mirror of https://github.com/python/cpython
1516 lines
53 KiB
Python
1516 lines
53 KiB
Python
import datetime
|
||
import textwrap
|
||
import unittest
|
||
from email import errors
|
||
from email import policy
|
||
from email.message import Message
|
||
from test.test_email import TestEmailBase, parameterize
|
||
from email import headerregistry
|
||
from email.headerregistry import Address, Group
|
||
|
||
|
||
DITTO = object()
|
||
|
||
|
||
class TestHeaderRegistry(TestEmailBase):
|
||
|
||
def test_arbitrary_name_unstructured(self):
|
||
factory = headerregistry.HeaderRegistry()
|
||
h = factory('foobar', 'test')
|
||
self.assertIsInstance(h, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h, headerregistry.UnstructuredHeader)
|
||
|
||
def test_name_case_ignored(self):
|
||
factory = headerregistry.HeaderRegistry()
|
||
# Whitebox check that test is valid
|
||
self.assertNotIn('Subject', factory.registry)
|
||
h = factory('Subject', 'test')
|
||
self.assertIsInstance(h, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h, headerregistry.UniqueUnstructuredHeader)
|
||
|
||
class FooBase:
|
||
def __init__(self, *args, **kw):
|
||
pass
|
||
|
||
def test_override_default_base_class(self):
|
||
factory = headerregistry.HeaderRegistry(base_class=self.FooBase)
|
||
h = factory('foobar', 'test')
|
||
self.assertIsInstance(h, self.FooBase)
|
||
self.assertIsInstance(h, headerregistry.UnstructuredHeader)
|
||
|
||
class FooDefault:
|
||
parse = headerregistry.UnstructuredHeader.parse
|
||
|
||
def test_override_default_class(self):
|
||
factory = headerregistry.HeaderRegistry(default_class=self.FooDefault)
|
||
h = factory('foobar', 'test')
|
||
self.assertIsInstance(h, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h, self.FooDefault)
|
||
|
||
def test_override_default_class_only_overrides_default(self):
|
||
factory = headerregistry.HeaderRegistry(default_class=self.FooDefault)
|
||
h = factory('subject', 'test')
|
||
self.assertIsInstance(h, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h, headerregistry.UniqueUnstructuredHeader)
|
||
|
||
def test_dont_use_default_map(self):
|
||
factory = headerregistry.HeaderRegistry(use_default_map=False)
|
||
h = factory('subject', 'test')
|
||
self.assertIsInstance(h, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h, headerregistry.UnstructuredHeader)
|
||
|
||
def test_map_to_type(self):
|
||
factory = headerregistry.HeaderRegistry()
|
||
h1 = factory('foobar', 'test')
|
||
factory.map_to_type('foobar', headerregistry.UniqueUnstructuredHeader)
|
||
h2 = factory('foobar', 'test')
|
||
self.assertIsInstance(h1, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h1, headerregistry.UnstructuredHeader)
|
||
self.assertIsInstance(h2, headerregistry.BaseHeader)
|
||
self.assertIsInstance(h2, headerregistry.UniqueUnstructuredHeader)
|
||
|
||
|
||
class TestHeaderBase(TestEmailBase):
|
||
|
||
factory = headerregistry.HeaderRegistry()
|
||
|
||
def make_header(self, name, value):
|
||
return self.factory(name, value)
|
||
|
||
|
||
class TestBaseHeaderFeatures(TestHeaderBase):
|
||
|
||
def test_str(self):
|
||
h = self.make_header('subject', 'this is a test')
|
||
self.assertIsInstance(h, str)
|
||
self.assertEqual(h, 'this is a test')
|
||
self.assertEqual(str(h), 'this is a test')
|
||
|
||
def test_substr(self):
|
||
h = self.make_header('subject', 'this is a test')
|
||
self.assertEqual(h[5:7], 'is')
|
||
|
||
def test_has_name(self):
|
||
h = self.make_header('subject', 'this is a test')
|
||
self.assertEqual(h.name, 'subject')
|
||
|
||
def _test_attr_ro(self, attr):
|
||
h = self.make_header('subject', 'this is a test')
|
||
with self.assertRaises(AttributeError):
|
||
setattr(h, attr, 'foo')
|
||
|
||
def test_name_read_only(self):
|
||
self._test_attr_ro('name')
|
||
|
||
def test_defects_read_only(self):
|
||
self._test_attr_ro('defects')
|
||
|
||
def test_defects_is_tuple(self):
|
||
h = self.make_header('subject', 'this is a test')
|
||
self.assertEqual(len(h.defects), 0)
|
||
self.assertIsInstance(h.defects, tuple)
|
||
# Make sure it is still true when there are defects.
|
||
h = self.make_header('date', '')
|
||
self.assertEqual(len(h.defects), 1)
|
||
self.assertIsInstance(h.defects, tuple)
|
||
|
||
# XXX: FIXME
|
||
#def test_CR_in_value(self):
|
||
# # XXX: this also re-raises the issue of embedded headers,
|
||
# # need test and solution for that.
|
||
# value = '\r'.join(['this is', ' a test'])
|
||
# h = self.make_header('subject', value)
|
||
# self.assertEqual(h, value)
|
||
# self.assertDefectsEqual(h.defects, [errors.ObsoleteHeaderDefect])
|
||
|
||
def test_RFC2047_value_decoded(self):
|
||
value = '=?utf-8?q?this_is_a_test?='
|
||
h = self.make_header('subject', value)
|
||
self.assertEqual(h, 'this is a test')
|
||
|
||
|
||
class TestDateHeader(TestHeaderBase):
|
||
|
||
datestring = 'Sun, 23 Sep 2001 20:10:55 -0700'
|
||
utcoffset = datetime.timedelta(hours=-7)
|
||
tz = datetime.timezone(utcoffset)
|
||
dt = datetime.datetime(2001, 9, 23, 20, 10, 55, tzinfo=tz)
|
||
|
||
def test_parse_date(self):
|
||
h = self.make_header('date', self.datestring)
|
||
self.assertEqual(h, self.datestring)
|
||
self.assertEqual(h.datetime, self.dt)
|
||
self.assertEqual(h.datetime.utcoffset(), self.utcoffset)
|
||
self.assertEqual(h.defects, ())
|
||
|
||
def test_set_from_datetime(self):
|
||
h = self.make_header('date', self.dt)
|
||
self.assertEqual(h, self.datestring)
|
||
self.assertEqual(h.datetime, self.dt)
|
||
self.assertEqual(h.defects, ())
|
||
|
||
def test_date_header_properties(self):
|
||
h = self.make_header('date', self.datestring)
|
||
self.assertIsInstance(h, headerregistry.UniqueDateHeader)
|
||
self.assertEqual(h.max_count, 1)
|
||
self.assertEqual(h.defects, ())
|
||
|
||
def test_resent_date_header_properties(self):
|
||
h = self.make_header('resent-date', self.datestring)
|
||
self.assertIsInstance(h, headerregistry.DateHeader)
|
||
self.assertEqual(h.max_count, None)
|
||
self.assertEqual(h.defects, ())
|
||
|
||
def test_no_value_is_defect(self):
|
||
h = self.make_header('date', '')
|
||
self.assertEqual(len(h.defects), 1)
|
||
self.assertIsInstance(h.defects[0], errors.HeaderMissingRequiredValue)
|
||
|
||
def test_datetime_read_only(self):
|
||
h = self.make_header('date', self.datestring)
|
||
with self.assertRaises(AttributeError):
|
||
h.datetime = 'foo'
|
||
|
||
def test_set_date_header_from_datetime(self):
|
||
m = Message(policy=policy.default)
|
||
m['Date'] = self.dt
|
||
self.assertEqual(m['Date'], self.datestring)
|
||
self.assertEqual(m['Date'].datetime, self.dt)
|
||
|
||
|
||
@parameterize
|
||
class TestContentTypeHeader(TestHeaderBase):
|
||
|
||
def content_type_as_value(self,
|
||
source,
|
||
content_type,
|
||
maintype,
|
||
subtype,
|
||
*args):
|
||
l = len(args)
|
||
parmdict = args[0] if l>0 else {}
|
||
defects = args[1] if l>1 else []
|
||
decoded = args[2] if l>2 and args[2] is not DITTO else source
|
||
header = 'Content-Type:' + ' ' if source else ''
|
||
folded = args[3] if l>3 else header + source + '\n'
|
||
h = self.make_header('Content-Type', source)
|
||
self.assertEqual(h.content_type, content_type)
|
||
self.assertEqual(h.maintype, maintype)
|
||
self.assertEqual(h.subtype, subtype)
|
||
self.assertEqual(h.params, parmdict)
|
||
self.assertDefectsEqual(h.defects, defects)
|
||
self.assertEqual(h, decoded)
|
||
self.assertEqual(h.fold(policy=policy.default), folded)
