cpython/Lib/mimify.py

446 lines
12 KiB
Python
Raw Normal View History

#! /usr/bin/env python
'''Mimification and unmimification of mail messages.
decode quoted-printable parts of a mail message or encode using
quoted-printable.
Usage:
mimify(input, output)
unmimify(input, output)
to encode and decode respectively. Input and output may be the name
of a file or an open file object. Only a readline() method is used
on the input file, only a write() method is used on the output file.
When using file names, the input and output file names may be the
same.
Interactive usage:
mimify.py -e [infile [outfile]]
mimify.py -d [infile [outfile]]
to encode and decode respectively. Infile defaults to standard
input and outfile to standard output.
'''
# Configure
MAXLEN = 200 # if lines longer than this, encode as quoted-printable
CHARSET = 'ISO-8859-1' # default charset for non-US-ASCII mail
QUOTE = '> ' # string replies are quoted with
# End configure
import regex, regsub, string
1996-06-10 16:04:02 -03:00
qp = regex.compile('^content-transfer-encoding:[ \t]*quoted-printable',
regex.casefold)
1997-04-11 12:22:56 -03:00
base64_re = regex.compile('^content-transfer-encoding:[ \t]*base64',
1996-06-10 16:04:02 -03:00
regex.casefold)
mp = regex.compile('^content-type:[\000-\377]*multipart/[\000-\377]*boundary="?\\([^;"\n]*\\)',
regex.casefold)
chrset = regex.compile('^\\(content-type:.*charset="\\)\\(us-ascii\\|iso-8859-[0-9]+\\)\\("[\000-\377]*\\)',
regex.casefold)
he = regex.compile('^-*$')
mime_code = regex.compile('=\\([0-9a-f][0-9a-f]\\)', regex.casefold)
mime_head = regex.compile('=\\?iso-8859-1\\?q\\?\\([^?]+\\)\\?=',
regex.casefold)
repl = regex.compile('^subject:[ \t]+re: ', regex.casefold)
class File:
'''A simple fake file object that knows about limited
read-ahead and boundaries.
The only supported method is readline().'''
def __init__(self, file, boundary):
self.file = file
self.boundary = boundary
self.peek = None
def readline(self):
if self.peek is not None:
return ''
line = self.file.readline()
if not line:
return line
if self.boundary:
if line == self.boundary + '\n':
self.peek = line
return ''
if line == self.boundary + '--\n':
self.peek = line
return ''
return line
class HeaderFile:
def __init__(self, file):
self.file = file
self.peek = None
def readline(self):
if self.peek is not None:
line = self.peek
self.peek = None
else:
line = self.file.readline()
if not line:
return line
if he.match(line) >= 0:
return line
while 1:
self.peek = self.file.readline()
if len(self.peek) == 0 or \
(self.peek[0] != ' ' and self.peek[0] != '\t'):
return line
line = line + self.peek
self.peek = None
def mime_decode(line):
'''Decode a single line of quoted-printable text to 8bit.'''
newline = ''
while 1:
i = mime_code.search(line)
if i < 0:
break
newline = newline + line[:i] + \
chr(string.atoi(mime_code.group(1), 16))
line = line[i+3:]
return newline + line
def mime_decode_header(line):
'''Decode a header line to 8bit.'''
newline = ''
while 1:
i = mime_head.search(line)
if i < 0:
break
match = mime_head.group(0, 1)
newline = newline + line[:i] + mime_decode(match[1])
line = line[i + len(match[0]):]
return newline + line
1997-04-11 12:22:56 -03:00
def unmimify_part(ifile, ofile, decode_base64 = 0):
'''Convert a quoted-printable part of a MIME mail message to 8bit.'''
