mylar/lib/cherrypy/test/test_encoding.py

364 lines
16 KiB
Python
Executable File

import gzip
import sys
import cherrypy
from cherrypy._cpcompat import BytesIO, IncompleteRead, ntob, ntou
europoundUnicode = ntou('\x80\xa3')
sing = u"\u6bdb\u6cfd\u4e1c: Sing, Little Birdie?"
sing8 = sing.encode('utf-8')
sing16 = sing.encode('utf-16')
from cherrypy.test import helper
class EncodingTests(helper.CPWebCase):
def setup_server():
class Root:
def index(self, param):
assert param == europoundUnicode, "%r != %r" % (param, europoundUnicode)
yield europoundUnicode
index.exposed = True
def mao_zedong(self):
return sing
mao_zedong.exposed = True
def utf8(self):
return sing8
utf8.exposed = True
utf8._cp_config = {'tools.encode.encoding': 'utf-8'}
def cookies_and_headers(self):
# if the headers have non-ascii characters and a cookie has
# any part which is unicode (even ascii), the response
# should not fail.
cherrypy.response.cookie['candy'] = 'bar'
cherrypy.response.cookie['candy']['domain'] = 'cherrypy.org'
cherrypy.response.headers['Some-Header'] = 'My d\xc3\xb6g has fleas'
return 'Any content'
cookies_and_headers.exposed = True
def reqparams(self, *args, **kwargs):
return ntob(', ').join([": ".join((k, v)).encode('utf8')
for k, v in cherrypy.request.params.items()])
reqparams.exposed = True
def nontext(self, *args, **kwargs):
cherrypy.response.headers['Content-Type'] = 'application/binary'
return '\x00\x01\x02\x03'
nontext.exposed = True
nontext._cp_config = {'tools.encode.text_only': False,
'tools.encode.add_charset': True,
}
class GZIP:
def index(self):
yield "Hello, world"
index.exposed = True
def noshow(self):
# Test for ticket #147, where yield showed no exceptions (content-
# encoding was still gzip even though traceback wasn't zipped).
raise IndexError()
yield "Here be dragons"
noshow.exposed = True
# Turn encoding off so the gzip tool is the one doing the collapse.
noshow._cp_config = {'tools.encode.on': False}
def noshow_stream(self):
# Test for ticket #147, where yield showed no exceptions (content-
# encoding was still gzip even though traceback wasn't zipped).
raise IndexError()
yield "Here be dragons"
noshow_stream.exposed = True
noshow_stream._cp_config = {'response.stream': True}
class Decode:
def extra_charset(self, *args, **kwargs):
return ', '.join([": ".join((k, v))
for k, v in cherrypy.request.params.items()])
extra_charset.exposed = True
extra_charset._cp_config = {
'tools.decode.on': True,
'tools.decode.default_encoding': ['utf-16'],
}
def force_charset(self, *args, **kwargs):
return ', '.join([": ".join((k, v))
for k, v in cherrypy.request.params.items()])
force_charset.exposed = True
force_charset._cp_config = {
'tools.decode.on': True,
'tools.decode.encoding': 'utf-16',
}
root = Root()
root.gzip = GZIP()
root.decode = Decode()
cherrypy.tree.mount(root, config={'/gzip': {'tools.gzip.on': True}})
setup_server = staticmethod(setup_server)
def test_query_string_decoding(self):
europoundUtf8 = europoundUnicode.encode('utf-8')
self.getPage(ntob('/?param=') + europoundUtf8)
self.assertBody(europoundUtf8)
# Encoded utf8 query strings MUST be parsed correctly.
# Here, q is the POUND SIGN U+00A3 encoded in utf8 and then %HEX
self.getPage("/reqparams?q=%C2%A3")
# The return value will be encoded as utf8.
self.assertBody(ntob("q: \xc2\xa3"))
# Query strings that are incorrectly encoded MUST raise 404.
# Here, q is the POUND SIGN U+00A3 encoded in latin1 and then %HEX
self.getPage("/reqparams?q=%A3")
self.assertStatus(404)
self.assertErrorPage(404,
"The given query string could not be processed. Query "
"strings for this resource must be encoded with 'utf8'.")
