cpython/Lib/test/test_email/test_message.py
R David Murray 29d1bc0842 #24277: The new email API is no longer provisional.
This is a wholesale reorganization and editing of the email documentation to
make the new API the standard one, and the old API the 'legacy' one.  The
default is still the compat32 policy, for backward compatibility.  We will
change that eventually.
2016-09-07 21:15:59 -04:00

803 lines
27 KiB
Python

import unittest
import textwrap
from email import policy, message_from_string
from email.message import EmailMessage, MIMEPart
from test.test_email import TestEmailBase, parameterize
# Helper.
def first(iterable):
return next(filter(lambda x: x is not None, iterable), None)
class Test(TestEmailBase):
policy = policy.default
def test_error_on_setitem_if_max_count_exceeded(self):
m = self._str_msg("")
m['To'] = 'abc@xyz'
with self.assertRaises(ValueError):
m['To'] = 'xyz@abc'
def test_rfc2043_auto_decoded_and_emailmessage_used(self):
m = message_from_string(textwrap.dedent("""\
Subject: Ayons asperges pour le =?utf-8?q?d=C3=A9jeuner?=
From: =?utf-8?q?Pep=C3=A9?= Le Pew <pepe@example.com>
To: "Penelope Pussycat" <"penelope@example.com">
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
sample text
"""), policy=policy.default)
self.assertEqual(m['subject'], "Ayons asperges pour le déjeuner")
self.assertEqual(m['from'], "Pepé Le Pew <pepe@example.com>")
self.assertIsInstance(m, EmailMessage)
@parameterize
class TestEmailMessageBase:
policy = policy.default
# The first argument is a triple (related, html, plain) of indices into the
# list returned by 'walk' called on a Message constructed from the third.
# The indices indicate which part should match the corresponding part-type
# when passed to get_body (ie: the "first" part of that type in the
# message). The second argument is a list of indices into the 'walk' list
# of the attachments that should be returned by a call to
# 'iter_attachments'. The third argument is a list of indices into 'walk'
# that should be returned by a call to 'iter_parts'. Note that the first
# item returned by 'walk' is the Message itself.
message_params = {
'empty_message': (
(None, None, 0),
(),
(),
""),
'non_mime_plain': (
(None, None, 0),
(),
(),
textwrap.dedent("""\
To: foo@example.com
simple text body
""")),
'mime_non_text': (
(None, None, None),
(),
(),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: image/jpg
bogus body.
""")),
'plain_html_alternative': (
(None, 2, 1),
(),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/alternative; boundary="==="
preamble
--===
Content-Type: text/plain
simple body
--===
Content-Type: text/html
<p>simple body</p>
--===--
""")),
'plain_html_mixed': (
(None, 2, 1),
(),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
preamble
--===
Content-Type: text/plain
simple body
--===
Content-Type: text/html
<p>simple body</p>
--===--
""")),
'plain_html_attachment_mixed': (
(None, None, 1),
(2,),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: text/plain
simple body
--===
Content-Type: text/html
Content-Disposition: attachment
<p>simple body</p>
--===--
""")),
'html_text_attachment_mixed': (
(None, 2, None),
(1,),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: text/plain
Content-Disposition: AtTaChment
simple body
--===
Content-Type: text/html
<p>simple body</p>
--===--
""")),
'html_text_attachment_inline_mixed': (
(None, 2, 1),
(),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: text/plain
Content-Disposition: InLine
simple body
--===
Content-Type: text/html
Content-Disposition: inline
<p>simple body</p>
--===--
""")),
# RFC 2387
'related': (
(0, 1, None),
(2,),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/related; boundary="==="; type=text/html
--===
Content-Type: text/html
<p>simple body</p>
--===
Content-Type: image/jpg
Content-ID: <image1>
bogus data
--===--
""")),
# This message structure will probably never be seen in the wild, but
# it proves we distinguish between text parts based on 'start'. The
# content would not, of course, actually work :)
'related_with_start': (
(0, 2, None),
(1,),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/related; boundary="==="; type=text/html;
start="<body>"
--===
Content-Type: text/html
Content-ID: <include>
useless text
--===
Content-Type: text/html
Content-ID: <body>
<p>simple body</p>
<!--#include file="<include>"-->
--===--
""")),
'mixed_alternative_plain_related': (
(3, 4, 2),
(6, 7),
(1, 6, 7),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: multipart/alternative; boundary="+++"
--+++
Content-Type: text/plain
simple body
--+++
Content-Type: multipart/related; boundary="___"
