mylar/mylar/rsscheck.py

1182 lines
51 KiB
Python
Executable File

#!/usr/bin/python
import os, sys
import re
import feedparser
import requests
import cfscrape
import urlparse
import ftpsshup
from datetime import datetime, timedelta
import gzip
import time
import random
from StringIO import StringIO
import mylar
from mylar import db, logger, ftpsshup, helpers, auth32p, utorrent
import torrent.clients.transmission as transmission
import torrent.clients.deluge as deluge
import torrent.clients.qbittorrent as qbittorrent
def _start_newznab_attr(self, attrsD):
context = self._getContext()
context.setdefault('newznab', feedparser.FeedParserDict())
context['newznab'].setdefault('tags', feedparser.FeedParserDict())
name = attrsD.get('name')
value = attrsD.get('value')
if name == 'category':
context['newznab'].setdefault('categories', []).append(value)
else:
context['newznab'][name] = value
feedparser._FeedParserMixin._start_newznab_attr = _start_newznab_attr
def torrents(pickfeed=None, seriesname=None, issue=None, feedinfo=None):
if pickfeed is None:
return
srchterm = None
if seriesname:
srchterm = re.sub(' ', '%20', seriesname)
if issue:
srchterm += '%20' + str(issue)
if mylar.CONFIG.TPSE_PROXY:
if mylar.CONFIG.TPSE_PROXY.endswith('/'):
tpse_url = mylar.CONFIG.TPSE_PROXY
else:
tpse_url = mylar.CONFIG.TPSE_PROXY + '/'
else:
#switched to https.
tpse_url = mylar.TPSEURL
#this is for the public trackers included thus far in order to properly cycle throught the correct ones depending on the search request
# TPSE = search only
# DEM = rss feed
# WWT = rss feed
if pickfeed == 'TPSE-SEARCH':
pickfeed = '2'
loopit = 1
elif pickfeed == 'TPSE':
#we need to cycle through both DEM + WWT feeds
loopit = 2
else:
loopit = 1
lp = 0
totalcount = 0
title = []
link = []
description = []
seriestitle = []
feeddata = []
myDB = db.DBConnection()
torthetpse = []
torthe32p = []
torinfo = {}
while (lp < loopit):
if lp == 0 and loopit == 2:
pickfeed = '6' #DEM RSS
elif lp == 1 and loopit == 2:
pickfeed = '999' #WWT RSS
feedtype = None
if pickfeed == "1" and mylar.CONFIG.ENABLE_32P: # 32pages new releases feed.
feed = 'https://32pag.es/feeds.php?feed=torrents_all&user=' + feedinfo['user'] + '&auth=' + feedinfo['auth'] + '&passkey=' + feedinfo['passkey'] + '&authkey=' + feedinfo['authkey']
feedtype = ' from the New Releases RSS Feed for comics'
verify = bool(mylar.CONFIG.VERIFY_32P)
elif pickfeed == "2" and srchterm is not None: # TP.SE search / RSS
feed = tpse_url + 'rss/' + str(srchterm) + '/'
verify = bool(mylar.CONFIG.TPSE_VERIFY)
elif pickfeed == "3": # TP.SE rss feed (3101 = comics category) / non-RSS
feed = tpse_url + '?hl=en&safe=off&num=50&start=0&orderby=best&s=&filter=3101'
feedtype = ' from the New Releases RSS Feed for comics from TP.SE'
verify = bool(mylar.CONFIG.TPSE_VERIFY)
elif pickfeed == "4": #32p search
if any([mylar.CONFIG.USERNAME_32P is None, mylar.CONFIG.USERNAME_32P == '', mylar.CONFIG.PASSWORD_32P is None, mylar.CONFIG.PASSWORD_32P == '']):
logger.error('[RSS] Warning - you NEED to enter in your 32P Username and Password to use this option.')
lp=+1
continue
if mylar.CONFIG.MODE_32P == 0:
logger.warn('[32P] Searching is not available in 32p Legacy mode. Switch to Auth mode to use the search functionality.')
