From 5419f42c6f82c81eddac47956910eb1ecc3c639e Mon Sep 17 00:00:00 2001 From: Oleg Broytman Date: Fri, 24 Jan 2014 00:56:43 +0400 Subject: [PATCH] Make indent 4 spaces --- mimedecode.py | 492 +++++++++++++++++++++++++------------------------- setup.py | 22 +-- 2 files changed, 257 insertions(+), 257 deletions(-) diff --git a/mimedecode.py b/mimedecode.py index d6930b7..21ab371 100755 --- a/mimedecode.py +++ b/mimedecode.py @@ -9,9 +9,9 @@ import sys, os import email try: - from cStringIO import StringIO + from cStringIO import StringIO except ImportError: - from StringIO import StringIO + from StringIO import StringIO import socket @@ -21,342 +21,342 @@ me = os.path.basename(sys.argv[0]) def version(exit=1): - sys.stdout.write("""\ + sys.stdout.write("""\ Broytman mimedecode.py version %s, %s """ % (__version__, __copyright__)) - if exit: sys.exit(0) + if exit: sys.exit(0) def usage(code=0): - version(0) - sys.stdout.write("""\ + version(0) + sys.stdout.write("""\ Usage: %s [-h|--help] [-V|--version] [-cCDP] [-f charset] [-d header] [-p header:param] [-beit mask] [filename] """ % me) - sys.exit(code) + sys.exit(code) def output(s, outfile=sys.stdout): - outfile.write(s) + outfile.write(s) def output_headers(msg, outfile=sys.stdout): - unix_from = msg.get_unixfrom() - if unix_from: - output(unix_from + os.linesep) - for key, value in msg.items(): - output("%s: %s\n" % (key, value), outfile) - output("\n", outfile) # End of headers + unix_from = msg.get_unixfrom() + if unix_from: + output(unix_from + os.linesep) + for key, value in msg.items(): + output("%s: %s\n" % (key, value), outfile) + output("\n", outfile) # End of headers def recode(s, charset): - return unicode(s, charset, "replace").encode(GlobalOptions.default_encoding, "replace") + return unicode(s, charset, "replace").encode(GlobalOptions.default_encoding, "replace") def recode2(s, charset): - if charset and charset.lower() <> GlobalOptions.default_encoding: - s = recode(s, charset) - return s + if charset and charset.lower() <> GlobalOptions.default_encoding: + s = recode(s, charset) + return s def _decode_header(s): - """Return a decoded string according to RFC 2047. - NOTE: This is almost the same as email.Utils.decode. - """ - from types import ListType - import email.Header - - L = email.Header.decode_header(s) - if not isinstance(L, ListType): - # s wasn't decoded - return s - - rtn = [] - for atom, charset in L: - if charset is None: - rtn.append(atom) - else: - rtn.append(recode2(atom, charset)) - rtn.append(' ') - del rtn[-1] # remove the last space - - # Now that we've decoded everything, we just need to join all the parts - # together into the final string. - return ''.join(rtn) + """Return a decoded string according to RFC 2047. + NOTE: This is almost the same as email.Utils.decode. + """ + from types import ListType + import email.Header + + L = email.Header.decode_header(s) + if not isinstance(L, ListType): + # s wasn't decoded + return s + + rtn = [] + for atom, charset in L: + if charset is None: + rtn.append(atom) + else: + rtn.append(recode2(atom, charset)) + rtn.append(' ') + del rtn[-1] # remove the last space + + # Now that we've decoded everything, we just need to join all the parts + # together into the final string. + return ''.join(rtn) def decode_header(msg, header): - "Decode mail header (if exists) and put it back, if it was encoded" + "Decode mail header (if exists) and put it back, if it was encoded" - if msg.has_key(header): - value = msg[header] - new_value = _decode_header(value) - if new_value <> value: # do not bother to touch msg if not changed - set_header(msg, header, new_value) + if msg.has_key(header): + value = msg[header] + new_value = _decode_header(value) + if new_value <> value: # do not bother to touch msg if not changed + set_header(msg, header, new_value) def _decode_header_param(s): - return recode2(s[2], s[0]) + return recode2(s[2], s[0]) def decode_header_param(msg, header, param): - "Decode mail header's parameter (if exists) and put it back, if it was encoded" + "Decode mail header's parameter (if exists) and put it back, if it was encoded" - if msg.has_key(header): - value = msg.get_param(param, header=header) - if value: - from types import TupleType - if isinstance(value, TupleType): - new_value = _decode_header_param(value) - else: - new_value = _decode_header(value) - if new_value <> value: # do not bother to touch msg if not changed - msg.set_param(param, new_value, header) + if msg.has_key(header): + value = msg.get_param(param, header=header) + if