foo+bar=baz", # plus in query param name "https://example.com/?foo=bar+baz", # plus in query param value # double-encoded percent sign in all percent-encodable positions: "http://(%2525):(%2525)@example.com/(%2525)/?(%2525)=(%2525)#(%2525)", # colon in first part of schemeless relative url "first_seg_rel_path__colon%3Anotok/second_seg__colon%3Aok", ) class TestURL(HyperlinkTestCase): """ Tests for L{URL}. """ def assertUnicoded(self, u): # type: (URL) -> None """ The given L{URL}'s components should be L{unicode}. @param u: The L{URL} to test. """ self.assertTrue( isinstance(u.scheme, unicode) or u.scheme is None, repr(u) ) self.assertTrue(isinstance(u.host, unicode) or u.host is None, repr(u)) for seg in u.path: self.assertEqual(type(seg), unicode, repr(u)) for (_k, v) in u.query: self.assertEqual(type(seg), unicode, repr(u)) self.assertTrue(v is None or isinstance(v, unicode), repr(u)) self.assertEqual(type(u.fragment), unicode, repr(u)) def assertURL( self, u, # type: URL scheme, # type: Text host, # type: Text path, # type: Iterable[Text] query, # type: Iterable[Tuple[Text, Optional[Text]]] fragment, # type: Text port, # type: Optional[int] userinfo="", # type: Text ): # type: (...) -> None """ The given L{URL} should have the given components. @param u: The actual L{URL} to examine. @param scheme: The expected scheme. @param host: The expected host. @param path: The expected path. @param query: The expected query. @param fragment: The expected fragment. @param port: The expected port. @param userinfo: The expected userinfo. """ actual = ( u.scheme, u.host, u.path, u.query, u.fragment, u.port, u.userinfo, ) expected = ( scheme, host, tuple(path), tuple(query), fragment, port, u.userinfo, ) self.assertEqual(actual, expected) def test_initDefaults(self): # type: () -> None """ L{URL} should have appropriate default values. """ def check(u): # type: (URL) -> None self.assertUnicoded(u) self.assertURL(u, "http", "", [], [], "", 80, "") check(URL("http", "")) check(URL("http", "", [], [])) check(URL("http", "", [], [], "")) def test_init(self): # type: () -> None """ L{URL} should accept L{unicode} parameters. """ u = URL("s", "h", ["p"], [("k", "v"), ("k", None)], "f") self.assertUnicoded(u) self.assertURL(u, "s", "h", ["p"], [("k", "v"), ("k", None)], "f", None) self.assertURL( URL("http", "\xe0", ["\xe9"], [("\u03bb", "\u03c0")], "\u22a5"), "http", "\xe0", ["\xe9"], [("\u03bb", "\u03c0")], "\u22a5", 80, ) def test_initPercent(self): # type: () -> None """ L{URL} should accept (and not interpret) percent characters. """ u = URL("s", "%68", ["%70"], [("%6B", "%76"), ("%6B", None)], "%66") self.assertUnicoded(u) self.assertURL( u, "s", "%68", ["%70"], [("%6B", "%76"), ("%6B", None)], "%66", None ) def test_repr(self): # type: () -> None """ L{URL.__repr__} will display the canonical form of the URL, wrapped in a L{URL.from_text} invocation, so that it is C{eval}-able but still easy to read. """ self.assertEqual( repr( URL( scheme="http", host="foo", path=["bar"], query=[("baz", None), ("k", "v")], fragment="frob", ) ), "URL.from_text(%s)" % (repr("http://foo/bar?baz&k=v#frob"),), ) def test_from_text(self): # type: () -> None """ Round-tripping L{URL.from_text} with C{str} results in an equivalent URL. """ urlpath = URL.from_text(BASIC_URL) self.assertEqual(BASIC_URL, urlpath.to_text()) def test_roundtrip(self): # type: () -> None """ L{URL.to_text} should invert L{URL.from_text}. """ for test in ROUNDTRIP_TESTS: result = URL.from_text(test).to_text(with_password=True) self.assertEqual(test, result) def test_roundtrip_double_iri(self): # type: () -> None for test in ROUNDTRIP_TESTS: url = URL.from_text(test) iri = url.to_iri() double_iri = iri.to_iri() assert iri == double_iri iri_text = iri.to_text(with_password=True) double_iri_text = double_iri.to_text(with_password=True) assert iri_text == double_iri_text return def test_equality(self): # type: () -> None """ Two URLs decoded using L{URL.from_text} will be equal (C{==}) if they decoded same URL string, and unequal (C{!=}) if they decoded different strings. """ urlpath = URL.from_text(BASIC_URL) self.assertEqual(urlpath, URL.from_text(BASIC_URL)) self.assertNotEqual( urlpath, URL.from_text( "ftp://www.anotherinvaliddomain.com/" "foo/bar/baz/?zot=21&zut" ), ) def test_fragmentEquality(self): # type: () -> None """ An URL created with the empty string for a fragment compares equal to an URL created with an unspecified fragment. """ self.assertEqual(URL(fragment=""), URL()) self.assertEqual( URL.from_text("http://localhost/#"), URL.from_text("http://localhost/"), ) def test_child(self): # type: () -> None """ L{URL.child} appends a new path segment, but does not affect the query or fragment. """ urlpath = URL.from_text(BASIC_URL) self.assertEqual( "http://www.foo.com/a/nice/path/gong?zot=23&zut", urlpath.child("gong").to_text(), ) self.assertEqual( "http://www.foo.com/a/nice/path/gong%2F?zot=23&zut", urlpath.child("gong/").to_text(), ) self.assertEqual( "http://www.foo.com/a/nice/path/gong%2Fdouble?zot=23&zut", urlpath.child("gong/double").to_text(), ) self.assertEqual( "http://www.foo.com/a/nice/path/gong%2Fdouble%2F?zot=23&zut", urlpath.child("gong/double/").to_text(), ) def test_multiChild(self): # type: () -> None """ L{URL.child} receives multiple segments as C{*args} and appends each in turn. """ url = URL.from_text("http://example.com/a/b") self.assertEqual( url.child("c", "d", "e").to_text(), "http://example.com/a/b/c/d/e" ) def test_childInitRoot(self): # type: () -> None """ L{URL.child} of a L{URL} without a path produces a L{URL} with a single path segment. """ childURL = URL(host="www.foo.com").child("c") self.assertTrue(childURL.rooted) self.assertEqual("http://www.foo.com/c", childURL.to_text()) def test_emptyChild(self): # type: () -> None """ L{URL.child} without any new segments returns the original L{URL}. """ url = URL(host="www.foo.com") self.assertEqual(url.child(), url) def test_sibling(self): # type: () -> None """ L{URL.sibling} of a L{URL} replaces the last path segment, but does not affect the query or fragment. """ urlpath = URL.from_text(BASIC_URL) self.assertEqual( "http://www.foo.com/a/nice/path/sister?zot=23&zut", urlpath.sibling("sister").to_text(), ) # Use an url without trailing '/' to check child removal. url_text = "http://www.foo.com/a/nice/path?zot=23&zut" urlpath = URL.from_text(url_text) self.assertEqual( "http://www.foo.com/a/nice/sister?zot=23&zut", urlpath.sibling("sister").to_text(), ) def test_click(self): # type: () -> None """ L{URL.click} interprets the given string as a relative URI-reference and returns a new L{URL} interpreting C{self} as the base absolute URI. """ urlpath = URL.from_text(BASIC_URL) # A null uri should be valid (return here). self.assertEqual( "http://www.foo.com/a/nice/path/?zot=23&zut", urlpath.click("").to_text(), ) # A simple relative path remove the query. self.assertEqual( "http://www.foo.com/a/nice/path/click", urlpath.click("click").to_text(), ) # An absolute path replace path and query. self.assertEqual( "http://www.foo.com/click", urlpath.click("/click").to_text() ) # Replace just the query. self.assertEqual( "http://www.foo.com/a/nice/path/?burp", urlpath.click("?burp").to_text(), ) # One full url to another should not generate '//' between authority. # and path self.assertTrue( "//foobar" not in urlpath.click("http://www.foo.com/foobar").to_text() ) # From a url with no query clicking a url with a query, the query # should be handled properly. u = URL.from_text("http://www.foo.com/me/noquery") self.assertEqual( "http://www.foo.com/me/17?spam=158", u.click("/me/17?spam=158").to_text(), ) # Check that everything from the path onward is removed when the click # link has no path. u = URL.from_text("http://localhost/foo?abc=def") self.assertEqual( u.click("http://www.python.org").to_text(), "http://www.python.org" ) # https://twistedmatrix.com/trac/ticket/8184 u = URL.from_text("http://hatnote.com/a/b/../c/./d/e/..") res = "http://hatnote.com/a/c/d/" self.assertEqual(u.click("").to_text(), res) # test click default arg is same as empty string above self.assertEqual(u.click().to_text(), res) # test click on a URL instance u = URL.fromText("http://localhost/foo/?abc=def") u2 = URL.from_text("bar") u3 = u.click(u2) self.assertEqual(u3.to_text(), "http://localhost/foo/bar") def test_clickRFC3986(self): # type: () -> None """ L{URL.click} should correctly resolve the examples in RFC 3986. """ base = URL.from_text(relativeLinkBaseForRFC3986) for (ref, expected) in relativeLinkTestsForRFC3986: self.assertEqual(base.click(ref).to_text(), expected) def test_clickSchemeRelPath(self): # type: () -> None """ L{URL.click} should not accept schemes with relative paths. """ base = URL.from_text(relativeLinkBaseForRFC3986) self.assertRaises(NotImplementedError, base.click, "g:h") self.assertRaises(NotImplementedError, base.click, "http:h") def test_cloneUnchanged(self): # type: () -> None """ Verify that L{URL.replace} doesn't change any of the arguments it is passed. """ urlpath = URL.from_text("https://x:1/y?z=1#A") self.assertEqual( urlpath.replace( urlpath.scheme, urlpath.host, urlpath.path, urlpath.query, urlpath.fragment, urlpath.port, ), urlpath, ) self.assertEqual(urlpath.replace(), urlpath) def test_clickCollapse(self): # type: () -> None """ L{URL.click} collapses C{.} and C{..