lp=+1
continue
return
elif pickfeed == "5" and srchterm is not None: # demonoid search / non-RSS
feed = mylar.DEMURL + "files/?category=10&subcategory=All&language=0&seeded=2&external=2&query=" + str(srchterm) + "&uid=0&out=rss"
verify = bool(mylar.CONFIG.TPSE_VERIFY)
elif pickfeed == "6": # demonoid rss feed
feed = mylar.DEMURL + 'rss/10.xml'
feedtype = ' from the New Releases RSS Feed from Demonoid'
verify = bool(mylar.CONFIG.TPSE_VERIFY)
elif pickfeed == "999": #WWT rss feed
feed = mylar.WWTURL + 'rss.php?cat=132,50'
feedtype = ' from the New Releases RSS Feed from WorldWideTorrents'
elif int(pickfeed) >= 7 and feedinfo is not None:
#personal 32P notification feeds.
#get the info here
feed = 'https://32pag.es/feeds.php?feed=' + feedinfo['feed'] + '&user=' + feedinfo['user'] + '&auth=' + feedinfo['auth'] + '&passkey=' + feedinfo['passkey'] + '&authkey=' + feedinfo['authkey'] + '&name=' + feedinfo['feedname']
feedtype = ' from your Personal Notification Feed : ' + feedinfo['feedname']
verify = bool(mylar.CONFIG.VERIFY_32P)
else:
logger.error('invalid pickfeed denoted...')
return
if pickfeed == '2' or pickfeed == '3':
picksite = 'TPSE'
#if pickfeed == '2':
# feedme = tpse.
elif pickfeed == '5' or pickfeed == '6':
picksite = 'DEM'
#if pickfeed == '5':
# feedme = dem.
elif pickfeed == '999':
picksite = 'WWT'
elif pickfeed == '1' or pickfeed == '4' or int(pickfeed) > 7:
picksite = '32P'
if all([pickfeed != '4', pickfeed != '3', pickfeed != '5', pickfeed != '999']):
payload = None
ddos_protection = round(random.uniform(0,15),2)
time.sleep(ddos_protection)
try:
cf_cookievalue = None
scraper = cfscrape.create_scraper()
if pickfeed == '2':
cf_cookievalue, cf_user_agent = scraper.get_tokens(feed)
headers = {'Accept-encoding': 'gzip',
'User-Agent': cf_user_agent}
if cf_cookievalue:
r = scraper.get(feed, verify=verify, cookies=cf_cookievalue, headers=headers)
else:
r = scraper.get(feed, verify=verify)#requests.get(feed, params=payload, verify=verify)
except Exception, e:
logger.warn('Error fetching RSS Feed Data from %s: %s' % (picksite, e))
lp+=1
continue
feedme = feedparser.parse(r.content)
#logger.info(feedme) #<-- uncomment this to see what Mylar is retrieving from the feed
i = 0
if pickfeed == '4':
for entry in searchresults['entries']:
justdigits = entry['file_size'] #size not available in follow-list rss feed
seeddigits = entry['seeders'] #number of seeders not available in follow-list rss feed
if int(seeddigits) >= int(mylar.CONFIG.MINSEEDS):
torthe32p.append({
'site': picksite,
'title': entry['torrent_seriesname'].lstrip() + ' ' + entry['torrent_seriesvol'] + ' #' + entry['torrent_seriesiss'],
'volume': entry['torrent_seriesvol'], # not stored by mylar yet.
'issue': entry['torrent_seriesiss'], # not stored by mylar yet.
