import unittest from test import test_support import os, socket import StringIO import urllib2 from urllib2 import Request, OpenerDirector # XXX # Request # CacheFTPHandler (hard to write) # parse_keqv_list, parse_http_list (I'm leaving this for Anthony Baxter # and Greg Stein, since they're doing Digest Authentication) # Authentication stuff (ditto) # ProxyHandler, CustomProxy, CustomProxyHandler (I don't use a proxy) # GopherHandler (haven't used gopher for a decade or so...) class TrivialTests(unittest.TestCase): def test_trivial(self): # A couple trivial tests self.assertRaises(ValueError, urllib2.urlopen, 'bogus url') # XXX Name hacking to get this to work on Windows. fname = os.path.abspath(urllib2.__file__).replace('\\', '/') if fname[1:2] == ":": fname = fname[2:] # And more hacking to get it to work on MacOS. This assumes # urllib.pathname2url works, unfortunately... if os.name == 'mac': fname = '/' + fname.replace(':', '/') elif os.name == 'riscos': import string fname = os.expand(fname) fname = fname.translate(string.maketrans("/.", "./")) file_url = "file://%s" % fname f = urllib2.urlopen(file_url) buf = f.read() f.close() def test_statudict(self): # test the new-in-2.5 httpresponses dictionary self.assertEquals(urllib2.httpresponses[404], "Not Found") def test_parse_http_list(self): tests = [('a,b,c', ['a', 'b', 'c']), ('path"o,l"og"i"cal, example', ['path"o,l"og"i"cal', 'example']), ('a, b, "c", "d", "e,f", g, h', ['a', 'b', '"c"', '"d"', '"e,f"', 'g', 'h']), ('a="b\\"c", d="e\\,f", g="h\\\\i"', ['a="b"c"', 'd="e,f"', 'g="h\\i"'])] for string, list in tests: self.assertEquals(urllib2.parse_http_list(string), list) class MockOpener: addheaders = [] def open(self, req, data=None): self.req, self.data = req, data def error(self, proto, *args): self.proto, self.args = proto, args class MockFile: def read(self, count=None): pass def readline(self, count=None): pass def close(self): pass class MockHeaders(dict): def getheaders(self, name): return self.values() class MockResponse(StringIO.StringIO): def __init__(self, code, msg, headers, data, url=None): StringIO.StringIO.__init__(self, data) self.code, self.msg, self.headers, self.url = code, msg, headers, url def info(self): return self.headers def geturl(self): return self.url class MockCookieJar: def add_cookie_header(self, request): self.ach_req = request def extract_cookies(self, response, request): self.ec_req, self.ec_r = request, response class FakeMethod: def __init__(self, meth_name, action, handle): self.meth_name = meth_name self.handle = handle self.action = action def __call__(self, *args): return self.handle(self.meth_name, self.action, *args) class MockHandler: def __init__(self, methods): self._define_methods(methods) def _define_methods(self, methods): for spec in methods: if len(spec) == 2: name, action = spec else: name, action = spec, None meth = FakeMethod(name, action, self.handle) setattr(self.__class__, name, meth) def handle(self, fn_name, action, *args, **kwds): self.parent.calls.append((self, fn_name, args, kwds)) if action is None: return None elif action == "return self": return self elif action == "return response": res = MockResponse(200, "OK", {}, "") return res elif action == "return request": return Request("http://blah/") elif action.startswith("error"): code = action[action.rfind(" ")+1:] try: code = int(code) except ValueError: pass res = MockResponse(200, "OK", {}, "") return self.parent.error("http", args[0], res, code, "", {}) elif action == "raise": raise urllib2.URLError("blah") assert False def close(self): pass def add_parent(self, parent): self.parent = parent self.parent.calls = [] def __lt__(self, other): if not hasattr(other, "handler_order"): # No handler_order, leave in original order. Yuck. return True return self.handler_order < other.handler_order def add_ordered_mock_handlers(opener, meth_spec): """Create MockHandlers and add them to an OpenerDirector. meth_spec: list of lists of tuples and strings defining methods to define on