multipart = None
quoted_printable = 0
1997-04-11 12:22:56 -03:00
is_base64 = 0
is_repl = 0
if ifile.boundary and ifile.boundary[:2] == QUOTE:
prefix = QUOTE
else:
prefix = ''
# read header
hfile = HeaderFile(ifile)
while 1:
line = hfile.readline()
if not line:
return
if prefix and line[:len(prefix)] == prefix:
line = line[len(prefix):]
pref = prefix
else:
pref = ''
line = mime_decode_header(line)
if qp.match(line) >= 0:
quoted_printable = 1
continue # skip this header
1997-04-11 12:22:56 -03:00
if decode_base64 and base64_re.match(line) >= 0:
is_base64 = 1
continue
ofile.write(pref + line)
if not prefix and repl.match(line) >= 0:
# we're dealing with a reply message
is_repl = 1
if mp.match(line) >= 0:
multipart = '--' + mp.group(1)
if he.match(line) >= 0:
break
if is_repl and (quoted_printable or multipart):
is_repl = 0
# read body
while 1:
line = ifile.readline()
if not line:
return
line = regsub.gsub(mime_head, '\\1', line)
if prefix and line[:len(prefix)] == prefix:
line = line[len(prefix):]
pref = prefix
else:
pref = ''
## if is_repl and len(line) >= 4 and line[:4] == QUOTE+'--' and line[-3:] != '--\n':
## multipart = line[:-1]
while multipart:
if line == multipart + '--\n':
ofile.write(pref + line)
multipart = None
line = None
break
if line == multipart + '\n':
ofile.write(pref + line)
nifile = File(ifile, multipart)
1997-04-11 12:22:56 -03:00
unmimify_part(nifile, ofile, decode_base64)
line = nifile.peek
continue
# not a boundary between parts
break
if line and quoted_printable:
while line[-2:] == '=\n':
line = line[:-2]
newline = ifile.readline()
if newline[:len(QUOTE)] == QUOTE:
newline = newline[len(QUOTE):]
line = line + newline
line = mime_decode(line)
1997-04-11 12:22:56 -03:00
if line and is_base64 and not pref:
import base64
line = base64.decodestring(line)
if line:
ofile.write(pref + line)
1997-04-11 12:22:56 -03:00
def unmimify(infile, outfile, decode_base64 = 0):
'''Convert quoted-printable parts of a MIME mail message to 8bit.'''
if type(infile) == type(''):
ifile = open(infile)
if type(outfile) == type('') and infile == outfile:
import os
d, f = os.path.split(infile)
os.rename(infile, os.path.join(d, ',' + f))
else:
ifile = infile
if type(outfile) == type(''):
ofile = open(outfile, 'w')
else:
ofile = outfile
nifile = File(ifile, None)
1997-04-11 12:22:56 -03:00
unmimify_part(nifile, ofile, decode_base64)
ofile.flush()
mime_char = regex.compile('[=\240-\377]') # quote these chars in body
mime_header_char = regex.compile('[=?\240-\377]') # quote these in header
def mime_encode(line, header):
'''Code a single line as quoted-printable.
If header is set, quote some extra characters.'''
if header:
reg = mime_header_char
else:
reg = mime_char
newline = ''
if len(line) >= 5 and line[:5] == 'From ':
# quote 'From ' at the start of a line for stupid mailers
newline = string.upper('=%02x' % ord('F'))
line = line[1:]
while 1:
i = reg.search(line)
if i < 0:
break
newline = newline + line[:i] + \
string.upper('=%02x' % ord(line[i]))
line = line[i+1:]
line = newline + line
newline = ''
while len(line) >= 75:
i = 73
while line[i] == '=' or line[i-1] == '=':
i = i - 1
i = i + 1
newline = newline + line[:i] + '=\n'
line = line[i:]
return newline + line
mime_header = regex.compile('\\([ \t(]\\|^\\)\\([-a-zA-Z0-9_+]*[\240-\377][-a-zA-Z0-9_+\240-\377]*\\)\\([ \t)]\\|$\\)')
def mime_encode_header(line):
'''Code a single header line as quoted-printable.'''
newline = ''
while 1:
i = mime_header.search(line)
if i < 0:
break
newline = newline + line[:i] + mime_header.group(1) + \
'=?' + CHARSET + '?Q?' + \
mime_encode(mime_header.group(2), 1) + \
'?=' + mime_header.group(3)
line = line[i+len(mime_header.group(0)):]
return newline + line
mv = regex.compile('^mime-version:', regex.casefold)
cte = regex.compile('^content-transfer-encoding:', regex.casefold)
iso_char = regex.compile('[\240-\377]')
def mimify_part(ifile, ofile, is_mime):
'''Convert an 8bit part of a MIME mail message to quoted-printable.'''
