diff options
author | Guido van Rossum <guido@python.org> | 2007-08-30 01:15:14 (GMT) |
---|---|---|
committer | Guido van Rossum <guido@python.org> | 2007-08-30 01:15:14 (GMT) |
commit | 8b3febef2f96c35e9aad9db2ef499db040fdefae (patch) | |
tree | 6bc3322d80780a8d57d845b350aad9fbe250d5de /Lib/email/test/test_email.py | |
parent | 21b731fb7798218a0e59e6db204d1d43d2a1e820 (diff) | |
download | cpython-8b3febef2f96c35e9aad9db2ef499db040fdefae.zip cpython-8b3febef2f96c35e9aad9db2ef499db040fdefae.tar.gz cpython-8b3febef2f96c35e9aad9db2ef499db040fdefae.tar.bz2 |
Copying the email package back, despite its failings.
Diffstat (limited to 'Lib/email/test/test_email.py')
-rw-r--r-- | Lib/email/test/test_email.py | 3307 |
1 files changed, 3307 insertions, 0 deletions
diff --git a/Lib/email/test/test_email.py b/Lib/email/test/test_email.py new file mode 100644 index 0000000..fc8224a --- /dev/null +++ b/Lib/email/test/test_email.py @@ -0,0 +1,3307 @@ +# Copyright (C) 2001-2007 Python Software Foundation +# Contact: email-sig@python.org +# email package unit tests + +import os +import sys +import time +import base64 +import difflib +import unittest +import warnings + +from io import StringIO +from itertools import chain + +import email + +from email.charset import Charset +from email.header import Header, decode_header, make_header +from email.parser import Parser, HeaderParser +from email.generator import Generator, DecodedGenerator +from email.message import Message +from email.mime.application import MIMEApplication +from email.mime.audio import MIMEAudio +from email.mime.text import MIMEText +from email.mime.image import MIMEImage +from email.mime.base import MIMEBase +from email.mime.message import MIMEMessage +from email.mime.multipart import MIMEMultipart +from email import utils +from email import errors +from email import encoders +from email import iterators +from email import base64mime +from email import quoprimime + +from test.test_support import findfile, run_unittest +from email.test import __file__ as landmark + + +NL = '\n' +EMPTYSTRING = '' +SPACE = ' ' + + + +def openfile(filename, *args, **kws): + path = os.path.join(os.path.dirname(landmark), 'data', filename) + return open(path, *args, **kws) + + + +# Base test class +class TestEmailBase(unittest.TestCase): + def ndiffAssertEqual(self, first, second): + """Like failUnlessEqual except use ndiff for readable output.""" + if first != second: + sfirst = str(first) + ssecond = str(second) + rfirst = [repr(line) for line in sfirst.splitlines()] + rsecond = [repr(line) for line in ssecond.splitlines()] + diff = difflib.ndiff(rfirst, rsecond) + raise self.failureException(NL + NL.join(diff)) + + def _msgobj(self, filename): + with openfile(findfile(filename)) as fp: + return email.message_from_file(fp) + + + +# Test various aspects of the Message class's API +class TestMessageAPI(TestEmailBase): + def test_get_all(self): + eq = self.assertEqual + msg = self._msgobj('msg_20.txt') + eq(msg.get_all('cc'), ['ccc@zzz.org', 'ddd@zzz.org', 'eee@zzz.org']) + eq(msg.get_all('xx', 'n/a'), 'n/a') + + def test_getset_charset(self): + eq = self.assertEqual + msg = Message() + eq(msg.get_charset(), None) + charset = Charset('iso-8859-1') + msg.set_charset(charset) + eq(msg['mime-version'], '1.0') + eq(msg.get_content_type(), 'text/plain') + eq(msg['content-type'], 'text/plain; charset="iso-8859-1"') + eq(msg.get_param('charset'), 'iso-8859-1') + eq(msg['content-transfer-encoding'], 'quoted-printable') + eq(msg.get_charset().input_charset, 'iso-8859-1') + # Remove the charset + msg.set_charset(None) + eq(msg.get_charset(), None) + eq(msg['content-type'], 'text/plain') + # Try adding a charset when there's already MIME headers present + msg = Message() + msg['MIME-Version'] = '2.0' + msg['Content-Type'] = 'text/x-weird' + msg['Content-Transfer-Encoding'] = 'quinted-puntable' + msg.set_charset(charset) + eq(msg['mime-version'], '2.0') + eq(msg['content-type'], 'text/x-weird; charset="iso-8859-1"') + eq(msg['content-transfer-encoding'], 'quinted-puntable') + + def test_set_charset_from_string(self): + eq = self.assertEqual + msg = Message() + msg.set_charset('us-ascii') + eq(msg.get_charset().input_charset, 'us-ascii') + eq(msg['content-type'], 'text/plain; charset="us-ascii"') + + def test_set_payload_with_charset(self): + msg = Message() + charset = Charset('iso-8859-1') + msg.set_payload('This is a string payload', charset) + self.assertEqual(msg.get_charset().input_charset, 'iso-8859-1') + + def test_get_charsets(self): + eq = self.assertEqual + + msg = self._msgobj('msg_08.txt') + charsets = msg.get_charsets() + eq(charsets, [None, 'us-ascii', 'iso-8859-1', 'iso-8859-2', 'koi8-r']) + + msg = self._msgobj('msg_09.txt') + charsets = msg.get_charsets('dingbat') + eq(charsets, ['dingbat', 'us-ascii', 'iso-8859-1', 'dingbat', + 'koi8-r']) + + msg = self._msgobj('msg_12.txt') + charsets = msg.get_charsets() + eq(charsets, [None, 'us-ascii', 'iso-8859-1', None, 'iso-8859-2', + 'iso-8859-3', 'us-ascii', 'koi8-r']) + + def test_get_filename(self): + eq = self.assertEqual + + msg = self._msgobj('msg_04.txt') + filenames = [p.get_filename() for p in msg.get_payload()] + eq(filenames, ['msg.txt', 'msg.txt']) + + msg = self._msgobj('msg_07.txt') + subpart = msg.get_payload(1) + eq(subpart.get_filename(), 'dingusfish.gif') + + def test_get_filename_with_name_parameter(self): + eq = self.assertEqual + + msg = self._msgobj('msg_44.txt') + filenames = [p.get_filename() for p in msg.get_payload()] + eq(filenames, ['msg.txt', 'msg.txt']) + + def test_get_boundary(self): + eq = self.assertEqual + msg = self._msgobj('msg_07.txt') + # No quotes! + eq(msg.get_boundary(), 'BOUNDARY') + + def test_set_boundary(self): + eq = self.assertEqual + # This one has no existing boundary parameter, but the Content-Type: + # header appears fifth. + msg = self._msgobj('msg_01.txt') + msg.set_boundary('BOUNDARY') + header, value = msg.items()[4] + eq(header.lower(), 'content-type') + eq(value, 'text/plain; charset="us-ascii"; boundary="BOUNDARY"') + # This one has a Content-Type: header, with a boundary, stuck in the + # middle of its headers. Make sure the order is preserved; it should + # be fifth. + msg = self._msgobj('msg_04.txt') + msg.set_boundary('BOUNDARY') + header, value = msg.items()[4] + eq(header.lower(), 'content-type') + eq(value, 'multipart/mixed; boundary="BOUNDARY"') + # And this one has no Content-Type: header at all. + msg = self._msgobj('msg_03.txt') + self.assertRaises(errors.HeaderParseError, + msg.set_boundary, 'BOUNDARY') + + def test_get_decoded_payload(self): + eq = self.assertEqual + msg = self._msgobj('msg_10.txt') + # The outer message is a multipart + eq(msg.get_payload(decode=True), None) + # Subpart 1 is 7bit encoded + eq(msg.get_payload(0).get_payload(decode=True), + b'This is a 7bit encoded message.\n') + # Subpart 2 is quopri + eq(msg.get_payload(1).get_payload(decode=True), + b'\xa1This is a Quoted Printable encoded message!\n') + # Subpart 3 is base64 + eq(msg.get_payload(2).get_payload(decode=True), + b'This is a Base64 encoded message.') + # Subpart 4 has no Content-Transfer-Encoding: header. + eq(msg.get_payload(3).get_payload(decode=True), + b'This has no Content-Transfer-Encoding: header.\n') + + def test_get_decoded_uu_payload(self): + eq = self.assertEqual + msg = Message() + msg.set_payload('begin 666 -\n+:&5L;&\\@=V]R;&0 \n \nend\n') + for cte in ('x-uuencode', 'uuencode', 'uue', 'x-uue'): + msg['content-transfer-encoding'] = cte + eq(msg.get_payload(decode=True), b'hello world') + # Now try some bogus data + msg.set_payload('foo') + eq(msg.get_payload(decode=True), b'foo') + + def test_decoded_generator(self): + eq = self.assertEqual + msg = self._msgobj('msg_07.txt') + with openfile('msg_17.txt') as fp: + text = fp.read() + s = StringIO() + g = DecodedGenerator(s) + g.flatten(msg) + eq(s.getvalue(), text) + + def test__contains__(self): + msg = Message() + msg['From'] = 'Me' + msg['to'] = 'You' + # Check for case insensitivity + self.failUnless('from' in msg) + self.failUnless('From' in msg) + self.failUnless('FROM' in msg) + self.failUnless('to' in msg) + self.failUnless('To' in msg) + self.failUnless('TO' in msg) + + def test_as_string(self): + eq = self.ndiffAssertEqual + msg = self._msgobj('msg_01.txt') + with openfile('msg_01.txt') as fp: + text = fp.read() + eq(text, str(msg)) + fullrepr = msg.as_string(unixfrom=True) + lines = fullrepr.split('\n') + self.failUnless(lines[0].startswith('From ')) + eq(text, NL.join(lines[1:])) + + def test_bad_param(self): + msg = email.message_from_string("Content-Type: blarg; baz; boo\n") + self.assertEqual(msg.get_param('baz'), '') + + def test_missing_filename(self): + msg = email.message_from_string("From: foo\n") + self.assertEqual(msg.get_filename(), None) + + def test_bogus_filename(self): + msg = email.message_from_string( + "Content-Disposition: blarg; filename\n") + self.assertEqual(msg.get_filename(), '') + + def test_missing_boundary(self): + msg = email.message_from_string("From: foo\n") + self.assertEqual(msg.get_boundary(), None) + + def test_get_params(self): + eq = self.assertEqual + msg = email.message_from_string( + 'X-Header: foo=one; bar=two; baz=three\n') + eq(msg.get_params(header='x-header'), + [('foo', 'one'), ('bar', 'two'), ('baz', 'three')]) + msg = email.message_from_string( + 'X-Header: foo; bar=one; baz=two\n') + eq(msg.get_params(header='x-header'), + [('foo', ''), ('bar', 'one'), ('baz', 'two')]) + eq(msg.get_params(), None) + msg = email.message_from_string( + 'X-Header: foo; bar="one"; baz=two\n') + eq(msg.get_params(header='x-header'), + [('foo', ''), ('bar', 'one'), ('baz', 'two')]) + + def test_get_param_liberal(self): + msg = Message() + msg['Content-Type'] = 'Content-Type: Multipart/mixed; boundary = "CPIMSSMTPC06p5f3tG"' + self.assertEqual(msg.get_param('boundary'), 'CPIMSSMTPC06p5f3tG') + + def test_get_param(self): + eq = self.assertEqual + msg = email.message_from_string( + "X-Header: foo=one; bar=two; baz=three\n") + eq(msg.get_param('bar', header='x-header'), 'two') + eq(msg.get_param('quuz', header='x-header'), None) + eq(msg.get_param('quuz'), None) + msg = email.message_from_string( + 'X-Header: foo; bar="one"; baz=two\n') + eq(msg.get_param('foo', header='x-header'), '') + eq(msg.get_param('bar', header='x-header'), 'one') + eq(msg.get_param('baz', header='x-header'), 'two') + # XXX: We are not RFC-2045 compliant! We cannot parse: + # msg["Content-Type"] = 'text/plain; weird="hey; dolly? [you] @ <\\"home\\">?"' + # msg.get_param("weird") + # yet. + + def test_get_param_funky_continuation_lines(self): + msg = self._msgobj('msg_22.txt') + self.assertEqual(msg.get_payload(1).get_param('name'), 'wibble.JPG') + + def test_get_param_with_semis_in_quotes(self): + msg = email.message_from_string( + 'Content-Type: image/pjpeg; name="Jim&&Jill"\n') + self.assertEqual(msg.get_param('name'), 'Jim&&Jill') + self.assertEqual(msg.get_param('name', unquote=False), + '"Jim&&Jill"') + + def test_field_containment(self): + unless = self.failUnless + msg = email.message_from_string('Header: exists') + unless('header' in msg) + unless('Header' in msg) + unless('HEADER' in msg) + self.failIf('headerx' in msg) + + def test_set_param(self): + eq = self.assertEqual + msg = Message() + msg.set_param('charset', 'iso-2022-jp') + eq(msg.get_param('charset'), 'iso-2022-jp') + msg.set_param('importance', 'high value') + eq(msg.get_param('importance'), 'high value') + eq(msg.get_param('importance', unquote=False), '"high value"') + eq(msg.get_params(), [('text/plain', ''), + ('charset', 'iso-2022-jp'), + ('importance', 'high value')]) + eq(msg.get_params(unquote=False), [('text/plain', ''), + ('charset', '"iso-2022-jp"'), + ('importance', '"high value"')]) + msg.set_param('charset', 'iso-9999-xx', header='X-Jimmy') + eq(msg.get_param('charset', header='X-Jimmy'), 'iso-9999-xx') + + def test_del_param(self): + eq = self.assertEqual + msg = self._msgobj('msg_05.txt') + eq(msg.get_params(), + [('multipart/report', ''), ('report-type', 'delivery-status'), + ('boundary', 'D1690A7AC1.996856090/mail.example.com')]) + old_val = msg.get_param("report-type") + msg.del_param("report-type") + eq(msg.get_params(), + [('multipart/report', ''), + ('boundary', 'D1690A7AC1.996856090/mail.example.com')]) + msg.set_param("report-type", old_val) + eq(msg.get_params(), + [('multipart/report', ''), + ('boundary', 'D1690A7AC1.996856090/mail.example.com'), + ('report-type', old_val)]) + + def test_del_param_on_other_header(self): + msg = Message() + msg.add_header('Content-Disposition', 'attachment', filename='bud.gif') + msg.del_param('filename', 'content-disposition') + self.assertEqual(msg['content-disposition'], 'attachment') + + def test_set_type(self): + eq = self.assertEqual + msg = Message() + self.assertRaises(ValueError, msg.set_type, 'text') + msg.set_type('text/plain') + eq(msg['content-type'], 'text/plain') + msg.set_param('charset', 'us-ascii') + eq(msg['content-type'], 'text/plain; charset="us-ascii"') + msg.set_type('text/html') + eq(msg['content-type'], 'text/html; charset="us-ascii"') + + def