|
||
|
||
content_type_params = {
|
||
|
||
# Examples from RFC 2045.
|
||
|
||
'RFC_2045_1': (
|
||
'text/plain; charset=us-ascii (Plain text)',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'us-ascii'},
|
||
[],
|
||
'text/plain; charset="us-ascii"'),
|
||
|
||
'RFC_2045_2': (
|
||
'text/plain; charset=us-ascii',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'us-ascii'},
|
||
[],
|
||
'text/plain; charset="us-ascii"'),
|
||
|
||
'RFC_2045_3': (
|
||
'text/plain; charset="us-ascii"',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'us-ascii'}),
|
||
|
||
# RFC 2045 5.2 says syntactically invalid values are to be treated as
|
||
# text/plain.
|
||
|
||
'no_subtype_in_content_type': (
|
||
'text/',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{},
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'no_slash_in_content_type': (
|
||
'foo',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{},
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'junk_text_in_content_type': (
|
||
'<crazy "stuff">',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{},
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'too_many_slashes_in_content_type': (
|
||
'image/jpeg/foo',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{},
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
# But unknown names are OK. We could make non-IANA names a defect, but
|
||
# by not doing so we make ourselves future proof. The fact that they
|
||
# are unknown will be detectable by the fact that they don't appear in
|
||
# the mime_registry...and the application is free to extend that list
|
||
# to handle them even if the core library doesn't.
|
||
|
||
'unknown_content_type': (
|
||
'bad/names',
|
||
'bad/names',
|
||
'bad',
|
||
'names'),
|
||
|
||
# The content type is case insensitive, and CFWS is ignored.
|
||
|
||
'mixed_case_content_type': (
|
||
'ImAge/JPeg',
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg'),
|
||
|
||
'spaces_in_content_type': (
|
||
' text / plain ',
|
||
'text/plain',
|
||
'text',
|
||
'plain'),
|
||
|
||
'cfws_in_content_type': (
|
||
'(foo) text (bar)/(baz)plain(stuff)',
|
||
'text/plain',
|
||
'text',
|
||
'plain'),
|
||
|
||
# test some parameters (more tests could be added for parameters
|
||
# associated with other content types, but since parameter parsing is
|
||
# generic they would be redundant for the current implementation).
|
||
|
||
'charset_param': (
|
||
'text/plain; charset="utf-8"',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'utf-8'}),
|
||
|
||
'capitalized_charset': (
|
||
'text/plain; charset="US-ASCII"',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'US-ASCII'}),
|
||
|
||
'unknown_charset': (
|
||
'text/plain; charset="fOo"',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'fOo'}),
|
||
|
||
'capitalized_charset_param_name_and_comment': (
|
||
'text/plain; (interjection) Charset="utf-8"',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'utf-8'},
|
||
[],
|
||
# Should the parameter name be lowercased here?
|
||
'text/plain; Charset="utf-8"'),
|
||
|
||
# Since this is pretty much the ur-mimeheader, we'll put all the tests
|
||
# that exercise the parameter parsing and formatting here.
|
||
#
|
||
# XXX: question: is minimal quoting preferred?
|
||
|
||
'unquoted_param_value': (
|
||
'text/plain; title=foo',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'title': 'foo'},
|
||
[],
|
||
'text/plain; title="foo"'),
|
||
|
||
'param_value_with_tspecials': (
|
||
'text/plain; title="(bar)foo blue"',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'title': '(bar)foo blue'}),
|
||
|
||
'param_with_extra_quoted_whitespace': (
|
||
'text/plain; title=" a loong way \t home "',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'title': ' a loong way \t home '}),
|
||
|
||
'bad_params': (
|
||
'blarg; baz; boo',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'baz': '', 'boo': ''},
|
||
[errors.InvalidHeaderDefect]*3),
|
||
|
||
'spaces_around_param_equals': (
|
||
'Multipart/mixed; boundary = "CPIMSSMTPC06p5f3tG"',
|
||
'multipart/mixed',
|
||
'multipart',
|
||
'mixed',
|
||
{'boundary': 'CPIMSSMTPC06p5f3tG'},
|
||
[],
|
||
'Multipart/mixed; boundary="CPIMSSMTPC06p5f3tG"'),
|
||
|
||
'spaces_around_semis': (
|
||
('image/jpeg; name="wibble.JPG" ; x-mac-type="4A504547" ; '
|
||
'x-mac-creator="474B4F4E"'),
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg',
|
||
{'name': 'wibble.JPG',
|
||
'x-mac-type': '4A504547',
|
||
'x-mac-creator': '474B4F4E'},
|
||
[],
|
||
('image/jpeg; name="wibble.JPG"; x-mac-type="4A504547"; '
|
||
'x-mac-creator="474B4F4E"'),
|
||
# XXX: it could be that we will eventually prefer to fold starting
|
||
# from the decoded value, in which case these spaces and similar
|
||
# spaces in other tests will be wrong.
|
||
('Content-Type: image/jpeg; name="wibble.JPG" ; '
|
||
'x-mac-type="4A504547" ;\n'