handlers. eg: [["http_error", "ftp_open"], ["http_open"]] defines methods .http_error() and .ftp_open() on one handler, and .http_open() on another. These methods just record their arguments and return None. Using a tuple instead of a string causes the method to perform some action (see MockHandler.handle()), eg: [["http_error"], [("http_open", "return request")]] defines .http_error() on one handler (which simply returns None), and .http_open() on another handler, which returns a Request object. """ handlers = [] count = 0 for meths in meth_spec: class MockHandlerSubclass(MockHandler): pass h = MockHandlerSubclass(meths) h.handler_order = count h.add_parent(opener) count = count + 1 handlers.append(h) opener.add_handler(h) return handlers class OpenerDirectorTests(unittest.TestCase): def test_handled(self): # handler returning non-None means no more handlers will be called o = OpenerDirector() meth_spec = [ ["http_open", "ftp_open", "http_error_302"], ["ftp_open"], [("http_open", "return self")], [("http_open", "return self")], ] handlers = add_ordered_mock_handlers(o, meth_spec) req = Request("http://example.com/") r = o.open(req) # Second .http_open() gets called, third doesn't, since second returned # non-None. Handlers without .http_open() never get any methods called # on them. # In fact, second mock handler defining .http_open() returns self # (instead of response), which becomes the OpenerDirector's return # value. self.assertEqual(r, handlers[2]) calls = [(handlers[0], "http_open"), (handlers[2], "http_open")] for expected, got in zip(calls, o.calls): handler, name, args, kwds = got self.assertEqual((handler, name), expected) self.assertEqual(args, (req,)) def test_handler_order(self): o = OpenerDirector() handlers = [] for meths, handler_order in [ ([("http_open", "return self")], 500), (["http_open"], 0), ]: class MockHandlerSubclass(MockHandler): pass h = MockHandlerSubclass(meths) h.handler_order = handler_order handlers.append(h) o.add_handler(h) r = o.open("http://example.com/") # handlers called in reverse order, thanks to their sort order self.assertEqual(o.calls[0][0], handlers[1]) self.assertEqual(o.calls[1][0], handlers[0]) def test_raise(self): # raising URLError stops processing of request o = OpenerDirector() meth_spec = [ [("http_open", "raise")], [("http_open", "return self")], ] handlers = add_ordered_mock_handlers(o, meth_spec) req = Request("http://example.com/") self.assertRaises(urllib2.URLError, o.open, req) self.assertEqual(o.calls, [(handlers[0], "http_open", (req,), {})]) ## def test_error(self): ## # XXX this doesn't actually seem to be used in standard library, ## # but should really be tested anyway... def test_http_error(self): # XXX http_error_default # http errors are a special case o = OpenerDirector() meth_spec = [ [("http_open", "error 302")], [("http_error_400", "raise"), "http_open"], [("http_error_302", "return response"), "http_error_303", "http_error"], [("http_error_302")], ] handlers = add_ordered_mock_handlers(o, meth_spec) class Unknown: def __eq__(self, other): return True req = Request("http://example.com/") r = o.open(req) assert len(o.calls) == 2 calls = [(handlers[0], "http_open", (req,)), (handlers[2], "http_error_302", (req, Unknown(), 302, "", {}))] for expected, got in zip(calls, o.calls): handler, method_name, args = expected self.assertEqual((handler, method_name), got[:2]) self.assertEqual(args, got[2]) def test_processors(self): # *_request / *_response methods get called appropriately o = OpenerDirector() meth_spec = [ [("http_request", "return request"), ("http_response", "return response")], [("http_request", "return request"), ("http_response", "return response")], ] handlers = add_ordered_mock_handlers(o, meth_spec) req = Request("http://example.com/") r = o.open(req) # processor methods are called on *all* handlers that define them, # not just the first handler that handles the request calls = [ (handlers[0], "http_request"), (handlers[1], "http_request"), (handlers[0], "http_response"), (handlers[1], "http_response")] for