test_set_type_on_other_header(self): + msg = Message() + msg['X-Content-Type'] = 'text/plain' + msg.set_type('application/octet-stream', 'X-Content-Type') + self.assertEqual(msg['x-content-type'], 'application/octet-stream') + + def test_get_content_type_missing(self): + msg = Message() + self.assertEqual(msg.get_content_type(), 'text/plain') + + def test_get_content_type_missing_with_default_type(self): + msg = Message() + msg.set_default_type('message/rfc822') + self.assertEqual(msg.get_content_type(), 'message/rfc822') + + def test_get_content_type_from_message_implicit(self): + msg = self._msgobj('msg_30.txt') + self.assertEqual(msg.get_payload(0).get_content_type(), + 'message/rfc822') + + def test_get_content_type_from_message_explicit(self): + msg = self._msgobj('msg_28.txt') + self.assertEqual(msg.get_payload(0).get_content_type(), + 'message/rfc822') + + def test_get_content_type_from_message_text_plain_implicit(self): + msg = self._msgobj('msg_03.txt') + self.assertEqual(msg.get_content_type(), 'text/plain') + + def test_get_content_type_from_message_text_plain_explicit(self): + msg = self._msgobj('msg_01.txt') + self.assertEqual(msg.get_content_type(), 'text/plain') + + def test_get_content_maintype_missing(self): + msg = Message() + self.assertEqual(msg.get_content_maintype(), 'text') + + def test_get_content_maintype_missing_with_default_type(self): + msg = Message() + msg.set_default_type('message/rfc822') + self.assertEqual(msg.get_content_maintype(), 'message') + + def test_get_content_maintype_from_message_implicit(self): + msg = self._msgobj('msg_30.txt') + self.assertEqual(msg.get_payload(0).get_content_maintype(), 'message') + + def test_get_content_maintype_from_message_explicit(self): + msg = self._msgobj('msg_28.txt') + self.assertEqual(msg.get_payload(0).get_content_maintype(), 'message') + + def test_get_content_maintype_from_message_text_plain_implicit(self): + msg = self._msgobj('msg_03.txt') + self.assertEqual(msg.get_content_maintype(), 'text') + + def test_get_content_maintype_from_message_text_plain_explicit(self): + msg = self._msgobj('msg_01.txt') + self.assertEqual(msg.get_content_maintype(), 'text') + + def test_get_content_subtype_missing(self): + msg = Message() + self.assertEqual(msg.get_content_subtype(), 'plain') + + def test_get_content_subtype_missing_with_default_type(self): + msg = Message() + msg.set_default_type('message/rfc822') + self.assertEqual(msg.get_content_subtype(), 'rfc822') + + def test_get_content_subtype_from_message_implicit(self): + msg = self._msgobj('msg_30.txt') + self.assertEqual(msg.get_payload(0).get_content_subtype(), 'rfc822') + + def test_get_content_subtype_from_message_explicit(self): + msg = self._msgobj('msg_28.txt') + self.assertEqual(msg.get_payload(0).get_content_subtype(), 'rfc822') + + def test_get_content_subtype_from_message_text_plain_implicit(self): + msg = self._msgobj('msg_03.txt') + self.assertEqual(msg.get_content_subtype(), 'plain') + + def test_get_content_subtype_from_message_text_plain_explicit(self): + msg = self._msgobj('msg_01.txt') + self.assertEqual(msg.get_content_subtype(), 'plain') + + def test_get_content_maintype_error(self): + msg = Message() + msg['Content-Type'] = 'no-slash-in-this-string' + self.assertEqual(msg.get_content_maintype(), 'text') + + def test_get_content_subtype_error(self): + msg = Message() + msg['Content-Type'] = 'no-slash-in-this-string' + self.assertEqual(msg.get_content_subtype(), 'plain') + + def test_replace_header(self): + eq = self.assertEqual + msg = Message() + msg.add_header('First', 'One') + msg.add_header('Second', 'Two') + msg.add_header('Third', 'Three') + eq(msg.keys(), ['First', 'Second', 'Third']) + eq(msg.values(), ['One', 'Two', 'Three']) + msg.replace_header('Second', 'Twenty') + eq(msg.keys(), ['First', 'Second', 'Third']) + eq(msg.values(), ['One', 'Twenty', 'Three']) + msg.add_header('First', 'Eleven') + msg.replace_header('First', 'One Hundred') + eq(msg.keys(), ['First', 'Second', 'Third', 'First']) + eq(msg.values(), ['One Hundred', 'Twenty', 'Three', 'Eleven']) + self.assertRaises(KeyError, msg.replace_header, 'Fourth', 'Missing') + + def test_broken_base64_payload(self): + x = 'AwDp0P7//y6LwKEAcPa/6Q=9' + msg = Message() + msg['content-type'] = 'audio/x-midi' + msg['content-transfer-encoding'] = 'base64' + msg.set_payload(x) + self.assertEqual(msg.get_payload(decode=True), + bytes(ord(c) for c in x)) + + + +# Test the email.encoders module +class TestEncoders(unittest.TestCase): + def test_encode_empty_payload(self): + eq = self.assertEqual + msg = Message() + msg.set_charset('us-ascii') + eq(msg['content-transfer-encoding'], '7bit') + + def test_default_cte(self): + eq = self.assertEqual + msg = MIMEText('hello world') + eq(msg['content-transfer-encoding'], '7bit') + + def test_default_cte(self): + eq = self.assertEqual + # With no explicit _charset its us-ascii, and all are 7-bit + msg = MIMEText('hello world') + eq(msg['content-transfer-encoding'], '7bit') + # Similar, but with 8-bit data + msg = MIMEText('hello \xf8 world') + eq(msg['content-transfer-encoding'], '8bit') + # And now with a different charset + msg = MIMEText('hello \xf8 world', _charset='iso-8859-1') + eq(msg['content-transfer-encoding'], 'quoted-printable') + + + +# Test long header wrapping +class TestLongHeaders(TestEmailBase): + def test_split_long_continuation(self): + eq = self.ndiffAssertEqual + msg = email.message_from_string("""\ +Subject: bug demonstration +\t12345678911234567892123456789312345678941234567895123456789612345678971234567898112345678911234567892123456789112345678911234567892123456789 +\tmore text + +test +""") + sfp = StringIO() + g = Generator(sfp) + g.flatten(msg) + eq(sfp.getvalue(), """\ +Subject: bug demonstration +\t12345678911234567892123456789312345678941234567895123456789612345678971234567898112345678911234567892123456789112345678911234567892123456789 +\tmore text + +test +""") + + def test_another_long_almost_unsplittable_header(self): + eq = self.ndiffAssertEqual + hstr = """\ +bug demonstration +\t12345678911234567892123456789312345678941234567895123456789612345678971234567898112345678911234567892123456789112345678911234567892123456789 +\tmore text""" + h = Header(hstr, continuation_ws='\t') + eq(h.encode(), """\ +bug demonstration +\t12345678911234567892123456789312345678941234567895123456789612345678971234567898112345678911234567892123456789112345678911234567892123456789 +\tmore text""") + h = Header(hstr.replace('\t', ' ')) + eq(h.encode(), """\ +bug demonstration + 12345678911234567892123456789312345678941234567895123456789612345678971234567898112345678911234567892123456789112345678911234567892123456789 + more text""") + + def test_long_nonstring(self): + eq = self.ndiffAssertEqual + g = Charset("iso-8859-1") + cz = Charset("iso-8859-2") + utf8 = Charset("utf-8") + g_head = (b'Die Mieter treten hier ein werden mit einem Foerderband ' + b'komfortabel den Korridor entlang, an s\xfcdl\xfcndischen ' + b'Wandgem\xe4lden vorbei, gegen die rotierenden Klingen ' + b'bef\xf6rdert. ') + cz_head = (b'Finan\xe8ni metropole se hroutily pod tlakem jejich ' + b'd\xf9vtipu.. ') + utf8_head = ('\u6b63\u78ba\u306b\u8a00\u3046\u3068\u7ffb\u8a33\u306f' + '\u3055\u308c\u3066\u3044\u307e\u305b\u3093\u3002\u4e00' + '\u90e8\u306f\u30c9\u30a4\u30c4\u8a9e\u3067\u3059\u304c' + '\u3001\u3042\u3068\u306f\u3067\u305f\u3089\u3081\u3067' + '\u3059\u3002\u5b9f\u969b\u306b\u306f\u300cWenn ist das ' + 'Nunstuck git und Slotermeyer? Ja! Beiherhund das Oder ' + 'die Flipperwaldt gersput.\u300d\u3068\u8a00\u3063\u3066' + '\u3044\u307e\u3059\u3002') + h = Header(g_head, g, header_name='Subject') + h.append(cz_head, cz) + h.append(utf8_head, utf8) + msg = Message() + msg['Subject'] = h + sfp = StringIO() + g = Generator(sfp) + g.flatten(msg) + eq(sfp.getvalue(), """\ +Subject: =?iso-8859-1?q?Die_Mieter_treten_hier_ein_werden_mit_einem_Foerd?= + =?iso-8859-1?q?erband_komfortabel_den_Korridor_entlang=2C_an_s=FCdl=FCndi?= + =?iso-8859-1?q?schen_Wandgem=E4lden_vorbei=2C_gegen_die_rotierenden_Kling?= + =?iso-8859-1?q?en_bef=F6rdert=2E_?= =?iso-8859-2?q?Finan=E8ni_met?= + =?iso-8859-2?q?ropole_se_hroutily_pod_tlakem_jejich_d=F9vtipu=2E=2E_?= + =?utf-8?b?5q2j56K644Gr6KiA44GG44Go57+76Kiz44Gv44GV44KM44Gm44GE?= + =?utf-8?b?44G+44Gb44KT44CC5LiA6YOo44Gv44OJ44Kk44OE6Kqe44Gn44GZ44GM44CB?= + =?utf-8?b?44GC44Go44Gv44Gn44Gf44KJ44KB44Gn44GZ44CC5a6f6Zqb44Gr44Gv44CM?= + =?utf-8?q?Wenn_ist_das_Nunstuck_git_und_Slotermeyer=3F_Ja!_Beiherhund_das?= + =?utf-8?b?IE9kZXIgZGllIEZsaXBwZXJ3YWxkdCBnZXJzcHV0LuOAjeOBqOiogOOBow==?= + =?utf-8?b?44Gm44GE44G+44GZ44CC?= + +""") + eq(h.encode(), """\ +=?iso-8859-1?q?Die_Mieter_treten_hier_ein_werden_mit_einem_Foerd?= + =?iso-8859-1?q?erband_komfortabel_den_Korridor_entlang=2C_an_s=FCdl=FCndi?= + =?iso-8859-1?q?schen_Wandgem=E4lden_vorbei=2C_gegen_die_rotierenden_Kling?= + =?iso-8859-1?q?en_bef=F6rdert=2E_?= =?iso-8859-2?q?Finan=E8ni_met?= + =?iso-8859-2?q?ropole_se_hroutily_pod_tlakem_jejich_d=F9vtipu=2E=2E_?= + =?utf-8?b?5q2j56K644Gr6KiA44GG44Go57+76Kiz44Gv44GV44KM44Gm44GE?= + =?utf-8?b?44G+44Gb44KT44CC5LiA6YOo44Gv44OJ44Kk44OE6Kqe44Gn44GZ44GM44CB?= + =?utf-8?b?44GC44Go44Gv44Gn44Gf44KJ44KB44Gn44GZ44CC5a6f6Zqb44Gr44Gv44CM?= + =?utf-8?q?Wenn_ist_das_Nunstuck_git_und_Slotermeyer=3F_Ja!_Beiherhund_das?= + =?utf-8?b?IE9kZXIgZGllIEZsaXBwZXJ3YWxkdCBnZXJzcHV0LuOAjeOBqOiogOOBow==?= + =?utf-8?b?44Gm44GE44G+44GZ44CC?=""") + + def test_long_header_encode(self): + eq = self.ndiffAssertEqual + h = Header('wasnipoop; giraffes="very-long-necked-animals"; ' + 'spooge="yummy"; hippos="gargantuan"; marshmallows="gooey"', + header_name='X-Foobar-Spoink-Defrobnit') + eq(h.encode(), '''\ +wasnipoop; giraffes="very-long-necked-animals"; + spooge="yummy"; hippos="gargantuan"; marshmallows="gooey"''') + + def test_long_header_encode_with_tab_continuation_is_just_a_hint(self): + eq = self.ndiffAssertEqual + h = Header('wasnipoop; giraffes="very-long-necked-animals"; ' + 'spooge="yummy"; hippos="gargantuan"; marshmallows="gooey"', + header_name='X-Foobar-Spoink-Defrobnit', + continuation_ws='\t') + eq(h.encode(), '''\ +wasnipoop; giraffes="very-long-necked-animals"; + spooge="yummy"; hippos="gargantuan"; marshmallows="gooey"''') + + def test_long_header_encode_with_tab_continuation(self): + eq = self.ndiffAssertEqual + h = Header('wasnipoop; giraffes="very-long-necked-animals";\t' + 'spooge="yummy"; hippos="gargantuan"; marshmallows="gooey"', + header_name='X-Foobar-Spoink-Defrobnit', + continuation_ws='\t') + eq(h.encode(), '''\ +wasnipoop; giraffes="very-long-necked-animals"; +\tspooge="yummy"; hippos="gargantuan"; marshmallows="gooey"''') + + def test_header_splitter(self): + eq = self.ndiffAssertEqual + msg = MIMEText('') + # It'd be great if we could use add_header() here, but that doesn't + # guarantee an order of the parameters. + msg['X-Foobar-Spoink-Defrobnit'] = ( + 'wasnipoop; giraffes="very-long-necked-animals"; ' + 'spooge="yummy"; hippos="gargantuan"; marshmallows="gooey"') + sfp = StringIO() + g = Generator(sfp) + g.flatten(msg) + eq(sfp.getvalue(), '''\ +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit +X-Foobar-Spoink-Defrobnit: wasnipoop; giraffes="very-long-necked-animals"; + spooge="yummy"; hippos="gargantuan"; marshmallows="gooey" + +''') + + def test_no_semis_header_splitter(self): + eq = self.ndiffAssertEqual + msg = Message() + msg['From'] = 'test@dom.ain' + msg['References'] = SPACE.join('<%d@dom.ain>' % i for i in range(10)) + msg.set_payload('Test') + sfp = StringIO() + g = Generator(sfp) + g.flatten(msg) + eq(sfp.getvalue(), """\ +From: test@dom.ain +References: <0@dom.ain> <1@dom.ain> <2@dom.ain> <3@dom.ain> <4@dom.ain> + <5@dom.ain> <6@dom.ain> <7@dom.ain> <8@dom.ain> <9@dom.ain> + +Test""") + + def test_no_split_long_header(self): + eq = self.ndiffAssertEqual + hstr = 'References: ' + 'x' * 80 + h = Header(hstr, continuation_ws='\t') + eq(h.encode(), """\ +References: xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx""") + + def test_splitting_multiple_long_lines(self): + eq = self.ndiffAssertEqual + hstr = """\ +from babylon.socal-raves.org (localhost [127.0.0.1]); by babylon.socal-raves.org (Postfix) with ESMTP id B570E51B81; for <mailman-admin@babylon.socal-raves.org>; Sat, 2 Feb 2002 17:00:06 -0800 (PST) +\tfrom babylon.socal-raves.org (localhost [127.0.0.1]); by babylon.socal-raves.org (Postfix) with ESMTP id B570E51B81; for <mailman-admin@babylon.socal-raves.org>; Sat, 2 Feb 2002 17:00:06 -0800 (PST) +\tfrom babylon.socal-raves.org (localhost [127.0.0.1]); by babylon.socal-raves.org (Postfix) with ESMTP id B570E51B81; for <mailman-admin@babylon.socal-raves.org>; Sat, 2 Feb 2002 17:00:06 -0800 (PST) +""" + h = Header(hstr, continuation_ws='\t') + eq(h.encode(), """\ +from babylon.socal-raves.org (localhost [127.0.0.1]); + by babylon.socal-raves.org (Postfix) with ESMTP id B570E51B81; + for <mailman-admin@babylon.socal-raves.org>; + Sat, 2 Feb 2002 17:00:06 -0800 (PST) +\tfrom babylon.socal-raves.org (localhost [127.0.0.1]); + by babylon.socal-raves.org (Postfix) with ESMTP id B570E51B81; + for <mailman-admin@babylon.socal-raves.org>; + Sat, 2 Feb 2002 17:00:06 -0800 (PST) +\tfrom babylon.socal-raves.org (localhost [127.0.0.1]); + by babylon.socal-raves.org (Postfix) with ESMTP id B570E51B81; + for <mailman-admin@babylon.socal-raves.org>; + Sat, 2 Feb 2002 17:00:06 -0800 (PST)""") + + def test_splitting_first_line_only_is_long(self): + eq = self.ndiffAssertEqual + hstr = """\ +from modemcable093.139-201-24.que.mc.videotron.ca ([24.201.139.93] helo=cthulhu.gerg.ca) +\tby kronos.mems-exchange.org with esmtp (Exim 4.05) +\tid 17k4h5-00034i-00 +\tfor test@mems-exchange.org; Wed, 28 Aug 2002 11:25:20 -0400""" + h = Header(hstr, maxlinelen=78, header_name='Received', + continuation_ws='\t') + eq(h.encode(), """\ +from modemcable093.139-201-24.que.mc.videotron.ca ([24.201.139.93] + helo=cthulhu.gerg.ca) +\tby kronos.mems-exchange.org with esmtp (Exim 4.05) +\tid 17k4h5-00034i-00 +\tfor test@mems-exchange.org; Wed, 28 Aug 2002 11:25:20 -0400""") + + def test_long_8bit_header(self): + eq = self.ndiffAssertEqual + msg = Message() + h = Header('Britische Regierung gibt', 'iso-8859-1', + header_name='Subject') + h.append('gr\xfcnes Licht f\xfcr Offshore-Windkraftprojekte') + msg['Subject'] = h + eq(msg.as_string(), """\ +Subject: =?iso-8859-1?q?Britische_Regierung_gibt_gr=FCnes_Licht_f=FCr?