'link': entry['torrent_id'], #just the id for the torrent
'pubdate': entry['pubdate'],
'size': entry['file_size'],
'seeders': entry['seeders'],
'files': entry['num_files']
})
i += 1
elif pickfeed == '3':
#TP.SE RSS FEED (parse)
pass
elif pickfeed == '5':
#DEMONOID SEARCH RESULT (parse)
pass
elif pickfeed == "999":
try:
feedme = feedparser.parse(feed)
except Exception, e:
logger.warn('Error fetching RSS Feed Data from %s: %s' % (picksite, e))
lp+=1
continue
#WWT / FEED
for entry in feedme.entries:
tmpsz = entry.description
tmpsz_st = tmpsz.find('Size:') + 6
if 'GB' in tmpsz[tmpsz_st:]:
szform = 'GB'
sz = 'G'
elif 'MB' in tmpsz[tmpsz_st:]:
szform = 'MB'
sz = 'M'
linkwwt = urlparse.parse_qs(urlparse.urlparse(entry.link).query)['id']
feeddata.append({
'site': picksite,
'title': entry.title,
'link': ''.join(linkwwt),
'pubdate': entry.updated,
'size': helpers.human2bytes(str(tmpsz[tmpsz_st:tmpsz.find(szform, tmpsz_st) -1]) + str(sz)) #+ 2 is for the length of the MB/GB in the size.
})
i+=1
else:
for entry in feedme['entries']:
#TP.SE RSS SEARCH RESULT
if pickfeed == "2":
try:
tmpenc = feedme.entries[i].enclosures[0]
except AttributeError:
logger.warn('Unable to retrieve results - probably just hitting it too fast...')
continue
id = urlparse.urlparse(feedme.entries[i].link).path.rpartition('/')[0]
torthetpse.append({
'site': picksite,
'title': feedme.entries[i].title,
'id': re.sub('/', '', id).strip(), #make sure to remove the leading '/'
'link': tmpenc['url'], #should change this to magnet
'pubdate': feedme.entries[i].updated,
'size': tmpenc['length']
})
#DEMONOID / FEED
elif pickfeed == "6":
tmpsz = feedme.entries[i].description
tmpsz_st = tmpsz.find('Size')
if tmpsz_st != -1:
tmpsize = tmpsz[tmpsz_st:tmpsz_st+14]
if any(['GB' in tmpsize, 'MB' in tmpsize, 'KB' in tmpsize, 'TB' in tmpsize]):
tmp1 = tmpsz.find('MB', tmpsz_st)
if tmp1 == -1:
tmp1 = tmpsz.find('GB', tmpsz_st)
if tmp1 == -1:
tmp1 = tmpsz.find('TB', tmpsz_st)
if tmp1 == -1:
tmp1 = tmpsz.find('KB', tmpsz_st)
tmpsz_end = tmp1 + 2
tmpsz_st += 7
else:
tmpsz_st = tmpsz.rfind('|')
if tmpsz_st != -1:
tmpsize = tmpsz[tmpsz_st:tmpsz_st+14]
if any(['GB' in tmpsize, 'MB' in tmpsize, 'KB' in tmpsize, 'TB' in tmpsize]):
tmp1 = tmpsz.find('MB', tmpsz_st)
if tmp1 == -1:
tmp1 = tmpsz.find('GB', tmpsz_st)
if tmp1 == -1:
tmp1 = tmpsz.find('TB', tmpsz_st)
if tmp1 == -1:
tmp1 = tmpsz.find('KB', tmpsz_st)
tmpsz_end = tmp1 + 2
tmpsz_st += 2
if 'KB' in tmpsz[tmpsz_st:tmpsz_end]:
szform = 'KB'
sz = 'K'
elif 'GB' in tmpsz[tmpsz_st:tmpsz_end]:
szform = 'GB'
sz = 'G'
elif 'MB' in tmpsz[tmpsz_st:tmpsz_end]:
szform = 'MB'
sz = 'M'
elif 'TB' in tmpsz[tmpsz_st:tmpsz_end]:
szform = 'TB'
sz = 'T'
tsize = helpers.human2bytes(str(tmpsz[tmpsz_st:tmpsz.find(szform, tmpsz_st) -1]) + str(sz))
#timestamp is in YYYY-MM-DDTHH:MM:SS+TZ :/
dt = feedme.entries[i].updated
try:
pd = datetime.strptime(dt[0:19], '%Y-%m-%dT%H:%M:%S')
pdate = pd.strftime('%a, %d %b %Y %H:%M:%S') + ' ' + re.sub(':', '', dt[19:]).strip()
#if dt[19]=='+':
# pdate+=timedelta(hours=int(dt[20:22]), minutes=int(dt[23:]))
#elif dt[19]=='-':
# pdate-=timedelta(hours=int(dt[20:22]), minutes=int(dt[23:]))
except:
pdate = feedme.entries[i].updated
feeddata.append({
'site': picksite,
'title': feedme.entries[i].title,
'link': str(urlparse.urlparse(feedme.entries[i].link)[2].rpartition('/')[0].rsplit('/',2)[2]),
'pubdate': pdate,
'size': tsize,
})
#32p / FEEDS
elif pickfeed == "1" or int(pickfeed) > 7:
tmpdesc = feedme.entries[i].description
st_pub = feedme.entries[i].title.find('(')
st_end = feedme.entries[i].title.find(')')
pub = feedme.entries[i].title[st_pub +1:st_end] # +1 to not include (
#logger.fdebug('publisher: ' + re.sub("'",'', pub).strip()) #publisher sometimes is given within quotes for some reason, strip 'em.