|
||
' x-mac-creator="474B4F4E"\n'),
|
||
),
|
||
|
||
'semis_inside_quotes': (
|
||
'image/jpeg; name="Jim&&Jill"',
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg',
|
||
{'name': 'Jim&&Jill'}),
|
||
|
||
'single_quotes_inside_quotes': (
|
||
'image/jpeg; name="Jim \'Bob\' Jill"',
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg',
|
||
{'name': "Jim 'Bob' Jill"}),
|
||
|
||
'double_quotes_inside_quotes': (
|
||
r'image/jpeg; name="Jim \"Bob\" Jill"',
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg',
|
||
{'name': 'Jim "Bob" Jill'},
|
||
[],
|
||
r'image/jpeg; name="Jim \"Bob\" Jill"'),
|
||
|
||
# XXX: This test works except for the refolding of the header. I'll
|
||
# deal with that bug when I deal with the other folding bugs.
|
||
#'non_ascii_in_params': (
|
||
# ('foo\xa7/bar; b\xa7r=two; '
|
||
# 'baz=thr\xa7e'.encode('latin-1').decode('us-ascii',
|
||
# 'surrogateescape')),
|
||
# 'foo\uFFFD/bar',
|
||
# 'foo\uFFFD',
|
||
# 'bar',
|
||
# {'b\uFFFDr': 'two', 'baz': 'thr\uFFFDe'},
|
||
# [errors.UndecodableBytesDefect]*3,
|
||
# 'foo<6F>/bar; b<>r="two"; baz="thr<68>e"',
|
||
# ),
|
||
|
||
# RFC 2231 parameter tests.
|
||
|
||
'rfc2231_segmented_normal_values': (
|
||
'image/jpeg; name*0="abc"; name*1=".html"',
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg',
|
||
{'name': "abc.html"},
|
||
[],
|
||
'image/jpeg; name="abc.html"'),
|
||
|
||
'quotes_inside_rfc2231_value': (
|
||
r'image/jpeg; bar*0="baz\"foobar"; bar*1="\"baz"',
|
||
'image/jpeg',
|
||
'image',
|
||
'jpeg',
|
||
{'bar': 'baz"foobar"baz'},
|
||
[],
|
||
r'image/jpeg; bar="baz\"foobar\"baz"'),
|
||
|
||
# XXX: This test works except for the refolding of the header. I'll
|
||
# deal with that bug when I deal with the other folding bugs.
|
||
#'non_ascii_rfc2231_value': (
|
||
# ('text/plain; charset=us-ascii; '
|
||
# "title*=us-ascii'en'This%20is%20"
|
||
# 'not%20f\xa7n').encode('latin-1').decode('us-ascii',
|
||
# 'surrogateescape'),
|
||
# 'text/plain',
|
||
# 'text',
|
||
# 'plain',
|
||
# {'charset': 'us-ascii', 'title': 'This is not f\uFFFDn'},
|
||
# [errors.UndecodableBytesDefect],
|
||
# 'text/plain; charset="us-ascii"; title="This is not f<>n"'),
|
||
|
||
'rfc2231_encoded_charset': (
|
||
'text/plain; charset*=ansi-x3.4-1968\'\'us-ascii',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'us-ascii'},
|
||
[],
|
||
'text/plain; charset="us-ascii"'),
|
||
|
||
# This follows the RFC: no double quotes around encoded values.
|
||
'rfc2231_encoded_no_double_quotes': (
|
||
("text/plain;"
|
||
"\tname*0*=''This%20is%20;"
|
||
"\tname*1*=%2A%2A%2Afun%2A%2A%2A%20;"
|
||
'\tname*2="is it not.pdf"'),
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'This is ***fun*** is it not.pdf'},
|
||
[],
|
||
'text/plain; name="This is ***fun*** is it not.pdf"',
|
||
('Content-Type: text/plain;\tname*0*=\'\'This%20is%20;\n'
|
||
'\tname*1*=%2A%2A%2Afun%2A%2A%2A%20;\tname*2="is it not.pdf"\n'),
|
||
),
|
||
|
||
# Make sure we also handle it if there are spurrious double qoutes.
|
||
'rfc2231_encoded_with_double_quotes': (
|
||
("text/plain;"
|
||
'\tname*0*="us-ascii\'\'This%20is%20even%20more%20";'
|
||
'\tname*1*="%2A%2A%2Afun%2A%2A%2A%20";'
|
||
'\tname*2="is it not.pdf"'),
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'This is even more ***fun*** is it not.pdf'},
|
||
[errors.InvalidHeaderDefect]*2,
|
||
'text/plain; name="This is even more ***fun*** is it not.pdf"',
|
||
('Content-Type: text/plain;\t'
|
||
'name*0*="us-ascii\'\'This%20is%20even%20more%20";\n'
|
||
'\tname*1*="%2A%2A%2Afun%2A%2A%2A%20";\tname*2="is it not.pdf"\n'),
|
||
),
|
||
|
||
'rfc2231_single_quote_inside_double_quotes': (
|
||
('text/plain; charset=us-ascii;'
|
||
'\ttitle*0*="us-ascii\'en\'This%20is%20really%20";'
|
||
'\ttitle*1*="%2A%2A%2Afun%2A%2A%2A%20";'
|
||
'\ttitle*2="isn\'t it!"'),
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'us-ascii', 'title': "This is really ***fun*** isn't it!"},
|
||
[errors.InvalidHeaderDefect]*2,
|
||
('text/plain; charset="us-ascii"; '
|
||
'title="This is really ***fun*** isn\'t it!"'),
|
||
('Content-Type: text/plain; charset=us-ascii;\n'
|
||
'\ttitle*0*="us-ascii\'en\'This%20is%20really%20";\n'
|
||
'\ttitle*1*="%2A%2A%2Afun%2A%2A%2A%20";\ttitle*2="isn\'t it!"\n'),
|
||
),
|
||
|
||
'rfc2231_single_quote_in_value_with_charset_and_lang': (
|
||
('application/x-foo;'
|
||
"\tname*0*=\"us-ascii'en-us'Frank's\"; name*1*=\" Document\""),
|
||
'application/x-foo',
|
||
'application',
|
||
'x-foo',
|
||
{'name': "Frank's Document"},
|
||
[errors.InvalidHeaderDefect]*2,
|
||
'application/x-foo; name="Frank\'s Document"',
|
||
('Content-Type: application/x-foo;\t'
|
||
'name*0*="us-ascii\'en-us\'Frank\'s";\n'
|
||
' name*1*=" Document"\n'),
|
||
),
|
||
|
||
'rfc2231_single_quote_in_non_encoded_value': (
|
||
('application/x-foo;'
|
||
"\tname*0=\"us-ascii'en-us'Frank's\"; name*1=\" Document\""),
|
||
'application/x-foo',
|
||
'application',
|
||
'x-foo',
|
||
{'name': "us-ascii'en-us'Frank's Document"},
|
||
[],
|
||
'application/x-foo; name="us-ascii\'en-us\'Frank\'s Document"',
|
||
('Content-Type: application/x-foo;\t'
|
||
'name*0="us-ascii\'en-us\'Frank\'s";\n'
|
||
' name*1=" Document"\n'),
|
||
),
|
||
|
||
'rfc2231_no_language_or_charset': (
|
||
'text/plain; NAME*0*=english_is_the_default.html',
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'english_is_the_default.html'},
|
||
[errors.InvalidHeaderDefect],
|
||
'text/plain; NAME="english_is_the_default.html"'),
|
||
|
||
'rfc2231_encoded_no_charset': (
|
||
("text/plain;"
|
||
'\tname*0*="\'\'This%20is%20even%20more%20";'
|
||
'\tname*1*="%2A%2A%2Afun%2A%2A%2A%20";'
|
||
'\tname*2="is it.pdf"'),
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'This is even more ***fun*** is it.pdf'},
|
||
[errors.InvalidHeaderDefect]*2,
|
||
'text/plain; name="This is even more ***fun*** is it.pdf"',
|
||
('Content-Type: text/plain;\t'
|
||
'name*0*="\'\'This%20is%20even%20more%20";\n'