= + =?iso-8859-1?q?Offshore-Windkraftprojekte?= + +""") + + def test_long_8bit_header_no_charset(self): + eq = self.ndiffAssertEqual + msg = Message() + msg['Reply-To'] = 'Britische Regierung gibt gr\xfcnes Licht f\xfcr Offshore-Windkraftprojekte <a-very-long-address@example.com>' + eq(msg.as_string(), """\ +Reply-To: Britische Regierung gibt gr\xfcnes Licht f\xfcr Offshore-Windkraftprojekte <a-very-long-address@example.com> + +""") + + def test_long_to_header(self): + eq = self.ndiffAssertEqual + to = ('"Someone Test #A" <someone@eecs.umich.edu>,' + '<someone@eecs.umich.edu>,' + '"Someone Test #B" <someone@umich.edu>, ' + '"Someone Test #C" <someone@eecs.umich.edu>, ' + '"Someone Test #D" <someone@eecs.umich.edu>') + msg = Message() + msg['To'] = to + eq(msg.as_string(maxheaderlen=78), '''\ +To: "Someone Test #A" <someone@eecs.umich.edu>, <someone@eecs.umich.edu>, +\t"Someone Test #B" <someone@umich.edu>, +\t"Someone Test #C" <someone@eecs.umich.edu>, +\t"Someone Test #D" <someone@eecs.umich.edu> + +''') + + def test_long_line_after_append(self): + eq = self.ndiffAssertEqual + s = 'This is an example of string which has almost the limit of header length.' + h = Header(s) + h.append('Add another line.') + eq(h.encode(), """\ +This is an example of string which has almost the limit of header length. + Add another line.""") + + def test_shorter_line_with_append(self): + eq = self.ndiffAssertEqual + s = 'This is a shorter line.' + h = Header(s) + h.append('Add another sentence. (Surprise?)') + eq(h.encode(), + 'This is a shorter line. Add another sentence. (Surprise?)') + + def test_long_field_name(self): + eq = self.ndiffAssertEqual + fn = 'X-Very-Very-Very-Long-Header-Name' + gs = "Die Mieter treten hier ein werden mit einem Foerderband komfortabel den Korridor entlang, an s\xfcdl\xfcndischen Wandgem\xe4lden vorbei, gegen die rotierenden Klingen bef\xf6rdert. " + h = Header(gs, 'iso-8859-1', header_name=fn) + # BAW: this seems broken because the first line is too long + eq(h.encode(), """\ +=?iso-8859-1?q?Die_Mieter_treten_hier_?= + =?iso-8859-1?q?ein_werden_mit_einem_Foerderband_komfortabel_den_Korridor_?= + =?iso-8859-1?q?entlang=2C_an_s=FCdl=FCndischen_Wandgem=E4lden_vorbei=2C_g?= + =?iso-8859-1?q?egen_die_rotierenden_Klingen_bef=F6rdert=2E_?=""") + + def test_long_received_header(self): + h = ('from FOO.TLD (vizworld.acl.foo.tld [123.452.678.9]) ' + 'by hrothgar.la.mastaler.com (tmda-ofmipd) with ESMTP; ' + 'Wed, 05 Mar 2003 18:10:18 -0700') + msg = Message() + msg['Received-1'] = Header(h, continuation_ws='\t') + msg['Received-2'] = h + self.ndiffAssertEqual(msg.as_string(maxheaderlen=78), """\ +Received-1: from FOO.TLD (vizworld.acl.foo.tld [123.452.678.9]) by +\throthgar.la.mastaler.com (tmda-ofmipd) with ESMTP; +\tWed, 05 Mar 2003 18:10:18 -0700 +Received-2: from FOO.TLD (vizworld.acl.foo.tld [123.452.678.9]) by +\throthgar.la.mastaler.com (tmda-ofmipd) with ESMTP; +\tWed, 05 Mar 2003 18:10:18 -0700 + +""") + + def test_string_headerinst_eq(self): + h = ('<15975.17901.207240.414604@sgigritzmann1.mathematik.' + 'tu-muenchen.de> (David Bremner\'s message of ' + '"Thu, 6 Mar 2003 13:58:21 +0100")') + msg = Message() + msg['Received-1'] = Header(h, header_name='Received-1', + continuation_ws='\t') + msg['Received-2'] = h + self.ndiffAssertEqual(msg.as_string(maxheaderlen=78), """\ +Received-1: <15975.17901.207240.414604@sgigritzmann1.mathematik.tu-muenchen.de> +\t(David Bremner's message of "Thu, 6 Mar 2003 13:58:21 +0100") +Received-2: <15975.17901.207240.414604@sgigritzmann1.mathematik.tu-muenchen.de> +\t(David Bremner's message of "Thu, 6 Mar 2003 13:58:21 +0100") + +""") + + def test_long_unbreakable_lines_with_continuation(self): + eq = self.ndiffAssertEqual + msg = Message() + t = """\ +iVBORw0KGgoAAAANSUhEUgAAADAAAAAwBAMAAAClLOS0AAAAGFBMVEUAAAAkHiJeRUIcGBi9 + locQDQ4zJykFBAXJfWDjAAACYUlEQVR4nF2TQY/jIAyFc6lydlG5x8Nyp1Y69wj1PN2I5gzp""" + msg['Face-1'] = t + msg['Face-2'] = Header(t, header_name='Face-2') + eq(msg.as_string(maxheaderlen=78), """\ +Face-1: iVBORw0KGgoAAAANSUhEUgAAADAAAAAwBAMAAAClLOS0AAAAGFBMVEUAAAAkHiJeRUIcGBi9 + locQDQ4zJykFBAXJfWDjAAACYUlEQVR4nF2TQY/jIAyFc6lydlG5x8Nyp1Y69wj1PN2I5gzp +Face-2: iVBORw0KGgoAAAANSUhEUgAAADAAAAAwBAMAAAClLOS0AAAAGFBMVEUAAAAkHiJeRUIcGBi9 + locQDQ4zJykFBAXJfWDjAAACYUlEQVR4nF2TQY/jIAyFc6lydlG5x8Nyp1Y69wj1PN2I5gzp + +""") + + def test_another_long_multiline_header(self): + eq = self.ndiffAssertEqual + m = ('Received: from siimage.com ' + '([172.25.1.3]) by zima.siliconimage.com with ' + 'Microsoft SMTPSVC(5.0.2195.4905);' + '\tWed, 16 Oct 2002 07:41:11 -0700') + msg = email.message_from_string(m) + eq(msg.as_string(maxheaderlen=78), '''\ +Received: from siimage.com ([172.25.1.3]) by zima.siliconimage.com with +\tMicrosoft SMTPSVC(5.0.2195.4905); Wed, 16 Oct 2002 07:41:11 -0700 + +''') + + def test_long_lines_with_different_header(self): + eq = self.ndiffAssertEqual + h = ('List-Unsubscribe: ' + '<http://lists.sourceforge.net/lists/listinfo/spamassassin-talk>,' + ' <mailto:spamassassin-talk-request@lists.sourceforge.net' + '?subject=unsubscribe>') + msg = Message() + msg['List'] = h + msg['List'] = Header(h, header_name='List') + eq(msg.as_string(maxheaderlen=78), """\ +List: List-Unsubscribe: <http://lists.sourceforge.net/lists/listinfo/spamassassin-talk>, +\t<mailto:spamassassin-talk-request@lists.sourceforge.net?subject=unsubscribe> +List: List-Unsubscribe: <http://lists.sourceforge.net/lists/listinfo/spamassassin-talk>, + <mailto:spamassassin-talk-request@lists.sourceforge.net?subject=unsubscribe> + +""") + + + +# Test mangling of "From " lines in the body of a message +class TestFromMangling(unittest.TestCase): + def setUp(self): + self.msg = Message() + self.msg['From'] = 'aaa@bbb.org' + self.msg.set_payload("""\ +From the desk of A.A.A.: +Blah blah blah +""") + + def test_mangled_from(self): + s = StringIO() + g = Generator(s, mangle_from_=True) + g.flatten(self.msg) + self.assertEqual(s.getvalue(), """\ +From: aaa@bbb.org + +>From the desk of A.A.A.: +Blah blah blah +""") + + def test_dont_mangle_from(self): + s = StringIO() + g = Generator(s, mangle_from_=False) + g.flatten(self.msg) + self.assertEqual(s.getvalue(), """\ +From: aaa@bbb.org + +From the desk of A.A.A.: +Blah blah blah +""") + + + +# Test the basic MIMEAudio class +class TestMIMEAudio(unittest.TestCase): + def setUp(self): + # Make sure we pick up the audiotest.au that lives in email/test/data. + # In Python, there's an audiotest.au living in Lib/test but that isn't + # included in some binary distros that don't include the test + # package. The trailing empty string on the .join() is significant + # since findfile() will do a dirname(). + datadir = os.path.join(os.path.dirname(landmark), 'data', '') + with open(findfile('audiotest.au', datadir), 'rb') as fp: + self._audiodata = fp.read() + self._au = MIMEAudio(self._audiodata) + + def test_guess_minor_type(self): + self.assertEqual(self._au.get_content_type(), 'audio/basic') + + def test_encoding(self): + payload = self._au.get_payload() + self.assertEqual(base64.decodestring(payload), self._audiodata) + + def test_checkSetMinor(self): + au = MIMEAudio(self._audiodata, 'fish') + self.assertEqual(au.get_content_type(), 'audio/fish') + + def test_add_header(self): + eq = self.assertEqual + unless = self.failUnless + self._au.add_header('Content-Disposition', 'attachment', + filename='audiotest.au') + eq(self._au['content-disposition'], + 'attachment; filename="audiotest.au"') + eq(self._au.get_params(header='content-disposition'), + [('attachment', ''), ('filename', 'audiotest.au')]) + eq(self._au.get_param('filename', header='content-disposition'), + 'audiotest.au') + missing = [] + eq(self._au.get_param('attachment', header='content-disposition'), '') + unless(self._au.get_param('foo', failobj=missing, + header='content-disposition') is missing) + # Try some missing stuff + unless(self._au.get_param('foobar', missing) is missing) + unless(self._au.get_param('attachment', missing, + header='foobar') is missing) + + + +# Test the basic MIMEImage class +class TestMIMEImage(unittest.TestCase): + def setUp(self): + with openfile('PyBanner048.gif', 'rb') as fp: + self._imgdata = fp.read() + self._im = MIMEImage(self._imgdata) + + def test_guess_minor_type(self): + self.assertEqual(self._im.get_content_type(), 'image/gif') + + def test_encoding(self): + payload = self._im.get_payload() + self.assertEqual(base64.decodestring(payload), self._imgdata) + + def test_checkSetMinor(self): + im = MIMEImage(self._imgdata, 'fish') + self.assertEqual(im.get_content_type(), 'image/fish') + + def test_add_header(self): + eq = self.assertEqual + unless = self.failUnless + self._im.add_header('Content-Disposition', 'attachment', + filename='dingusfish.gif') + eq(self._im['content-disposition'], + 'attachment; filename="dingusfish.gif"') + eq(self._im.get_params(header='content-disposition'), + [('attachment', ''), ('filename', 'dingusfish.gif')]) + eq(self._im.get_param('filename', header='content-disposition'), + 'dingusfish.gif') + missing = [] + eq(self._im.get_param('attachment', header='content-disposition'), '') + unless(self._im.get_param('foo', failobj=missing, + header='content-disposition') is missing) + # Try some missing stuff + unless(self._im.get_param('foobar', missing) is missing) + unless(self._im.get_param('attachment', missing, + header='foobar') is missing) + + + +# Test the basic MIMEApplication class +class TestMIMEApplication(unittest.TestCase): + def test_headers(self): + eq = self.assertEqual + msg = MIMEApplication('\xfa\xfb\xfc\xfd\xfe\xff') + eq(msg.get_content_type(), 'application/octet-stream') + eq(msg['content-transfer-encoding'], 'base64') + + def test_body(self): + eq = self.assertEqual + bytes = '\xfa\xfb\xfc\xfd\xfe\xff' + msg = MIMEApplication(bytes) + eq(msg.get_payload(), '+vv8/f7/') + eq(msg.get_payload(decode=True), bytes) + + + +# Test the basic MIMEText class +class TestMIMEText(unittest.TestCase): + def setUp(self): + self._msg = MIMEText('hello there') + + def test_types(self): + eq = self.assertEqual + unless = self.failUnless + eq(self._msg.get_content_type(), 'text/plain') + eq(self._msg.get_param('charset'), 'us-ascii') + missing = [] + unless(self._msg.get_param('foobar', missing) is missing) + unless(self._msg.get_param('charset', missing, header='foobar') + is missing) + + def test_payload(self): + self.assertEqual(self._msg.get_payload(), 'hello there') + self.failUnless(not self._msg.is_multipart()) + + def test_charset(self): + eq = self.assertEqual + msg = MIMEText('hello there', _charset='us-ascii') + eq(msg.get_charset().input_charset, 'us-ascii') + eq(msg['content-type'], 'text/plain; charset="us-ascii"') + + + +# Test complicated multipart/* messages +class TestMultipart(TestEmailBase): + def setUp(self): + with openfile('PyBanner048.gif', 'rb') as fp: + data = fp.read() + container = MIMEBase('multipart', 'mixed', boundary='BOUNDARY') + image = MIMEImage(data, name='dingusfish.gif') + image.add_header('content-disposition', 'attachment', + filename='dingusfish.gif') + intro = MIMEText('''\ +Hi there, + +This is the dingus fish. +''') + container.attach(intro) + container.attach(image) + container['From'] = 'Barry <barry@digicool.com>' + container['To'] = 'Dingus Lovers <cravindogs@cravindogs.com>' + container['Subject'] = 'Here is your dingus fish' + + now = 987809702.54848599 + timetuple = time.localtime(now) + if timetuple[-1] == 0: + tzsecs = time.timezone + else: + tzsecs = time.altzone + if tzsecs > 0: + sign = '-' + else: + sign = '+' + tzoffset = ' %s%04d' % (sign, tzsecs / 36) + container['Date'] = time.strftime( + '%a, %d %b %Y %H:%M:%S', + time.localtime(now)) + tzoffset + self._msg = container + self._im = image + self._txt = intro + + def test_hierarchy(self): + # convenience + eq = self.assertEqual + unless = self.failUnless + raises = self.assertRaises + # tests + m = self._msg + unless(m.is_multipart()) + eq(m.get_content_type(), 'multipart/mixed') + eq(len(m.get_payload()), 2) + raises(IndexError, m.get_payload, 2) + m0 = m.get_payload(0) + m1 = m.get_payload(1) + unless(m0 is self._txt) + unless(m1 is self._im) + eq(m.get_payload(), [m0, m1]) + unless(not m0.is_multipart()) + unless(not m1.is_multipart()) + + def test_empty_multipart_idempotent(self): + text = """\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + + +--BOUNDARY + + +--BOUNDARY-- +""" + msg = Parser().parsestr(text) + self.ndiffAssertEqual(text, msg.as_string()) + + def test_no_parts_in_a_multipart_with_none_epilogue(self): + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.set_boundary('BOUNDARY') + self.ndiffAssertEqual(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + +--BOUNDARY + +--BOUNDARY--''') + + def test_no_parts_in_a_multipart_with_empty_epilogue(self): + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.preamble = '' + outer.epilogue = '' + outer.set_boundary('BOUNDARY') + self.ndiffAssertEqual(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + + +--BOUNDARY + +--BOUNDARY-- +''') + + def test_one_part_in_a_multipart(self): + eq = self.ndiffAssertEqual + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.set_boundary('BOUNDARY') + msg = MIMEText('hello world') + outer.attach(msg) + eq(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +hello world +--BOUNDARY--''') + + def test_seq_parts_in_a_multipart_with_empty_preamble(self): + eq = self.ndiffAssertEqual + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.preamble = '' + msg = MIMEText('hello world') + outer.attach(msg) + outer.set_boundary('BOUNDARY') + eq(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + + +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +hello world +--BOUNDARY--''') + + + def test_seq_parts_in_a_multipart_with_none_preamble(self): + eq = self.ndiffAssertEqual + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.preamble = None + msg = MIMEText('hello world') + outer.attach(msg) + outer.set_boundary('BOUNDARY') + eq(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +hello world +--BOUNDARY--''') + + + def test_seq_parts_in_a_multipart_with_none_epilogue(self): + eq = self.ndiffAssertEqual + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.epilogue = None + msg = MIMEText('hello world') + outer.attach(msg) + outer.set_boundary('BOUNDARY') + eq(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +hello world +--BOUNDARY--''') + + + def test_seq_parts_in_a_multipart_with_empty_epilogue(self): + eq = self.ndiffAssertEqual + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.epilogue = '' + msg = MIMEText('hello world') + outer.attach(msg) + outer.set_boundary('BOUNDARY') + eq(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +hello world +--BOUNDARY-- +''') + + + def test_seq_parts_in_a_multipart_with_nl_epilogue(self): + eq = self.ndiffAssertEqual + outer = MIMEBase('multipart', 'mixed') + outer['Subject'] = 'A subject' + outer['To'] = 'aperson@dom.ain' + outer['From'] = 'bperson@dom.ain' + outer.epilogue = '\n' + msg = MIMEText('hello world') + outer.attach(msg) + outer.set_boundary('BOUNDARY') + eq(outer.as_string(), '''\ +Content-Type: multipart/mixed; boundary="BOUNDARY" +MIME-Version: 1.0 +Subject: A subject +To: aperson@dom.ain +From: bperson@dom.ain + +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +hello world +--BOUNDARY-- + +''') + + def test_message_external_body(self): + eq = self.assertEqual + msg = self._msgobj('msg_36.txt') + eq(len(msg.get_payload()), 2) + msg1 = msg.get_payload(1) + eq(msg1.get_content_type(), 'multipart/alternative') + eq(len(msg1.get_payload()), 2) + for subpart in msg1.get_payload(): + eq(subpart.get_content_type(), 'message/external-body') + eq(len(subpart.get_payload()), 1) + subsubpart = subpart.get_payload(0) + eq(subsubpart.get_content_type(), 'text/plain') + + def test_double_boundary(self): + # msg_37.txt is a multipart that contains two dash-boundary's in a + # row. Our interpretation of RFC 2046 calls for ignoring the second + # and subsequent boundaries. + msg = self._msgobj('msg_37.txt') + self.assertEqual(len(msg.get_payload()), 3) + + def test_nested_inner_contains_outer_boundary(self): + eq = self.ndiffAssertEqual + # msg_38.txt has an inner part that contains outer boundaries. My + # interpretation of RFC 2046 (based on sections 5.1 and 5.1.2) say + # these are illegal and should be interpreted as unterminated inner + # parts. + msg = self._msgobj('msg_38.txt') + sfp = StringIO() + iterators._structure(msg, sfp) + eq(sfp.getvalue(), """\ +multipart/mixed + multipart/mixed + multipart/alternative + text/plain + text/plain + text/plain + text/plain +""") + + def test_nested_with_same_boundary(self): + eq = self.ndiffAssertEqual + # msg 39.txt is similarly evil in that it's got inner parts that use + # the same boundary as outer parts. Again, I believe the way this is + # parsed is closest to the spirit of RFC 2046 + msg = self._msgobj('msg_39.txt') + sfp = StringIO() + iterators._structure(msg, sfp) + eq(sfp.getvalue(), """\ +multipart/mixed + multipart/mixed + multipart/alternative + application/octet-stream + application/octet-stream + text/plain +""") + + def test_boundary_in_non_multipart(self): + msg = self._msgobj('msg_40.txt') + self.assertEqual(msg.as_string(), '''\ +MIME-Version: 1.0 +Content-Type: text/html; boundary="--961284236552522269" + +----961284236552522269 +Content-Type: text/html; +Content-Transfer-Encoding: 7Bit + +<html></html> + +----961284236552522269-- +''') + + def test_boundary_with_leading_space(self): + eq = self.assertEqual + msg = email.message_from_string('''\ +MIME-Version: 1.0 +Content-Type: multipart/mixed; boundary=" XXXX" + +-- XXXX +Content-Type: text/plain + + +-- XXXX +Content-Type: text/plain + +-- XXXX-- +''') + self.failUnless(msg.is_multipart()) + eq(msg.get_boundary(), ' XXXX') + eq(len(msg.get_payload()), 2) + + def test_boundary_without_trailing_newline(self): + m = Parser().parsestr("""\ +Content-Type: multipart/mixed; boundary="===============0012394164==" +MIME-Version: 1.0 + +--===============0012394164== +Content-Type: image/file1.jpg +MIME-Version: 1.0 +Content-Transfer-Encoding: base64 + +YXNkZg== +--===============0012394164==--""") + self.assertEquals(m.get_payload(0).get_payload(), 'YXNkZg==') + + + +# Test some badly formatted messages +class TestNonConformant(TestEmailBase): + def test_parse_missing_minor_type(self): + eq = self.assertEqual + msg = self._msgobj('msg_14.txt') + eq(msg.get_content_type(), 'text/plain') + eq(msg.get_content_maintype(), 'text') + eq(msg.get_content_subtype(), 'plain') + + def test_same_boundary_inner_outer(self): + unless = self.failUnless + msg = self._msgobj('msg_15.txt') + # XXX We can probably eventually do better + inner = msg.get_payload(0) + unless(hasattr(inner, 'defects')) + self.assertEqual(len(inner.defects), 1) + unless(isinstance(inner.defects[0], + errors.StartBoundaryNotFoundDefect)) + + def test_multipart_no_boundary(self): + unless = self.failUnless + msg = self._msgobj('msg_25.txt') + unless(isinstance(msg.get_payload(), str)) + self.assertEqual(len(msg.defects), 2) + unless(isinstance(msg.defects[0], errors.NoBoundaryInMultipartDefect)) + unless(isinstance(msg.defects[1], + errors.MultipartInvariantViolationDefect)) + + def test_invalid_content_type(self): + eq = self.assertEqual + neq = self.ndiffAssertEqual + msg = Message() + # RFC 2045, $5.2 says invalid yields text/plain + msg['Content-Type'] = 'text' + eq(msg.get_content_maintype(), 'text') + eq(msg.get_content_subtype(), 'plain') + eq(msg.get_content_type(), 'text/plain') + # Clear the old value and try something /really/ invalid + del msg['content-type'] + msg['Content-Type'] = 'foo' + eq(msg.get_content_maintype(), 'text') + eq(msg.get_content_subtype(), 'plain') + eq(msg.get_content_type(), 'text/plain') + # Still, make sure that the message is idempotently generated + s = StringIO() + g = Generator(s) + g.flatten(msg) + neq(s.getvalue(), 'Content-Type: foo\n\n') + + def test_no_start_boundary(self): + eq = self.ndiffAssertEqual + msg = self._msgobj('msg_31.txt') + eq(msg.get_payload(), """\ +--BOUNDARY +Content-Type: text/plain + +message 1 + +--BOUNDARY +Content-Type: text/plain + +message 2 + +--BOUNDARY-- +""") + + def test_no_separating_blank_line(self): + eq = self.ndiffAssertEqual + msg = self._msgobj('msg_35.txt') + eq(msg.as_string(), """\ +From: aperson@dom.ain +To: bperson@dom.ain +Subject: here's something interesting + +counter to RFC 2822, there's no separating newline here +""") + + def test_lying_multipart(self): + unless = self.failUnless + msg = self._msgobj('msg_41.txt') + unless(hasattr(msg, 'defects')) + self.assertEqual(len(msg.defects), 2) + unless(isinstance(msg.defects[0], errors.NoBoundaryInMultipartDefect)) + unless(isinstance(msg.defects[1], + errors.MultipartInvariantViolationDefect)) + + def test_missing_start_boundary(self): + outer = self._msgobj('msg_42.txt') + # The message structure is: + # + # multipart/mixed + # text/plain + # message/rfc822 + # multipart/mixed [*] + # + # [*] This message is missing its start boundary + bad = outer.get_payload(1).get_payload(0) + self.assertEqual(len(bad.defects), 1) + self.failUnless(isinstance(bad.defects[0], + errors.StartBoundaryNotFoundDefect)) + + def test_first_line_is_continuation_header(self): + eq = self.assertEqual + m = ' Line 1\nLine 2\nLine 3' + msg = email.message_from_string(m) + eq(msg.keys(), []) + eq(msg.get_payload(), 'Line 2\nLine 3') + eq(len(msg.defects), 1) + self.failUnless(isinstance(msg.defects[0], + errors.FirstHeaderLineIsContinuationDefect)) + eq(msg.defects[0].line, ' Line 1\n') + + + +# Test RFC 2047 header encoding and decoding +class TestRFC2047(unittest.TestCase): + def test_rfc2047_multiline(self): + eq = self.assertEqual + s = """Re: =?mac-iceland?q?r=8Aksm=9Arg=8Cs?= baz + foo bar =?mac-iceland?q?r=8Aksm=9Arg=8Cs?=""" + dh = decode_header(s) + eq(dh, [ + (b'Re:', None), + (b'r\x8aksm\x9arg\x8cs', 'mac-iceland'), + (b'baz foo bar', None), + (b'r\x8aksm\x9arg\x8cs', 'mac-iceland')]) + header = make_header(dh) + eq(str(header), + 'Re: r\xe4ksm\xf6rg\xe5s baz foo bar r\xe4ksm\xf6rg\xe5s') + eq(header.encode(), + """Re: =?mac-iceland?q?r=8Aksm=9Arg=8Cs?= baz foo bar + =?mac-iceland?q?r=8Aksm=9Arg=8Cs?=""") + + def test_whitespace_eater_unicode(self): + eq = self.assertEqual + s = '=?ISO-8859-1?Q?Andr=E9?= Pirard <pirard@dom.ain>' + dh = decode_header(s) + eq(dh, [(b'Andr\xe9', 'iso-8859-1'), + (b'Pirard <pirard@dom.ain>', None)]) + header = str(make_header(dh)) + eq(header, 'Andr\xe9 Pirard <pirard@dom.ain>') + + def test_whitespace_eater_unicode_2(self): + eq = self.assertEqual + s = 'The =?iso-8859-1?b?cXVpY2sgYnJvd24gZm94?= jumped over the =?iso-8859-1?b?bGF6eSBkb2c=?=' + dh = decode_header(s) + eq(dh, [(b'The', None), (b'quick brown fox', 'iso-8859-1'), + (b'jumped over the', None), (b'lazy dog', 'iso-8859-1')]) + hu = str(make_header(dh)) + eq(hu, 'The quick brown fox jumped over the lazy dog') + + def test_rfc2047_missing_whitespace(self): + s = 'Sm=?ISO-8859-1?B?9g==?=rg=?ISO-8859-1?B?5Q==?=sbord' + dh = decode_header(s) + self.assertEqual(dh, [(s, None)]) + + def test_rfc2047_with_whitespace(self): + s = 'Sm =?ISO-8859-1?B?9g==?= rg =?ISO-8859-1?B?5Q==?= sbord' + dh = decode_header(s) + self.assertEqual(dh, [(b'Sm', None), (b'\xf6', 'iso-8859-1'), + (b'rg', None), (b'\xe5', 'iso-8859-1'), + (b'sbord', None)]) + + + +# Test the MIMEMessage class +class TestMIMEMessage(TestEmailBase): + def setUp(self): + with openfile('msg_11.txt') as fp: + self._text = fp.read() + + def test_type_error(self): + self.assertRaises(TypeError, MIMEMessage, 'a plain string') + + def test_valid_argument(self): + eq = self.assertEqual + unless = self.failUnless + subject = 'A sub-message' + m = Message() + m['Subject'] = subject + r = MIMEMessage(m) + eq(r.get_content_type(), 'message/rfc822') + payload = r.get_payload() + unless(isinstance(payload, list)) + eq(len(payload), 1) + subpart = payload[0] + unless(subpart is m) + eq(subpart['subject'], subject) + + def test_bad_multipart(self): + eq = self.assertEqual + msg1 = Message() + msg1['Subject'] = 'subpart 1' + msg2 = Message() + msg2['Subject'] = 'subpart 2' + r = MIMEMessage(msg1) + self.assertRaises(errors.MultipartConversionError, r.attach, msg2) + + def test_generate(self): + # First craft the message to be encapsulated + m = Message() + m['Subject'] = 'An enclosed message' + m.set_payload('Here is the body of the message.\n') + r = MIMEMessage(m) + r['Subject'] = 'The enclosing message' + s = StringIO() + g = Generator(s) + g.flatten(r) + self.assertEqual(s.getvalue(), """\ +Content-Type: message/rfc822 +MIME-Version: 1.0 +Subject: The enclosing message + +Subject: An enclosed message + +Here is the body of the message. +""") + + def test_parse_message_rfc822(self): + eq = self.assertEqual + unless = self.failUnless + msg = self._msgobj('msg_11.txt') + eq(msg.get_content_type(), 'message/rfc822') + payload = msg.get_payload() + unless(isinstance(payload, list)) + eq(len(payload), 1) + submsg = payload[0] + self.failUnless(isinstance(submsg, Message)) + eq(submsg['subject'], 'An enclosed message') + eq(submsg.get_payload(), 'Here is the body of the message.