vol_find = feedme.entries[i].title.find('vol.')
series = feedme.entries[i].title[st_end +1:vol_find].strip()
series = re.sub('&amp;', '&', series).strip()
#logger.fdebug('series title: ' + series)
iss_st = feedme.entries[i].title.find(' - ', vol_find)
vol = re.sub('\.', '', feedme.entries[i].title[vol_find:iss_st]).strip()
#logger.fdebug('volume #: ' + str(vol))
issue = feedme.entries[i].title[iss_st +3:].strip()
#logger.fdebug('issue # : ' + str(issue))
try:
justdigits = feedme.entries[i].torrent_contentlength
except:
justdigits = '0'
seeddigits = 0
#if '0-Day Comics Pack' in series:
# logger.info('Comic Pack detected : ' + series)
# itd = True
if int(mylar.CONFIG.MINSEEDS) >= int(seeddigits):
#new releases has it as '&id', notification feeds have it as %ampid (possibly even &amp;id
link = feedme.entries[i].link
link = re.sub('&amp','&', link)
link = re.sub('&amp;','&', link)
linkst = link.find('&id')
linken = link.find('&', linkst +1)
if linken == -1:
linken = len(link)
newlink = re.sub('&id=', '', link[linkst:linken]).strip()
feeddata.append({
'site': picksite,
'title': series.lstrip() + ' ' + vol + ' #' + issue,
'volume': vol, # not stored by mylar yet.
'issue': issue, # not stored by mylar yet.
'link': newlink, #just the id for the torrent
'pubdate': feedme.entries[i].updated,
'size': justdigits
})
i += 1
if feedtype is None:
logger.info('[' + picksite + '] there were ' + str(i) + ' results..')
else:
logger.info('[' + picksite + '] there were ' + str(i) + ' results' + feedtype)
totalcount += i
lp += 1
if not seriesname:
#rss search results
rssdbupdate(feeddata, totalcount, 'torrent')
else:
#backlog (parsing) search results
if pickfeed == '4':
torinfo['entries'] = torthe32p
else:
torinfo['entries'] = torthetpse
return torinfo
return
def nzbs(provider=None, forcerss=False):
feedthis = []
def _parse_feed(site, url, verify):
logger.fdebug('[RSS] Fetching items from ' + site)
payload = None
headers = {'User-Agent': str(mylar.USER_AGENT)}
try:
r = requests.get(url, params=payload, verify=verify, headers=headers)
except Exception, e:
logger.warn('Error fetching RSS Feed Data from %s: %s' % (site, e))
return
feedme = feedparser.parse(r.content)
feedthis.append({"site": site,
"feed": feedme})
newznab_hosts = []
logger.info('config.newznab: %s' % mylar.CONFIG.NEWZNAB)
logger.info('extra.newznabs: %s' % mylar.CONFIG.EXTRA_NEWZNABS)
if mylar.CONFIG.NEWZNAB == 1:
for newznab_host in mylar.CONFIG.EXTRA_NEWZNABS:
logger.fdebug('[RSS] newznab name: ' + str(newznab_host[0]) + ' - enabled: ' + str(newznab_host[5]))
if str(newznab_host[5]) == '1':
newznab_hosts.append(newznab_host)
providercount = len(newznab_hosts) + int(mylar.CONFIG.EXPERIMENTAL == 1) + int(mylar.CONFIG.NZBSU == 1) + int(mylar.CONFIG.DOGNZB == 1)
logger.fdebug('[RSS] You have enabled ' + str(providercount) + ' NZB RSS search providers.')
