1 | """Various tools used by MIME-reading or MIME-writing programs.""" |
---|
2 | |
---|
3 | |
---|
4 | import os |
---|
5 | import sys |
---|
6 | import tempfile |
---|
7 | from warnings import filterwarnings, catch_warnings |
---|
8 | with catch_warnings(): |
---|
9 | if sys.py3kwarning: |
---|
10 | filterwarnings("ignore", ".*rfc822 has been removed", DeprecationWarning) |
---|
11 | import rfc822 |
---|
12 | |
---|
13 | from warnings import warnpy3k |
---|
14 | warnpy3k("in 3.x, mimetools has been removed in favor of the email package", |
---|
15 | stacklevel=2) |
---|
16 | |
---|
17 | __all__ = ["Message","choose_boundary","encode","decode","copyliteral", |
---|
18 | "copybinary"] |
---|
19 | |
---|
20 | class Message(rfc822.Message): |
---|
21 | """A derived class of rfc822.Message that knows about MIME headers and |
---|
22 | contains some hooks for decoding encoded and multipart messages.""" |
---|
23 | |
---|
24 | def __init__(self, fp, seekable = 1): |
---|
25 | rfc822.Message.__init__(self, fp, seekable) |
---|
26 | self.encodingheader = \ |
---|
27 | self.getheader('content-transfer-encoding') |
---|
28 | self.typeheader = \ |
---|
29 | self.getheader('content-type') |
---|
30 | self.parsetype() |
---|
31 | self.parseplist() |
---|
32 | |
---|
33 | def parsetype(self): |
---|
34 | str = self.typeheader |
---|
35 | if str is None: |
---|
36 | str = 'text/plain' |
---|
37 | if ';' in str: |
---|
38 | i = str.index(';') |
---|
39 | self.plisttext = str[i:] |
---|
40 | str = str[:i] |
---|
41 | else: |
---|
42 | self.plisttext = '' |
---|
43 | fields = str.split('/') |
---|
44 | for i in range(len(fields)): |
---|
45 | fields[i] = fields[i].strip().lower() |
---|
46 | self.type = '/'.join(fields) |
---|
47 | self.maintype = fields[0] |
---|
48 | self.subtype = '/'.join(fields[1:]) |
---|
49 | |
---|
50 | def parseplist(self): |
---|
51 | str = self.plisttext |
---|
52 | self.plist = [] |
---|
53 | while str[:1] == ';': |
---|
54 | str = str[1:] |
---|
55 | if ';' in str: |
---|
56 | # XXX Should parse quotes! |
---|
57 | end = str.index(';') |
---|
58 | else: |
---|
59 | end = len(str) |
---|
60 | f = str[:end] |
---|
61 | if '=' in f: |
---|
62 | i = f.index('=') |
---|
63 | f = f[:i].strip().lower() + \ |
---|
64 | '=' + f[i+1:].strip() |
---|
65 | self.plist.append(f.strip()) |
---|
66 | str = str[end:] |
---|
67 | |
---|
68 | def getplist(self): |
---|
69 | return self.plist |
---|
70 | |
---|
71 | def getparam(self, name): |
---|
72 | name = name.lower() + '=' |
---|
73 | n = len(name) |
---|
74 | for p in self.plist: |
---|
75 | if p[:n] == name: |
---|
76 | return rfc822.unquote(p[n:]) |
---|
77 | return None |
---|
78 | |
---|
79 | def getparamnames(self): |
---|
80 | result = [] |
---|
81 | for p in self.plist: |
---|
82 | i = p.find('=') |
---|
83 | if i >= 0: |
---|
84 | result.append(p[:i].lower()) |
---|
85 | return result |
---|
86 | |
---|
87 | def getencoding(self): |
---|
88 | if self.encodingheader is None: |
---|
89 | return '7bit' |
---|
90 | return self.encodingheader.lower() |
---|
91 | |
---|
92 | def gettype(self): |
---|
93 | return self.type |
---|
94 | |
---|
95 | def getmaintype(self): |
---|
96 | return self.maintype |
---|
97 | |
---|
98 | def getsubtype(self): |
---|
99 | return self.subtype |
---|
100 | |
---|
101 | |
---|
102 | |
---|
103 | |
---|
104 | # Utility functions |
---|
105 | # ----------------- |
---|
106 | |
---|
107 | try: |
---|
108 | import thread |
---|
109 | except ImportError: |
---|
110 | import dummy_thread as thread |
---|
111 | _counter_lock = thread.allocate_lock() |
---|
112 | del thread |
---|
113 | |
---|
114 | _counter = 0 |
---|
115 | def _get_next_counter(): |
---|
116 | global _counter |
---|
117 | _counter_lock.acquire() |
---|
118 | _counter += 1 |
---|
119 | result = _counter |
---|
120 | _counter_lock.release() |
---|
121 | return result |
---|
122 | |
---|
123 | _prefix = None |
---|
124 | |
---|
125 | def choose_boundary(): |
---|
126 | """Return a string usable as a multipart boundary. |
---|
127 | |
---|
128 | The string chosen is unique within a single program run, and |
---|
129 | incorporates the user id (if available), process id (if available), |
---|
130 | and current time. So it's very unlikely the returned string appears |
---|
131 | in message text, but there's no guarantee. |
---|
132 | |
---|