\n') + + def test_dsn(self): + eq = self.assertEqual + unless = self.failUnless + # msg 16 is a Delivery Status Notification, see RFC 1894 + msg = self._msgobj('msg_16.txt') + eq(msg.get_content_type(), 'multipart/report') + unless(msg.is_multipart()) + eq(len(msg.get_payload()), 3) + # Subpart 1 is a text/plain, human readable section + subpart = msg.get_payload(0) + eq(subpart.get_content_type(), 'text/plain') + eq(subpart.get_payload(), """\ +This report relates to a message you sent with the following header fields: + + Message-id: <002001c144a6$8752e060$56104586@oxy.edu> + Date: Sun, 23 Sep 2001 20:10:55 -0700 + From: "Ian T. Henry" <henryi@oxy.edu> + To: SoCal Raves <scr@socal-raves.org> + Subject: [scr] yeah for Ians!! + +Your message cannot be delivered to the following recipients: + + Recipient address: jangel1@cougar.noc.ucla.edu + Reason: recipient reached disk quota + +""") + # Subpart 2 contains the machine parsable DSN information. It + # consists of two blocks of headers, represented by two nested Message + # objects. + subpart = msg.get_payload(1) + eq(subpart.get_content_type(), 'message/delivery-status') + eq(len(subpart.get_payload()), 2) + # message/delivery-status should treat each block as a bunch of + # headers, i.e. a bunch of Message objects. + dsn1 = subpart.get_payload(0) + unless(isinstance(dsn1, Message)) + eq(dsn1['original-envelope-id'], '0GK500B4HD0888@cougar.noc.ucla.edu') + eq(dsn1.get_param('dns', header='reporting-mta'), '') + # Try a missing one <wink> + eq(dsn1.get_param('nsd', header='reporting-mta'), None) + dsn2 = subpart.get_payload(1) + unless(isinstance(dsn2, Message)) + eq(dsn2['action'], 'failed') + eq(dsn2.get_params(header='original-recipient'), + [('rfc822', ''), ('jangel1@cougar.noc.ucla.edu', '')]) + eq(dsn2.get_param('rfc822', header='final-recipient'), '') + # Subpart 3 is the original message + subpart = msg.get_payload(2) + eq(subpart.get_content_type(), 'message/rfc822') + payload = subpart.get_payload() + unless(isinstance(payload, list)) + eq(len(payload), 1) + subsubpart = payload[0] + unless(isinstance(subsubpart, Message)) + eq(subsubpart.get_content_type(), 'text/plain') + eq(subsubpart['message-id'], + '<002001c144a6$8752e060$56104586@oxy.edu>') + + def test_epilogue(self): + eq = self.ndiffAssertEqual + with openfile('msg_21.txt') as fp: + text = fp.read() + msg = Message() + msg['From'] = 'aperson@dom.ain' + msg['To'] = 'bperson@dom.ain' + msg['Subject'] = 'Test' + msg.preamble = 'MIME message' + msg.epilogue = 'End of MIME message\n' + msg1 = MIMEText('One') + msg2 = MIMEText('Two') + msg.add_header('Content-Type', 'multipart/mixed', boundary='BOUNDARY') + msg.attach(msg1) + msg.attach(msg2) + sfp = StringIO() + g = Generator(sfp) + g.flatten(msg) + eq(sfp.getvalue(), text) + + def test_no_nl_preamble(self): + eq = self.ndiffAssertEqual + msg = Message() + msg['From'] = 'aperson@dom.ain' + msg['To'] = 'bperson@dom.ain' + msg['Subject'] = 'Test' + msg.preamble = 'MIME message' + msg.epilogue = '' + msg1 = MIMEText('One') + msg2 = MIMEText('Two') + msg.add_header('Content-Type', 'multipart/mixed', boundary='BOUNDARY') + msg.attach(msg1) + msg.attach(msg2) + eq(msg.as_string(), """\ +From: aperson@dom.ain +To: bperson@dom.ain +Subject: Test +Content-Type: multipart/mixed; boundary="BOUNDARY" + +MIME message +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +One +--BOUNDARY +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +Two +--BOUNDARY-- +""") + + def test_default_type(self): + eq = self.assertEqual + with openfile('msg_30.txt') as fp: + msg = email.message_from_file(fp) + container1 = msg.get_payload(0) + eq(container1.get_default_type(), 'message/rfc822') + eq(container1.get_content_type(), 'message/rfc822') + container2 = msg.get_payload(1) + eq(container2.get_default_type(), 'message/rfc822') + eq(container2.get_content_type(), 'message/rfc822') + container1a = container1.get_payload(0) + eq(container1a.get_default_type(), 'text/plain') + eq(container1a.get_content_type(), 'text/plain') + container2a = container2.get_payload(0) + eq(container2a.get_default_type(), 'text/plain') + eq(container2a.get_content_type(), 'text/plain') + + def test_default_type_with_explicit_container_type(self): + eq = self.assertEqual + with openfile('msg_28.txt') as fp: + msg = email.message_from_file(fp) + container1 = msg.get_payload(0) + eq(container1.get_default_type(), 'message/rfc822') + eq(container1.get_content_type(), 'message/rfc822') + container2 = msg.get_payload(1) + eq(container2.get_default_type(), 'message/rfc822') + eq(container2.get_content_type(), 'message/rfc822') + container1a = container1.get_payload(0) + eq(container1a.get_default_type(), 'text/plain') + eq(container1a.get_content_type(), 'text/plain') + container2a = container2.get_payload(0) + eq(container2a.get_default_type(), 'text/plain') + eq(container2a.get_content_type(), 'text/plain') + + def test_default_type_non_parsed(self): + eq = self.assertEqual + neq = self.ndiffAssertEqual + # Set up container + container = MIMEMultipart('digest', 'BOUNDARY') + container.epilogue = '' + # Set up subparts + subpart1a = MIMEText('message 1\n') + subpart2a = MIMEText('message 2\n') + subpart1 = MIMEMessage(subpart1a) + subpart2 = MIMEMessage(subpart2a) + container.attach(subpart1) + container.attach(subpart2) + eq(subpart1.get_content_type(), 'message/rfc822') + eq(subpart1.get_default_type(), 'message/rfc822') + eq(subpart2.get_content_type(), 'message/rfc822') + eq(subpart2.get_default_type(), 'message/rfc822') + neq(container.as_string(0), '''\ +Content-Type: multipart/digest; boundary="BOUNDARY" +MIME-Version: 1.0 + +--BOUNDARY +Content-Type: message/rfc822 +MIME-Version: 1.0 + +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +message 1 + +--BOUNDARY +Content-Type: message/rfc822 +MIME-Version: 1.0 + +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +message 2 + +--BOUNDARY-- +''') + del subpart1['content-type'] + del subpart1['mime-version'] + del subpart2['content-type'] + del subpart2['mime-version'] + eq(subpart1.get_content_type(), 'message/rfc822') + eq(subpart1.get_default_type(), 'message/rfc822') + eq(subpart2.get_content_type(), 'message/rfc822') + eq(subpart2.get_default_type(), 'message/rfc822') + neq(container.as_string(0), '''\ +Content-Type: multipart/digest; boundary="BOUNDARY" +MIME-Version: 1.0 + +--BOUNDARY + +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +message 1 + +--BOUNDARY + +Content-Type: text/plain; charset="us-ascii" +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit + +message 2 + +--BOUNDARY-- +''') + + def test_mime_attachments_in_constructor(self): + eq = self.assertEqual + text1 = MIMEText('') + text2 = MIMEText('') + msg = MIMEMultipart(_subparts=(text1, text2)) + eq(len(msg.get_payload()), 2) + eq(msg.get_payload(0), text1) + eq(msg.get_payload(1), text2) + + + +# A general test of parser->model->generator idempotency. IOW, read a message +# in, parse it into a message object tree, then without touching the tree, +# regenerate the plain text. The original text and the transformed text +# should be identical. Note: that we ignore the Unix-From since that may +# contain a changed date. +class TestIdempotent(TestEmailBase): + def _msgobj(self, filename): + with openfile(filename) as fp: + data = fp.read() + msg = email.message_from_string(data) + return msg, data + + def _idempotent(self, msg, text): + eq = self.ndiffAssertEqual + s = StringIO() + g = Generator(s, maxheaderlen=0) + g.flatten(msg) + eq(text, s.getvalue()) + + def test_parse_text_message(self): + eq = self.assertEquals + msg, text = self._msgobj('msg_01.txt') + eq(msg.get_content_type(), 'text/plain') + eq(msg.get_content_maintype(), 'text') + eq(msg.get_content_subtype(), 'plain') + eq(msg.get_params()[1], ('charset', 'us-ascii')) + eq(msg.get_param('charset'), 'us-ascii') + eq(msg.preamble, None) + eq(msg.epilogue, None) + self._idempotent(msg, text) + + def test_parse_untyped_message(self): + eq = self.assertEquals + msg, text = self._msgobj('msg_03.txt') + eq(msg.get_content_type(), 'text/plain') + eq(msg.get_params(), None) + eq(msg.get_param('charset'), None) + self._idempotent(msg, text) + + def test_simple_multipart(self): + msg, text = self._msgobj('msg_04.txt') + self._idempotent(msg, text) + + def test_MIME_digest(self): + msg, text = self._msgobj('msg_02.txt') + self._idempotent(msg, text) + + def test_long_header(self): + msg, text = self._msgobj('msg_27.txt') + self._idempotent(msg, text) + + def test_MIME_digest_with_part_headers(self): + msg, text = self._msgobj('msg_28.txt') + self._idempotent(msg, text) + + def test_mixed_with_image(self): + msg, text = self._msgobj('msg_06.txt') + self._idempotent(msg, text) + + def test_multipart_report(self): + msg, text = self._msgobj('msg_05.txt') + self._idempotent(msg, text) + + def test_dsn(self): + msg, text = self._msgobj('msg_16.txt') + self._idempotent(msg, text) + + def test_preamble_epilogue(self): + msg, text = self._msgobj('msg_21.txt') + self._idempotent(msg, text) + + def test_multipart_one_part(self): + msg, text = self._msgobj('msg_23.txt') + self._idempotent(msg, text) + + def test_multipart_no_parts(self): + msg, text = self._msgobj('msg_24.txt') + self._idempotent(msg, text) + + def test_no_start_boundary(self): + msg, text = self._msgobj('msg_31.txt') + self._idempotent(msg, text) + + def test_rfc2231_charset(self): + msg, text = self._msgobj('msg_32.txt') + self._idempotent(msg, text) + + def test_more_rfc2231_parameters(self): + msg, text = self._msgobj('msg_33.txt') + self._idempotent(msg, text) + + def test_text_plain_in_a_multipart_digest(self): + msg, text = self._msgobj('msg_34.txt') + self._idempotent(msg, text) + + def test_nested_multipart_mixeds(self): + msg, text = self._msgobj('msg_12a.txt') + self._idempotent(msg, text) + + def test_message_external_body_idempotent(self): + msg, text = self._msgobj('msg_36.txt') + self._idempotent(msg, text) + + def test_content_type(self): + eq = self.assertEquals + unless = self.failUnless + # Get a message object and reset the seek pointer for other tests + msg, text = self._msgobj('msg_05.txt') + eq(msg.get_content_type(), 'multipart/report') + # Test the Content-Type: parameters + params = {} + for pk, pv in msg.get_params(): + params[pk] = pv + eq(params['report-type'], 'delivery-status') + eq(params['boundary'], 'D1690A7AC1.996856090/mail.example.com') + eq(msg.preamble, 'This is a MIME-encapsulated message.\n') + eq(msg.epilogue, '\n') + eq(len(msg.get_payload()), 3) + # Make sure the subparts are what we expect + msg1 = msg.get_payload(0) + eq(msg1.get_content_type(), 'text/plain') + eq(msg1.get_payload(), 'Yadda yadda yadda\n') + msg2 = msg.get_payload(1) + eq(msg2.get_content_type(), 'text/plain') + eq(msg2.get_payload(), 'Yadda yadda yadda\n') + msg3 = msg.get_payload(2) + eq(msg3.get_content_type(), 'message/rfc822') + self.failUnless(isinstance(msg3, Message)) + payload = msg3.get_payload() + unless(isinstance(payload, list)) + eq(len(payload), 1) + msg4 = payload[0] + unless(isinstance(msg4, Message)) + eq(msg4.get_payload(), 'Yadda yadda yadda\n') + + def test_parser(self): + eq = self.assertEquals + unless = self.failUnless + msg, text = self._msgobj('msg_06.txt') + # Check some of the outer headers + eq(msg.get_content_type(), 'message/rfc822') + # Make sure the payload is a list of exactly one sub-Message, and that + # that submessage has a type of text/plain + payload = msg.get_payload() + unless(isinstance(payload, list)) + eq(len(payload), 1) + msg1 = payload[0] + self.failUnless(isinstance(msg1, Message)) + eq(msg1.get_content_type(), 'text/plain') + self.failUnless(isinstance(msg1.get_payload(), str)) + eq(msg1.get_payload(), '\n') + + + +# Test various other bits of the package's functionality +class TestMiscellaneous(TestEmailBase): + def test_message_from_string(self): + with openfile('msg_01.txt') as fp: + text = fp.read() + msg = email.message_from_string(text) + s = StringIO() + # Don't wrap/continue long headers since we're trying to test + # idempotency. + g = Generator(s, maxheaderlen=0) + g.flatten(msg) + self.assertEqual(text, s.getvalue()) + + def test_message_from_file(self): + with openfile('msg_01.txt') as fp: + text = fp.read() + fp.seek(0) + msg = email.message_from_file(fp) + s = StringIO() + # Don't wrap/continue long headers since we're trying to test + # idempotency. + g = Generator(s, maxheaderlen=0) + g.flatten(msg) + self.assertEqual(text, s.getvalue()) + + def test_message_from_string_with_class(self): + unless = self.failUnless + with openfile('msg_01.txt') as fp: + text = fp.read() + + # Create a subclass + class MyMessage(Message): + pass + + msg = email.message_from_string(text, MyMessage) + unless(isinstance(msg, MyMessage)) + # Try something more complicated + with openfile('msg_02.txt') as fp: + text = fp.read() + msg = email.message_from_string(text, MyMessage) + for subpart in msg.walk(): + unless(isinstance(subpart, MyMessage)) + + def test_message_from_file_with_class(self): + unless = self.failUnless + # Create a subclass + class MyMessage(Message): + pass + + with openfile('msg_01.txt') as fp: + msg = email.message_from_file(fp, MyMessage) + unless(isinstance(msg, MyMessage)) + # Try something more complicated + with openfile('msg_02.txt') as fp: + msg = email.message_from_file(fp, MyMessage) + for subpart in msg.walk(): + unless(isinstance(subpart, MyMessage)) + + def test__all__(self): + module = __import__('email') + # Can't use sorted() here due to Python 2.3 compatibility + all = module.