--___
Content-Type: text/html
<p>simple body</p>
--___
Content-Type: image/jpg
Content-ID: <image1@cid>
bogus jpg body
--___--
--+++--
--===
Content-Type: image/jpg
Content-Disposition: attachment
bogus jpg body
--===
Content-Type: image/jpg
Content-Disposition: AttacHmenT
another bogus jpg body
--===--
""")),
# This structure suggested by Stephen J. Turnbull...may not exist/be
# supported in the wild, but we want to support it.
'mixed_related_alternative_plain_html': (
(1, 4, 3),
(6, 7),
(1, 6, 7),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: multipart/related; boundary="+++"
--+++
Content-Type: multipart/alternative; boundary="___"
--___
Content-Type: text/plain
simple body
--___
Content-Type: text/html
<p>simple body</p>
--___--
--+++
Content-Type: image/jpg
Content-ID: <image1@cid>
bogus jpg body
--+++--
--===
Content-Type: image/jpg
Content-Disposition: attachment
bogus jpg body
--===
Content-Type: image/jpg
Content-Disposition: attachment
another bogus jpg body
--===--
""")),
# Same thing, but proving we only look at the root part, which is the
# first one if there isn't any start parameter. That is, this is a
# broken related.
'mixed_related_alternative_plain_html_wrong_order': (
(1, None, None),
(6, 7),
(1, 6, 7),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: multipart/related; boundary="+++"
--+++
Content-Type: image/jpg
Content-ID: <image1@cid>
bogus jpg body
--+++
Content-Type: multipart/alternative; boundary="___"
--___
Content-Type: text/plain
simple body
--___
Content-Type: text/html
<p>simple body</p>
--___--
--+++--
--===
Content-Type: image/jpg
Content-Disposition: attachment
bogus jpg body
--===
Content-Type: image/jpg
Content-Disposition: attachment
another bogus jpg body
--===--
""")),
'message_rfc822': (
(None, None, None),
(),
(),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: message/rfc822
To: bar@example.com
From: robot@examp.com
this is a message body.
""")),
'mixed_text_message_rfc822': (
(None, None, 1),
(2,),
(1, 2),
textwrap.dedent("""\
To: foo@example.com
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="==="
--===
Content-Type: text/plain
Your message has bounced, ser.
--===
Content-Type: message/rfc822
To: bar@example.com
From: robot@examp.com
this is a message body.
--===--
""")),
}
def message_as_get_body(self, body_parts, attachments, parts, msg):
m = self._str_msg(msg)
allparts = list(m.walk())
expected = [None if n is None else allparts[n] for n in body_parts]
related = 0; html = 1; plain = 2
self.assertEqual(m.get_body(), first(expected))
self.assertEqual(m.get_body(preferencelist=(
'related', 'html', 'plain')),
first(expected))
self.assertEqual(m.get_body(preferencelist=('related', 'html')),
first(expected[related:html+1]))
self.assertEqual(m.get_body(preferencelist=('related', 'plain')),
first([expected[related], expected[plain]]))
self.assertEqual(m.get_body(preferencelist=('html', 'plain')),
first(expected[html:plain+1]))
self.assertEqual(m.get_body(preferencelist=['related']),
expected[related])
self.assertEqual(m.get_body(preferencelist=['html']), expected[html])
self.assertEqual(m.get_body(preferencelist=['plain']), expected[plain])
self.assertEqual(m.get_body(preferencelist=('plain', 'html')),
first(expected[plain:html-1:-1]))
self.assertEqual(m.get_body(preferencelist=('plain', 'related')),
first([expected[plain], expected[related]]))
self.assertEqual(m.get_body(preferencelist=('html', 'related')),
first(expected[html::-1]))
self.assertEqual(m.get_body(preferencelist=('plain', 'html', 'related')),
first(expected[::-1]))
self.assertEqual(m.get_body(preferencelist=('html', 'plain', 'related')),