|
||
'\tname*1*="%2A%2A%2Afun%2A%2A%2A%20";\tname*2="is it.pdf"\n'),
|
||
),
|
||
|
||
# XXX: see below...the first name line here should be *0 not *0*.
|
||
'rfc2231_partly_encoded': (
|
||
("text/plain;"
|
||
'\tname*0*="\'\'This%20is%20even%20more%20";'
|
||
'\tname*1*="%2A%2A%2Afun%2A%2A%2A%20";'
|
||
'\tname*2="is it.pdf"'),
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'This is even more ***fun*** is it.pdf'},
|
||
[errors.InvalidHeaderDefect]*2,
|
||
'text/plain; name="This is even more ***fun*** is it.pdf"',
|
||
('Content-Type: text/plain;\t'
|
||
'name*0*="\'\'This%20is%20even%20more%20";\n'
|
||
'\tname*1*="%2A%2A%2Afun%2A%2A%2A%20";\tname*2="is it.pdf"\n'),
|
||
),
|
||
|
||
'rfc2231_partly_encoded_2': (
|
||
("text/plain;"
|
||
'\tname*0*="\'\'This%20is%20even%20more%20";'
|
||
'\tname*1="%2A%2A%2Afun%2A%2A%2A%20";'
|
||
'\tname*2="is it.pdf"'),
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'This is even more %2A%2A%2Afun%2A%2A%2A%20is it.pdf'},
|
||
[errors.InvalidHeaderDefect],
|
||
'text/plain; name="This is even more %2A%2A%2Afun%2A%2A%2A%20is it.pdf"',
|
||
('Content-Type: text/plain;\t'
|
||
'name*0*="\'\'This%20is%20even%20more%20";\n'
|
||
'\tname*1="%2A%2A%2Afun%2A%2A%2A%20";\tname*2="is it.pdf"\n'),
|
||
),
|
||
|
||
'rfc2231_unknown_charset_treated_as_ascii': (
|
||
"text/plain; name*0*=bogus'xx'ascii_is_the_default",
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'name': 'ascii_is_the_default'},
|
||
[],
|
||
'text/plain; name="ascii_is_the_default"'),
|
||
|
||
'rfc2231_bad_character_in_charset_parameter_value': (
|
||
"text/plain; charset*=ascii''utf-8%E2%80%9D",
|
||
'text/plain',
|
||
'text',
|
||
'plain',
|
||
{'charset': 'utf-8\uFFFD\uFFFD\uFFFD'},
|
||
[errors.UndecodableBytesDefect],
|
||
'text/plain; charset="utf-8\uFFFD\uFFFD\uFFFD"'),
|
||
|
||
'rfc2231_encoded_then_unencoded_segments': (
|
||
('application/x-foo;'
|
||
'\tname*0*="us-ascii\'en-us\'My";'
|
||
'\tname*1=" Document";'
|
||
'\tname*2=" For You"'),
|
||
'application/x-foo',
|
||
'application',
|
||
'x-foo',
|
||
{'name': 'My Document For You'},
|
||
[errors.InvalidHeaderDefect],
|
||
'application/x-foo; name="My Document For You"',
|
||
('Content-Type: application/x-foo;\t'
|
||
'name*0*="us-ascii\'en-us\'My";\n'
|
||
'\tname*1=" Document";\tname*2=" For You"\n'),
|
||
),
|
||
|
||
# My reading of the RFC is that this is an invalid header. The RFC
|
||
# says that if charset and language information is given, the first
|
||
# segment *must* be encoded.
|
||
'rfc2231_unencoded_then_encoded_segments': (
|
||
('application/x-foo;'
|
||
'\tname*0=us-ascii\'en-us\'My;'
|
||
'\tname*1*=" Document";'
|
||
'\tname*2*=" For You"'),
|
||
'application/x-foo',
|
||
'application',
|
||
'x-foo',
|
||
{'name': 'My Document For You'},
|
||
[errors.InvalidHeaderDefect]*3,
|
||
'application/x-foo; name="My Document For You"',
|
||
("Content-Type: application/x-foo;\tname*0=us-ascii'en-us'My;\t"
|
||
# XXX: the newline is in the wrong place, come back and fix
|
||
# this when the rest of tests pass.
|
||
'name*1*=" Document"\n;'
|
||
'\tname*2*=" For You"\n'),
|
||
),
|
||
|
||
# XXX: I would say this one should default to ascii/en for the
|
||
# "encoded" segment, since the first segment is not encoded and is
|
||
# in double quotes, making the value a valid non-encoded string. The
|
||
# old parser decodes this just like the previous case, which may be the
|
||
# better Postel rule, but could equally result in borking headers that
|
||
# intentially have quoted quotes in them. We could get this 98% right
|
||
# if we treat it as a quoted string *unless* it matches the
|
||
# charset'lang'value pattern exactly *and* there is at least one
|
||
# encoded segment. Implementing that algorithm will require some
|
||
# refactoring, so I haven't done it (yet).
|
||
|
||
'rfc2231_qouted_unencoded_then_encoded_segments': (
|
||
('application/x-foo;'
|
||
'\tname*0="us-ascii\'en-us\'My";'
|
||
'\tname*1*=" Document";'
|
||
'\tname*2*=" For You"'),
|
||
'application/x-foo',
|
||
'application',
|
||
'x-foo',
|
||
{'name': "us-ascii'en-us'My Document For You"},
|
||
[errors.InvalidHeaderDefect]*2,
|
||
'application/x-foo; name="us-ascii\'en-us\'My Document For You"',
|
||
('Content-Type: application/x-foo;\t'
|
||
'name*0="us-ascii\'en-us\'My";\n'
|
||
'\tname*1*=" Document";\tname*2*=" For You"\n'),
|
||
),
|
||
|
||
}
|
||
|
||
|
||
@parameterize
|
||
class TestContentTransferEncoding(TestHeaderBase):
|
||
|
||
def cte_as_value(self,
|
||
source,
|
||
cte,
|
||
*args):
|
||
l = len(args)
|
||
defects = args[0] if l>0 else []
|
||
decoded = args[1] if l>1 and args[1] is not DITTO else source
|
||
header = 'Content-Transfer-Encoding:' + ' ' if source else ''
|
||
folded = args[2] if l>2 else header + source + '\n'
|
||
h = self.make_header('Content-Transfer-Encoding', source)
|
||
self.assertEqual(h.cte, cte)
|
||
self.assertDefectsEqual(h.defects, defects)
|
||
self.assertEqual(h, decoded)
|
||
self.assertEqual(h.fold(policy=policy.default), folded)
|
||
|
||
cte_params = {
|
||
|
||
'RFC_2183_1': (
|
||
'base64',
|
||
'base64',),
|
||
|
||
'no_value': (
|
||
'',
|
||
'7bit',
|
||
[errors.HeaderMissingRequiredValue],
|
||
'',
|
||
'Content-Transfer-Encoding:\n',
|
||
),
|
||
|
||
'junk_after_cte': (
|
||
'7bit and a bunch more',
|
||
'7bit',
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
}
|
||
|
||
|
||
@parameterize
|
||
class TestContentDisposition(TestHeaderBase):
|
||
|
||
def content_disp_as_value(self,
|
||
source,
|
||
content_disposition,
|
||
*args):
|
||
l = len(args)
|
||
parmdict = args[0] if l>0 else {}
|
||
defects = args[1] if l>1 else []
|
||
decoded = args[2] if l>2 and args[2] is not DITTO else source
|
||
header = 'Content-Disposition:' + ' ' if source else ''
|
||
folded = args[3] if l>3 else header + source + '\n'
|
||
h = self.make_header('Content-Disposition', source)
|
||
self.assertEqual(h.content_disposition, content_disposition)
|
||
self.assertEqual(h.params, parmdict)
|
||
self.assertDefectsEqual(h.defects, defects)
|
||
self.assertEqual(h, decoded)
|
||
self.assertEqual(h.fold(policy=policy.default), folded)