__all__[:] + all.sort() + self.assertEqual(all, [ + 'base64mime', 'charset', 'encoders', 'errors', 'generator', + 'header', 'iterators', 'message', 'message_from_file', + 'message_from_string', 'mime', 'parser', + 'quoprimime', 'utils', + ]) + + def test_formatdate(self): + now = time.time() + self.assertEqual(utils.parsedate(utils.formatdate(now))[:6], + time.gmtime(now)[:6]) + + def test_formatdate_localtime(self): + now = time.time() + self.assertEqual( + utils.parsedate(utils.formatdate(now, localtime=True))[:6], + time.localtime(now)[:6]) + + def test_formatdate_usegmt(self): + now = time.time() + self.assertEqual( + utils.formatdate(now, localtime=False), + time.strftime('%a, %d %b %Y %H:%M:%S -0000', time.gmtime(now))) + self.assertEqual( + utils.formatdate(now, localtime=False, usegmt=True), + time.strftime('%a, %d %b %Y %H:%M:%S GMT', time.gmtime(now))) + + def test_parsedate_none(self): + self.assertEqual(utils.parsedate(''), None) + + def test_parsedate_compact(self): + # The FWS after the comma is optional + self.assertEqual(utils.parsedate('Wed,3 Apr 2002 14:58:26 +0800'), + utils.parsedate('Wed, 3 Apr 2002 14:58:26 +0800')) + + def test_parsedate_no_dayofweek(self): + eq = self.assertEqual + eq(utils.parsedate_tz('25 Feb 2003 13:47:26 -0800'), + (2003, 2, 25, 13, 47, 26, 0, 1, -1, -28800)) + + def test_parsedate_compact_no_dayofweek(self): + eq = self.assertEqual + eq(utils.parsedate_tz('5 Feb 2003 13:47:26 -0800'), + (2003, 2, 5, 13, 47, 26, 0, 1, -1, -28800)) + + def test_parsedate_acceptable_to_time_functions(self): + eq = self.assertEqual + timetup = utils.parsedate('5 Feb 2003 13:47:26 -0800') + t = int(time.mktime(timetup)) + eq(time.localtime(t)[:6], timetup[:6]) + eq(int(time.strftime('%Y', timetup)), 2003) + timetup = utils.parsedate_tz('5 Feb 2003 13:47:26 -0800') + t = int(time.mktime(timetup[:9])) + eq(time.localtime(t)[:6], timetup[:6]) + eq(int(time.strftime('%Y', timetup[:9])), 2003) + + def test_parseaddr_empty(self): + self.assertEqual(utils.parseaddr('<>'), ('', '')) + self.assertEqual(utils.formataddr(utils.parseaddr('<>')), '') + + def test_noquote_dump(self): + self.assertEqual( + utils.formataddr(('A Silly Person', 'person@dom.ain')), + 'A Silly Person <person@dom.ain>') + + def test_escape_dump(self): + self.assertEqual( + utils.formataddr(('A (Very) Silly Person', 'person@dom.ain')), + r'"A \(Very\) Silly Person" <person@dom.ain>') + a = r'A \(Special\) Person' + b = 'person@dom.ain' + self.assertEqual(utils.parseaddr(utils.formataddr((a, b))), (a, b)) + + def test_escape_backslashes(self): + self.assertEqual( + utils.formataddr(('Arthur \Backslash\ Foobar', 'person@dom.ain')), + r'"Arthur \\Backslash\\ Foobar" <person@dom.ain>') + a = r'Arthur \Backslash\ Foobar' + b = 'person@dom.ain' + self.assertEqual(utils.parseaddr(utils.formataddr((a, b))), (a, b)) + + def test_name_with_dot(self): + x = 'John X. Doe <jxd@example.com>' + y = '"John X. Doe" <jxd@example.com>' + a, b = ('John X. Doe', 'jxd@example.com') + self.assertEqual(utils.parseaddr(x), (a, b)) + self.assertEqual(utils.parseaddr(y), (a, b)) + # formataddr() quotes the name if there's a dot in it + self.assertEqual(utils.formataddr((a, b)), y) + + def test_multiline_from_comment(self): + x = """\ +Foo +\tBar <foo@example.com>""" + self.assertEqual(utils.parseaddr(x), ('Foo Bar', 'foo@example.com')) + + def test_quote_dump(self): + self.assertEqual( + utils.formataddr(('A Silly; Person', 'person@dom.ain')), + r'"A Silly; Person" <person@dom.ain>') + + def test_fix_eols(self): + eq = self.assertEqual + eq(utils.fix_eols('hello'), 'hello') + eq(utils.fix_eols('hello\n'), 'hello\r\n') + eq(utils.fix_eols('hello\r'), 'hello\r\n') + eq(utils.fix_eols('hello\r\n'), 'hello\r\n') + eq(utils.fix_eols('hello\n\r'), 'hello\r\n\r\n') + + def test_charset_richcomparisons(self): + eq = self.assertEqual + ne = self.failIfEqual + cset1 = Charset() + cset2 = Charset() + eq(cset1, 'us-ascii') + eq(cset1, 'US-ASCII') + eq(cset1, 'Us-AsCiI') + eq('us-ascii', cset1) + eq('US-ASCII', cset1) + eq('Us-AsCiI', cset1) + ne(cset1, 'usascii') + ne(cset1, 'USASCII') + ne(cset1, 'UsAsCiI') + ne('usascii', cset1) + ne('USASCII', cset1) + ne('UsAsCiI', cset1) + eq(cset1, cset2) + eq(cset2, cset1) + + def test_getaddresses(self): + eq = self.assertEqual + eq(utils.getaddresses(['aperson@dom.ain (Al Person)', + 'Bud Person <bperson@dom.ain>']), + [('Al Person', 'aperson@dom.ain'), + ('Bud Person', 'bperson@dom.ain')]) + + def test_getaddresses_nasty(self): + eq = self.assertEqual + eq(utils.getaddresses(['foo: ;']), [('', '')]) + eq(utils.getaddresses( + ['[]*-- =~$']), + [('', ''), ('', ''), ('', '*--')]) + eq(utils.getaddresses( + ['foo: ;', '"Jason R. Mastaler" <jason@dom.ain>']), + [('', ''), ('Jason R. Mastaler', 'jason@dom.ain')]) + + def test_getaddresses_embedded_comment(self): + """Test proper handling of a nested comment""" + eq = self.assertEqual + addrs = utils.getaddresses(['User ((nested comment)) <foo@bar.com>']) + eq(addrs[0][1], 'foo@bar.com') + + def test_utils_quote_unquote(self): + eq = self.assertEqual + msg = Message() + msg.add_header('content-disposition', 'attachment', + filename='foo\\wacky"name') + eq(msg.get_filename(), 'foo\\wacky"name') + + def test_get_body_encoding_with_bogus_charset(self): + charset = Charset('not a charset') + self.assertEqual(charset.get_body_encoding(), 'base64') + + def test_get_body_encoding_with_uppercase_charset(self): + eq = self.assertEqual + msg = Message() + msg['Content-Type'] = 'text/plain; charset=UTF-8' + eq(msg['content-type'], 'text/plain; charset=UTF-8') + charsets = msg.get_charsets() + eq(len(charsets), 1) + eq(charsets[0], 'utf-8') + charset = Charset(charsets[0]) + eq(charset.get_body_encoding(), 'base64') + msg.set_payload('hello world', charset=charset) + eq(msg.get_payload(), 'aGVsbG8gd29ybGQ=\n') + eq(msg.get_payload(decode=True), b'hello world') + eq(msg['content-transfer-encoding'], 'base64') + # Try another one + msg = Message() + msg['Content-Type'] = 'text/plain; charset="US-ASCII"' + charsets = msg.get_charsets() + eq(len(charsets), 1) + eq(charsets[0], 'us-ascii') + charset = Charset(charsets[0]) + eq(charset.get_body_encoding(), encoders.encode_7or8bit) + msg.set_payload('hello world', charset=charset) + eq(msg.get_payload(), 'hello world') + eq(msg['content-transfer-encoding'], '7bit') + + def test_charsets_case_insensitive(self): + lc = Charset('us-ascii') + uc = Charset('US-ASCII') + self.assertEqual(lc.get_body_encoding(), uc.get_body_encoding()) + + def test_partial_falls_inside_message_delivery_status(self): + eq = self.ndiffAssertEqual + # The Parser interface provides chunks of data to FeedParser in 8192 + # byte gulps. SF bug #1076485 found one of those chunks inside + # message/delivery-status header block, which triggered an + # unreadline() of NeedMoreData. + msg = self._msgobj('msg_43.txt') + sfp = StringIO() + iterators._structure(msg, sfp) + eq(sfp.getvalue(), """\ +multipart/report + text/plain + message/delivery-status + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/plain + text/rfc822-headers +""") + + + +# Test the iterator/generators +class TestIterators(TestEmailBase): + def test_body_line_iterator(self): + eq = self.assertEqual + neq = self.ndiffAssertEqual + # First a simple non-multipart message + msg = self._msgobj('msg_01.txt') + it = iterators.body_line_iterator(msg) + lines = list(it) + eq(len(lines), 6) + neq(EMPTYSTRING.join(lines), msg.get_payload()) + # Now a more complicated multipart + msg = self._msgobj('msg_02.txt') + it = iterators.body_line_iterator(msg) + lines = list(it) + eq(len(lines), 43) + with openfile('msg_19.txt') as fp: + neq(EMPTYSTRING.join(lines), fp.read()) + + def test_typed_subpart_iterator(self): + eq = self.assertEqual + msg = self._msgobj('msg_04.txt') + it = iterators.typed_subpart_iterator(msg, 'text') + lines = [] + subparts = 0 + for subpart in it: + subparts += 1 + lines.append(subpart.get_payload()) + eq(subparts, 2) + eq(EMPTYSTRING.join(lines), """\ +a simple kind of mirror +to reflect upon our own +a simple kind of mirror +to reflect upon our own +""") + + def test_typed_subpart_iterator_default_type(self): + eq = self.assertEqual + msg = self._msgobj('msg_03.txt') + it = iterators.typed_subpart_iterator(msg, 'text', 'plain') + lines = [] + subparts = 0 + for subpart in it: + subparts += 1 + lines.append(subpart.get_payload()) + eq(subparts, 1) + eq(EMPTYSTRING.join(lines), """\ + +Hi, + +Do you like this message? + +-Me +""") + + + +class TestParsers(TestEmailBase): + def test_header_parser(self): + eq = self.assertEqual + # Parse only the headers of a complex multipart MIME document + with openfile('msg_02.txt') as fp: + msg = HeaderParser().parse(fp) + eq(msg['from'], 'ppp-request@zzz.org') + eq(msg['to'], 'ppp@zzz.org') + eq(msg.get_content_type(), 'multipart/mixed') + self.failIf(msg.is_multipart()) + self.failUnless(isinstance(msg.get_payload(), str)) + + def test_whitespace_continuation(self): + eq = self.assertEqual + # This message contains a line after the Subject: header that has only + # whitespace, but it is not empty! + msg = email.message_from_string("""\ +From: aperson@dom.ain +To: bperson@dom.ain +Subject: the next line has a space on it +\x20 +Date: Mon, 8 Apr 2002 15:09:19 -0400 +Message-ID: spam + +Here's the message body +""") + eq(msg['subject'], 'the next line has a space on it\n ') + eq(msg['message-id'], 'spam') + eq(msg.get_payload(), "Here's the message body\n") + + def test_whitespace_continuation_last_header(self): + eq = self.assertEqual + # Like the previous test, but the subject line is the last + # header. + msg = email.message_from_string("""\ +From: aperson@dom.ain +To: bperson@dom.ain +Date: Mon, 8 Apr 2002 15:09:19 -0400 +Message-ID: spam +Subject: the next line has a space on it +\x20 + +Here's the message body +""") + eq(msg['subject'], 'the next line has a space on it\n ') + eq(msg['message-id'], 'spam') + eq(msg.get_payload(), "Here's the message body\n") + + def test_crlf_separation(self): + eq = self.assertEqual + # XXX When Guido fixes TextIOWrapper.read() to act just like + # .readlines(), open this in 'rb' mode with newlines='\n'. + with openfile('msg_26.txt', mode='rb') as fp: + msg = Parser().parse(fp) + eq(len(msg.get_payload()), 2) + part1 = msg.get_payload(0) + eq(part1.get_content_type(), 'text/plain') + eq(part1.get_payload(), 'Simple email with attachment.\r\n\r\n') + part2 = msg.get_payload(1) + eq(part2.get_content_type(), 'application/riscos') + + def test_multipart_digest_with_extra_mime_headers(self): + eq = self.assertEqual + neq = self.ndiffAssertEqual + with openfile('msg_28.txt') as fp: + msg = email.message_from_file(fp) + # Structure is: + # multipart/digest + # message/rfc822 + # text/plain + # message/rfc822 + # text/plain + eq(msg.is_multipart(), 1) + eq(len(msg.get_payload()), 2) + part1 = msg.get_payload(0) + eq(part1.get_content_type(), 'message/rfc822') + eq(part1.is_multipart(), 1) + eq(len(part1.get_payload()), 1) + part1a = part1.get_payload(0) + eq(part1a.is_multipart(), 0) + eq(part1a.get_content_type(), 'text/plain') + neq(part1a.get_payload(), 'message 1\n') + # next message/rfc822 + part2 = msg.get_payload(1) + eq(part2.get_content_type(), 'message/rfc822') + eq(part2.is_multipart(), 1) + eq(len(part2.get_payload()), 1) + part2a = part2.get_payload(0) + eq(part2a.is_multipart(), 0) + eq(part2a.get_content_type(), 'text/plain') + neq(part2a.get_payload(), 'message 2\n') + + def test_three_lines(self): + # A bug report by Andrew McNamara + lines = ['From: Andrew Person <aperson@dom.ain', + 'Subject: Test', + 'Date: Tue, 20 Aug 2002 16:43:45 +1000'] + msg = email.message_from_string(NL.join(lines)) + self.assertEqual(msg['date'], 'Tue, 20 Aug 2002 16:43:45 +1000') + + def test_strip_line_feed_and_carriage_return_in_headers(self): + eq = self.assertEqual + # For [ 1002475 ] email message parser doesn't handle \r\n correctly + value1 = 'text' + value2 = 'more text' + m = 'Header: %s\r\nNext-Header: %s\r\n\r\nBody\r\n\r\n' % ( + value1, value2) + msg = email.message_from_string(m) + eq(msg.get('Header'), value1) + eq(msg.get('Next-Header'), value2) + + def test_rfc2822_header_syntax(self): + eq = self.assertEqual + m = '>From: foo\nFrom: bar\n!"#QUX;~: zoo\n\nbody' + msg = email.message_from_string(m) + eq(len(msg), 3) + eq(sorted(field for field in msg), ['!"#QUX;~', '>From', 'From']) + eq(msg.get_payload(), 'body') + + def test_rfc2822_space_not_allowed_in_header(self): + eq = self.assertEqual + m = '>From foo@example.com 11:25:53\nFrom: bar\n!"