first([expected[html],
expected[plain],
expected[related]]))
def message_as_iter_attachment(self, body_parts, attachments, parts, msg):
m = self._str_msg(msg)
allparts = list(m.walk())
attachments = [allparts[n] for n in attachments]
self.assertEqual(list(m.iter_attachments()), attachments)
def message_as_iter_parts(self, body_parts, attachments, parts, msg):
m = self._str_msg(msg)
allparts = list(m.walk())
parts = [allparts[n] for n in parts]
self.assertEqual(list(m.iter_parts()), parts)
class _TestContentManager:
def get_content(self, msg, *args, **kw):
return msg, args, kw
def set_content(self, msg, *args, **kw):
self.msg = msg
self.args = args
self.kw = kw
def test_get_content_with_cm(self):
m = self._str_msg('')
cm = self._TestContentManager()
self.assertEqual(m.get_content(content_manager=cm), (m, (), {}))
msg, args, kw = m.get_content('foo', content_manager=cm, bar=1, k=2)
self.assertEqual(msg, m)
self.assertEqual(args, ('foo',))
self.assertEqual(kw, dict(bar=1, k=2))
def test_get_content_default_cm_comes_from_policy(self):
p = policy.default.clone(content_manager=self._TestContentManager())
m = self._str_msg('', policy=p)
self.assertEqual(m.get_content(), (m, (), {}))
msg, args, kw = m.get_content('foo', bar=1, k=2)
self.assertEqual(msg, m)
self.assertEqual(args, ('foo',))
self.assertEqual(kw, dict(bar=1, k=2))
def test_set_content_with_cm(self):
m = self._str_msg('')
cm = self._TestContentManager()
m.set_content(content_manager=cm)
self.assertEqual(cm.msg, m)
self.assertEqual(cm.args, ())
self.assertEqual(cm.kw, {})
m.set_content('foo', content_manager=cm, bar=1, k=2)
self.assertEqual(cm.msg, m)
self.assertEqual(cm.args, ('foo',))
self.assertEqual(cm.kw, dict(bar=1, k=2))
def test_set_content_default_cm_comes_from_policy(self):
cm = self._TestContentManager()
p = policy.default.clone(content_manager=cm)
m = self._str_msg('', policy=p)
m.set_content()
self.assertEqual(cm.msg, m)
self.assertEqual(cm.args, ())
self.assertEqual(cm.kw, {})
m.set_content('foo', bar=1, k=2)
self.assertEqual(cm.msg, m)
self.assertEqual(cm.args, ('foo',))
self.assertEqual(cm.kw, dict(bar=1, k=2))
# outcome is whether xxx_method should raise ValueError error when called
# on multipart/subtype. Blank outcome means it depends on xxx (add
# succeeds, make raises). Note: 'none' means there are content-type
# headers but payload is None...this happening in practice would be very
# unusual, so treating it as if there were content seems reasonable.
# method subtype outcome
subtype_params = (
('related', 'no_content', 'succeeds'),
('related', 'none', 'succeeds'),
('related', 'plain', 'succeeds'),
('related', 'related', ''),
('related', 'alternative', 'raises'),
('related', 'mixed', 'raises'),
('alternative', 'no_content', 'succeeds'),
('alternative', 'none', 'succeeds'),
('alternative', 'plain', 'succeeds'),
('alternative', 'related', 'succeeds'),
('alternative', 'alternative', ''),
('alternative', 'mixed', 'raises'),
('mixed', 'no_content', 'succeeds'),
('mixed', 'none', 'succeeds'),
('mixed', 'plain', 'succeeds'),
('mixed', 'related', 'succeeds'),
('mixed', 'alternative', 'succeeds'),
('mixed', 'mixed', ''),
)
def _make_subtype_test_message(self, subtype):
m = self.message()
payload = None
msg_headers = [
('To', 'foo@bar.com'),
('From', 'bar@foo.com'),
]
if subtype != 'no_content':
('content-shadow', 'Logrus'),
msg_headers.append(('X-Random-Header', 'Corwin'))
if subtype == 'text':
payload = ''
msg_headers.append(('Content-Type', 'text/plain'))
m.set_payload('')
elif subtype != 'no_content':
payload = []