if mylar.CONFIG.EXPERIMENTAL == 1:
max_entries = "250" if forcerss else "50"
_parse_feed('experimental', 'http://nzbindex.nl/rss/alt.binaries.comics.dcp/?sort=agedesc&max=' + max_entries + '&more=1', False)
if mylar.CONFIG.NZBSU == 1:
num_items = "&num=100" if forcerss else "" # default is 25
_parse_feed('nzb.su', 'http://api.nzb.su/rss?t=7030&dl=1&i=' + (mylar.CONFIG.NZBSU_UID or '1') + '&r=' + mylar.CONFIG.NZBSU_APIKEY + num_items, bool(mylar.CONFIG.NZBSU_VERIFY))
if mylar.CONFIG.DOGNZB == 1:
num_items = "&num=100" if forcerss else "" # default is 25
_parse_feed('dognzb', 'https://dognzb.cr/rss.cfm?r=' + mylar.CONFIG.DOGNZB_APIKEY + '&t=7030' + num_items, bool(mylar.CONFIG.DOGNZB_VERIFY))
for newznab_host in newznab_hosts:
site = newznab_host[0].rstrip()
(newznabuid, _, newznabcat) = (newznab_host[4] or '').partition('#')
newznabuid = newznabuid or '1'
newznabcat = newznabcat or '7030'
if site[-10:] == '[nzbhydra]':
#to allow nzbhydra to do category search by most recent (ie. rss)
_parse_feed(site, newznab_host[1].rstrip() + '/api?t=search&cat=' + str(newznabcat) + '&dl=1&i=' + str(newznabuid) + '&num=100&apikey=' + newznab_host[3].rstrip(), bool(newznab_host[2]))
else:
# 11-21-2014: added &num=100 to return 100 results (or maximum) - unsure of cross-reliablity
_parse_feed(site, newznab_host[1].rstrip() + '/rss?t=' + str(newznabcat) + '&dl=1&i=' + str(newznabuid) + '&num=100&r=' + newznab_host[3].rstrip(), bool(newznab_host[2]))
feeddata = []
for ft in feedthis:
site = ft['site']
logger.fdebug('[RSS] (' + site + ') now being updated...')
for entry in ft['feed'].entries:
if site == 'dognzb':
#because the rss of dog doesn't carry the enclosure item, we'll use the newznab size value
size = 0
if 'newznab' in entry and 'size' in entry['newznab']:
size = entry['newznab']['size']
else:
# experimental, nzb.su, newznab
size = entry.enclosures[0]['length']
# Link
if site == 'experimental':
link = entry.enclosures[0]['url']
else:
# dognzb, nzb.su, newznab
link = entry.link
#Remove the API keys from the url to allow for possible api key changes
if site == 'dognzb':
link = re.sub(mylar.CONFIG.DOGNZB_APIKEY, '', link).strip()
else:
link = link[:link.find('&i=')].strip()
feeddata.append({'Site': site,
'Title': entry.title,
'Link': link,
'Pubdate': entry.updated,
'Size': size})
# logger.fdebug(" Site: " + site)
# logger.fdebug(" Title: " + entry.title)
# logger.fdebug(" Link: " + link)
# logger.fdebug(" pubdate: " + entry.updated)
# logger.fdebug(" size: " + size)
logger.info('[RSS] (' + site + ') ' + str(len(ft['feed'].entries)) + ' entries indexed.')
i = len(feeddata)
if i:
logger.info('[RSS] ' + str(i) + ' entries have been indexed and are now going to be stored for caching.')
rssdbupdate(feeddata, i, 'usenet')
return
def rssdbupdate(feeddata, i, type):
rsschktime = 15
myDB = db.DBConnection()
#let's add the entries into the db so as to save on searches
#also to build up the ID's ;)
for dataval in feeddata:
if type == 'torrent':
#we just store the torrent ID's now.