|
||
|
||
content_disp_params = {
|
||
|
||
# Examples from RFC 2183.
|
||
|
||
'RFC_2183_1': (
|
||
'inline',
|
||
'inline',),
|
||
|
||
'RFC_2183_2': (
|
||
('attachment; filename=genome.jpeg;'
|
||
' modification-date="Wed, 12 Feb 1997 16:29:51 -0500";'),
|
||
'attachment',
|
||
{'filename': 'genome.jpeg',
|
||
'modification-date': 'Wed, 12 Feb 1997 16:29:51 -0500'},
|
||
[],
|
||
('attachment; filename="genome.jpeg"; '
|
||
'modification-date="Wed, 12 Feb 1997 16:29:51 -0500"'),
|
||
('Content-Disposition: attachment; filename=genome.jpeg;\n'
|
||
' modification-date="Wed, 12 Feb 1997 16:29:51 -0500";\n'),
|
||
),
|
||
|
||
'no_value': (
|
||
'',
|
||
None,
|
||
{},
|
||
[errors.HeaderMissingRequiredValue],
|
||
'',
|
||
'Content-Disposition:\n'),
|
||
|
||
'invalid_value': (
|
||
'ab./k',
|
||
'ab.',
|
||
{},
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'invalid_value_with_params': (
|
||
'ab./k; filename="foo"',
|
||
'ab.',
|
||
{'filename': 'foo'},
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
}
|
||
|
||
|
||
@parameterize
|
||
class TestMIMEVersionHeader(TestHeaderBase):
|
||
|
||
def version_string_as_MIME_Version(self,
|
||
source,
|
||
decoded,
|
||
version,
|
||
major,
|
||
minor,
|
||
defects):
|
||
h = self.make_header('MIME-Version', source)
|
||
self.assertEqual(h, decoded)
|
||
self.assertEqual(h.version, version)
|
||
self.assertEqual(h.major, major)
|
||
self.assertEqual(h.minor, minor)
|
||
self.assertDefectsEqual(h.defects, defects)
|
||
if source:
|
||
source = ' ' + source
|
||
self.assertEqual(h.fold(policy=policy.default),
|
||
'MIME-Version:' + source + '\n')
|
||
|
||
version_string_params = {
|
||
|
||
# Examples from the RFC.
|
||
|
||
'RFC_2045_1': (
|
||
'1.0',
|
||
'1.0',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[]),
|
||
|
||
'RFC_2045_2': (
|
||
'1.0 (produced by MetaSend Vx.x)',
|
||
'1.0 (produced by MetaSend Vx.x)',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[]),
|
||
|
||
'RFC_2045_3': (
|
||
'(produced by MetaSend Vx.x) 1.0',
|
||
'(produced by MetaSend Vx.x) 1.0',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[]),
|
||
|
||
'RFC_2045_4': (
|
||
'1.(produced by MetaSend Vx.x)0',
|
||
'1.(produced by MetaSend Vx.x)0',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[]),
|
||
|
||
# Other valid values.
|
||
|
||
'1_1': (
|
||
'1.1',
|
||
'1.1',
|
||
'1.1',
|
||
1,
|
||
1,
|
||
[]),
|
||
|
||
'2_1': (
|
||
'2.1',
|
||
'2.1',
|
||
'2.1',
|
||
2,
|
||
1,
|
||
[]),
|
||
|
||
'whitespace': (
|
||
'1 .0',
|
||
'1 .0',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[]),
|
||
|
||
'leading_trailing_whitespace_ignored': (
|
||
' 1.0 ',
|
||
' 1.0 ',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[]),
|
||
|
||
# Recoverable invalid values. We can recover here only because we
|
||
# already have a valid value by the time we encounter the garbage.
|
||
# Anywhere else, and we don't know where the garbage ends.
|
||
|
||
'non_comment_garbage_after': (
|
||
'1.0 <abc>',
|
||
'1.0 <abc>',
|
||
'1.0',
|
||
1,
|
||
0,
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
# Unrecoverable invalid values. We *could* apply more heuristics to
|
||
# get someing out of the first two, but doing so is not worth the
|
||
# effort.
|
||
|
||
'non_comment_garbage_before': (
|
||
'<abc> 1.0',
|
||
'<abc> 1.0',
|
||
None,
|
||
None,
|
||
None,
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'non_comment_garbage_inside': (
|
||
'1.<abc>0',
|
||
'1.<abc>0',
|
||
None,
|
||
None,
|
||
None,
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'two_periods': (
|
||
'1..0',
|
||
'1..0',
|
||
None,
|
||
None,
|
||
None,
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'2_x': (
|
||
'2.x',
|
||
'2.x',
|
||
None, # This could be 2, but it seems safer to make it None.
|
||
None,
|
||
None,
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'foo': (
|
||
'foo',
|
||
'foo',
|
||
None,
|
||
None,
|
||
None,
|
||
[errors.InvalidHeaderDefect]),
|
||
|
||
'missing': (
|
||
'',
|
||
'',
|
||
None,
|
||
None,
|
||
None,
|
||
[errors.HeaderMissingRequiredValue]),
|
||
|
||
}
|
||
|
||
|
||
@parameterize
|
||
class TestAddressHeader(TestHeaderBase):
|
||
|
||
example_params = {
|
||
|
||
'empty':
|
||
('<>',
|
||
[errors.InvalidHeaderDefect],
|
||
'<>',
|
||
'',
|
||
'<>',
|
||
'',
|
||
'',
|
||
None),
|
||
|
||
'address_only':
|
||
('zippy@pinhead.com',
|
||
[],
|
||
'zippy@pinhead.com',
|
||
'',
|
||
'zippy@pinhead.com',
|
||
'zippy',
|
||
'pinhead.com',
|
||
None),
|
||
|
||
'name_and_address':
|
||
('Zaphrod Beblebrux <zippy@pinhead.com>',
|
||
[],
|
||
'Zaphrod Beblebrux <zippy@pinhead.com>',
|
||
'Zaphrod Beblebrux',
|
||
'zippy@pinhead.com',
|
||
'zippy',
|
||
'pinhead.com',
|
||
None),
|
||
|
||
'quoted_local_part':
|
||
('Zaphrod Beblebrux <"foo bar"@pinhead.com>',
|
||
[],
|
||
'Zaphrod Beblebrux <"foo bar"@pinhead.com>',
|
||
'Zaphrod Beblebrux',
|
||
'"foo bar"@pinhead.com',
|
||
'foo bar',
|
||
'pinhead.com',
|
||
None),
|
||
|
||
'quoted_parens_in_name':
|
||
(r'"A \(Special\) Person" <person@dom.ain>',
|
||
[],
|
||
'"A (Special) Person" <person@dom.ain>',
|
||
'A (Special) Person',
|
||
'person@dom.ain',
|
||
'person',
|
||
'dom.ain',
|
||
None),
|
||
|
||
'quoted_backslashes_in_name':
|
||
(r'"Arthur \\Backslash\\ Foobar" <person@dom.ain>',
|
||
[],
|
||
r'"Arthur \\Backslash\\ Foobar" <person@dom.ain>',
|
||
r'Arthur \Backslash\ Foobar',
|
||
'person@dom.ain',
|
||
'person',
|
||
'dom.ain',
|
||
None),
|
||
|
||
'name_with_dot':
|
||
('John X. Doe <jxd@example.com>',
|
||
[errors.ObsoleteHeaderDefect],
|
||
'"John X. Doe" <jxd@example.com>',
|
||
'John X. Doe',
|
||
'jxd@example.com',
|
||
'jxd',
|
||
'example.com',
|
||
None),
|
||
|
||
'quoted_strings_in_local_part':
|
||
('""example" example"@example.com',
|
||
[errors.InvalidHeaderDefect]*3,
|
||
'"example example"@example.com',
|
||
'',
|
||
'"example example"@example.com',
|
||
'example example',
|
||
'example.com',
|
||
None),
|
||
|
||
'escaped_quoted_strings_in_local_part':
|
||
(r'"\"example\" example"@example.com',
|
||
[],
|
||
r'"\"example\" example"@example.com',
|
||
'',
|
||
r'"\"example\" example"@example.com',
|
||
r'"example" example',
|
||
'example.com',
|
||
None),
|
||
|
||
'escaped_escapes_in_local_part':
|
||
(r'"\\"example\\" example"@example.com',
|
||
[errors.InvalidHeaderDefect]*5,
|
||
r'"\\example\\\\ example"@example.com',
|
||
'',
|
||
r'"\\example\\\\ example"@example.com',
|
||
r'\example\\ example',
|
||
'example.com',
|
||
None),
|
||
|
||
'spaces_in_unquoted_local_part_collapsed':
|
||
('merwok wok @example.com',
|
||
[errors.InvalidHeaderDefect]*2,
|
||
'"merwok wok"@example.com',
|
||
'',
|
||
'"merwok wok"@example.com',
|
||
'merwok wok',
|
||
'example.com',
|
||
None),
|
||
|
||
'spaces_around_dots_in_local_part_removed':