} according to RFC 3986 section 5.2.4. """ tests = [ ["http://localhost/", ".", "http://localhost/"], ["http://localhost/", "..", "http://localhost/"], ["http://localhost/a/b/c", ".", "http://localhost/a/b/"], ["http://localhost/a/b/c", "..", "http://localhost/a/"], ["http://localhost/a/b/c", "./d/e", "http://localhost/a/b/d/e"], ["http://localhost/a/b/c", "../d/e", "http://localhost/a/d/e"], ["http://localhost/a/b/c", "/./d/e", "http://localhost/d/e"], ["http://localhost/a/b/c", "/../d/e", "http://localhost/d/e"], [ "http://localhost/a/b/c/", "../../d/e/", "http://localhost/a/d/e/", ], ["http://localhost/a/./c", "../d/e", "http://localhost/d/e"], ["http://localhost/a/./c/", "../d/e", "http://localhost/a/d/e"], [ "http://localhost/a/b/c/d", "./e/../f/../g", "http://localhost/a/b/c/g", ], ["http://localhost/a/b/c", "d//e", "http://localhost/a/b/d//e"], ] for start, click, expected in tests: actual = URL.from_text(start).click(click).to_text() self.assertEqual( actual, expected, "{start}.click({click}) => {actual} not {expected}".format( start=start, click=repr(click), actual=actual, expected=expected, ), ) def test_queryAdd(self): # type: () -> None """ L{URL.add} adds query parameters. """ self.assertEqual( "http://www.foo.com/a/nice/path/?foo=bar", URL.from_text("http://www.foo.com/a/nice/path/") .add("foo", "bar") .to_text(), ) self.assertEqual( "http://www.foo.com/?foo=bar", URL(host="www.foo.com").add("foo", "bar").to_text(), ) urlpath = URL.from_text(BASIC_URL) self.assertEqual( "http://www.foo.com/a/nice/path/?zot=23&zut&burp", urlpath.add("burp").to_text(), ) self.assertEqual( "http://www.foo.com/a/nice/path/?zot=23&zut&burp=xxx", urlpath.add("burp", "xxx").to_text(), ) self.assertEqual( "http://www.foo.com/a/nice/path/?zot=23&zut&burp=xxx&zing", urlpath.add("burp", "xxx").add("zing").to_text(), ) # Note the inversion! self.assertEqual( "http://www.foo.com/a/nice/path/?zot=23&zut&zing&burp=xxx", urlpath.add("zing").add("burp", "xxx").to_text(), ) # Note the two values for the same name. self.assertEqual( "http://www.foo.com/a/nice/path/?zot=23&zut&burp=xxx&zot=32", urlpath.add("burp", "xxx").add("zot", "32").to_text(), ) def test_querySet(self): # type: () -> None """ L{URL.set} replaces query parameters by name. """ urlpath = URL.from_text(BASIC_URL) self.assertEqual( "http://www.foo.com/a/nice/path/?zot=32&zut", urlpath.set("zot", "32").to_text(), ) # Replace name without value with name/value and vice-versa. self.assertEqual( "http://www.foo.com/a/nice/path/?zot&zut=itworked", urlpath.set("zot").set("zut", "itworked").to_text(), ) # Q: what happens when the query has two values and we replace? # A: we replace both values with a single one self.assertEqual( "http://www.foo.com/a/nice/path/?zot=32&zut", urlpath.add("zot", "xxx").set("zot", "32").to_text(), ) def test_queryRemove(self): # type: () -> None """ L{URL.remove} removes instances of a query parameter. """ url = URL.from_text("https://example.com/a/b/?foo=1&bar=2&foo=3") self.assertEqual( url.remove("foo"), URL.from_text("https://example.com/a/b/?bar=2") ) self.assertEqual( url.remove(name="foo", value="1"), URL.from_text("https://example.com/a/b/?bar=2&foo=3"), ) self.assertEqual( url.remove(name="foo", limit=1), URL.from_text("https://example.com/a/b/?bar=2&foo=3"), ) self.assertEqual( url.remove(name="foo", value="1", limit=0), URL.from_text("https://example.com/a/b/?foo=1&bar=2&foo=3"), ) def test_parseEqualSignInParamValue(self): # type: () -> None """ Every C{=}-sign after the first in a query parameter is simply included in the value of the parameter. """ u = URL.from_text("http://localhost/?=x=x=x") self.assertEqual(u.get(""), ["x=x=x"]) self.assertEqual(u.to_text(), "http://localhost/?=x=x=x") u = URL.from_text("http://localhost/?foo=x=x=x&bar=y") self.assertEqual(u.query, (("foo", "x=x=x"), ("bar", "y"))) self.assertEqual(u.to_text(), "http://localhost/?foo=x=x=x&bar=y") u = URL.from_text( "https://example.com/?argument=3&argument=4&operator=%3D" ) iri = u.to_iri() self.assertEqual(iri.get("operator"), ["="]) # assert that the equals is not unnecessarily escaped self.assertEqual(iri.to_uri().get("operator"), ["="]) def test_empty(self): # type: () -> None """ An empty L{URL} should serialize as the empty string. """ self.assertEqual(URL().to_text(), "") def test_justQueryText(self): # type: () -> None """ An L{URL} with query text should serialize as just query text. """ u = URL(query=[("hello", "world")]) self.assertEqual(u.to_text(), "?hello=world") def test_identicalEqual(self): # type: () -> None """ L{URL} compares equal to itself. """ u = URL.from_text("http://localhost/") self.assertEqual(u, u) def test_similarEqual(self): # type: () -> None """ URLs with equivalent components should compare equal. """ u1 = URL.from_text("http://u@localhost:8080/p/a/t/h?q=p#f") u2 = URL.from_text("http://u@localhost:8080/p/a/t/h?q=p#f") self.assertEqual(u1, u2) def test_differentNotEqual(self): # type: () -> None """ L{URL}s that refer to different resources are both unequal (C{!