def test_urlencoded_decoding(self):
# Test the decoding of an application/x-www-form-urlencoded entity.
europoundUtf8 = europoundUnicode.encode('utf-8')
body=ntob("param=") + europoundUtf8
self.getPage('/', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(europoundUtf8)
# Encoded utf8 entities MUST be parsed and decoded correctly.
# Here, q is the POUND SIGN U+00A3 encoded in utf8
body = ntob("q=\xc2\xa3")
self.getPage('/reqparams', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(ntob("q: \xc2\xa3"))
# ...and in utf16, which is not in the default attempt_charsets list:
body = ntob("\xff\xfeq\x00=\xff\xfe\xa3\x00")
self.getPage('/reqparams', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded;charset=utf-16"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(ntob("q: \xc2\xa3"))
# Entities that are incorrectly encoded MUST raise 400.
# Here, q is the POUND SIGN U+00A3 encoded in utf16, but
# the Content-Type incorrectly labels it utf-8.
body = ntob("\xff\xfeq\x00=\xff\xfe\xa3\x00")
self.getPage('/reqparams', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded;charset=utf-8"),
("Content-Length", str(len(body))),
],
body=body),
self.assertStatus(400)
self.assertErrorPage(400,
"The request entity could not be decoded. The following charsets "
"were attempted: ['utf-8']")
def test_decode_tool(self):
# An extra charset should be tried first, and succeed if it matches.
# Here, we add utf-16 as a charset and pass a utf-16 body.
body = ntob("\xff\xfeq\x00=\xff\xfe\xa3\x00")
self.getPage('/decode/extra_charset', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(ntob("q: \xc2\xa3"))
# An extra charset should be tried first, and continue to other default
# charsets if it doesn't match.
# Here, we add utf-16 as a charset but still pass a utf-8 body.
body = ntob("q=\xc2\xa3")
self.getPage('/decode/extra_charset', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(ntob("q: \xc2\xa3"))
# An extra charset should error if force is True and it doesn't match.
# Here, we force utf-16 as a charset but still pass a utf-8 body.
body = ntob("q=\xc2\xa3")
self.getPage('/decode/force_charset', method='POST',
headers=[("Content-Type", "application/x-www-form-urlencoded"),
("Content-Length", str(len(body))),
],
body=body),
self.assertErrorPage(400,
"The request entity could not be decoded. The following charsets "
"were attempted: ['utf-16']")
def test_multipart_decoding(self):
# Test the decoding of a multipart entity when the charset (utf16) is
# explicitly given.
body=ntob('\r\n'.join(['--X',
'Content-Type: text/plain;charset=utf-16',
'Content-Disposition: form-data; name="text"',
'',
'\xff\xfea\x00b\x00\x1c c\x00',
'--X',
'Content-Type: text/plain;charset=utf-16',
'Content-Disposition: form-data; name="submit"',
'',
'\xff\xfeC\x00r\x00e\x00a\x00t\x00e\x00',
'--X--']))
self.getPage('/reqparams', method='POST',
headers=[("Content-Type", "multipart/form-data;boundary=X"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(ntob("text: ab\xe2\x80\x9cc, submit: Create"))
def test_multipart_decoding_no_charset(self):
# Test the decoding of a multipart entity when the charset (utf8) is
# NOT explicitly given, but is in the list of charsets to attempt.
body=ntob('\r\n'.join(['--X',
'Content-Disposition: form-data; name="text"',
'',
'\xe2\x80\x9c',
'--X',
'Content-Disposition: form-data; name="submit"',
'',
'Create',
'--X--']))
self.getPage('/reqparams', method='POST',
headers=[("Content-Type", "multipart/form-data;boundary=X"),
("Content-Length", str(len(body))),
],
body=body),
self.assertBody(ntob("text: \xe2\x80\x9c, submit: Create"))
def test_multipart_decoding_no_successful_charset(self):
# Test the decoding of a multipart entity when the charset (utf16) is
# NOT explicitly given, and is NOT in the list of charsets to attempt.
body=ntob('\r\n'.join(['--X',
'Content-Disposition: form-data; name="text"',
'',
'\xff\xfea\x00b\x00\x1c c\x00',
'--X',
'Content-Disposition: form-data; name="submit"',
'',
'\xff\xfeC\x00r\x00e\x00a\x00t\x00e\x00',
'--X--']))
self.getPage('/reqparams', method='POST',
headers=[("Content-Type", "multipart/form-data;boundary=X"),
("Content-Length", str(len(body))),
],
body=body),
self.assertStatus(400)
self.assertErrorPage(400,
"The request entity could not be decoded. The following charsets "
"were attempted: ['us-ascii', 'utf-8']")
def test_nontext(self):
self.getPage('/nontext')
self.assertHeader('Content-Type', 'application/binary;charset=utf-8')
self.assertBody('\x00\x01\x02\x03')
def testEncoding(self):
# Default encoding should be utf-8
self.getPage('/mao_zedong')
self.assertBody(sing8)