|
||
('merwok. wok . wok@example.com',
|
||
[errors.ObsoleteHeaderDefect],
|
||
'merwok.wok.wok@example.com',
|
||
'',
|
||
'merwok.wok.wok@example.com',
|
||
'merwok.wok.wok',
|
||
'example.com',
|
||
None),
|
||
|
||
}
|
||
|
||
# XXX: Need many more examples, and in particular some with names in
|
||
# trailing comments, which aren't currently handled. comments in
|
||
# general are not handled yet.
|
||
|
||
def example_as_address(self, source, defects, decoded, display_name,
|
||
addr_spec, username, domain, comment):
|
||
h = self.make_header('sender', source)
|
||
self.assertEqual(h, decoded)
|
||
self.assertDefectsEqual(h.defects, defects)
|
||
a = h.address
|
||
self.assertEqual(str(a), decoded)
|
||
self.assertEqual(len(h.groups), 1)
|
||
self.assertEqual([a], list(h.groups[0].addresses))
|
||
self.assertEqual([a], list(h.addresses))
|
||
self.assertEqual(a.display_name, display_name)
|
||
self.assertEqual(a.addr_spec, addr_spec)
|
||
self.assertEqual(a.username, username)
|
||
self.assertEqual(a.domain, domain)
|
||
# XXX: we have no comment support yet.
|
||
#self.assertEqual(a.comment, comment)
|
||
|
||
def example_as_group(self, source, defects, decoded, display_name,
|
||
addr_spec, username, domain, comment):
|
||
source = 'foo: {};'.format(source)
|
||
gdecoded = 'foo: {};'.format(decoded) if decoded else 'foo:;'
|
||
h = self.make_header('to', source)
|
||
self.assertEqual(h, gdecoded)
|
||
self.assertDefectsEqual(h.defects, defects)
|
||
self.assertEqual(h.groups[0].addresses, h.addresses)
|
||
self.assertEqual(len(h.groups), 1)
|
||
self.assertEqual(len(h.addresses), 1)
|
||
a = h.addresses[0]
|
||
self.assertEqual(str(a), decoded)
|
||
self.assertEqual(a.display_name, display_name)
|
||
self.assertEqual(a.addr_spec, addr_spec)
|
||
self.assertEqual(a.username, username)
|
||
self.assertEqual(a.domain, domain)
|
||
|
||
def test_simple_address_list(self):
|
||
value = ('Fred <dinsdale@python.org>, foo@example.com, '
|
||
'"Harry W. Hastings" <hasty@example.com>')
|
||
h = self.make_header('to', value)
|
||
self.assertEqual(h, value)
|
||
self.assertEqual(len(h.groups), 3)
|
||
self.assertEqual(len(h.addresses), 3)
|
||
for i in range(3):
|
||
self.assertEqual(h.groups[i].addresses[0], h.addresses[i])
|
||
self.assertEqual(str(h.addresses[0]), 'Fred <dinsdale@python.org>')
|
||
self.assertEqual(str(h.addresses[1]), 'foo@example.com')
|
||
self.assertEqual(str(h.addresses[2]),
|
||
'"Harry W. Hastings" <hasty@example.com>')
|
||
self.assertEqual(h.addresses[2].display_name,
|
||
'Harry W. Hastings')
|
||
|
||
def test_complex_address_list(self):
|
||
examples = list(self.example_params.values())
|
||
source = ('dummy list:;, another: (empty);,' +
|
||
', '.join([x[0] for x in examples[:4]]) + ', ' +
|
||
r'"A \"list\"": ' +
|
||
', '.join([x[0] for x in examples[4:6]]) + ';,' +
|
||
', '.join([x[0] for x in examples[6:]])
|
||
)
|
||
# XXX: the fact that (empty) disappears here is a potential API design
|
||
# bug. We don't currently have a way to preserve comments.
|
||
expected = ('dummy list:;, another:;, ' +
|
||
', '.join([x[2] for x in examples[:4]]) + ', ' +
|
||
r'"A \"list\"": ' +
|
||
', '.join([x[2] for x in examples[4:6]]) + ';, ' +
|
||
', '.join([x[2] for x in examples[6:]])
|
||
)
|
||
|
||
h = self.make_header('to', source)
|
||
self.assertEqual(h.split(','), expected.split(','))
|
||
self.assertEqual(h, expected)
|
||
self.assertEqual(len(h.groups), 7 + len(examples) - 6)
|
||
self.assertEqual(h.groups[0].display_name, 'dummy list')
|
||
self.assertEqual(h.groups[1].display_name, 'another')
|
||
self.assertEqual(h.groups[6].display_name, 'A "list"')
|
||
self.assertEqual(len(h.addresses), len(examples))
|
||
for i in range(4):
|
||
self.assertIsNone(h.groups[i+2].display_name)
|
||
self.assertEqual(str(h.groups[i+2].addresses[0]), examples[i][2])
|
||
for i in range(7, 7 + len(examples) - 6):
|
||
self.assertIsNone(h.groups[i].display_name)
|
||
self.assertEqual(str(h.groups[i].addresses[0]), examples[i-1][2])
|
||
for i in range(len(examples)):
|
||
self.assertEqual(str(h.addresses[i]), examples[i][2])
|
||
self.assertEqual(h.addresses[i].addr_spec, examples[i][4])
|
||
|
||
def test_address_read_only(self):
|
||
h = self.make_header('sender', 'abc@xyz.com')
|
||
with self.assertRaises(AttributeError):
|
||
h.address = 'foo'
|
||
|
||
def test_addresses_read_only(self):
|
||
h = self.make_header('sender', 'abc@xyz.com')
|
||
with self.assertRaises(AttributeError):
|
||
h.addresses = 'foo'
|
||
|
||
def test_groups_read_only(self):
|
||
h = self.make_header('sender', 'abc@xyz.com')
|
||
with self.assertRaises(AttributeError):
|
||
h.groups = 'foo'
|
||
|
||
def test_addresses_types(self):
|
||
source = 'me <who@example.com>'
|
||
h = self.make_header('to', source)
|
||
self.assertIsInstance(h.addresses, tuple)
|
||
self.assertIsInstance(h.addresses[0], Address)
|
||
|
||
def test_groups_types(self):
|
||