=}) and also not equal (not C{==}). """ u1 = URL.from_text("http://localhost/a") u2 = URL.from_text("http://localhost/b") self.assertFalse(u1 == u2, "%r != %r" % (u1, u2)) self.assertNotEqual(u1, u2) def test_otherTypesNotEqual(self): # type: () -> None """ L{URL} is not equal (C{==}) to other types. """ u = URL.from_text("http://localhost/") self.assertFalse(u == 42, "URL must not equal a number.") self.assertFalse(u == object(), "URL must not equal an object.") self.assertNotEqual(u, 42) self.assertNotEqual(u, object()) def test_identicalNotUnequal(self): # type: () -> None """ Identical L{URL}s are not unequal (C{!=}) to each other. """ u = URL.from_text("http://u@localhost:8080/p/a/t/h?q=p#f") self.assertFalse(u != u, "%r == itself" % u) def test_similarNotUnequal(self): # type: () -> None """ Structurally similar L{URL}s are not unequal (C{!=}) to each other. """ u1 = URL.from_text("http://u@localhost:8080/p/a/t/h?q=p#f") u2 = URL.from_text("http://u@localhost:8080/p/a/t/h?q=p#f") self.assertFalse(u1 != u2, "%r == %r" % (u1, u2)) def test_differentUnequal(self): # type: () -> None """ Structurally different L{URL}s are unequal (C{!=}) to each other. """ u1 = URL.from_text("http://localhost/a") u2 = URL.from_text("http://localhost/b") self.assertTrue(u1 != u2, "%r == %r" % (u1, u2)) def test_otherTypesUnequal(self): # type: () -> None """ L{URL} is unequal (C{!=}) to other types. """ u = URL.from_text("http://localhost/") self.assertTrue(u != 42, "URL must differ from a number.") self.assertTrue(u != object(), "URL must be differ from an object.") def test_asURI(self): # type: () -> None """ L{URL.asURI} produces an URI which converts any URI unicode encoding into pure US-ASCII and returns a new L{URL}. """ unicodey = ( "http://\N{LATIN SMALL LETTER E WITH ACUTE}.com/" "\N{LATIN SMALL LETTER E}\N{COMBINING ACUTE ACCENT}" "?\N{LATIN SMALL LETTER A}\N{COMBINING ACUTE ACCENT}=" "\N{LATIN SMALL LETTER I}\N{COMBINING ACUTE ACCENT}" "#\N{LATIN SMALL LETTER U}\N{COMBINING ACUTE ACCENT}" ) iri = URL.from_text(unicodey) uri = iri.asURI() self.assertEqual(iri.host, "\N{LATIN SMALL LETTER E WITH ACUTE}.com") self.assertEqual( iri.path[0], "\N{LATIN SMALL LETTER E}\N{COMBINING ACUTE ACCENT}" ) self.assertEqual(iri.to_text(), unicodey) expectedURI = "http://xn--9ca.com/%C3%A9?%C3%A1=%C3%AD#%C3%BA" actualURI = uri.to_text() self.assertEqual( actualURI, expectedURI, "%r != %r" % (actualURI, expectedURI) ) def test_asIRI(self): # type: () -> None """ L{URL.asIRI} decodes any percent-encoded text in the URI, making it more suitable for reading by humans, and returns a new L{URL}. """ asciiish = "http://xn--9ca.com/%C3%A9?%C3%A1=%C3%AD#%C3%BA" uri = URL.from_text(asciiish) iri = uri.asIRI() self.assertEqual(uri.host, "xn--9ca.com") self.assertEqual(uri.path[0], "%C3%A9") self.assertEqual(uri.to_text(), asciiish) expectedIRI = ( "http://\N{LATIN SMALL LETTER E WITH ACUTE}.com/" "\N{LATIN SMALL LETTER E WITH ACUTE}" "?\N{LATIN SMALL LETTER A WITH ACUTE}=" "\N{LATIN SMALL LETTER I WITH ACUTE}" "#\N{LATIN SMALL LETTER U WITH ACUTE}" ) actualIRI = iri.to_text() self.assertEqual( actualIRI, expectedIRI, "%r != %r" % (actualIRI, expectedIRI) ) def test_badUTF8AsIRI(self): # type: () -> None """ Bad UTF-8 in a path segment, query parameter, or fragment results in that portion of the URI remaining percent-encoded in the IRI. """ urlWithBinary = "http://xn--9ca.com/%00%FF/%C3%A9" uri = URL.from_text(urlWithBinary) iri = uri.asIRI() expectedIRI = ( "http://\N{LATIN SMALL LETTER E WITH ACUTE}.com/" "%00%FF/" "\N{LATIN SMALL LETTER E WITH ACUTE}" ) actualIRI = iri.to_text() self.assertEqual( actualIRI, expectedIRI, "%r != %r" % (actualIRI, expectedIRI) ) def test_alreadyIRIAsIRI(self): # type: () -> None """ A L{URL} composed of non-ASCII text will result in non-ASCII text. """ unicodey = ( "http://\N{LATIN SMALL LETTER E WITH ACUTE}.com/" "\N{LATIN SMALL LETTER E}\N{COMBINING ACUTE ACCENT}" "?\N{LATIN SMALL LETTER A}\N{COMBINING ACUTE ACCENT}=" "\N{LATIN SMALL LETTER I}\N{COMBINING ACUTE ACCENT}" "#\N{LATIN SMALL LETTER U}\N{COMBINING ACUTE ACCENT}" ) iri = URL.from_text(unicodey) alsoIRI = iri.asIRI() self.assertEqual(alsoIRI.to_text(), unicodey) def test_alreadyURIAsURI(self): # type: () -> None """ A L{URL} composed of encoded text will remain encoded. """ expectedURI = "http://xn--9ca.com/%C3%A9?