value: + from types import TupleType + if isinstance(value, TupleType): + new_value = _decode_header_param(value) + else: + new_value = _decode_header(value) + if new_value <> value: # do not bother to touch msg if not changed + msg.set_param(param, new_value, header) def decode_headers(msg): - "Decode message headers according to global options" + "Decode message headers according to global options" - for header in GlobalOptions.decode_headers: - decode_header(msg, header) + for header in GlobalOptions.decode_headers: + decode_header(msg, header) - for header, param in GlobalOptions.decode_header_params: - decode_header_param(msg, header, param) + for header, param in GlobalOptions.decode_header_params: + decode_header_param(msg, header, param) def set_header(msg, header, value): - "Replace header" + "Replace header" - if msg.has_key(header): - msg.replace_header(header, value) - else: - msg[header] = value + if msg.has_key(header): + msg.replace_header(header, value) + else: + msg[header] = value def set_content_type(msg, newtype, charset=None): - msg.set_type(newtype) + msg.set_type(newtype) - if charset: - msg.set_param("charset", charset, "Content-Type") + if charset: + msg.set_param("charset", charset, "Content-Type") caps = None # Globally stored mailcap database; initialized only if needed def decode_body(msg, s): - "Decode body to plain text using first copiousoutput filter from mailcap" + "Decode body to plain text using first copiousoutput filter from mailcap" - import mailcap, tempfile + import mailcap, tempfile - global caps - if caps is None: - caps = mailcap.getcaps() + global caps + if caps is None: + caps = mailcap.getcaps() - content_type = msg.get_content_type() - filename = tempfile.mktemp() - command = None + content_type = msg.get_content_type() + filename = tempfile.mktemp() + command = None - entries = mailcap.lookup(caps, content_type, "view") - for entry in entries: - if entry.has_key('copiousoutput'): - if entry.has_key('test'): - test = mailcap.subst(entry['test'], content_type, filename) - if test and os.system(test) != 0: - continue - command = mailcap.subst(entry["view"], content_type, filename) - break + entries = mailcap.lookup(caps, content_type, "view") + for entry in entries: + if entry.has_key('copiousoutput'): + if entry.has_key('test'): + test = mailcap.subst(entry['test'], content_type, filename) + if test and os.system(test) != 0: + continue + command = mailcap.subst(entry["view"], content_type, filename) + break - if not command: - return s + if not command: + return s - file = open(filename, 'w') - file.write(s) - file.close() + file = open(filename, 'w') + file.write(s) + file.close() - pipe = os.popen(command, 'r') - s = pipe.read() - pipe.close() - os.remove(filename) + pipe = os.popen(command, 'r') + s = pipe.read() + pipe.close() + os.remove(filename) - set_content_type(msg, "text/plain") - msg["X-MIME-Autoconverted"] = "from %s to text/plain by %s id %s" % (content_type, host_name, command.split()[0]) + set_content_type(msg, "text/plain") + msg["X-MIME-Autoconverted"] = "from %s to text/plain by %s id %s" % (content_type, host_name, command.split()[0]) - return s + return s def recode_charset(msg, s): - "Recode charset of the message to the default charset" + "Recode charset of the message to the default charset" - save_charset = charset = msg.get_content_charset() - if charset and charset.lower() <> GlobalOptions.default_encoding: - s = recode2(s, charset) - content_type = msg.get_content_type() - set_content_type(msg, content_type, GlobalOptions.default_encoding) - msg["X-MIME-Autoconverted"] = "from %s to %s by %s id %s" % (save_charset, GlobalOptions.default_encoding, host_name, me) - return s + save_charset = charset = msg.get_content_charset() + if charset and charset.lower() <> GlobalOptions.default_encoding: + s = recode2(s, charset) + content_type = msg.get_content_type() + set_content_type(msg, content_type, GlobalOptions.default_encoding) + msg["X-MIME-Autoconverted"] = "from %s to %s by %s id %s" % (save_charset, GlobalOptions.default_encoding, host_name, me) + return s def totext(msg, instring): - "Convert instring content to text" + "Convert instring content to text" - if msg.is_multipart(): # Recursively decode all parts of the multipart message - newfile = StringIO(str(msg)) - newfile.seek(0) - decode_file(newfile) - return + if msg.is_multipart(): # Recursively decode all parts of the multipart message + newfile = StringIO(str(msg)) + newfile.seek(0) + decode_file(newfile) + return - # Decode body