source = 'me <who@example.com>'
|
||
h = self.make_header('to', source)
|
||
self.assertIsInstance(h.groups, tuple)
|
||
self.assertIsInstance(h.groups[0], Group)
|
||
|
||
def test_set_from_Address(self):
|
||
h = self.make_header('to', Address('me', 'foo', 'example.com'))
|
||
self.assertEqual(h, 'me <foo@example.com>')
|
||
|
||
def test_set_from_Address_list(self):
|
||
h = self.make_header('to', [Address('me', 'foo', 'example.com'),
|
||
Address('you', 'bar', 'example.com')])
|
||
self.assertEqual(h, 'me <foo@example.com>, you <bar@example.com>')
|
||
|
||
def test_set_from_Address_and_Group_list(self):
|
||
h = self.make_header('to', [Address('me', 'foo', 'example.com'),
|
||
Group('bing', [Address('fiz', 'z', 'b.com'),
|
||
Address('zif', 'f', 'c.com')]),
|
||
Address('you', 'bar', 'example.com')])
|
||
self.assertEqual(h, 'me <foo@example.com>, bing: fiz <z@b.com>, '
|
||
'zif <f@c.com>;, you <bar@example.com>')
|
||
self.assertEqual(h.fold(policy=policy.default.clone(max_line_length=40)),
|
||
'to: me <foo@example.com>,\n'
|
||
' bing: fiz <z@b.com>, zif <f@c.com>;,\n'
|
||
' you <bar@example.com>\n')
|
||
|
||
def test_set_from_Group_list(self):
|
||
h = self.make_header('to', [Group('bing', [Address('fiz', 'z', 'b.com'),
|
||
Address('zif', 'f', 'c.com')])])
|
||
self.assertEqual(h, 'bing: fiz <z@b.com>, zif <f@c.com>;')
|
||
|
||
|
||
class TestAddressAndGroup(TestEmailBase):
|
||
|
||
def _test_attr_ro(self, obj, attr):
|
||
with self.assertRaises(AttributeError):
|
||
setattr(obj, attr, 'foo')
|
||
|
||
def test_address_display_name_ro(self):
|
||
self._test_attr_ro(Address('foo', 'bar', 'baz'), 'display_name')
|
||
|
||
def test_address_username_ro(self):
|
||
self._test_attr_ro(Address('foo', 'bar', 'baz'), 'username')
|
||
|
||
def test_address_domain_ro(self):
|
||
self._test_attr_ro(Address('foo', 'bar', 'baz'), 'domain')
|
||
|
||
def test_group_display_name_ro(self):
|
||
self._test_attr_ro(Group('foo'), 'display_name')
|
||
|
||
def test_group_addresses_ro(self):
|
||
self._test_attr_ro(Group('foo'), 'addresses')
|
||
|
||
def test_address_from_username_domain(self):
|
||
a = Address('foo', 'bar', 'baz')
|
||
self.assertEqual(a.display_name, 'foo')
|
||
self.assertEqual(a.username, 'bar')
|
||
self.assertEqual(a.domain, 'baz')
|
||
self.assertEqual(a.addr_spec, 'bar@baz')
|
||
self.assertEqual(str(a), 'foo <bar@baz>')
|
||
|
||
def test_address_from_addr_spec(self):
|
||
a = Address('foo', addr_spec='bar@baz')
|
||
self.assertEqual(a.display_name, 'foo')
|
||
self.assertEqual(a.username, 'bar')
|
||
self.assertEqual(a.domain, 'baz')
|
||
self.assertEqual(a.addr_spec, 'bar@baz')
|
||
self.assertEqual(str(a), 'foo <bar@baz>')
|
||
|
||
def test_address_with_no_display_name(self):
|
||
a = Address(addr_spec='bar@baz')
|
||
self.assertEqual(a.display_name, '')
|
||
self.assertEqual(a.username, 'bar')
|
||
self.assertEqual(a.domain, 'baz')
|
||
self.assertEqual(a.addr_spec, 'bar@baz')
|
||
self.assertEqual(str(a), 'bar@baz')
|
||
|
||
def test_null_address(self):
|
||
a = Address()
|
||
self.assertEqual(a.display_name, '')
|
||
self.assertEqual(a.username, '')
|
||
self.assertEqual(a.domain, '')
|
||
self.assertEqual(a.addr_spec, '<>')
|
||
self.assertEqual(str(a), '<>')
|
||
|
||
def test_domain_only(self):
|
||
# This isn't really a valid address.
|
||
a = Address(domain='buzz')
|
||
self.assertEqual(a.display_name, '')
|
||
self.assertEqual(a.username, '')
|
||
self.assertEqual(a.domain, 'buzz')
|
||
self.assertEqual(a.addr_spec, '@buzz')
|
||
self.assertEqual(str(a), '@buzz')
|
||
|
||
def test_username_only(self):
|
||
# This isn't really a valid address.
|
||
a = Address(username='buzz')
|
||
self.assertEqual(a.display_name, '')
|
||
self.assertEqual(a.username, 'buzz')
|
||
self.assertEqual(a.domain, '')
|
||
self.assertEqual(a.addr_spec, 'buzz')
|
||
self.assertEqual(str(a), 'buzz')
|
||
|
||
def test_display_name_only(self):
|
||
a = Address('buzz')
|
||
self.assertEqual(a.display_name, 'buzz')
|
||
self.assertEqual(a.username, '')
|
||
self.assertEqual(a.domain, '')
|
||
self.assertEqual(a.addr_spec, '<>')
|
||
self.assertEqual(str(a), 'buzz <>')
|
||
|
||
def test_quoting(self):
|
||
# Ideally we'd check every special individually, but I'm not up for
|
||
# writing that many tests.
|
||
a = Address('Sara J.', 'bad name', 'example.com')
|
||
self.assertEqual(a.display_name, 'Sara J.')
|
||
self.assertEqual(a.username, 'bad name')
|
||
self.assertEqual(a.domain, 'example.com')
|
||
self.assertEqual(a.addr_spec, '"bad name"@example.com')
|
||
self.assertEqual(str(a), '"Sara J." <"bad name"@example.com>')
|
||
|
||
def test_il8n(self):
|
||
a = Address('Éric', 'wok', 'exàmple.com')
|
||
self.assertEqual(a.display_name, 'Éric')
|
||
self.assertEqual(a.username, 'wok')
|
||
self.assertEqual(a.domain, 'exàmple.com')
|
||
self.assertEqual(a.addr_spec, 'wok@exàmple.com')
|
||
self.assertEqual(str(a), 'Éric <wok@exàmple.com>')