# Ask for utf-16.
self.getPage('/mao_zedong', [('Accept-Charset', 'utf-16')])
self.assertHeader('Content-Type', 'text/html;charset=utf-16')
self.assertBody(sing16)
# Ask for multiple encodings. ISO-8859-1 should fail, and utf-16
# should be produced.
self.getPage('/mao_zedong', [('Accept-Charset',
'iso-8859-1;q=1, utf-16;q=0.5')])
self.assertBody(sing16)
# The "*" value should default to our default_encoding, utf-8
self.getPage('/mao_zedong', [('Accept-Charset', '*;q=1, utf-7;q=.2')])
self.assertBody(sing8)
# Only allow iso-8859-1, which should fail and raise 406.
self.getPage('/mao_zedong', [('Accept-Charset', 'iso-8859-1, *;q=0')])
self.assertStatus("406 Not Acceptable")
self.assertInBody("Your client sent this Accept-Charset header: "
"iso-8859-1, *;q=0. We tried these charsets: "
"iso-8859-1.")
# Ask for x-mac-ce, which should be unknown. See ticket #569.
self.getPage('/mao_zedong', [('Accept-Charset',
'us-ascii, ISO-8859-1, x-mac-ce')])
self.assertStatus("406 Not Acceptable")
self.assertInBody("Your client sent this Accept-Charset header: "
"us-ascii, ISO-8859-1, x-mac-ce. We tried these "
"charsets: ISO-8859-1, us-ascii, x-mac-ce.")
# Test the 'encoding' arg to encode.
self.getPage('/utf8')
self.assertBody(sing8)
self.getPage('/utf8', [('Accept-Charset', 'us-ascii, ISO-8859-1')])
self.assertStatus("406 Not Acceptable")
def testGzip(self):
zbuf = BytesIO()
zfile = gzip.GzipFile(mode='wb', fileobj=zbuf, compresslevel=9)
zfile.write(ntob("Hello, world"))
zfile.close()
self.getPage('/gzip/', headers=[("Accept-Encoding", "gzip")])
self.assertInBody(zbuf.getvalue()[:3])
self.assertHeader("Vary", "Accept-Encoding")
self.assertHeader("Content-Encoding", "gzip")
# Test when gzip is denied.
self.getPage('/gzip/', headers=[("Accept-Encoding", "identity")])
self.assertHeader("Vary", "Accept-Encoding")
self.assertNoHeader("Content-Encoding")
self.assertBody("Hello, world")
self.getPage('/gzip/', headers=[("Accept-Encoding", "gzip;q=0")])
self.assertHeader("Vary", "Accept-Encoding")
self.assertNoHeader("Content-Encoding")
self.assertBody("Hello, world")
self.getPage('/gzip/', headers=[("Accept-Encoding", "*;q=0")])
self.assertStatus(406)
self.assertNoHeader("Content-Encoding")
self.assertErrorPage(406, "identity, gzip")
# Test for ticket #147
self.getPage('/gzip/noshow', headers=[("Accept-Encoding", "gzip")])
self.assertNoHeader('Content-Encoding')
self.assertStatus(500)
self.assertErrorPage(500, pattern="IndexError\n")
# In this case, there's nothing we can do to deliver a
# readable page, since 1) the gzip header is already set,
# and 2) we may have already written some of the body.
# The fix is to never stream yields when using gzip.
if (cherrypy.server.protocol_version == "HTTP/1.0" or
getattr(cherrypy.server, "using_apache", False)):
self.getPage('/gzip/noshow_stream',
headers=[("Accept-Encoding", "gzip")])
self.assertHeader('Content-Encoding', 'gzip')
self.assertInBody('\x1f\x8b\x08\x00')
else:
# The wsgiserver will simply stop sending data, and the HTTP client
# will error due to an incomplete chunk-encoded stream.
self.assertRaises((ValueError, IncompleteRead), self.getPage,
'/gzip/noshow_stream',
headers=[("Accept-Encoding", "gzip")])
def test_UnicodeHeaders(self):
self.getPage('/cookies_and_headers')
self.assertBody('Any content')