X-Git-Url: https://git.sommitrealweird.co.uk/rss2maildir.git/blobdiff_plain/63dc58ea5dc82d4ab0984101798fd8a813187450..55d5cb2f1f1831a5c8d5846d460e26fb99adc2c8:/rss2maildir.py diff --git a/rss2maildir.py b/rss2maildir.py index cece696..a09cff2 100755 --- a/rss2maildir.py +++ b/rss2maildir.py @@ -1,12 +1,225 @@ #!/usr/bin/python +# coding=utf8 + +# rss2maildir.py - RSS feeds to Maildir 1 email per item +# Copyright (C) 2007 Brett Parker +# +# This program is free software: you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation, either version 3 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see . import mailbox import sys import os +import stat +import urllib + +import feedparser + +from email.MIMEMultipart import MIMEMultipart +from email.MIMEText import MIMEText + +import datetime +import random +import string +import textwrap + +import socket from optparse import OptionParser from ConfigParser import SafeConfigParser +from base64 import b64encode +import md5 + +import cgi +import dbm + +from HTMLParser import HTMLParser + +entities = { + "amp": "&", + "lt": "<", + "gt": ">", + "pound": "£", + "copy": "©", + "apos": "'", + "quote": "\"", + "nbsp": " ", + } + +class HTML2Text(HTMLParser): + + def __init__(self): + self.inheadingone = False + self.inheadingtwo = False + self.inotherheading = False + self.inparagraph = True + self.inblockquote = False + self.inlink = False + self.text = "" + self.currentparagraph = "" + self.headingtext = "" + self.blockquote = "" + HTMLParser.__init__(self) + + def handle_starttag(self, tag, attrs): + if tag.lower() == "h1": + self.inheadingone = True + self.inparagraph = False + elif tag.lower() == "h2": + self.inheadingtwo = True + self.inparagraph = False + elif tag.lower() in ["h3", "h4", "h5", "h6"]: + self.inotherheading = True + self.inparagraph = False + elif tag.lower() == "a": + self.inlink = True + elif tag.lower() == "br": + if self.inparagraph: + self.text = self.text + "\n".join(textwrap.wrap(self.currentparagraph, 70)) + self.currentparagraph = "" + elif self.inblockquote: + self.text = self.text + "\n> " + "\n> ".join([a.strip() for a in textwrap.wrap(self.blockquote, 68)]) + self.blockquote = "" + else: + self.text = self.text + "\n" + elif tag.lower() == "blockquote": + self.inblockquote = True + self.text = self.text + "\n" + elif tag.lower() == "p": + if self.text != "": + self.text = self.text + "\n\n" + self.currentparagraph = "" + self.inparagraph = True + + def handle_startendtag(self, tag, attrs): + if tag.lower() == "br": + if self.inparagraph: + self.text = self.text + "\n".join(textwrap.wrap(self.currentparagraph, 70)) + self.currentparagraph = "" + elif self.inblockquote: + self.text = self.text + "\n> " + "\n> ".join([a.strip() for a in textwrap.wrap(self.blockquote, 68)]) + self.blockquote = "" + else: + self.text = self.text + "\n" + + def handle_endtag(self, tag): + if tag.lower() == "h1": + self.inheadingone = False + self.text = self.text + self.headingtext + "\n" + "=" * len(self.headingtext) + self.headingtext = "" + elif tag.lower() == "h2": + self.inheadingtwo = False + self.text = self.text + self.headingtext + "\n" + "-" * len(self.headingtext) + self.headingtext = "" + elif tag.lower() in ["h3", "h4", "h5", "h6"]: + self.inotherheading = False + self.text = self.text + self.headingtext + "\n" + "~" * len(self.headingtext) + self.headingtext = "" + elif tag.lower() == "p": + self.text = self.text + "\n".join(textwrap.wrap(self.currentparagraph, 