i, (handler, name, args, kwds) in enumerate(o.calls): if i < 2: # *_request self.assertEqual((handler, name), calls[i]) self.assertEqual(len(args), 1) self.assert_(isinstance(args[0], Request)) else: # *_response self.assertEqual((handler, name), calls[i]) self.assertEqual(len(args), 2) self.assert_(isinstance(args[0], Request)) # response from opener.open is None, because there's no # handler that defines http_open to handle it self.assert_(args[1] is None or isinstance(args[1], MockResponse)) def sanepathname2url(path): import urllib urlpath = urllib.pathname2url(path) if os.name == "nt" and urlpath.startswith("///"): urlpath = urlpath[2:] # XXX don't ask me about the mac... return urlpath class HandlerTests(unittest.TestCase): def test_ftp(self): class MockFTPWrapper: def __init__(self, data): self.data = data def retrfile(self, filename, filetype): self.filename, self.filetype = filename, filetype return StringIO.StringIO(self.data), len(self.data) class NullFTPHandler(urllib2.FTPHandler): def __init__(self, data): self.data = data def connect_ftp(self, user, passwd, host, port, dirs): self.user, self.passwd = user, passwd self.host, self.port = host, port self.dirs = dirs self.ftpwrapper = MockFTPWrapper(self.data) return self.ftpwrapper import ftplib, socket data = "rheum rhaponicum" h = NullFTPHandler(data) o = h.parent = MockOpener() for url, host, port, type_, dirs, filename, mimetype in [ ("ftp://localhost/foo/bar/baz.html", "localhost", ftplib.FTP_PORT, "I", ["foo", "bar"], "baz.html", "text/html"), ("ftp://localhost:80/foo/bar/", "localhost", 80, "D", ["foo", "bar"], "", None), ("ftp://localhost/baz.gif;type=a", "localhost", ftplib.FTP_PORT, "A", [], "baz.gif", None), # XXX really this should guess image/gif ]: r = h.ftp_open(Request(url)) # ftp authentication not yet implemented by FTPHandler self.assert_(h.user == h.passwd == "") self.assertEqual(h.host, socket.gethostbyname(host)) self.assertEqual(h.port, port) self.assertEqual(h.dirs, dirs) self.assertEqual(h.ftpwrapper.filename, filename) self.assertEqual(h.ftpwrapper.filetype, type_) headers = r.info() self.assertEqual(headers.get("Content-type"), mimetype) self.assertEqual(int(headers["Content-length"]), len(data)) def test_file(self): import time, rfc822, socket h = urllib2.FileHandler() o = h.parent = MockOpener() TESTFN = test_support.TESTFN urlpath = sanepathname2url(os.path.abspath(TESTFN)) towrite = "hello, world\n" for url in [ "file://localhost%s" % urlpath, "file://%s" % urlpath, "file://%s%s" % (socket.gethostbyname('localhost'), urlpath), "file://%s%s" % (socket.gethostbyname(socket.gethostname()), urlpath), ]: f = open(TESTFN, "wb") try: try: f.write(towrite) finally: f.close() r = h.file_open(Request(url)) try: data = r.read() headers = r.info() newurl = r.geturl() finally: r.close() stats = os.stat(TESTFN) modified = rfc822.formatdate(stats.st_mtime) finally: os.remove(TESTFN) self.assertEqual(data, towrite) self.assertEqual(headers["Content-type"], "text/plain") self.assertEqual(headers["Content-length"], "13") self.assertEqual(headers["Last-modified"], modified) for url in [ "file://localhost:80%s" % urlpath, # XXXX bug: these fail with socket.gaierror, should be URLError ## "file://%s:80%s/%s" % (socket.gethostbyname('localhost'), ## os.getcwd(), TESTFN), ## "file://somerandomhost.ontheinternet.com%s/%s" % ## (os.getcwd(), TESTFN), ]: try: f = open(TESTFN, "wb") try: f.write(towrite) finally: f.close() self.assertRaises(urllib2.URLError, h.file_open, Request(url)) finally: os.remove(TESTFN) h = urllib2.FileHandler() o = h.parent = MockOpener() # XXXX why does // mean ftp (and /// mean not ftp!), and where # is file: scheme specified? I think this is really a bug, and # what was intended was to distinguish between URLs like: # file:/blah.txt (a file) # file://localhost/blah.txt (a file) # file:///blah.txt (a file) # file://ftp.example.com/blah.txt (an ftp URL) for url, ftp in [ ("file://ftp.example.com//foo.txt", True), ("file://ftp.example.com///foo.txt", False), # XXXX bug: fails with OSError, should be URLError ("file://ftp.example.com/foo.txt", False), ]: req = Request(url) try: h.file_open(req) # XXXX remove OSError when bug fixed except (urllib2.URLError, OSError): self.assert_(not ftp) else: self.assert_(o.req is req) self.assertEqual(req.type, "ftp") def test_http(self): class MockHTTPResponse: def __init__(self, fp, msg, status, reason): self.fp = fp self.msg = msg self.status = status self.reason = reason def read(self): return '' class MockHTTPClass: def __init__(self): self.req_headers = [] self.data = None self.raise_on_endheaders = False def __call__(self, host): self.host = host return self def set_debuglevel(self, level): self.level = level def request(self, method, url, body=None, headers={}): self.method = method self.selector = url self.req_headers += headers.items() if body: self.data = body if self.raise_on_endheaders: import socket raise socket.error() def getresponse(self): return MockHTTPResponse(MockFile(), {}, 200, "OK") h = urllib2.AbstractHTTPHandler() o = h.parent = MockOpener() url = "http://example.com/" for method, data in [("GET", None), ("POST", "blah")]: req = Request(url, data, {"Foo": "bar"}) req.add_unredirected_header("Spam", "eggs") http = MockHTTPClass() r = h.do_open(http, req) # result attributes r.read; r.readline # wrapped MockFile methods r.info; r.geturl # addinfourl methods r.code, r.msg == 200, "OK" # added from MockHTTPClass.getreply() hdrs = r.info() hdrs.get; hdrs.has_key # r.info() gives dict from .getreply() self.assertEqual(r.geturl(), url) self.assertEqual(http.host, "example.com") self.assertEqual(http.level, 0) self.assertEqual(http.method, method) self.assertEqual(http.selector, "/") self.assertEqual(http.req_headers, [("Connection", "close"), ("Foo", "bar"), ("Spam", "eggs")]) self.assertEqual(http.data, data) # check socket.error converted to URLError http.raise_on_endheaders = True self.assertRaises(urllib2.URLError, h.do_open, http, req) # check adding of standard headers o.addheaders = [("Spam", "eggs")] for data in "", None: # POST, GET req = Request("http://example.com/", data) r = MockResponse(200, "OK", {}, "") newreq = h.do_request_(req) if data is None: # GET self.assert_("Content-length" not in req.unredirected_hdrs) self.assert_("Content-type" not in req.unredirected_hdrs) else: # POST self.assertEqual(req.unredirected_hdrs["Content-length"], "0") self.assertEqual(req.unredirected_hdrs["Content-type"], "application/x-www-form-urlencoded") # XXX the details of Host could be better tested self.assertEqual(req.unredirected_hdrs["Host"], "example.com") self.assertEqual(req.unredirected_hdrs["Spam"], "eggs") # don't clobber existing headers req.add_unredirected_header("Content-length", "foo") req.add_unredirected_header("Content-type", "bar") req.add_unredirected_header("Host", "baz") req.add_unredirected_header("Spam", "foo") newreq = h.do_request_(req) self.assertEqual(req.unredirected_hdrs["Content-length"], "foo") self.assertEqual(req.unredirected_hdrs["Content-type"], "bar") self.assertEqual(req.unredirected_hdrs["Host"], "baz") self.assertEqual(req.unredirected_hdrs["Spam"], "foo") def test_errors(self): h = urllib2.HTTPErrorProcessor() o = h.parent = MockOpener() url = "http://example.com/" req = Request(url) # 200 OK is passed through r = MockResponse(200, "OK", {}, "", url) newr = h.http_response(req, r) self.assert_(r is newr) self.assert_(not hasattr(o, "proto")) # o.error not called # anything else calls o.error (and MockOpener returns None, here) r = MockResponse(201, "Created", {}, "", url) self.assert_(h.http_response(req, r) is None) self.assertEqual(o.proto, "http") # o.error called self.assertEqual(o.args, (req, r, 201, "Created", {})) def test_cookies(self): cj = MockCookieJar() h = urllib2.HTTPCookieProcessor(cj) o = h.parent = MockOpener() req = Request("http://example.com/") r = MockResponse(200, "OK", {}, "") newreq = h.http_request(req) self.assert_(cj.ach_req