and recode charset - s = decode_body(msg, instring) - if GlobalOptions.recode_charset: - s = recode_charset(msg, s) + # Decode body and recode charset + s = decode_body(msg, instring) + if GlobalOptions.recode_charset: + s = recode_charset(msg, s) - output_headers(msg) - output(s) + output_headers(msg) + output(s) def decode_part(msg): - "Decode one part of the message" - - decode_headers(msg) - encoding = msg["Content-Transfer-Encoding"] - - if encoding in (None, '', '7bit', '8bit', 'binary'): - outstring = str(msg.get_payload()) - else: # Decode from transfer ecoding to text or binary form - outstring = str(msg.get_payload(decode=1)) - set_header(msg, "Content-Transfer-Encoding", "8bit") - msg["X-MIME-Autoconverted"] = "from %s to 8bit by %s id %s" % (encoding, host_name, me) - - # Test all mask lists and find what to do with this content type - masks = [] - ctype = msg.get_content_type() - if ctype: - masks.append(ctype) - mtype = msg.get_content_maintype() - if mtype: - masks.append(mtype + '/*') - masks.append('*/*') - - for content_type in masks: - if content_type in GlobalOptions.totext_mask: - totext(msg, outstring) - return - elif content_type in GlobalOptions.binary_mask: - output_headers(msg) - output(outstring) - return - elif content_type in GlobalOptions.ignore_mask: - output_headers(msg) - output("\nMessage body of type `%s' skipped.\n" % content_type) - return - elif content_type in GlobalOptions.error_mask: - raise ValueError, "content type `%s' prohibited" % content_type - - # Neither content type nor masks were listed - decode by default - totext(msg, outstring) + "Decode one part of the message" + + decode_headers(msg) + encoding = msg["Content-Transfer-Encoding"] + + if encoding in (None, '', '7bit', '8bit', 'binary'): + outstring = str(msg.get_payload()) + else: # Decode from transfer ecoding to text or binary form + outstring = str(msg.get_payload(decode=1)) + set_header(msg, "Content-Transfer-Encoding", "8bit") + msg["X-MIME-Autoconverted"] = "from %s to 8bit by %s id %s" % (encoding, host_name, me) + + # Test all mask lists and find what to do with this content type + masks = [] + ctype = msg.get_content_type() + if ctype: + masks.append(ctype) + mtype = msg.get_content_maintype() + if mtype: + masks.append(mtype + '/*') + masks.append('*/*') + + for content_type in masks: + if content_type in GlobalOptions.totext_mask: + totext(msg, outstring) + return + elif content_type in GlobalOptions.binary_mask: + output_headers(msg) + output(outstring) + return + elif content_type in GlobalOptions.ignore_mask: + output_headers(msg) + output("\nMessage body of type `%s' skipped.\n" % content_type) + return + elif content_type in GlobalOptions.error_mask: + raise ValueError, "content type `%s' prohibited" % content_type + + # Neither content type nor masks were listed - decode by default + totext(msg, outstring) def decode_file(infile): - "Decode the entire message" + "Decode the entire message" - msg = email.message_from_file(infile) - boundary = msg.get_boundary() + msg = email.message_from_file(infile) + boundary = msg.get_boundary() - if msg.is_multipart(): - decode_headers(msg) - output_headers(msg) + if msg.is_multipart(): + decode_headers(msg) + output_headers(msg) - if msg.preamble: # Preserve the first part, it is probably not a RFC822-message - output(msg.preamble) # Usually it is just a few lines of text (MIME warning) + if msg.preamble: # Preserve the first part, it is probably not a RFC822-message + output(msg.preamble) # Usually it is just a few lines of text (MIME warning) - for subpart in msg.get_payload(): - output("\n--%s\n" % boundary) - decode_part(subpart) + for subpart in msg.get_payload(): + output("\n--%s\n" % boundary) + decode_part(subpart) - output("\n--%s--\n" % boundary) + output("\n--%s--\n" % boundary) - if msg.epilogue: - output(msg.epilogue) + if msg.epilogue: + output(msg.epilogue) - else: - if msg.has_key("Content-Type"): # Simple one-part message - decode it - decode_part(msg) + else: + if msg.has_key("Content-Type"): # Simple one-part message - decode it + decode_part(msg) - else: # Not a message, just text - copy it literally - output(str(msg)) + else: # Not a message, just text - copy it literally + output(str(msg)) class GlobalOptions: - from m_lib.defenc import default_encoding - recode_charset = 1 # recode charset of message body + from m_lib.defenc import default_encoding + recode_charset = 1 # recode charset of message body - decode_headers = ["From", "Subject"] # A list of headers to decode - decode_header_params = [ - ("Content-Type", "name"), - ("Content-Disposition", "filename"), - ] # A list of headers' parameters to decode + decode_headers = ["From", "Subject"] # A list of headers to decode + decode_header_params = [ + ("Content-Type", "name"), + ("Content-Disposition", "filename"), + ] # A list of headers' parameters to decode - totext_mask = [] # A list of content-types to decode - binary_mask = [] # A list to pass through - ignore_mask = [] # Ignore (skip, do not decode and do not include into output) - error_mask = [] # Raise error if encounter one of these + totext_mask = [] # A list of content-types to decode + binary_mask = [] # A list to pass through + ignore_mask = [] # Ignore (skip, do not decode and do not include into output) + error_mask = [] # Raise error if encounter one of these def init(): - from getopt import getopt, GetoptError - - try: - options, arguments = getopt(sys.argv[1:], 'hVcCDPf:d:p:b:e:i:t:', - ['help', 'version']) - except GetoptError: - usage(1) - - for option, value in options: - if option == '-h': - usage() - elif option == '--help': - usage() - elif option == '-V': - version() - elif option == '--version': - version() - elif option == '-c': - GlobalOptions.recode_charset = 1 - elif option == '-C': - GlobalOptions.recode_charset = 0 - elif option == '-f': - GlobalOptions.default_encoding = value - elif option == '-d': - GlobalOptions.decode_headers.append(value) - elif option == '-D': - GlobalOptions.decode_headers = [] - elif option == '-p': - GlobalOptions.decode_header_params.append(value.split(':', 1)) - elif option == '-P': - GlobalOptions.decode_header_params = [] - elif option == '-t': - GlobalOptions.totext_mask.append(value) - elif option == '-b': - GlobalOptions.binary_mask.append(value) - elif option == '-i': - GlobalOptions.ignore_mask.append(value) - elif option == '-e': - GlobalOptions.error_mask.append(value) - else: - usage(1) - - return arguments + from getopt import getopt, GetoptError + + try: + options, arguments = getopt(sys.argv[1:], 'hVcCDPf:d:p:b:e:i:t:', + ['help', 'version']) + except GetoptError: + usage(1) + + for option, value in options: + if option == '-h': + usage() + elif option == '--help': + usage() + elif option == '-V': + version() + elif option == '--version': + version() + elif option == '-c': + GlobalOptions.recode_charset = 1 + elif option == '-C': + GlobalOptions.recode_charset = 0 + elif option == '-f': + GlobalOptions.default_encoding = value + elif option == '-d': + GlobalOptions.decode_headers.append(value) + elif option == '-D': + GlobalOptions.decode_headers = [] + elif option == '-p': + GlobalOptions.decode_header_params.append(value.split(':', 1)) + elif option == '-P': + GlobalOptions.decode_header_params = [] + elif option == '-t': + GlobalOptions.totext_mask.append(value) + elif option == '-b': + GlobalOptions.binary_mask.append(value) + elif option == '-i': + GlobalOptions.ignore_mask.append(value) + elif option == '-e': + GlobalOptions.error_mask.append(value) + else: + usage(1) + + return arguments if __name__ == "__main__": - arguments = init() - - la = len(arguments) - if la == 0: - infile = sys.stdin - elif la <> 1: - usage(1) - elif arguments[0] == '-': - infile = sys.stdin - else: - infile = open(arguments[0], 'r') - - decode_file(infile) + arguments = init() + + la = len(arguments) + if la == 0: + infile = sys.stdin + elif la <> 1: + usage(1) + elif arguments[0] == '-': + infile = sys.stdin + else: + infile = open(arguments[0], 'r') + + decode_file(infile) diff --git a/setup.py b/setup.py index f98336b..6514cf5 100755 --- a/setup.py +++ b/setup.py @@ -16,15 +16,15 @@ if is_setuptools: from mimedecode_version import __version__, __copyright__, __license__ setup(name = "mimedecode", - version = __version__, - description = "A program to decode MIME messages", - long_description = "A program to decode MIME messages. " + __copyright__, - author = "Oleg Broytman", - author_email = "phd@phdru.name", - url = "http://phdru.name/Software/Python/#mimedecode", - license = __license__, - platforms = "All", - py_modules = ['mimedecode_version'], - scripts = ['mimedecode.py'], - **kw + version = __version__, + description = "A program to decode MIME messages", + long_description = "A program to decode MIME messages. " + __copyright__, + author = "Oleg Broytman", + author_email = "phd@phdru.name", + url = "http://phdru.name/Software/Python/#mimedecode", + license = __license__, + platforms = "All", + py_modules = ['mimedecode_version'], + scripts = ['mimedecode.py'], + **kw ) -- 2.39.5