%C3%A1=%C3%AD#%C3%BA" uri = URL.from_text(expectedURI) actualURI = uri.asURI().to_text() self.assertEqual(actualURI, expectedURI) def test_userinfo(self): # type: () -> None """ L{URL.from_text} will parse the C{userinfo} portion of the URI separately from the host and port. """ url = URL.from_text( "http://someuser:somepassword@example.com/some-segment@ignore" ) self.assertEqual( url.authority(True), "someuser:somepassword@example.com" ) self.assertEqual(url.authority(False), "someuser:@example.com") self.assertEqual(url.userinfo, "someuser:somepassword") self.assertEqual(url.user, "someuser") self.assertEqual( url.to_text(), "http://someuser:@example.com/some-segment@ignore" ) self.assertEqual( url.replace(userinfo="someuser").to_text(), "http://someuser@example.com/some-segment@ignore", ) def test_portText(self): # type: () -> None """ L{URL.from_text} parses custom port numbers as integers. """ portURL = URL.from_text("http://www.example.com:8080/") self.assertEqual(portURL.port, 8080) self.assertEqual(portURL.to_text(), "http://www.example.com:8080/") def test_mailto(self): # type: () -> None """ Although L{URL} instances are mainly for dealing with HTTP, other schemes (such as C{mailto:}) should work as well. For example, L{URL.from_text}/L{URL.to_text} round-trips cleanly for a C{mailto:} URL representing an email address. """ self.assertEqual( URL.from_text("mailto:user@example.com").to_text(), "mailto:user@example.com", ) def test_httpWithoutHost(self): # type: () -> None """ An HTTP URL without a hostname, but with a path, should also round-trip cleanly. """ without_host = URL.from_text("http:relative-path") self.assertEqual(without_host.host, "") self.assertEqual(without_host.path, ("relative-path",)) self.assertEqual(without_host.uses_netloc, False) self.assertEqual(without_host.to_text(), "http:relative-path") def test_queryIterable(self): # type: () -> None """ When a L{URL} is created with a C{query} argument, the C{query} argument is converted into an N-tuple of 2-tuples, sensibly handling dictionaries. """ expected = (("alpha", "beta"),) url = URL(query=[("alpha", "beta")]) self.assertEqual(url.query, expected) url = URL(query={"alpha": "beta"}) self.assertEqual(url.query, expected) def test_pathIterable(self): # type: () -> None """ When a L{URL} is created with a C{path} argument, the C{path} is converted into a tuple. """ url = URL(path=["hello", "world"]) self.assertEqual(url.path, ("hello", "world")) def test_invalidArguments(self): # type: () -> None """ Passing an argument of the wrong type to any of the constructor arguments of L{URL} will raise a descriptive L{TypeError}. L{URL} typechecks very aggressively to ensure that its constitutent parts are all properly immutable and to prevent confusing errors when bad data crops up in a method call long after the code that called the constructor is off the stack. """ class Unexpected(object): def __str__(self): # type: () -> str return "wrong" def __repr__(self): # type: () -> str return "<unexpected>" defaultExpectation = "unicode" if bytes is str else "str" def assertRaised(raised, expectation, name): # type: (Any, Text, Text) -> None self.assertEqual( str(raised.exception), "expected {0} for {1}, got {2}".format( expectation, name, "<unexpected>" ), ) def check(param, expectation=defaultExpectation): # type: (Any, str) -> None with self.assertRaises(TypeError) as raised: URL(**{param: Unexpected()}) # type: ignore[arg-type] assertRaised(raised, expectation, param) check("scheme") check("host") check("fragment") check("rooted", "bool") check("userinfo") check("port", "int or NoneType") with self.assertRaises(TypeError) as raised: URL(path=[cast(Text, Unexpected())]) assertRaised(raised, defaultExpectation, "path segment") with self.assertRaises(TypeError) as raised: URL(query=[("name", cast(Text, Unexpected()))]) assertRaised( raised, defaultExpectation + " or NoneType", "query parameter value" ) with self.assertRaises(TypeError) as raised: URL(query=[(cast(Text, Unexpected()), "value")]) assertRaised(raised, defaultExpectation, "query parameter name") # No custom error message for this one, just want to make sure # non-2-tuples don't get through. with self.assertRaises(TypeError): URL(query=[cast(Tuple[Text, Text], Unexpected())]) with self.assertRaises(ValueError): URL(query=[cast(Tuple[Text, Text], ("k", "v", "vv"))]) with self.assertRaises(ValueError): URL(query=[cast(Tuple[Text, Text], ("k",))]) url = URL.from_text("https://valid.example.com/") with self.assertRaises(TypeError) as raised: url.child(cast(Text, Unexpected())) assertRaised(raised, defaultExpectation, "path segment") with self.assertRaises(TypeError) as raised: url.sibling(cast(Text, Unexpected())) assertRaised(raised, defaultExpectation, "path segment") with self.assertRaises(TypeError) as raised: url.click(cast(Text, Unexpected())) assertRaised(raised, defaultExpectation, "relative URL") def test_technicallyTextIsIterableBut(self): # type: () -> None """ Technically, L{str} (or L{unicode}, as appropriate) is iterable, but C{URL(path="foo")} resulting in C{URL.from_text("f/o/o")} is never what you want. """ with self.assertRaises(TypeError) as raised: URL(path="foo") self.assertEqual( str(raised.exception), "expected iterable of text for path, not: {0}".format(repr("foo")), ) def test_netloc(self): # type: () -> None url = URL(scheme="https") self.assertEqual(url.uses_netloc, True) self.assertEqual(url.to_text(), "https://") # scheme, no host, no