#QUX;~: zoo\n\nbody' + msg = email.message_from_string(m) + eq(len(msg.keys()), 0) + + def test_rfc2822_one_character_header(self): + eq = self.assertEqual + m = 'A: first header\nB: second header\nCC: third header\n\nbody' + msg = email.message_from_string(m) + headers = msg.keys() + headers.sort() + eq(headers, ['A', 'B', 'CC']) + eq(msg.get_payload(), 'body') + + + +class TestBase64(unittest.TestCase): + def test_len(self): + eq = self.assertEqual + eq(base64mime.base64_len('hello'), + len(base64mime.encode('hello', eol=''))) + for size in range(15): + if size == 0 : bsize = 0 + elif size <= 3 : bsize = 4 + elif size <= 6 : bsize = 8 + elif size <= 9 : bsize = 12 + elif size <= 12: bsize = 16 + else : bsize = 20 + eq(base64mime.base64_len('x'*size), bsize) + + def test_decode(self): + eq = self.assertEqual + eq(base64mime.decode(''), '') + eq(base64mime.decode('aGVsbG8='), b'hello') + eq(base64mime.decode('aGVsbG8=', 'X'), b'hello') + eq(base64mime.decode('aGVsbG8NCndvcmxk\n', 'X'), b'helloXworld') + + def test_encode(self): + eq = self.assertEqual + eq(base64mime.encode(''), '') + eq(base64mime.encode('hello'), 'aGVsbG8=\n') + # Test the binary flag + eq(base64mime.encode('hello\n'), 'aGVsbG8K\n') + eq(base64mime.encode('hello\n', 0), 'aGVsbG8NCg==\n') + # Test the maxlinelen arg + eq(base64mime.encode('xxxx ' * 20, maxlinelen=40), """\ +eHh4eCB4eHh4IHh4eHggeHh4eCB4eHh4IHh4eHgg +eHh4eCB4eHh4IHh4eHggeHh4eCB4eHh4IHh4eHgg +eHh4eCB4eHh4IHh4eHggeHh4eCB4eHh4IHh4eHgg +eHh4eCB4eHh4IA== +""") + # Test the eol argument + eq(base64mime.encode('xxxx ' * 20, maxlinelen=40, eol='\r\n'), """\ +eHh4eCB4eHh4IHh4eHggeHh4eCB4eHh4IHh4eHgg\r +eHh4eCB4eHh4IHh4eHggeHh4eCB4eHh4IHh4eHgg\r +eHh4eCB4eHh4IHh4eHggeHh4eCB4eHh4IHh4eHgg\r +eHh4eCB4eHh4IA==\r +""") + + def test_header_encode(self): + eq = self.assertEqual + he = base64mime.header_encode + eq(he('hello'), '=?iso-8859-1?b?aGVsbG8=?=') + eq(he('hello\nworld'), '=?iso-8859-1?b?aGVsbG8NCndvcmxk?=') + # Test the charset option + eq(he('hello', charset='iso-8859-2'), '=?iso-8859-2?b?aGVsbG8=?=') + eq(he('hello\nworld'), '=?iso-8859-1?b?aGVsbG8Kd29ybGQ=?=') + # Test the maxlinelen argument + eq(he('xxxx ' * 20, maxlinelen=40), """\ +=?iso-8859-1?b?eHh4eCB4eHh4IHh4eHggeHg=?= + =?iso-8859-1?b?eHggeHh4eCB4eHh4IHh4eHg=?= + =?iso-8859-1?b?IHh4eHggeHh4eCB4eHh4IHg=?= + =?iso-8859-1?b?eHh4IHh4eHggeHh4eCB4eHg=?= + =?iso-8859-1?b?eCB4eHh4IHh4eHggeHh4eCA=?= + =?iso-8859-1?b?eHh4eCB4eHh4IHh4eHgg?=""") + # Test the eol argument + eq(he('xxxx ' * 20, maxlinelen=40, eol='\r\n'), """\ +=?iso-8859-1?b?eHh4eCB4eHh4IHh4eHggeHg=?=\r + =?iso-8859-1?b?eHggeHh4eCB4eHh4IHh4eHg=?=\r + =?iso-8859-1?b?IHh4eHggeHh4eCB4eHh4IHg=?=\r + =?iso-8859-1?b?eHh4IHh4eHggeHh4eCB4eHg=?=\r + =?iso-8859-1?b?eCB4eHh4IHh4eHggeHh4eCA=?=\r + =?iso-8859-1?b?eHh4eCB4eHh4IHh4eHgg?=""") + + + +class TestQuopri(unittest.TestCase): + def setUp(self): + # Set of characters (as byte integers) that don't need to be encoded + # in headers. + self.hlit = list(chain( + range(ord('a'), ord('z') + 1), + range(ord('A'), ord('Z') + 1), + range(ord('0'), ord('9') + 1), + (c for c in b'!*+-/ '))) + # Set of characters (as byte integers) that do need to be encoded in + # headers. + self.hnon = [c for c in range(256) if c not in self.hlit] + assert len(self.hlit) + len(self.hnon) == 256 + # Set of characters (as byte integers) that don't need to be encoded + # in bodies. + self.blit = list(range(ord(' '), ord('~') + 1)) + self.blit.append(ord('\t')) + self.blit.remove(ord('=')) + # Set of characters (as byte integers) that do need to be encoded in + # bodies. + self.bnon = [c for c in range(256) if c not in self.blit] + assert len(self.blit) + len(self.bnon) == 256 + + def test_header_quopri_check(self): + for c in self.hlit: + self.failIf(quoprimime.header_quopri_check(c)) + for c in self.hnon: + self.failUnless(quoprimime.header_quopri_check(c)) + + def test_body_quopri_check(self): + for c in self.blit: + self.failIf(quoprimime.body_quopri_check(c)) + for c in self.bnon: + self.failUnless(quoprimime.body_quopri_check(c)) + + def test_header_quopri_len(self): + eq = self.assertEqual + eq(quoprimime.header_quopri_len(b'hello'), 5) + # RFC 2047 chrome is not included in header_quopri_len(). + eq(len(quoprimime.header_encode(b'hello', charset='xxx')), + quoprimime.header_quopri_len(b'hello') + + # =?xxx?q?...?= means 10 extra characters + 10) + eq(quoprimime.header_quopri_len(b'h@e@l@l@o@'), 20) + # RFC 2047 chrome is not included in header_quopri_len(). + eq(len(quoprimime.header_encode(b'h@e@l@l@o@', charset='xxx')), + quoprimime.header_quopri_len(b'h@e@l@l@o@') + + # =?xxx?q?...?= means 10 extra characters + 10) + for c in self.hlit: + eq(quoprimime.header_quopri_len(bytes([c])), 1, + 'expected length 1 for %r' % chr(c)) + for c in self.hnon: + eq(quoprimime.header_quopri_len(bytes([c])), 3, + 'expected length 3 for %r' % chr(c)) + + def test_body_quopri_len(self): + eq = self.assertEqual + bql = quoprimime.body_quopri_len + for c in self.blit: + eq(bql(c), 1) + for c in self.bnon: + eq(bql(c), 3) + + def test_quote_unquote_idempotent(self): + for x in range(256): + c = chr(x) + self.assertEqual(quoprimime.unquote(quoprimime.quote(c)), c) + + def test_header_encode(self): + eq = self.assertEqual + he = quoprimime.header_encode + eq(he(b'hello'), '=?iso-8859-1?q?hello?=') + eq(he(b'hello', charset='iso-8859-2'), '=?iso-8859-2?q?hello?=') + eq(he(b'hello\nworld'), '=?iso-8859-1?q?hello=0Aworld?=') + # Test a non-ASCII character + eq(he(b'hello\xc7there'), '=?iso-8859-1?q?hello=C7there?=') + + def test_decode(self): + eq = self.assertEqual + eq(quoprimime.decode(''), '') + eq(quoprimime.decode('hello'), 'hello') + eq(quoprimime.decode('hello', 'X'), 'hello') + eq(quoprimime.decode('hello\nworld', 'X'), 'helloXworld') + + def test_encode(self): + eq = self.assertEqual + eq(quoprimime.encode(''), '') + eq(quoprimime.encode('hello'), 'hello') + # Test the binary flag + eq(quoprimime.encode('hello\r\nworld'), 'hello\nworld') + eq(quoprimime.encode('hello\r\nworld', 0), 'hello\nworld') + # Test the maxlinelen arg + eq(quoprimime.encode('xxxx ' * 20, maxlinelen=40), """\ +xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx= + xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxx= +x xxxx xxxx xxxx xxxx=20""") + # Test the eol argument + eq(quoprimime.encode('xxxx ' * 20, maxlinelen=40, eol='\r\n'), """\ +xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx=\r + xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxx=\r +x xxxx xxxx xxxx xxxx=20""") + eq(quoprimime.encode("""\ +one line + +two line"""), """\ +one line + +two line""") + + + +# Test the Charset class +class TestCharset(unittest.TestCase): + def tearDown(self): + from email import charset as CharsetModule + try: + del CharsetModule.CHARSETS['fake'] + except KeyError: + pass + + def test_idempotent(self): + eq = self.assertEqual + # Make sure us-ascii = no Unicode conversion + c = Charset('us-ascii') + s = 'Hello World!' + sp = c.to_splittable(s) + eq(s, c.from_splittable(sp)) + # test 8-bit idempotency with us-ascii + s = '\xa4\xa2\xa4\xa4\xa4\xa6\xa4\xa8\xa4\xaa' + sp = c.to_splittable(s) + eq(s, c.from_splittable(sp)) + + def test_body_encode(self): + eq = self.assertEqual + # Try a charset with QP body encoding + c = Charset('iso-8859-1') + eq('hello w=F6rld', c.body_encode(b'hello w\xf6rld')) + # Try a charset with Base64 body encoding + c = Charset('utf-8') + eq('aGVsbG8gd29ybGQ=\n', c.body_encode(b'hello world')) + # Try a charset with None body encoding + c = Charset('us-ascii') + eq('hello world', c.body_encode(b'hello world')) + # Try the convert argument, where input codec != output codec + c = Charset('euc-jp') + # With apologies to Tokio Kikuchi ;) + try: + eq('\x1b$B5FCO;~IW\x1b(B', + c.body_encode(b'\xb5\xc6\xc3\xcf\xbb\xfe\xc9\xd7')) + eq('\xb5\xc6\xc3\xcf\xbb\xfe\xc9\xd7', + c.body_encode(b'\xb5\xc6\xc3\xcf\xbb\xfe\xc9\xd7', False)) + except LookupError: + # We probably don't have the Japanese codecs installed + pass + # Testing SF bug #625509, which we have to fake, since there are no + # built-in encodings where the header encoding is QP but the body + # encoding is not. + from email import charset as CharsetModule + CharsetModule.add_charset('fake', CharsetModule.QP, None) + c = Charset('fake') + eq('hello w\xf6rld', c.body_encode(b'hello w\xf6rld')) + + def test_unicode_charset_name(self): + charset = Charset('us-ascii') + self.assertEqual(str(charset), 'us-ascii') + self.assertRaises(errors.CharsetError, Charset, 'asc\xffii') + + + +# Test multilingual MIME headers. +class TestHeader(TestEmailBase): + def test_simple(self): + eq = self.ndiffAssertEqual + h = Header('Hello World!') + eq(h.encode(), 'Hello World!') + h.append(' Goodbye World!') + eq(h.encode(), 'Hello World! Goodbye World!') + + def test_simple_surprise(self): + eq = self.ndiffAssertEqual + h = Header('Hello World!') + eq(h.encode(), 'Hello World!') + h.append('Goodbye World!') + eq(h.encode(), 'Hello World! Goodbye World!') + + def test_header_needs_no_decoding(self): + h = 'no decoding needed' + self.assertEqual(decode_header(h), [(h, None)]) + + def test_long(self): + h = Header("I am the very model of a modern Major-General; I've information vegetable, animal, and mineral; I know the kings of England, and I quote the fights historical from Marathon to Waterloo, in order categorical; I'm very well acquainted, too, with matters mathematical; I understand equations, both the simple and quadratical; about binomial theorem I'm teeming with a lot o' news, with many cheerful facts about the square of the hypotenuse.", + maxlinelen=76) + for l in h.encode(splitchars=' ').split('\n '): + self.failUnless(len(l) <= 76) + + def test_multilingual(self): + eq = self.ndiffAssertEqual + g = Charset("iso-8859-1") + cz = Charset("iso-8859-2") + utf8 = Charset("utf-8") + g_head = (b'Die Mieter treten hier ein werden mit einem ' + b'Foerderband komfortabel den Korridor entlang, ' + b'an s\xfcdl\xfcndischen Wandgem\xe4lden vorbei, ' + b'gegen die rotierenden Klingen bef\xf6rdert. ') + cz_head = (b'Finan\xe8ni metropole se hroutily pod tlakem jejich ' + b'd\xf9vtipu.. ') + utf8_head = ('\u6b63\u78ba\u306b\u8a00\u3046\u3068\u7ffb\u8a33\u306f' + '\u3055\u308c\u3066\u3044\u307e\u305b\u3093\u3002\u4e00' + '\u90e8\u306f\u30c9\u30a4\u30c4\u8a9e\u3067\u3059\u304c' + '\u3001\u3042\u3068\u306f\u3067\u305f\u3089\u3081\u3067' + '\u3059\u3002\u5b9f\u969b\u306b\u306f\u300cWenn ist das ' + 'Nunstuck git und Slotermeyer? Ja! Beiherhund das Oder ' + 'die Flipperwaldt gersput.\u300d\u3068\u8a00\u3063\u3066' + '\u3044\u307e\u3059\u3002') + h = Header(g_head, g) + h.append(cz_head, cz) + h.append(utf8_head, utf8) + enc = h.encode() + eq(enc, """\ +=?iso-8859-1?q?Die_Mieter_treten_hier_ein_werden_mit_einem_Foerderband_ko?= + =?iso-8859-1?q?mfortabel_den_Korridor_entlang=2C_an_s=FCdl=FCndischen_Wan?= + =?iso-8859-1?q?dgem=E4lden_vorbei=2C_gegen_die_rotierenden_Klingen_bef=F6?= + =?iso-8859-1?q?rdert=2E_?= =?iso-8859-2?q?Finan=E8ni_metropole_se_hroutily?= + =?iso-8859-2?q?_pod_tlakem_jejich_d=F9vtipu=2E=2E_?= =?utf-8?b?5q2j56K6?= + =?utf-8?b?44Gr6KiA44GG44Go57+76Kiz44Gv44GV44KM44Gm44GE44G+44Gb44KT44CC?= + =?utf-8?b?5LiA6YOo44Gv44OJ44Kk44OE6Kqe44Gn44GZ44GM44CB44GC44Go44Gv44Gn?= + =?utf-8?b?44Gf44KJ44KB44Gn44GZ44CC5a6f6Zqb44Gr44Gv44CMV2VubiBpc3QgZGFz?= + =?utf-8?q?_Nunstuck_git_und_Slotermeyer=3F_Ja!_Beiherhund_das_Oder_die_Fl?= + =?utf-8?b?aXBwZXJ3YWxkdCBnZXJzcHV0LuOAjeOBqOiogOOBo+OBpuOBhOOBvuOBmQ==?= + =?utf-8?b?44CC?=""") + eq(decode_header(enc), + [(g_head, "iso-8859-1"), (cz_head, "iso-8859-2"), + (utf8_head, "utf-8")]) + ustr = str(h) + eq(ustr.encode('utf-8'), + 'Die Mieter treten hier ein werden mit einem Foerderband ' + 'komfortabel den Korridor entlang, an s\xc3\xbcdl\xc3\xbcndischen ' + 'Wandgem\xc3\xa4lden vorbei, gegen die rotierenden Klingen ' + 'bef\xc3\xb6rdert. Finan\xc4\x8dni metropole se hroutily pod ' + 'tlakem jejich d\xc5\xafvtipu.. \xe6\xad\xa3\xe7\xa2\xba\xe3\x81' + '\xab\xe8\xa8\x80\xe3\x81\x86\xe3\x81\xa8\xe7\xbf\xbb\xe8\xa8\xb3' + '\xe3\x81\xaf\xe3\x81\x95\xe3\x82\x8c\xe3\x81\xa6\xe3\x81\x84\xe3' + '\x81\xbe\xe3\x81\x9b\xe3\x82\x93\xe3\x80\x82\xe4\xb8\x80\xe9\x83' + '\xa8\xe3\x81\xaf\xe3\x83\x89\xe3\x82\xa4\xe3\x83\x84\xe8\xaa\x9e' + '\xe3\x81\xa7\xe3\x81\x99\xe3\x81\x8c\xe3\x80\x81\xe3\x81\x82\xe3' + '\x81\xa8\xe3\x81\xaf\xe3\x81\xa7\xe3\x81\x9f\xe3\x82\x89\xe3\x82' + '\x81\xe3\x81\xa7\xe3\x81\x99\xe3\x80\x82\xe5\xae\x9f\xe9\x9a\x9b' + '\xe3\x81\xab\xe3\x81\xaf\xe3\x80\x8cWenn ist das Nunstuck git ' + 'und Slotermeyer? Ja! Beiherhund das Oder die Flipperwaldt ' + 'gersput.\xe3\x80\x8d\xe3\x81\xa8\xe8\xa8\x80\xe3\x81\xa3\xe3\x81' + '\xa6\xe3\x81\x84\xe3\x81\xbe\xe3\x81\x99\xe3\x80\x82') + # Test make_header() + newh = make_header(decode_header(enc)) + eq(newh, enc) + + def test_empty_header_encode(self): + h = Header() + self.assertEqual(h.encode(), '') + + def test_header_ctor_default_args(self): + eq = self.ndiffAssertEqual + h = Header() + eq(h, '') + h.append('foo', Charset('iso-8859-1')) + eq(h, '=?iso-8859-1?q?foo?