newVal = {"Link": dataval['link'],
"Pubdate": dataval['pubdate'],
"Site": dataval['site'],
"Size": dataval['size']}
ctrlVal = {"Title": dataval['title']}
else:
newlink = dataval['Link']
newVal = {"Link": newlink,
"Pubdate": dataval['Pubdate'],
"Site": dataval['Site'],
"Size": dataval['Size']}
ctrlVal = {"Title": dataval['Title']}
myDB.upsert("rssdb", newVal, ctrlVal)
logger.fdebug('Completed adding new data to RSS DB. Next add in ' + str(mylar.CONFIG.RSS_CHECKINTERVAL) + ' minutes')
return
def torrentdbsearch(seriesname, issue, comicid=None, nzbprov=None, oneoff=False):
myDB = db.DBConnection()
seriesname_alt = None
if any([comicid is None, comicid == 'None', oneoff is True]):
pass
else:
logger.fdebug('ComicID: ' + str(comicid))
snm = myDB.selectone("SELECT * FROM comics WHERE comicid=?", [comicid]).fetchone()
if snm is None:
logger.fdebug('Invalid ComicID of ' + str(comicid) + '. Aborting search.')
return
else:
seriesname = snm['ComicName']
seriesname_alt = snm['AlternateSearch']
#remove 'and' and 'the':
tsearch_rem1 = re.sub("\\band\\b", "%", seriesname.lower())
tsearch_rem2 = re.sub("\\bthe\\b", "%", tsearch_rem1.lower())
tsearch_removed = re.sub('\s+', ' ', tsearch_rem2)
tsearch_seriesname = re.sub('[\'\!\@\#\$\%\:\-\;\/\\=\?\&\.\s\,]', '%', tsearch_removed)
if mylar.CONFIG.PREFERRED_QUALITY == 0:
tsearch = tsearch_seriesname + "%"
elif mylar.CONFIG.PREFERRED_QUALITY == 1:
tsearch = tsearch_seriesname + "%cbr%"
elif mylar.CONFIG.PREFERRED_QUALITY == 2:
tsearch = tsearch_seriesname + "%cbz%"
else:
tsearch = tsearch_seriesname + "%"
logger.fdebug('tsearch : ' + tsearch)
AS_Alt = []
tresults = []
tsearch = '%' + tsearch
if mylar.CONFIG.ENABLE_32P and nzbprov == '32P':
tresults = myDB.select("SELECT * FROM rssdb WHERE Title like ? AND Site='32P'", [tsearch])
if mylar.CONFIG.ENABLE_TPSE and nzbprov == 'TPSE':
tresults += myDB.select("SELECT * FROM rssdb WHERE Title like ? AND (Site='DEM' OR Site='WWT')", [tsearch])
logger.fdebug('seriesname_alt:' + str(seriesname_alt))
if seriesname_alt is None or seriesname_alt == 'None':
if tresults is None:
logger.fdebug('no Alternate name given. Aborting search.')