path, no netloc hack self.assertEqual(URL.from_text("https:").uses_netloc, False) # scheme, no host, absolute path, no netloc hack self.assertEqual(URL.from_text("https:/").uses_netloc, False) # scheme, no host, no path, netloc hack to indicate :// syntax self.assertEqual(URL.from_text("https://").uses_netloc, True) url = URL(scheme="https", uses_netloc=False) self.assertEqual(url.uses_netloc, False) self.assertEqual(url.to_text(), "https:") url = URL(scheme="git+https") self.assertEqual(url.uses_netloc, True) self.assertEqual(url.to_text(), "git+https://") url = URL(scheme="mailto") self.assertEqual(url.uses_netloc, False) self.assertEqual(url.to_text(), "mailto:") url = URL(scheme="ztp") self.assertEqual(url.uses_netloc, None) self.assertEqual(url.to_text(), "ztp:") url = URL.from_text("ztp://test.com") self.assertEqual(url.uses_netloc, True) url = URL.from_text("ztp:test:com") self.assertEqual(url.uses_netloc, False) def test_ipv6_with_port(self): # type: () -> None t = "https://[2001:0db8:85a3:0000:0000:8a2e:0370:7334]:80/" url = URL.from_text(t) assert url.host == "2001:0db8:85a3:0000:0000:8a2e:0370:7334" assert url.port == 80 assert SCHEME_PORT_MAP[url.scheme] != url.port def test_basic(self): # type: () -> None text = "https://user:pass@example.com/path/to/here?k=v#nice" url = URL.from_text(text) assert url.scheme == "https" assert url.userinfo == "user:pass" assert url.host == "example.com" assert url.path == ("path", "to", "here") assert url.fragment == "nice" text = "https://user:pass@127.0.0.1/path/to/here?k=v#nice" url = URL.from_text(text) assert url.scheme == "https" assert url.userinfo == "user:pass" assert url.host == "127.0.0.1" assert url.path == ("path", "to", "here") text = "https://user:pass@[::1]/path/to/here?k=v#nice" url = URL.from_text(text) assert url.scheme == "https" assert url.userinfo == "user:pass" assert url.host == "::1" assert url.path == ("path", "to", "here") def test_invalid_url(self): # type: () -> None self.assertRaises(URLParseError, URL.from_text, "#\n\n") def test_invalid_authority_url(self): # type: () -> None self.assertRaises(URLParseError, URL.from_text, "http://abc:\n\n/#") def test_invalid_ipv6(self): # type: () -> None invalid_ipv6_ips = [ "2001::0234:C1ab::A0:aabc:003F", "2001::1::3F", ":", "::::", "::256.0.0.1", ] for ip in invalid_ipv6_ips: url_text = "http://[" + ip + "]" self.assertRaises(socket.error, inet_pton, socket.AF_INET6, ip) self.assertRaises(URLParseError, URL.from_text, url_text) def test_invalid_port(self): # type: () -> None self.assertRaises(URLParseError, URL.from_text, "ftp://portmouth:smash") self.assertRaises( ValueError, URL.from_text, "http://reader.googlewebsite.com:neverforget", ) def test_idna(self): # type: () -> None u1 = URL.from_text("http://bücher.ch") self.assertEqual(u1.host, "bücher.ch") self.assertEqual(u1.to_text(), "http://bücher.ch") self.assertEqual(u1.to_uri().to_text(), "http://xn--bcher-kva.ch") u2 = URL.from_text("https://xn--bcher-kva.ch") self.assertEqual(u2.host, "xn--bcher-kva.ch") self.assertEqual(u2.to_text(), "https://xn--bcher-kva.ch") self.assertEqual(u2.to_iri().to_text(), "https://bücher.ch") def test_netloc_slashes(self): # type: () -> None # basic sanity checks url = URL.from_text("mailto:mahmoud@hatnote.com") self.assertEqual(url.scheme, "mailto") self.assertEqual(url.to_text(), "mailto:mahmoud@hatnote.com") url = URL.from_text("http://hatnote.com") self.assertEqual(url.scheme, "http") self.assertEqual(url.to_text(), "http://hatnote.com") # test that unrecognized schemes stay consistent with '//' url = URL.from_text("newscheme:a:b:c") self.assertEqual(url.scheme, "newscheme") self.assertEqual(url.to_text(), "newscheme:a:b:c") url = URL.from_text("newerscheme://a/b/c") self.assertEqual(url.scheme, "newerscheme") self.assertEqual(url.to_text(), "newerscheme://a/b/c") # test that reasonable guesses are made url = URL.from_text("git+ftp://gitstub.biz/glyph/lefkowitz") self.assertEqual(url.scheme, "git+ftp") self.assertEqual(url.to_text(), "git+ftp://gitstub.biz/glyph/lefkowitz") url = URL.from_text("what+mailto:freerealestate@enotuniq.org") self.assertEqual(url.scheme, "what+mailto") self.assertEqual( url.to_text(), "what+mailto:freerealestate@enotuniq.org" ) url = URL(scheme="ztp", path=("x", "y", "z"), rooted=True) self.assertEqual(url.to_text(), "ztp:/x/y/z") # also works when the input doesn't include '//' url = URL( scheme="git+ftp", path=("x", "y", "z", ""), rooted=True, uses_netloc=True, ) # broken bc urlunsplit self.assertEqual(url.to_text(), "git+ftp:///x/y/z/") # really why would this ever come up but ok url = URL.from_text("file:///path/to/heck") url2 = url.replace(scheme="mailto") self.assertEqual(url2.to_text(), "mailto:/path/to/heck") url_text = "unregisteredscheme:///a/b/c" url = URL.from_text(url_text) no_netloc_url = url.replace(uses_netloc=False) self.assertEqual(no_netloc_url.to_text(), "unregisteredscheme:/a/b/c") netloc_url = url.replace(uses_netloc=True) self.assertEqual(netloc_url.to_text(), url_text) return def test_rooted_to_relative(self): # type: () -> None """ On host-relative URLs, the C{rooted} flag can be updated to indicate that the path should no longer be treated as absolute. """ a = URL(path=["hello"]) self.assertEqual(a.to_text(), "hello") b = a.replace(rooted=True) self.assertEqual(b.to_text(), "/hello") self.assertNotEqual(a, b) def test_autorooted(self): # type: () -> None """ The C{rooted} flag can be updated in some cases, but it cannot be made to conflict with other facts surrounding the URL; for example, all URLs involving an authority (host) are inherently rooted because it is not syntactically possible to express otherwise; also, once an unrooted URL gains a path that starts with an empty string, that empty string is elided and it becomes rooted, because these cases are syntactically indistinguisable in real URL text. """ relative_path_rooted = URL(path=["", "foo"], rooted=False) self.assertEqual(relative_path_rooted.rooted, True) relative_flag_rooted = URL(path=["foo"], rooted=True) self.assertEqual(relative_flag_rooted.rooted, True) self.assertEqual(relative_path_rooted, relative_flag_rooted) attempt_unrooted_absolute = URL(host="foo", path=["bar"], rooted=False) normal_absolute = URL(host="foo", path=["bar"]) self.assertEqual(attempt_unrooted_absolute, normal_absolute) self.assertEqual(normal_absolute.rooted, True) self.assertEqual(attempt_unrooted_absolute.rooted, True) def test_rooted_with_port_but_no_host(self): # type: () -> None """ URLs which include a ``://`` netloc-separator for any reason are inherently rooted, regardless of the value or presence of the ``rooted`` constructor argument. They may include a netloc-separator because their constructor was directly invoked with an explicit host or port, or because they were parsed from a string which included the literal ``://`` separator. """ directly_constructed = URL(scheme="udp", port=4900, rooted=False) directly_constructed_implict = URL(scheme="udp", port=4900) directly_constructed_rooted = URL(scheme="udp", port=4900, rooted=True) self.assertEqual(directly_constructed.rooted, True) self.assertEqual(directly_constructed_implict.rooted, True) self.assertEqual(directly_constructed_rooted.rooted, True) parsed = URL.from_text("udp://:4900") self.assertEqual(str(directly_constructed), str(parsed)) self.assertEqual(str(directly_constructed_implict), str(parsed)) self.assertEqual(directly_constructed.asText(), parsed.asText()) self.assertEqual(directly_constructed, parsed) self.assertEqual(directly_constructed, directly_constructed_implict) self.assertEqual(directly_constructed, directly_constructed_rooted) self.assertEqual(directly_constructed_implict, parsed) self.assertEqual(directly_constructed_rooted, parsed) def test_wrong_constructor(self): # type: () -> None with self.assertRaises(ValueError): # whole URL not allowed URL(BASIC_URL) with self.assertRaises(ValueError): # explicitly bad scheme not allowed URL("HTTP_____more_like_imHoTTeP") def test_encoded_userinfo(self): # type: () -> None url = URL.from_text("http://user:pass@example.com") assert url.userinfo == "user:pass" url = url.replace(userinfo="us%20her:pass") iri = url.to_iri() assert ( iri.to_text(with_password=True) == "http://us her:pass@example.com" ) assert iri.to_text(with_password=False) == "http://us her:@example.com" assert ( iri.to_uri().to_text(with_password=True) == "http://us%20her:pass@example.com" ) def test_hash(self): # type: () -> None url_map = {} url1 = URL.from_text("http://blog.hatnote.com/ask?utm_source=geocity") assert hash(url1) == hash(url1) # sanity url_map[url1] = 1 url2 = URL.from_text("http://blog.hatnote.com/ask") url2 = url2.set("utm_source", "geocity") url_map[url2] = 2 assert len(url_map) == 1 assert list(url_map.values()) == [2] assert hash(URL()) == hash(URL()) # slightly more sanity def test_dir(self): # type: () -> None url = URL() res = dir(url) assert len(res) > 15 # twisted compat assert "fromText" not in res assert "asText" not in res assert "asURI" not in res assert "asIRI" not in res def test_twisted_compat(self): # type: () -> None url = URL.fromText("http://example.com/a%20té%C3%A9st") assert url.asText() == "http://example.com/a%20té%C3%A9st" assert url.asURI().asText() == "http://example.com/a%20t%C3%A9%C3%A9st" # TODO: assert url.asIRI().asText() == u'http://example.com/a%20téést' def test_set_ordering(self): # type: () -> None # TODO url = URL.from_text("http://example.com/?a=b&c") url = url.set("x", "x") url = url.add("x", "y") assert url.to_text() == "http://example.com/?a=b&x=x&c&x=y" # Would expect: # assert url.to_text() == u'http://example.com/?a=b&c&x=x&x=y' def test_schemeless_path(self): # type: () -> None "See issue #4" u1 = URL.from_text("urn%3Aietf%3Awg%3Aoauth%3A2.0%3Aoob") u2 = URL.from_text(u1.to_text()) assert u1 == u2 # sanity testing roundtripping u3 = URL.from_text(u1.to_iri().to_text()) assert u1 == u3 assert u2 == u3 # test that colons are ok past the first segment u4 = URL.from_text("first-segment/urn%3Aietf%3Awg%3Aoauth%3A2.0%3Aoob") u5 = u4.to_iri() assert u5.to_text() == "first-segment/urn:ietf:wg:oauth:2.0:oob" u6 = URL.from_text(u5.to_text()).to_uri() assert u5 == u6 # colons stay decoded bc they're not in the first seg def test_emoji_domain(self): # type: () -> None "See issue #7, affecting only narrow builds (2.6-3.3)" url = URL.from_text("https://xn--vi8hiv.ws") iri = url.to_iri() iri.to_text() # as long as we don't get ValueErrors, we're good def test_delim_in_param(self): # type: () -> None "Per issue #6 and #8" self.assertRaises(ValueError, URL, scheme="http", host="a/c") self.assertRaises(ValueError, URL, path=("?",)) self.assertRaises(ValueError, URL, path=("#",)) self.assertRaises(ValueError, URL, query=(("&", "test"))) def test_empty_paths_eq(self): # type: () -> None u1 = URL.from_text("http://example.com/") u2 = URL.from_text("http://example.com") assert u1 == u2 u1 = URL.from_text("http://example.com") u2 = URL.from_text("http://example.com") assert u1 == u2 u1 = URL.from_text("http://example.com") u2 = URL.from_text("http://example.com/") assert u1 == u2 u1 = URL.from_text("http://example.com/") u2 = URL.from_text("http://example.com/") assert u1 == u2 def test_from_text_type(self): # type: () -> None assert URL.from_text("#ok").fragment == "ok" # sanity self.assertRaises(TypeError, URL.from_text, b"bytes://x.y.z") self.assertRaises(TypeError, URL.from_text, object()) def test_from_text_bad_authority(self): # type: () -> None # bad ipv6 brackets self.assertRaises(URLParseError, URL.from_text, "http://[::1/") self.assertRaises(URLParseError, URL.from_text, "http://::1]/") self.assertRaises(URLParseError, URL.from_text, "http://[[::1]/") self.assertRaises(URLParseError, URL.from_text, "http://[::1]]/") # empty port self.assertRaises(URLParseError, URL.from_text, "http://127.0.0.1:") # non-integer port self.assertRaises(URLParseError, URL.from_text, "http://127.0.0.1:hi") # extra port colon (makes for an invalid host) self.assertRaises(URLParseError, URL.from_text, "http://127.0.0.1::80") def test_normalize(self): # type: () -> None url = URL.from_text("HTTP://Example.com/A%61/./../A%61?B%62=C%63#D%64") assert url.get("Bb") == [] assert url.get("B%62") == ["C%63"] assert len(url.path) == 4 # test that most expected normalizations happen norm_url = url.normalize() assert norm_url.scheme == "http" assert norm_url.host == "example.com" assert norm_url.path == ("Aa",) assert norm_url.get("Bb") == ["Cc"] assert norm_url.fragment == "Dd" assert norm_url.to_text() == "http://example.com/Aa?Bb=Cc#Dd" # test that flags work noop_norm_url = url.normalize( scheme=False, host=False, path=False, query=False, fragment=False ) assert noop_norm_url == url # test that empty paths get at least one slash slashless_url = URL.from_text("http://example.io") slashful_url = slashless_url.normalize() assert slashful_url.to_text() == "http://example.io/" # test case normalization for percent encoding delimited_url = URL.from_text("/a%2fb/cd%3f?k%3d=v%23#test") norm_delimited_url = delimited_url.normalize() assert norm_delimited_url.to_text() == "/a%2Fb/cd%3F?k%3D=v%23#test" # test invalid percent encoding during normalize assert ( URL(path=("", "%te%sts")).normalize(percents=False).to_text() == "/%te%sts" ) assert URL(path=("", "%te%sts")).normalize().to_text() == "/%25te%25sts" percenty_url = URL( scheme="ftp", path=["%%%", "%a%b"], query=[("%", "%%")], fragment="%", userinfo="%:%", ) assert ( percenty_url.to_text(with_password=True) == "ftp://%:%@/%%%/%a%b?%=%%#%" ) assert ( percenty_url.normalize().to_text(with_password=True) == "ftp://%25:%25@/%25%25%25/%25a%25b?%25=%25%25#%25" ) def test_str(self): # type: () -> None # see also issue #49 text = "http://example.com/á/y%20a%20y/?b=%25" url = URL.from_text(text) assert unicode(url) == text assert bytes(url) == b"http://example.com/%C3%A1/y%20a%20y/?b=%25" if PY2: assert isinstance(str(url), bytes) assert isinstance(unicode(url), unicode) else: assert isinstance(str(url), unicode) assert isinstance(bytes(url), bytes) def test_idna_corners(self): # type: () -> None url = URL.from_text("http://abé.com/") assert url.to_iri().host == "abé.com" assert url.to_uri().host == "xn--ab-cja.com" url = URL.from_text("http://ドメイン.テスト.co.jp#test") assert url.to_iri().host == "ドメイン.テスト.co.jp" assert url.to_uri().host == "xn--eckwd4c7c.xn--zckzah.co.jp" assert url.to_uri().get_decoded_url().host == "ドメイン.テスト.co.jp" text = "http://Example.com" assert ( URL.from_text(text).to_uri().get_decoded_url().host == "example.com" )