133 | The boundary contains dots so you have to quote it in the header.""" |
---|
134 | |
---|
135 | global _prefix |
---|
136 | import time |
---|
137 | if _prefix is None: |
---|
138 | import socket |
---|
139 | try: |
---|
140 | hostid = socket.gethostbyname(socket.gethostname()) |
---|
141 | except socket.gaierror: |
---|
142 | hostid = '127.0.0.1' |
---|
143 | try: |
---|
144 | uid = repr(os.getuid()) |
---|
145 | except AttributeError: |
---|
146 | uid = '1' |
---|
147 | try: |
---|
148 | pid = repr(os.getpid()) |
---|
149 | except AttributeError: |
---|
150 | pid = '1' |
---|
151 | _prefix = hostid + '.' + uid + '.' + pid |
---|
152 | return "%s.%.3f.%d" % (_prefix, time.time(), _get_next_counter()) |
---|
153 | |
---|
154 | |
---|
155 | # Subroutines for decoding some common content-transfer-types |
---|
156 | |
---|
157 | def decode(input, output, encoding): |
---|
158 | """Decode common content-transfer-encodings (base64, quopri, uuencode).""" |
---|
159 | if encoding == 'base64': |
---|
160 | import base64 |
---|
161 | return base64.decode(input, output) |
---|
162 | if encoding == 'quoted-printable': |
---|
163 | import quopri |
---|
164 | return quopri.decode(input, output) |
---|
165 | if encoding in ('uuencode', 'x-uuencode', 'uue', 'x-uue'): |
---|
166 | import uu |
---|
167 | return uu.decode(input, output) |
---|
168 | if encoding in ('7bit', '8bit'): |
---|
169 | return output.write(input.read()) |
---|
170 | if encoding in decodetab: |
---|
171 | pipethrough(input, decodetab[encoding], output) |
---|
172 | else: |
---|
173 | raise ValueError, \ |
---|
174 | 'unknown Content-Transfer-Encoding: %s' % encoding |
---|
175 | |
---|
176 | def encode(input, output, encoding): |
---|
177 | """Encode common content-transfer-encodings (base64, quopri, uuencode).""" |
---|
178 | if encoding == 'base64': |
---|
179 | import base64 |
---|
180 | return base64.encode(input, output) |
---|
181 | if encoding == 'quoted-printable': |
---|
182 | import quopri |
---|
183 | return quopri.encode(input, output, 0) |
---|
184 | if encoding in ('uuencode', 'x-uuencode', 'uue', 'x-uue'): |
---|
185 | import uu |
---|
186 | return uu.encode(input, output) |
---|
187 | if encoding in ('7bit', '8bit'): |
---|
188 | return output.write(input.read()) |
---|
189 | if encoding in encodetab: |
---|
190 | pipethrough(input, encodetab[encoding], output) |
---|
191 | else: |
---|
192 | raise ValueError, \ |
---|
193 | 'unknown Content-Transfer-Encoding: %s' % encoding |
---|
194 | |
---|
195 | # The following is no longer used for standard encodings |
---|
196 | |
---|
197 | # XXX This requires that uudecode and mmencode are in $PATH |
---|
198 | |
---|
199 | uudecode_pipe = '''( |
---|
200 | TEMP=/tmp/@uu.$$ |
---|
201 | sed "s%^begin [0-7][0-7]* .*%begin 600 $TEMP%" | uudecode |
---|
202 | cat $TEMP |
---|
203 | rm $TEMP |
---|
204 | )''' |
---|
205 | |
---|
206 | decodetab = { |
---|
207 | 'uuencode': uudecode_pipe, |
---|
208 | 'x-uuencode': uudecode_pipe, |
---|
209 | 'uue': uudecode_pipe, |
---|
210 | 'x-uue': uudecode_pipe, |
---|
211 | 'quoted-printable': 'mmencode -u -q', |
---|
212 | 'base64': 'mmencode -u -b', |
---|
213 | } |
---|
214 | |
---|
215 | encodetab = { |
---|
216 | 'x-uuencode': 'uuencode tempfile', |
---|
217 | 'uuencode': 'uuencode tempfile', |
---|
218 | 'x-uue': 'uuencode tempfile', |
---|
219 | 'uue': 'uuencode tempfile', |
---|
220 | 'quoted-printable': 'mmencode -q', |
---|
221 | 'base64': 'mmencode -b', |
---|
222 | } |
---|
223 | |
---|
224 | def pipeto(input, command): |
---|
225 | pipe = os.popen(command, 'w') |
---|
226 | copyliteral(input, pipe) |
---|
227 | pipe.close() |
---|
228 | |
---|
229 | def pipethrough(input, command, output): |
---|
230 | (fd, tempname) = tempfile.mkstemp() |
---|
231 | temp = os.fdopen(fd, 'w') |
---|
232 | copyliteral(input, temp) |
---|
233 | temp.close() |
---|
234 | pipe = os.popen(command + ' <' + tempname, 'r') |
---|
235 | copybinary(pipe, output) |
---|
236 | pipe.close() |
---|
237 | os.unlink(tempname) |
---|
238 | |
---|
239 | def copyliteral(input, output): |
---|
240 | while 1: |
---|
241 | line = input.readline() |
---|
242 | if not line: break |
---|
243 | output.write(line) |
---|
244 | |
---|
245 | def copybinary(input, output): |
---|
246 | BUFSIZE = 8192 |
---|
247 | while 1: |
---|
248 | line = input.read(BUFSIZE) |
---|
249 | if not line: break |
---|
250 | output.write(line) |
---|