is req is newreq) self.assertEquals(req.get_origin_req_host(), "example.com") self.assert_(not req.is_unverifiable()) newr = h.http_response(req, r) self.assert_(cj.ec_req is req) self.assert_(cj.ec_r is r is newr) def test_redirect(self): from_url = "http://example.com/a.html" to_url = "http://example.com/b.html" h = urllib2.HTTPRedirectHandler() o = h.parent = MockOpener() # ordinary redirect behaviour for code in 301, 302, 303, 307: for data in None, "blah\nblah\n": method = getattr(h, "http_error_%s" % code) req = Request(from_url, data) req.add_header("Nonsense", "viking=withhold") req.add_unredirected_header("Spam", "spam") try: method(req, MockFile(), code, "Blah", MockHeaders({"location": to_url})) except urllib2.HTTPError: # 307 in response to POST requires user OK self.assert_(code == 307 and data is not None) self.assertEqual(o.req.get_full_url(), to_url) try: self.assertEqual(o.req.get_method(), "GET") except AttributeError: self.assert_(not o.req.has_data()) self.assertEqual(o.req.headers["Nonsense"], "viking=withhold") self.assert_("Spam" not in o.req.headers) self.assert_("Spam" not in o.req.unredirected_hdrs) # loop detection req = Request(from_url) def redirect(h, req, url=to_url): h.http_error_302(req, MockFile(), 302, "Blah", MockHeaders({"location": url})) # Note that the *original* request shares the same record of # redirections with the sub-requests caused by the redirections. # detect infinite loop redirect of a URL to itself req = Request(from_url, origin_req_host="example.com") count = 0 try: while 1: redirect(h, req, "http://example.com/") count = count + 1 except urllib2.HTTPError: # don't stop until max_repeats, because cookies may introduce state self.assertEqual(count, urllib2.HTTPRedirectHandler.max_repeats) # detect endless non-repeating chain of redirects req = Request(from_url, origin_req_host="example.com") count = 0 try: while 1: redirect(h, req, "http://example.com/%d" % count) count = count + 1 except urllib2.HTTPError: self.assertEqual(count, urllib2.HTTPRedirectHandler.max_redirections) def test_cookie_redirect(self): class MockHTTPHandler(urllib2.HTTPHandler): def __init__(self): self._count = 0 def http_open(self, req): import mimetools from StringIO import StringIO if self._count == 0: self._count = self._count + 1 msg = mimetools.Message( StringIO("Location: http://www.cracker.com/\r\n\r\n")) return self.parent.error( "http", req, MockFile(), 302, "Found", msg) else: self.req = req msg = mimetools.Message(StringIO("\r\n\r\n")) return MockResponse(200, "OK", msg, "", req.get_full_url()) # cookies shouldn't leak into redirected requests from cookielib import CookieJar from urllib2 import build_opener, HTTPHandler, HTTPError, \ HTTPCookieProcessor from test_cookielib import interact_netscape cj = CookieJar() interact_netscape(cj, "http://www.example.com/", "spam=eggs") hh = MockHTTPHandler() cp = HTTPCookieProcessor(cj) o = build_opener(hh, cp) o.open("http://www.example.com/") self.assert_(not hh.req.has_header("Cookie")) class MiscTests(unittest.TestCase): def test_build_opener(self): class MyHTTPHandler(urllib2.HTTPHandler): pass class FooHandler(urllib2.BaseHandler): def foo_open(self): pass class BarHandler(urllib2.BaseHandler): def bar_open(self): pass build_opener = urllib2.build_opener o = build_opener(FooHandler, BarHandler) self.opener_has_handler(o, FooHandler) self.opener_has_handler(o, BarHandler) # can take a mix of classes and instances o = build_opener(FooHandler, BarHandler()) self.opener_has_handler(o, FooHandler) self.opener_has_handler(o, BarHandler) # subclasses of default handlers override default handlers o = build_opener(MyHTTPHandler) self.opener_has_handler(o, MyHTTPHandler) # a particular case of overriding: default handlers can be passed # in explicitly o = build_opener() self.opener_has_handler(o, urllib2.HTTPHandler) o = build_opener(urllib2.HTTPHandler) self.opener_has_handler(o, urllib2.HTTPHandler) o = build_opener(urllib2.HTTPHandler()) self.opener_has_handler(o, urllib2.HTTPHandler) def opener_has_handler(self, opener, handler_class): for h in opener.handlers: if h.