2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# Copyright (C) 2001-2006 Python Software Foundation
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# Author: Ben Gertzfield
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# Contact: email-sig@python.org
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								"""Quoted-printable content transfer encoding per RFCs 2045-2047.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								This module handles the content transfer encoding method defined in RFC 2045
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								to encode US ASCII-like 8-bit data called `quoted-printable'.  It is used to
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								safely encode text that is in a character set similar to the 7-bit US ASCII
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								character set, but that includes some 8-bit characters that are normally not
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								allowed in email bodies or headers.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								Quoted-printable is very space-inefficient for encoding binary files; use the
							 | 
						
					
						
							
								
									
										
										
										
											2009-07-12 16:43:19 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								email.base64mime module for that instead.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								This module provides an interface to encode and decode both headers and bodies
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								with quoted-printable encoding.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								RFC 2045 defines a method for including character set information in an
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								`encoded-word' in a header.  This method is commonly used for 8-bit real names
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								in To:/From:/Cc: etc. fields, as well as Subject: lines.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								This module does not do the line wrapping or end-of-line character
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								conversion necessary for proper internationalized headers; it only
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								does dumb encoding and decoding.  To deal with the various line
							 | 
						
					
						
							
								
									
										
										
										
											2009-07-12 16:43:19 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								wrapping issues, use the email.header module.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								"""
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								__all__ = [
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'body_decode',
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'body_encode',
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    'body_length',
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'decode',
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'decodestring',
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'header_decode',
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'header_encode',
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    'header_length',
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'quote',
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    'unquote',
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    ]
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								import re
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								from string import ascii_letters, digits, hexdigits
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								CRLF = '\r\n'
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								NL = '\n'
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								EMPTYSTRING = ''
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								# Build a mapping of octets to the expansion of that octet.  Since we're only
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# going to have 256 of these things, this isn't terribly inefficient
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# space-wise.  Remember that headers and bodies have different sets of safe
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# characters.  Initialize both maps with the full expansion, and then override
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# the safe bytes with the more compact form.
							 | 
						
					
						
							
								
									
										
										
										
											2014-01-13 13:19:21 -05:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								_QUOPRI_MAP = ['=%02X' % c for c in range(256)]
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								_QUOPRI_HEADER_MAP = _QUOPRI_MAP[:]
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								_QUOPRI_BODY_MAP = _QUOPRI_MAP[:]
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								# Safe header bytes which need no encoding.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 14:28:55 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								for c in b'-!*+/' + ascii_letters.encode('ascii') + digits.encode('ascii'):
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    _QUOPRI_HEADER_MAP[c] = chr(c)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# Headers have one other special encoding; spaces become underscores.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								_QUOPRI_HEADER_MAP[ord(' ')] = '_'
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 02:10:49 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								# Safe body bytes which need no encoding.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								for c in (b' !"#$%&\'()*+,-./0123456789:;<>'
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								          b'?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`'
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								          b'abcdefghijklmnopqrstuvwxyz{|}~\t'):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    _QUOPRI_BODY_MAP[c] = chr(c)
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2008-08-19 17:56:33 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# Helpers
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								def header_check(octet):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Return True if the octet should be escaped with header quopri."""
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return chr(octet) != _QUOPRI_HEADER_MAP[octet]
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								def body_check(octet):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Return True if the octet should be escaped with body quopri."""
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return chr(octet) != _QUOPRI_BODY_MAP[octet]
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								def header_length(bytearray):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Return a header quoted-printable encoding length.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    Note that this does not include any RFC 2047 chrome added by
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    `header_encode()`.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    :param bytearray: An array of bytes (a.k.a. octets).
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    :return: The length in bytes of the byte array when it is encoded with
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        quoted-printable for headers.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    return sum(len(_QUOPRI_HEADER_MAP[octet]) for octet in bytearray)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								def body_length(bytearray):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Return a body quoted-printable encoding length.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    :param bytearray: An array of bytes (a.k.a. octets).
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    :return: The length in bytes of the byte array when it is encoded with
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        quoted-printable for bodies.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return sum(len(_QUOPRI_BODY_MAP[octet]) for octet in bytearray)
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def _max_append(L, s, maxlen, extra=''):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if not isinstance(s, str):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        s = chr(s)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if not L:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        L.append(s.lstrip())
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    elif len(L[-1]) + len(s) <= maxlen:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        L[-1] += extra + s
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    else:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        L.append(s.lstrip())
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def unquote(s):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Turn a string in the form =AB to the ASCII character with value 0xab"""
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return chr(int(s[1:3], 16))
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def quote(c):
							 | 
						