msg_headers.append(('Content-Type', 'multipart/' + subtype))
msg_headers.append(('X-Trump', 'Random'))
m.set_payload(payload)
for name, value in msg_headers:
m[name] = value
return m, msg_headers, payload
def _check_disallowed_subtype_raises(self, m, method_name, subtype, method):
with self.assertRaises(ValueError) as ar:
getattr(m, method)()
exc_text = str(ar.exception)
self.assertIn(subtype, exc_text)
self.assertIn(method_name, exc_text)
def _check_make_multipart(self, m, msg_headers, payload):
count = 0
for name, value in msg_headers:
if not name.lower().startswith('content-'):
self.assertEqual(m[name], value)
count += 1
self.assertEqual(len(m), count+1) # +1 for new Content-Type
part = next(m.iter_parts())
count = 0
for name, value in msg_headers:
if name.lower().startswith('content-'):
self.assertEqual(part[name], value)
count += 1
self.assertEqual(len(part), count)
self.assertEqual(part.get_payload(), payload)
def subtype_as_make(self, method, subtype, outcome):
m, msg_headers, payload = self._make_subtype_test_message(subtype)
make_method = 'make_' + method
if outcome in ('', 'raises'):
self._check_disallowed_subtype_raises(m, method, subtype, make_method)
return
getattr(m, make_method)()
self.assertEqual(m.get_content_maintype(), 'multipart')
self.assertEqual(m.get_content_subtype(), method)
if subtype == 'no_content':
self.assertEqual(len(m.get_payload()), 0)
self.assertEqual(m.items(),
msg_headers + [('Content-Type',
'multipart/'+method)])
else:
self.assertEqual(len(m.get_payload()), 1)
self._check_make_multipart(m, msg_headers, payload)
def subtype_as_make_with_boundary(self, method, subtype, outcome):
# Doing all variation is a bit of overkill...
m = self.message()
if outcome in ('', 'raises'):
m['Content-Type'] = 'multipart/' + subtype
with self.assertRaises(ValueError) as cm:
getattr(m, 'make_' + method)()
return
if subtype == 'plain':
m['Content-Type'] = 'text/plain'
elif subtype != 'no_content':
m['Content-Type'] = 'multipart/' + subtype
getattr(m, 'make_' + method)(boundary="abc")
self.assertTrue(m.is_multipart())
self.assertEqual(m.get_boundary(), 'abc')
def test_policy_on_part_made_by_make_comes_from_message(self):
for method in ('make_related', 'make_alternative', 'make_mixed'):
m = self.message(policy=self.policy.clone(content_manager='foo'))
m['Content-Type'] = 'text/plain'
getattr(m, method)()
self.assertEqual(m.get_payload(0).policy.content_manager, 'foo')
class _TestSetContentManager:
def set_content(self, msg, content, *args, **kw):
msg['Content-Type'] = 'text/plain'
msg.set_payload(content)
def subtype_as_add(self, method, subtype, outcome):
m, msg_headers, payload = self._make_subtype_test_message(subtype)
cm = self._TestSetContentManager()
add_method = 'add_attachment' if method=='mixed' else 'add_' + method
if outcome == 'raises':
self._check_disallowed_subtype_raises(m, method, subtype, add_method)
return
getattr(m, add_method)('test', content_manager=cm)
self.assertEqual(m.get_content_maintype(), 'multipart')
self.assertEqual(m.get_content_subtype(), method)
if method == subtype or subtype == 'no_content':
self.assertEqual(len(m.get_payload()), 1)
for name, value in msg_headers:
self.assertEqual(m[name], value)
part = m.get_payload()[0]
else:
self.assertEqual(len(m.get_payload()), 2)
self._check_make_multipart(m, msg_headers, payload)
part = m.get_payload()[1]
self.assertEqual(part.get_content_type(), 'text/plain')
self.assertEqual(part.get_payload(), 'test')
if method=='mixed':
self.assertEqual(part['Content-Disposition'], 'attachment')
elif method=='related':
self.assertEqual(part['Content-Disposition'], 'inline')
else:
# Otherwise we don't guess.
self.assertIsNone(part['Content-Disposition'])
class _TestSetRaisingContentManager:
def set_content(self, msg, content, *args, **kw):
raise Exception('test')
def test_default_content_manager_for_add_comes_from_policy(self):
cm = self._TestSetRaisingContentManager()
m = self.message(policy=self.policy.clone(content_manager=cm))
for method in ('add_related', 'add_alternative', 'add_attachment'):
with self.assertRaises(Exception) as ar:
getattr(m, method)('')
self.assertEqual(str(ar.exception), 'test')
def message_as_clear(self, body_parts, attachments, parts, msg):
m = self._str_msg(msg)
m.clear()
self.assertEqual(len(m), 0)
self.assertEqual(list(m.items()), [])
self.assertIsNone(m.get_payload())
self.assertEqual(list(m.iter_parts()), [])
def message_as_clear_content(self, body_parts, attachments, parts, msg):
m = self._str_msg(msg)
expected_headers = [h for h in m.keys()
if not h.lower().startswith('content-')]
m.clear_content()
self.assertEqual(list(m.keys()), expected_headers)
self.assertIsNone(m.get_payload())
self.assertEqual(list(m.iter_parts()), [])
def test_is_attachment(self):
m = self._make_message()
self.assertFalse(m.is_attachment())
m['Content-Disposition'] = 'inline'
self.assertFalse(m.is_attachment())
m.replace_header('Content-Disposition', 'attachment')
self.assertTrue(m.is_attachment())
m.replace_header('Content-Disposition', 'AtTachMent')
self.assertTrue(m.is_attachment())
m.set_param('filename', 'abc.png', 'Content-Disposition')
self.assertTrue(m.is_attachment())
def test_iter_attachments_mutation(self):
# We had a bug where iter_attachments was mutating the list.
m = self._make_message()
m.set_content('arbitrary text as main part')
m.add_related('more text as a related part')
m.add_related('yet more text as a second "attachment"')
orig = m.get_payload().copy()
self.assertEqual(len(list(m.iter_attachments())), 2)
self.assertEqual(m.get_payload(), orig)
class TestEmailMessage(TestEmailMessageBase, TestEmailBase):
message = EmailMessage
def test_set_content_adds_MIME_Version(self):
m = self._str_msg('')
cm = self._TestContentManager()
self.assertNotIn('MIME-Version', m)
m.set_content(content_manager=cm)
self.assertEqual(m['MIME-Version'], '1.0')
class _MIME_Version_adding_CM:
def set_content(self, msg, *args, **kw):
msg['MIME-Version'] = '1.0'
def test_set_content_does_not_duplicate_MIME_Version(self):
m = self._str_msg('')
cm = self._MIME_Version_adding_CM()
self.assertNotIn('MIME-Version', m)
m.set_content(content_manager=cm)
self.assertEqual(m['MIME-Version'], '1.0')
def test_as_string_uses_max_header_length_by_default(self):
m = self._str_msg('Subject: long line' + ' ab'*50 + '\n\n')
self.assertEqual(len(m.as_string().strip().splitlines()), 3)
def test_as_string_allows_maxheaderlen(self):
m = self._str_msg('Subject: long line' + ' ab'*50 + '\n\n')
self.assertEqual(len(m.as_string(maxheaderlen=0).strip().splitlines()),
1)
self.assertEqual(len(m.as_string(maxheaderlen=34).strip().splitlines()),
6)
def test_str_defaults_to_policy_max_line_length(self):
m = self._str_msg('Subject: long line' + ' ab'*50 + '\n\n')
self.assertEqual(len(str(m).strip().splitlines()), 3)
def test_str_defaults_to_utf8(self):
m = EmailMessage()
m['Subject'] = 'unicöde'
self.assertEqual(str(m), 'Subject: unicöde\n\n')
class TestMIMEPart(TestEmailMessageBase, TestEmailBase):
# Doing the full test run here may seem a bit redundant, since the two
# classes are almost identical. But what if they drift apart? So we do
# the full tests so that any future drift doesn't introduce bugs.
message = MIMEPart
def test_set_content_does_not_add_MIME_Version(self):
m = self._str_msg('')
cm = self._TestContentManager()
self.assertNotIn('MIME-Version', m)
m.set_content(content_manager=cm)
self.assertNotIn('MIME-Version', m)
if __name__ == '__main__':
unittest.main()