1996-06-10 16:04:02 -03:00
has_cte = is_qp = is_base64 = 0
multipart = None
must_quote_body = must_quote_header = has_iso_chars = 0
header = []
header_end = ''
message = []
message_end = ''
# read header
hfile = HeaderFile(ifile)
while 1:
line = hfile.readline()
if not line:
break
if not must_quote_header and iso_char.search(line) >= 0:
must_quote_header = 1
if mv.match(line) >= 0:
is_mime = 1
if cte.match(line) >= 0:
has_cte = 1
if qp.match(line) >= 0:
is_qp = 1
1997-04-11 12:22:56 -03:00
elif base64_re.match(line) >= 0:
1996-06-10 16:04:02 -03:00
is_base64 = 1
if mp.match(line) >= 0:
multipart = '--' + mp.group(1)
if he.match(line) >= 0:
header_end = line
break
header.append(line)
# read body
while 1:
line = ifile.readline()
if not line:
break
if multipart:
if line == multipart + '--\n':
message_end = line
break
if line == multipart + '\n':
message_end = line
break
1996-06-10 16:04:02 -03:00
if is_base64:
message.append(line)
continue
if is_qp:
while line[-2:] == '=\n':
line = line[:-2]
newline = ifile.readline()
if newline[:len(QUOTE)] == QUOTE:
newline = newline[len(QUOTE):]
line = line + newline
line = mime_decode(line)
message.append(line)
if not has_iso_chars:
if iso_char.search(line) >= 0:
has_iso_chars = must_quote_body = 1
if not must_quote_body:
if len(line) > MAXLEN:
must_quote_body = 1
# convert and output header and body
for line in header:
if must_quote_header:
line = mime_encode_header(line)
if chrset.match(line) >= 0:
if has_iso_chars:
# change us-ascii into iso-8859-1
if string.lower(chrset.group(2)) == 'us-ascii':
line = chrset.group(1) + \
CHARSET + chrset.group(3)
else:
# change iso-8859-* into us-ascii
line = chrset.group(1) + 'us-ascii' + chrset.group(3)
if has_cte and cte.match(line) >= 0:
line = 'Content-Transfer-Encoding: '
1996-06-10 16:04:02 -03:00
if is_base64:
line = line + 'base64\n'
elif must_quote_body:
line = line + 'quoted-printable\n'
else:
line = line + '7bit\n'
ofile.write(line)
if (must_quote_header or must_quote_body) and not is_mime:
ofile.write('Mime-Version: 1.0\n')
ofile.write('Content-Type: text/plain; ')
if has_iso_chars:
ofile.write('charset="%s"\n' % CHARSET)
else:
ofile.write('charset="us-ascii"\n')
if must_quote_body and not has_cte:
ofile.write('Content-Transfer-Encoding: quoted-printable\n')
ofile.write(header_end)
for line in message:
if must_quote_body:
line = mime_encode(line, 0)
ofile.write(line)
ofile.write(message_end)
line = message_end
while multipart:
if line == multipart + '--\n':
# read bit after the end of the last part
while 1:
line = ifile.readline()
if not line:
return
if must_quote_body:
line = mime_encode(line, 0)
ofile.write(line)
if line == multipart + '\n':
nifile = File(ifile, multipart)
mimify_part(nifile, ofile, 1)
line = nifile.peek
ofile.write(line)
continue
def mimify(infile, outfile):
'''Convert 8bit parts of a MIME mail message to quoted-printable.'''
if type(infile) == type(''):
ifile = open(infile)
if type(outfile) == type('') and infile == outfile:
import os
d, f = os.path.split(infile)
os.rename(infile, os.path.join(d, ',' + f))
else:
ifile = infile
if type(outfile) == type(''):
ofile = open(outfile, 'w')
else:
ofile = outfile
nifile = File(ifile, None)
mimify_part(nifile, ofile, 0)
ofile.flush()
import sys
if __name__ == '__main__' or (len(sys.argv) > 0 and sys.argv[0] == 'mimify'):
import getopt
usage = 'Usage: mimify [-l len] -[ed] [infile [outfile]]'
1997-04-11 12:22:56 -03:00
decode_base64 = 0
opts, args = getopt.getopt(sys.argv[1:], 'l:edb')
if len(args) not in (0, 1, 2):
print usage
sys.exit(1)
1997-04-11 12:22:56 -03:00
if (('-e', '') in opts) == (('-d', '') in opts) or \
((('-b', '') in opts) and (('-d', '') not in opts)):
print usage
sys.exit(1)
for o, a in opts:
if o == '-e':
encode = mimify
elif o == '-d':
encode = unmimify
elif o == '-l':
try:
MAXLEN = string.atoi(a)
except:
print usage
sys.exit(1)
1997-04-11 12:22:56 -03:00
elif o == '-b':
decode_base64 = 1
if len(args) == 0:
1997-04-11 12:22:56 -03:00
encode_args = (sys.stdin, sys.stdout)
elif len(args) == 1:
1997-04-11 12:22:56 -03:00
encode_args = (args[0], sys.stdout)
else:
1997-04-11 12:22:56 -03:00
encode_args = (args[0], args[1])
if decode_base64:
encode_args = encode_args + (decode_base64,)
apply(encode, encode_args)