__class__ == handler_class: break else: self.assert_(False) class NetworkTests(unittest.TestCase): def setUp(self): if 0: # for debugging import logging logger = logging.getLogger("test_urllib2") logger.addHandler(logging.StreamHandler()) def test_range (self): req = urllib2.Request("http://www.python.org", headers={'Range': 'bytes=20-39'}) result = urllib2.urlopen(req) data = result.read() self.assertEqual(len(data), 20) # XXX The rest of these tests aren't very good -- they don't check much. # They do sometimes catch some major disasters, though. def test_ftp(self): urls = [ 'ftp://www.python.org/pub/python/misc/sousa.au', 'ftp://www.python.org/pub/tmp/blat', 'ftp://gatekeeper.research.compaq.com/pub/DEC/SRC' '/research-reports/00README-Legal-Rules-Regs', ] self._test_urls(urls, self._extra_handlers()) def test_gopher(self): urls = [ # Thanks to Fred for finding these! 'gopher://gopher.lib.ncsu.edu/11/library/stacks/Alex', 'gopher://gopher.vt.edu:10010/10/33', ] self._test_urls(urls, self._extra_handlers()) def test_file(self): TESTFN = test_support.TESTFN f = open(TESTFN, 'w') try: f.write('hi there\n') f.close() urls = [ 'file:'+sanepathname2url(os.path.abspath(TESTFN)), # XXX bug, should raise URLError #('file://nonsensename/etc/passwd', None, urllib2.URLError) ('file://nonsensename/etc/passwd', None, (OSError, socket.error)) ] self._test_urls(urls, self._extra_handlers()) finally: os.remove(TESTFN) def test_http(self): urls = [ 'http://www.espn.com/', # redirect 'http://www.python.org/Spanish/Inquistion/', ('http://www.python.org/cgi-bin/faqw.py', 'query=pythonistas&querytype=simple&casefold=yes&req=search', None), 'http://www.python.org/', ] self._test_urls(urls, self._extra_handlers()) # XXX Following test depends on machine configurations that are internal # to CNRI. Need to set up a public server with the right authentication # configuration for test purposes. ## def test_cnri(self): ## if socket.gethostname() == 'bitdiddle': ## localhost = 'bitdiddle.cnri.reston.va.us' ## elif socket.gethostname() == 'bitdiddle.concentric.net': ## localhost = 'localhost' ## else: ## localhost = None ## if localhost is not None: ## urls = [ ## 'file://%s/etc/passwd' % localhost, ## 'http://%s/simple/' % localhost, ## 'http://%s/digest/' % localhost, ## 'http://%s/not/found.h' % localhost, ## ] ## bauth = HTTPBasicAuthHandler() ## bauth.add_password('basic_test_realm', localhost, 'jhylton', ## 'password') ## dauth = HTTPDigestAuthHandler() ## dauth.add_password('digest_test_realm', localhost, 'jhylton', ## 'password') ## self._test_urls(urls, self._extra_handlers()+[bauth, dauth]) def _test_urls(self, urls, handlers): import socket import time import logging debug = logging.getLogger("test_urllib2").debug urllib2.install_opener(urllib2.build_opener(*handlers)) for url in urls: if isinstance(url, tuple): url, req, expected_err = url else: req = expected_err = None debug(url) try: f = urllib2.urlopen(url, req) except (IOError, socket.error, OSError), err: debug(err) if expected_err: self.assert_(isinstance(err, expected_err)) else: buf = f.read() f.close() debug("read %d bytes" % len(buf)) debug("******** next url coming up...") time.sleep(0.1) def _extra_handlers(self): handlers = [] handlers.append(urllib2.GopherHandler) cfh = urllib2.CacheFTPHandler() cfh.setTimeout(1) handlers.append(cfh) ## # XXX try out some custom proxy objects too! ## def at_cnri(req): ## host = req.get_host() ## debug(host) ## if host[-18:] == '.cnri.reston.va.us': ## return True ## p = CustomProxy('http', at_cnri, 'proxy.cnri.reston.va.us') ## ph = CustomProxyHandler(p) ## handlers.append(ph) return handlers def test_main(verbose=None): tests = (TrivialTests, OpenerDirectorTests, HandlerTests, MiscTests) if test_support.is_resource_enabled('network'): tests += (NetworkTests,) test_support.run_unittest(*tests) if __name__ == "__main__": test_main(verbose=True)