|
||
|
||
# XXX: there is an API design issue that needs to be solved here.
|
||
#def test_non_ascii_username_raises(self):
|
||
# with self.assertRaises(ValueError):
|
||
# Address('foo', 'wők', 'example.com')
|
||
|
||
def test_non_ascii_username_in_addr_spec_raises(self):
|
||
with self.assertRaises(ValueError):
|
||
Address('foo', addr_spec='wők@example.com')
|
||
|
||
def test_address_addr_spec_and_username_raises(self):
|
||
with self.assertRaises(TypeError):
|
||
Address('foo', username='bing', addr_spec='bar@baz')
|
||
|
||
def test_address_addr_spec_and_domain_raises(self):
|
||
with self.assertRaises(TypeError):
|
||
Address('foo', domain='bing', addr_spec='bar@baz')
|
||
|
||
def test_address_addr_spec_and_username_and_domain_raises(self):
|
||
with self.assertRaises(TypeError):
|
||
Address('foo', username='bong', domain='bing', addr_spec='bar@baz')
|
||
|
||
def test_space_in_addr_spec_username_raises(self):
|
||
with self.assertRaises(ValueError):
|
||
Address('foo', addr_spec="bad name@example.com")
|
||
|
||
def test_bad_addr_sepc_raises(self):
|
||
with self.assertRaises(ValueError):
|
||
Address('foo', addr_spec="name@ex[]ample.com")
|
||
|
||
def test_empty_group(self):
|
||
g = Group('foo')
|
||
self.assertEqual(g.display_name, 'foo')
|
||
self.assertEqual(g.addresses, tuple())
|
||
self.assertEqual(str(g), 'foo:;')
|
||
|
||
def test_empty_group_list(self):
|
||
g = Group('foo', addresses=[])
|
||
self.assertEqual(g.display_name, 'foo')
|
||
self.assertEqual(g.addresses, tuple())
|
||
self.assertEqual(str(g), 'foo:;')
|
||
|
||
def test_null_group(self):
|
||
g = Group()
|
||
self.assertIsNone(g.display_name)
|
||
self.assertEqual(g.addresses, tuple())
|
||
self.assertEqual(str(g), 'None:;')
|
||
|
||
def test_group_with_addresses(self):
|
||
addrs = [Address('b', 'b', 'c'), Address('a', 'b','c')]
|
||
g = Group('foo', addrs)
|
||
self.assertEqual(g.display_name, 'foo')
|
||
self.assertEqual(g.addresses, tuple(addrs))
|
||
self.assertEqual(str(g), 'foo: b <b@c>, a <b@c>;')
|
||
|
||
def test_group_with_addresses_no_display_name(self):
|
||
addrs = [Address('b', 'b', 'c'), Address('a', 'b','c')]
|
||
g = Group(addresses=addrs)
|
||
self.assertIsNone(g.display_name)
|
||
self.assertEqual(g.addresses, tuple(addrs))
|
||
self.assertEqual(str(g), 'None: b <b@c>, a <b@c>;')
|
||
|
||
def test_group_with_one_address_no_display_name(self):
|
||
addrs = [Address('b', 'b', 'c')]
|
||
g = Group(addresses=addrs)
|
||
self.assertIsNone(g.display_name)
|
||
self.assertEqual(g.addresses, tuple(addrs))
|
||
self.assertEqual(str(g), 'b <b@c>')
|
||
|
||
def test_display_name_quoting(self):
|
||
g = Group('foo.bar')
|
||
self.assertEqual(g.display_name, 'foo.bar')
|
||
self.assertEqual(g.addresses, tuple())
|
||
self.assertEqual(str(g), '"foo.bar":;')
|
||
|
||
def test_display_name_blanks_not_quoted(self):
|
||
g = Group('foo bar')
|
||
self.assertEqual(g.display_name, 'foo bar')
|
||
self.assertEqual(g.addresses, tuple())
|
||
self.assertEqual(str(g), 'foo bar:;')
|
||
|
||
def test_set_message_header_from_address(self):
|
||
a = Address('foo', 'bar', 'example.com')
|
||
m = Message(policy=policy.default)
|
||
m['To'] = a
|
||
self.assertEqual(m['to'], 'foo <bar@example.com>')
|
||
self.assertEqual(m['to'].addresses, (a,))
|
||
|
||
def test_set_message_header_from_group(self):
|
||
g = Group('foo bar')
|
||
m = Message(policy=policy.default)
|
||
m['To'] = g
|
||
self.assertEqual(m['to'], 'foo bar:;')
|
||
self.assertEqual(m['to'].addresses, g.addresses)
|
||
|
||
|
||
class TestFolding(TestHeaderBase):
|
||
|
||
def test_short_unstructured(self):
|
||
h = self.make_header('subject', 'this is a test')
|
||
self.assertEqual(h.fold(policy=policy.default),
|
||
'subject: this is a test\n')
|
||
|
||
def test_long_unstructured(self):
|
||
h = self.make_header('Subject', 'This is a long header '
|
||
'line that will need to be folded into two lines '
|
||
'and will demonstrate basic folding')
|
||
self.assertEqual(h.fold(policy=policy.default),
|
||
'Subject: This is a long header line that will '
|
||
'need to be folded into two lines\n'
|
||
' and will demonstrate basic folding\n')
|
||
|
||
def test_unstructured_short_max_line_length(self):
|
||
h = self.make_header('Subject', 'this is a short header '
|
||
'that will be folded anyway')
|
||
self.assertEqual(
|
||
h.fold(policy=policy.default.clone(max_line_length=20)),
|
||
textwrap.dedent("""\
|
||
Subject: this is a
|
||
short header that
|
||
will be folded
|
||
anyway
|
||
"""))
|
||
|
||
def test_fold_unstructured_single_word(self):
|
||
h = self.make_header('Subject', 'test')
|
||
self.assertEqual(h.fold(policy=policy.default), 'Subject: test\n')
|
||
|
||
def test_fold_unstructured_short(self):
|
||
h = self.make_header('Subject', 'test test test')
|
||
self.assertEqual(h.fold(policy=policy.default),
|
||
'Subject: test test test\n')
|
||
|
||
def test_fold_unstructured_with_overlong_word(self):
|
||
h = self.make_header('Subject', 'thisisaverylonglineconsistingofa'
|
||
'singlewordthatwontfit')
|
||
self.assertEqual(
|
||
h.fold(policy=policy.default.clone(max_line_length=20)),
|
||
'Subject: thisisaverylonglineconsistingofasinglewordthatwontfit\n')
|
||
|
||
def test_fold_unstructured_with_two_overlong_words(self):
|
||
h = self.make_header('Subject', 'thisisaverylonglineconsistingofa'
|
||
'singlewordthatwontfit plusanotherverylongwordthatwontfit')
|
||
self.assertEqual(
|
||
h.fold(policy=policy.default.clone(max_line_length=20)),
|
||
'Subject: thisisaverylonglineconsistingofasinglewordthatwontfit\n'
|
||
' plusanotherverylongwordthatwontfit\n')
|
||
|
||
def test_fold_unstructured_with_slightly_long_word(self):
|
||
h = self.make_header('Subject', 'thislongwordislessthanmaxlinelen')
|
||
self.assertEqual(
|
||
h.fold(policy=policy.default.clone(max_line_length=35)),
|
||
'Subject:\n thislongwordislessthanmaxlinelen\n')
|
||
|
||
def test_fold_unstructured_with_commas(self):
|
||
# The old wrapper would fold this at the commas.
|
||
h = self.make_header('Subject', "This header is intended to "
|
||
"demonstrate, in a fairly susinct way, that we now do "
|
||
"not give a , special treatment in unstructured headers.")
|
||
self.assertEqual(
|
||
h.fold(policy=policy.default.clone(max_line_length=60)),
|
||
textwrap.dedent("""\
|
||
Subject: This header is intended to demonstrate, in a fairly
|
||
susinct way, that we now do not give a , special treatment
|
||
in unstructured headers.
|
||
"""))
|
||
|
||
def test_fold_address_list(self):
|
||
h = self.make_header('To', '"Theodore H. Perfect" <yes@man.com>, '
|
||
'"My address is very long because my name is long" <foo@bar.com>, '
|
||
'"Only A. Friend" <no@yes.com>')
|
||
self.assertEqual(h.fold(policy=policy.default), textwrap.dedent("""\
|
||
To: "Theodore H. Perfect" <yes@man.com>,
|
||
"My address is very long because my name is long" <foo@bar.com>,
|
||
"Only A. Friend" <no@yes.com>
|
||
"""))
|
||
|
||
def test_fold_date_header(self):
|
||
h = self.make_header('Date', 'Sat, 2 Feb 2002 17:00:06 -0800')
|
||
self.assertEqual(h.fold(policy=policy.default),
|
||
'Date: Sat, 02 Feb 2002 17:00:06 -0800\n')
|
||
|
||
|
||
|
||
if __name__ == '__main__':
|
||
unittest.main()
|