					
						
							
								
									
										
										
										
											2014-01-13 13:19:21 -05:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    return _QUOPRI_MAP[ord(c)]
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2008-08-19 17:56:33 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def header_encode(header_bytes, charset='iso-8859-1'):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Encode a single header line with quoted-printable (like) encoding.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    Defined in RFC 2045, this `Q' encoding is similar to quoted-printable, but
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    used specifically for email header fields to allow charsets with mostly 7
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    bit characters (and some 8 bit) to remain more or less readable in non-RFC
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    2045 aware mail clients.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    charset names the character set to use in the RFC 2046 header.  It
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    defaults to iso-8859-1.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """
							 | 
						
					
						
							
								
									
										
										
										
											2011-03-23 15:25:55 -04:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    # Return empty headers as an empty string.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if not header_bytes:
							 | 
						
					
						
							
								
									
										
										
										
											2011-03-23 15:25:55 -04:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								        return ''
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # Iterate over every byte, encoding if necessary.
							 | 
						
					
						
							
								
									
										
										
										
											2014-01-13 13:19:21 -05:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    encoded = header_bytes.decode('latin1').translate(_QUOPRI_HEADER_MAP)
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # Now add the RFC chrome to each encoded chunk and glue the chunks
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # together.
							 | 
						
					
						
							
								
									
										
										
										
											2014-01-13 13:19:21 -05:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    return '=?%s?q?%s?=' % (charset, encoded)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2011-03-24 12:19:26 -04:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2014-01-13 13:19:21 -05:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								_QUOPRI_BODY_ENCODE_MAP = _QUOPRI_BODY_MAP[:]
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								for c in b'\r\n':
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    _QUOPRI_BODY_ENCODE_MAP[c] = chr(c)
							 | 
						
					
						
							
								
									
										
										
										
											2008-08-19 17:56:33 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 03:46:43 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								def body_encode(body, maxlinelen=76, eol=NL):
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Encode with quoted-printable, wrapping at maxlinelen characters.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    Each line of encoded text will end with eol, which defaults to "\\n".  Set
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    this to "\\r\\n" if you will be using the result of this function directly
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    in an email.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2011-03-24 12:19:26 -04:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    Each line will be wrapped at, at most, maxlinelen characters before the
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    eol string (maxlinelen defaults to 76 characters, the maximum value
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    permitted by RFC 2045).  Long lines will have the 'soft line break'
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    quoted-printable character "=" appended to them, so the decoded text will
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    be identical to the original text.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    The minimum maxlinelen is 4 to have room for a quoted character ("=XX")
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    followed by a soft line break.  Smaller values will generate a
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    ValueError.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """
							 | 
						
					
						
							
								
									
										
										
										