70)) + self.inparagraph = False + elif tag.lower() == "blockquote": + self.text = self.text + "\n> " + "\n> ".join([a.strip() for a in textwrap.wrap(self.blockquote, 68)]) + self.inblockquote = False + self.blockquote = "" + + def handle_data(self, data): + if self.inheadingone or self.inheadingtwo or self.inotherheading: + self.headingtext = self.headingtext + data.strip() + " " + elif self.inblockquote: + self.blockquote = self.blockquote + data.strip() + " " + elif self.inparagraph: + self.currentparagraph = self.currentparagraph + data.strip() + " " + else: + self.text = self.text + data.strip() + " " + + def handle_entityref(self, name): + if entities.has_key(name.lower()): + self.text = self.text + entities[name.lower()] + else: + self.text = self.text + "&" + name + ";" + + def gettext(self): + data = self.text + if self.inparagraph: + data = data + "\n".join(textwrap.wrap(self.currentparagraph, 70)) + return data + +def parse_and_deliver(maildir, url, statedir): + md = mailbox.Maildir(maildir) + fp = feedparser.parse(url) + db = dbm.open(os.path.join(statedir, "seen"), "c") + for item in fp["items"]: + # have we seen it before? + # need to work out what the content is first... + + if item.has_key("content"): + content = item["content"][0]["value"] + else: + content = item["summary"] + + md5sum = md5.md5(content.encode("utf8")).hexdigest() + + if db.has_key(url + "|" + item["link"]): + data = db[url + "|" + item["link"]] + data = cgi.parse_qs(data) + if data["contentmd5"][0] == md5sum: + continue + + try: + author = item["author"] + except: + author = url + + # create a basic email message + msg = MIMEMultipart("alternative") + messageid = "<" + datetime.datetime.now().strftime("%Y%m%d%H%M") + "." + "".join([random.choice(string.ascii_letters + string.digits) for a in range(0,6)]) + "@" + socket.gethostname() + ">" + msg.add_header("Message-ID", messageid) + msg.set_unixfrom("\"%s\" " %(url)) + msg.add_header("From", "\"%s\" " %(author)) + msg.add_header("To", "\"%s\" " %(url)) + createddate = datetime.datetime(*item["updated_parsed"][0:6]).strftime("%a, %e %b %Y %T -0000") + msg.add_header("Date", createddate) + msg.add_header("Subject", item["title"]) + msg.set_default_type("text/plain") + + htmlpart = MIMEText(content.encode("utf8"), "html", "utf8") + textparser = HTML2Text() + textparser.feed(content.encode("utf8")) + textcontent = textparser.gettext() + textpart = MIMEText(textcontent, "plain", "utf8") + msg.attach(textpart) + msg.attach(htmlpart) + + # start by working out the filename we should be writting to, we do + # this following the normal maildir style rules + fname = str(os.getpid()) + "." + socket.gethostname() + "." + "".join([random.choice(string.ascii_letters + string.digits) for a in range(0,10)]) + "." + datetime.datetime.now().strftime('%s') + fn = os.path.join(maildir, "tmp", fname) + fh = open(fn, "w") + fh.write(msg.as_string()) + fh.close() + # now move it in to the new directory + newfn = os.path.join(maildir, "new", fname) + os.link(fn, newfn) + os.unlink(fn) + + # now add to the database about the item + data = urllib.urlencode((("message-id", messageid), ("created", createddate), ("contentmd5", md5sum))) + db[url + "|" + item["link"]] = data + + db.close() + # first off, parse the command line arguments oparser = OptionParser() @@ -14,6 +227,10 @@ oparser.add_option( "-c", "--conf", dest="conf", help="location of config file" ) +oparser.add_option( + "-s", "--statedir", dest="statedir", + help="location