=') + + def test_explicit_maxlinelen(self): + eq = self.ndiffAssertEqual + hstr = ('A very long line that must get split to something other ' + 'than at the 76th character boundary to test the non-default ' + 'behavior') + h = Header(hstr) + eq(h.encode(), '''\ +A very long line that must get split to something other than at the 76th + character boundary to test the non-default behavior''') + eq(str(h), hstr) + h = Header(hstr, header_name='Subject') + eq(h.encode(), '''\ +A very long line that must get split to something other than at the + 76th character boundary to test the non-default behavior''') + eq(str(h), hstr) + h = Header(hstr, maxlinelen=1024, header_name='Subject') + eq(h.encode(), hstr) + eq(str(h), hstr) + + def test_long_splittables_with_trailing_spaces(self): + eq = self.ndiffAssertEqual + h = Header(charset='iso-8859-1', maxlinelen=20) + h.append('xxxx ' * 20) + eq(h.encode(), """\ +=?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx?= + =?iso-8859-1?q?xxxx_?=""") + h = Header(charset='iso-8859-1', maxlinelen=40) + h.append('xxxx ' * 20) + eq(h.encode(), """\ +=?iso-8859-1?q?xxxx_xxxx_xxxx_xxxx?= + =?iso-8859-1?q?xxxx_xxxx_xxxx_xxxx?= + =?iso-8859-1?q?xxxx_xxxx_xxxx_xxxx?= + =?iso-8859-1?q?xxxx_xxxx_xxxx_xxxx?= + =?iso-8859-1?q?xxxx_xxxx_xxxx_xxxx_?=""") + + def test_us_ascii_header(self): + eq = self.assertEqual + s = 'hello' + x = decode_header(s) + eq(x, [('hello', None)]) + h = make_header(x) + eq(s, h.encode()) + + def test_string_charset(self): + eq = self.assertEqual + h = Header() + h.append('hello', 'iso-8859-1') + eq(h, '=?iso-8859-1?q?hello?=') + +## def test_unicode_error(self): +## raises = self.assertRaises +## raises(UnicodeError, Header, u'[P\xf6stal]', 'us-ascii') +## raises(UnicodeError, Header, '[P\xf6stal]', 'us-ascii') +## h = Header() +## raises(UnicodeError, h.append, u'[P\xf6stal]', 'us-ascii') +## raises(UnicodeError, h.append, '[P\xf6stal]', 'us-ascii') +## raises(UnicodeError, Header, u'\u83ca\u5730\u6642\u592b', 'iso-8859-1') + + def test_utf8_shortest(self): + eq = self.assertEqual + h = Header('p\xf6stal', 'utf-8') + eq(h.encode(), '=?utf-8?q?p=C3=B6stal?=') + h = Header('\u83ca\u5730\u6642\u592b', 'utf-8') + eq(h.encode(), '=?utf-8?b?6I+K5Zyw5pmC5aSr?=') + + def test_bad_8bit_header(self): + raises = self.assertRaises + eq = self.assertEqual + x = b'Ynwp4dUEbay Auction Semiar- No Charge \x96 Earn Big' + raises(UnicodeError, Header, x) + h = Header() + raises(UnicodeError, h.append, x) + e = x.decode('utf-8', 'replace') + eq(str(Header(x, errors='replace')), e) + h.append(x, errors='replace') + eq(str(h), e) + + def test_encoded_adjacent_nonencoded(self): + eq = self.assertEqual + h = Header() + h.append('hello', 'iso-8859-1') + h.append('world') + s = h.encode() + eq(s, '=?iso-8859-1?q?hello?= world') + h = make_header(decode_header(s)) + eq(h.encode(), s) + + def test_whitespace_eater(self): + eq = self.assertEqual + s = 'Subject: =?koi8-r?b?8NLP18XSy8EgzsEgxsnOwczYztk=?= =?koi8-r?q?=CA?= zz.' + parts = decode_header(s) + eq(parts, [(b'Subject:', None), (b'\xf0\xd2\xcf\xd7\xc5\xd2\xcb\xc1 \xce\xc1 \xc6\xc9\xce\xc1\xcc\xd8\xce\xd9\xca', 'koi8-r'), (b'zz.', None)]) + hdr = make_header(parts) + eq(hdr.encode(), + 'Subject: =?koi8-r?b?8NLP18XSy8EgzsEgxsnOwczYztnK?= zz.') + + def test_broken_base64_header(self): + raises = self.assertRaises + s = 'Subject: =?EUC-KR?B?CSixpLDtKSC/7Liuvsax4iC6uLmwMcijIKHaILzSwd/H0SC8+LCjwLsgv7W/+Mj3IQ?=' + raises(errors.HeaderParseError, decode_header, s) + + + +# Test RFC 2231 header parameters (en/de)coding +class TestRFC2231(TestEmailBase): + def test_get_param(self): + eq = self.assertEqual + msg = self._msgobj('msg_29.txt') + eq(msg.get_param('title'), + ('us-ascii', 'en', 'This is even more ***fun*** isn\'t it!')) + eq(msg.get_param('title', unquote=False), + ('us-ascii', 'en', '"This is even more ***fun*** isn\'t it!"')) + + def test_set_param(self): + eq = self.ndiffAssertEqual + msg = Message() + msg.set_param('title', 'This is even more ***fun*** isn\'t it!', + charset='us-ascii') + eq(msg.get_param('title'), + ('us-ascii', '', 'This is even more ***fun*** isn\'t it!')) + msg.set_param('title', 'This is even more ***fun*** isn\'t it!', + charset='us-ascii', language='en') + eq(msg.get_param('title'), + ('us-ascii', 'en', 'This is even more ***fun*** isn\'t it!')) + msg = self._msgobj('msg_01.txt') + msg.set_param('title', 'This is even more ***fun*** isn\'t it!', + charset='us-ascii', language='en') + eq(msg.as_string(maxheaderlen=78), """\ +Return-Path: <bbb@zzz.org> +Delivered-To: bbb@zzz.org +Received: by mail.zzz.org (Postfix, from userid 889) +\tid 27CEAD38CC; Fri, 4 May 2001 14:05:44 -0400 (EDT) +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit +Message-ID: <15090.61304.110929.45684@aaa.zzz.org> +From: bbb@ddd.com (John X. Doe) +To: bbb@zzz.org +Subject: This is a test message +Date: Fri, 4 May 2001 14:05:44 -0400 +Content-Type: text/plain; charset=us-ascii; + title*="us-ascii'en'This%20is%20even%20more%20%2A%2A%2Afun%2A%2A%2A%20isn%27t%20it%21" + + +Hi, + +Do you like this message? + +-Me +""") + + def test_del_param(self): + eq = self.ndiffAssertEqual + msg = self._msgobj('msg_01.txt') + msg.set_param('foo', 'bar', charset='us-ascii', language='en') + msg.set_param('title', 'This is even more ***fun*** isn\'t it!', + charset='us-ascii', language='en') + msg.del_param('foo', header='Content-Type') + eq(msg.as_string(maxheaderlen=78), """\ +Return-Path: <bbb@zzz.org> +Delivered-To: bbb@zzz.org +Received: by mail.zzz.org (Postfix, from userid 889) +\tid 27CEAD38CC; Fri, 4 May 2001 14:05:44 -0400 (EDT) +MIME-Version: 1.0 +Content-Transfer-Encoding: 7bit +Message-ID: <15090.61304.110929.45684@aaa.zzz.org> +From: bbb@ddd.com (John X. Doe) +To: bbb@zzz.org +Subject: This is a test message +Date: Fri, 4 May 2001 14:05:44 -0400 +Content-Type: text/plain; charset="us-ascii"; + title*="us-ascii'en'This%20is%20even%20more%20%2A%2A%2Afun%2A%2A%2A%20isn%27t%20it%21" + + +Hi, + +Do you like this message? + +-Me +""") + + def test_rfc2231_get_content_charset(self): + eq = self.assertEqual + msg = self._msgobj('msg_32.txt') + eq(msg.get_content_charset(), 'us-ascii') + + def test_rfc2231_no_language_or_charset(self): + m = '''\ +Content-Transfer-Encoding: 8bit +Content-Disposition: inline; filename="file____C__DOCUMENTS_20AND_20SETTINGS_FABIEN_LOCAL_20SETTINGS_TEMP_nsmail.htm" +Content-Type: text/html; NAME*0=file____C__DOCUMENTS_20AND_20SETTINGS_FABIEN_LOCAL_20SETTINGS_TEM; NAME*1=P_nsmail.htm + +''' + msg = email.message_from_string(m) + param = msg.get_param('NAME') + self.failIf(isinstance(param, tuple)) + self.assertEqual( + param, + 'file____C__DOCUMENTS_20AND_20SETTINGS_FABIEN_LOCAL_20SETTINGS_TEMP_nsmail.htm') + + def test_rfc2231_no_language_or_charset_in_filename(self): + m = '''\ +Content-Disposition: inline; +\tfilename*0*="''This%20is%20even%20more%20"; +\tfilename*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tfilename*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual(msg.get_filename(), + 'This is even more ***fun*** is it not.pdf') + + def test_rfc2231_no_language_or_charset_in_filename_encoded(self): + m = '''\ +Content-Disposition: inline; +\tfilename*0*="''This%20is%20even%20more%20"; +\tfilename*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tfilename*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual(msg.get_filename(), + 'This is even more ***fun*** is it not.pdf') + + def test_rfc2231_partly_encoded(self): + m = '''\ +Content-Disposition: inline; +\tfilename*0="''This%20is%20even%20more%20"; +\tfilename*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tfilename*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual( + msg.get_filename(), + 'This%20is%20even%20more%20***fun*** is it not.pdf') + + def test_rfc2231_partly_nonencoded(self): + m = '''\ +Content-Disposition: inline; +\tfilename*0="This%20is%20even%20more%20"; +\tfilename*1="%2A%2A%2Afun%2A%2A%2A%20"; +\tfilename*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual( + msg.get_filename(), + 'This%20is%20even%20more%20%2A%2A%2Afun%2A%2A%2A%20is it not.pdf') + + def test_rfc2231_no_language_or_charset_in_boundary(self): + m = '''\ +Content-Type: multipart/alternative; +\tboundary*0*="''This%20is%20even%20more%20"; +\tboundary*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tboundary*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual(msg.get_boundary(), + 'This is even more ***fun*** is it not.pdf') + + def test_rfc2231_no_language_or_charset_in_charset(self): + # This is a nonsensical charset value, but tests the code anyway + m = '''\ +Content-Type: text/plain; +\tcharset*0*="This%20is%20even%20more%20"; +\tcharset*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tcharset*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual(msg.get_content_charset(), + 'this is even more ***fun*** is it not.pdf') + + def test_rfc2231_bad_encoding_in_filename(self): + m = '''\ +Content-Disposition: inline; +\tfilename*0*="bogus'xx'This%20is%20even%20more%20"; +\tfilename*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tfilename*2="is it not.pdf" + +''' + msg = email.message_from_string(m) + self.assertEqual(msg.get_filename(), + 'This is even more ***fun*** is it not.pdf') + + def test_rfc2231_bad_encoding_in_charset(self): + m = """\ +Content-Type: text/plain; charset*=bogus''utf-8%E2%80%9D + +""" + msg = email.message_from_string(m) + # This should return None because non-ascii characters in the charset + # are not allowed. + self.assertEqual(msg.get_content_charset(), None) + + def test_rfc2231_bad_character_in_charset(self): + m = """\ +Content-Type: text/plain; charset*=ascii''utf-8%E2%80%9D + +""" + msg = email.message_from_string(m) + # This should return None because non-ascii characters in the charset + # are not allowed. + self.assertEqual(msg.get_content_charset(), None) + + def test_rfc2231_bad_character_in_filename(self): + m = '''\ +Content-Disposition: inline; +\tfilename*0*="ascii'xx'This%20is%20even%20more%20"; +\tfilename*1*="%2A%2A%2Afun%2A%2A%2A%20"; +\tfilename*2*="is it not.pdf%E2" + +''' + msg = email.message_from_string(m) + self.assertEqual(msg.get_filename(), + 'This is even more ***fun*** is it not.pdf\ufffd') + + def test_rfc2231_unknown_encoding(self): + m = """\ +Content-Transfer-Encoding: 8bit +Content-Disposition: inline; filename*=X-UNKNOWN''myfile.txt + +""" + msg = email.message_from_string(m) + self.assertEqual(msg.get_filename(), 'myfile.txt') + + def test_rfc2231_single_tick_in_filename_extended(self): + eq = self.assertEqual + m = """\ +Content-Type: application/x-foo; +\tname*0*=\"Frank's\"; name*1*=\" Document\" + +""" + msg = email.message_from_string(m) + charset, language, s = msg.get_param('name') + eq(charset, None) + eq(language, None) + eq(s, "Frank's Document") + + def test_rfc2231_single_tick_in_filename(self): + m = """\ +Content-Type: application/x-foo; name*0=\"Frank's\"; name*1=\" Document\" + +""" + msg = email.message_from_string(m) + param = msg.get_param('name') + self.failIf(isinstance(param, tuple)) + self.assertEqual(param, "Frank's Document") + + def test_rfc2231_tick_attack_extended(self): + eq = self.assertEqual + m = """\ +Content-Type: application/x-foo; +\tname*0*=\"us-ascii'en-us'Frank's\"; name*1*=\" Document\" + +""" + msg = email.message_from_string(m) + charset, language, s = msg.get_param('name') + eq(charset, 'us-ascii') + eq(language, 'en-us') + eq(s, "Frank's Document") + + def test_rfc2231_tick_attack(self): + m = """\ +Content-Type: application/x-foo; +\tname*0=\"us-ascii'en-us'Frank's\"; name*1=\" Document\" + +""" + msg = email.message_from_string(m) + param = msg.get_param('name') + self.failIf(isinstance(param, tuple)) + self.assertEqual(param, "us-ascii'en-us'Frank's Document") + + def test_rfc2231_no_extended_values(self): + eq = self.assertEqual + m = """\ +Content-Type: application/x-foo; name=\"Frank's Document\" + +""" + msg = email.message_from_string(m) + eq(msg.get_param('name'), "Frank's Document") + + def test_rfc2231_encoded_then_unencoded_segments(self): + eq = self.assertEqual + m = """\ +Content-Type: application/x-foo; +\tname*0*=\"us-ascii'en-us'My\"; +\tname*1=\" Document\"; +\tname*2*=\" For You\" + +""" + msg = email.message_from_string(m) + charset, language, s = msg.get_param('name') + eq(charset, 'us-ascii') + eq(language, 'en-us') + eq(s, 'My Document For You') + + def test_rfc2231_unencoded_then_encoded_segments(self): + eq = self.assertEqual + m = """\ +Content-Type: application/x-foo; +\tname*0=\"us-ascii'en-us'My\"; +\tname*1*=\" Document\"; +\tname*2*=\" For You\" + +""" + msg = email.message_from_string(m) + charset, language, s = msg.get_param('name') + eq(charset, 'us-ascii') + eq(language, 'en-us') + eq(s, 'My Document For You') + + + +def _testclasses(): + mod = sys.modules[__name__] + return [getattr(mod, name) for name in dir(mod) if name.startswith('Test')] + + +def suite(): + suite = unittest.TestSuite() + for testclass in _testclasses(): + suite.addTest(unittest.makeSuite(testclass)) + return suite + + +def test_main(): + for testclass in _testclasses(): + run_unittest(testclass) + + + +if __name__ == '__main__': + unittest.main(defaultTest='suite') |