											2011-03-24 12:19:26 -04:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if maxlinelen < 4:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        raise ValueError("maxlinelen must be at least 4")
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if not body:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        return body
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2014-01-13 13:19:21 -05:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    # quote speacial characters
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    body = body.translate(_QUOPRI_BODY_ENCODE_MAP)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    soft_break = '=' + eol
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # leave space for the '=' at the end of a line
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    maxlinelen1 = maxlinelen - 1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    encoded_body = []
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    append = encoded_body.append
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    for line in body.splitlines():
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        # break up the line into pieces no longer than maxlinelen - 1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        start = 0
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        laststart = len(line) - 1 - maxlinelen
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        while start <= laststart:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            stop = start + maxlinelen1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            # make sure we don't break up an escape sequence
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            if line[stop - 2] == '=':
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                append(line[start:stop - 1])
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                start = stop - 2
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            elif line[stop - 1] == '=':
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                append(line[start:stop])
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                start = stop - 1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            else:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                append(line[start:stop] + '=')
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                start = stop
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        # handle rest of line, special case if line ends in whitespace
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        if line and line[-1] in ' \t':
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            room = start - laststart
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            if room >= 3:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                # It's a whitespace character at end-of-line, and we have room
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                # for the three-character quoted encoding.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                q = quote(line[-1])
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            elif room == 2:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                # There's room for the whitespace character and a soft break.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                q = line[-1] + soft_break
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            else:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                # There's room only for a soft break.  The quoted whitespace
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                # will be the only content on the subsequent line.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                q = soft_break + quote(line[-1])
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            append(line[start:-1] + q)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        else:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            append(line[start:])
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # add back final newline if present
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if body[-1] in CRLF:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        append('')
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return eol.join(encoded_body)
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2008-08-19 17:56:33 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# BAW: I'm not sure if the intent was for the signature of this function to be
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# the same as base64MIME.decode() or not...
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def decode(encoded, eol=NL):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Decode a quoted-printable string.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    Lines are separated with eol, which defaults to \\n.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    if not encoded:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        return encoded
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # BAW: see comment in encode() above.  Again, we're building up the
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # decoded string with string concatenation, which could be done much more
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # efficiently.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    decoded = ''
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    for line in encoded.splitlines():
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        line = line.rstrip()
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        if not line:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            decoded += eol
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            continue
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        i = 0
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        n = len(line)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        while i < n:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            c = line[i]
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            if c != '=':
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                decoded += c
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                i += 1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            # Otherwise, c == "=".  Are we at the end of the line?  If so, add
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            # a soft line break.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            elif i+1 == n:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                i += 1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                continue
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            # Decode if in form =AB
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            elif i+2 < n and line[i+1] in hexdigits and line[i+2] in hexdigits:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                decoded += unquote(line[i:i+3])
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                i += 3
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            # Otherwise, not in form =AB, pass literally
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            else:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                decoded += c
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                i += 1
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								            if i == n:
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								                decoded += eol
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    # Special case if original string did not end with eol
							 | 
						
					
						
							
								
									
										
										
										
											2011-03-23 15:25:55 -04:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    if encoded[-1] not in '\r\n' and decoded.endswith(eol):
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								        decoded = decoded[:-1]
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return decoded
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# For convenience and backwards compatibility w/ standard base64 module
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								body_decode = decode
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								decodestring = decode
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2008-08-19 17:56:33 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def _unquote_match(match):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Turn a match in the form =AB to the ASCII character with value 0xab"""
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    s = match.group(0)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    return unquote(s)
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								# Header decoding is done a bit differently
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								def header_decode(s):
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """Decode a string encoded with RFC 2045 MIME header `Q' encoding.
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    This function does not parse a full MIME header value encoded with
							 | 
						
					
						
							
								
									
										
										
										
											2016-08-10 05:25:16 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    quoted-printable (like =?iso-8859-1?q?Hello_World?=) -- please use
							 | 
						
					
						
							
								
									
										
										
										
											2009-07-12 16:43:19 +00:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    the high level email.header class for that functionality.
							 | 
						
					
						
							
								
									
										
										
										
											2007-08-30 01:15:14 +00:00
										 
									 
								 
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    """
							 | 
						
					
						
							| 
								
							 | 
							
								
							 | 
							
								
							 | 
							
							
								    s = s.replace('_', ' ')
							 | 
						
					
						
							
								
									
										
										
										
											2013-07-06 17:16:04 +02:00
										 
									 
								 
							 | 
							
								
									
										
									
								
							 | 
							
								
							 | 
							
							
								    return re.sub(r'=[a-fA-F0-9]{2}', _unquote_match, s, flags=re.ASCII)
							 |