of directory to store state in" + ) (options, args) = oparser.parse_args() @@ -50,10 +267,65 @@ scp = SafeConfigParser() scp.read(configfile) maildir_root = "RSSMaildir" +state_dir = "state" + +if options.statedir != None: + state_dir = options.statedir + try: + mode = os.stat(state_dir)[stat.ST_MODE] + if not stat.S_ISDIR(mode): + sys.stderr.write("State directory (%s) is not a directory\n" %(state_dir)) + sys.exit(1) + except: + # try to make the directory + try: + os.mkdir(state_dir) + except: + sys.stderr.write("Couldn't create statedir %s" %(state_dir)) + sys.exit(1) +elif scp.has_option("general", "state_dir"): + new_state_dir = scp.get("general", "state_dir") + try: + mode = os.stat(state_dir)[stat.ST_MODE] + if not stat.S_ISDIR(mode): + sys.stderr.write("State directory (%s) is not a directory\n" %(state_dir)) + sys.exit(1) + except: + # try to create it + try: + os.mkdir(new_state_dir) + state_dir = new_state_dir + except: + sys.stderr.write("Couldn't create state directory %s\n" %(new_state_dir)) + sys.exit(1) +else: + try: + mode = os.stat(state_dir)[stat.ST_MODE] + if not stat.S_ISDIR(mode): + sys.stderr.write("State directory %s is not a directory\n" %(state_dir)) + sys.exit(1) + except: + try: + os.mkdir(state_dir) + except: + sys.stderr.write("State directory %s could not be created\n" %(state_dir)) + sys.exit(1) if scp.has_option("general", "maildir_root"): maildir_root = scp.get("general", "maildir_root") +try: + mode = os.stat(maildir_root)[stat.ST_MODE] + if not stat.S_ISDIR(mode): + sys.stderr.write("Maildir Root %s is not a directory\n" %(maildir_root)) + sys.exit(1) +except: + try: + os.mkdir(maildir_root) + except: + sys.stderr.write("Couldn't create Maildir Root %s\n" %(maildir_root)) + sys.exit(1) + feeds = scp.sections() try: feeds.remove("general") @@ -61,6 +333,55 @@ except: pass for section in feeds: - print section - print "-" * len(section) - print scp.items(section) + # check if the directory exists + maildir = None + try: + maildir = scp.get(section, "maildir") + except: + maildir = section + + maildir = urllib.urlencode(((section, maildir),)).split("=")[1] + maildir = os.path.join(maildir_root, maildir) + + try: + exists = os.stat(maildir) + if stat.S_ISDIR(exists[stat.ST_MODE]): + # check if there's a new, cur and tmp directory + try: + mode = os.stat(os.path.join(maildir, "cur"))[stat.ST_MODE] + except: + os.mkdir(os.path.join(maildir, "cur")) + if not stat.S_ISDIR(mode): + sys.stderr.write("Broken maildir: %s\n" %(maildir)) + try: + mode = os.stat(os.path.join(maildir, "tmp"))[stat.ST_MODE] + except: + os.mkdir(os.path.join(maildir, "tmp")) + if not stat.S_ISDIR(mode): + sys.stderr.write("Broken maildir: %s\n" %(maildir)) + try: + mode = os.stat(os.path.join(maildir, "new"))[stat.ST_MODE] + if not stat.S_ISDIR(mode): + sys.stderr.write("Broken maildir: %s\n" %(maildir)) + except: + os.mkdir(os.path.join(maildir, "new")) + else: + sys.stderr.write("Broken maildir: %s\n" %(maildir)) + except: + try: + os.mkdir(maildir) + except: + sys.stderr.write("Couldn't create root maildir %s\n" %(maildir)) + sys.exit(1) + try: + os.mkdir(os.path.join(maildir, "new")) + os.mkdir(os.path.join(maildir, "cur")) + os.mkdir(os.path.join(maildir, "tmp")) + except: + sys.stderr.write("Couldn't create required maildir directories for %s\n" %(section,)) + sys.exit(1) + + # right - we've got the directories, we've got the section, we know the + # url... lets play! + + parse_and_deliver(maildir, section, state_dir)