return "no results"
else:
chkthealt = seriesname_alt.split('##')
if chkthealt == 0:
AS_Alternate = seriesname_alt
AS_Alt.append(seriesname_alt)
for calt in chkthealt:
AS_Alter = re.sub('##', '', calt)
u_altsearchcomic = AS_Alter.encode('ascii', 'ignore').strip()
AS_Altrem = re.sub("\\band\\b", "", u_altsearchcomic.lower())
AS_Altrem = re.sub("\\bthe\\b", "", AS_Altrem.lower())
AS_Alternate = re.sub('[\_\#\,\/\:\;\.\-\!\$\%\+\'\&\?\@\s]', '%', AS_Altrem)
AS_Altrem_mod = re.sub('[\&]', ' ', AS_Altrem)
AS_formatrem_seriesname = re.sub('[\'\!\@\#\$\%\:\;\/\\=\?\.\,]', '', AS_Altrem_mod)
AS_formatrem_seriesname = re.sub('\s+', ' ', AS_formatrem_seriesname)
if AS_formatrem_seriesname[:1] == ' ': AS_formatrem_seriesname = AS_formatrem_seriesname[1:]
AS_Alt.append(AS_formatrem_seriesname)
if mylar.CONFIG.PREFERRED_QUALITY == 0:
AS_Alternate += "%"
elif mylar.CONFIG.PREFERRED_QUALITY == 1:
AS_Alternate += "%cbr%"
elif mylar.CONFIG.PREFERRED_QUALITY == 2:
AS_Alternate += "%cbz%"
else:
AS_Alternate += "%"
AS_Alternate = '%' + AS_Alternate
if mylar.CONFIG.ENABLE_32P and nzbprov == '32P':
tresults += myDB.select("SELECT * FROM rssdb WHERE Title like ? AND Site='32P'", [AS_Alternate])
if mylar.CONFIG.ENABLE_TPSE and nzbprov == 'TPSE':
tresults += myDB.select("SELECT * FROM rssdb WHERE Title like ? AND (Site='DEM' OR Site='WWT')", [AS_Alternate])
if tresults is None:
logger.fdebug('torrent search returned no results for ' + seriesname)
return "no results"
extensions = ('cbr', 'cbz')
tortheinfo = []
torinfo = {}
for tor in tresults:
#&amp; have been brought into the title field incorretly occassionally - patched now, but to include those entries already in the
#cache db that have the incorrect entry, we'll adjust.
torTITLE = re.sub('&amp;', '&', tor['Title']).strip()
#torsplit = torTITLE.split(' ')
if mylar.CONFIG.PREFERRED_QUALITY == 1:
if 'cbr' in torTITLE:
logger.fdebug('Quality restriction enforced [ cbr only ]. Accepting result.')
else:
logger.fdebug('Quality restriction enforced [ cbr only ]. Rejecting result.')
continue
elif mylar.CONFIG.PREFERRED_QUALITY == 2:
if 'cbz' in torTITLE:
logger.fdebug('Quality restriction enforced [ cbz only ]. Accepting result.')
else:
logger.fdebug('Quality restriction enforced [ cbz only ]. Rejecting result.')
continue
logger.fdebug('tor-Title: ' + torTITLE)
#logger.fdebug('there are ' + str(len(torsplit)) + ' sections in this title')
i=0
if nzbprov is not None:
if nzbprov != tor['Site'] and not any([mylar.CONFIG.ENABLE_TPSE, tor['Site'] != 'WWT', tor['Site'] != 'DEM']):
logger.fdebug('this is a result from ' + str(tor['Site']) + ', not the site I am looking for of ' + str(nzbprov))
continue
#0 holds the title/issue and format-type.
#--- this was for old cbt feeds, no longer used for 32p
# while (i < len(torsplit)):
# #we'll rebuild the string here so that it's formatted accordingly to be passed back to the parser.
# logger.fdebug('section(' + str(i) + '): ' + torsplit[i])
# #remove extensions
# titletemp = torsplit[i]
# titletemp = re.sub('cbr', '', titletemp)
# titletemp = re.sub('cbz', '', titletemp)
# titletemp = re.sub('none', '', titletemp)
# if i == 0:
# rebuiltline = titletemp
# else:
# rebuiltline = rebuiltline + ' (' + titletemp + ')'
# i+=1
# logger.fdebug('rebuiltline is :' + rebuiltline)
#----
seriesname_mod = seriesname
foundname_mod = torTITLE #torsplit[0]
seriesname_mod = re.sub("\\band\\b", " ", seriesname_mod.lower())
foundname_mod = re.sub("\\band\\b", " ", foundname_mod.lower())
seriesname_mod = re.sub("\\bthe\\b", " ", seriesname_mod.lower())
foundname_mod = re.sub("\\bthe\\b", " ", foundname_mod.lower())
seriesname_mod = re.sub('[\&]', ' ', seriesname_mod)
foundname_mod = re.sub('[\&]', ' ', foundname_mod)
formatrem_seriesname = re.sub('[\'\!\@\#\$\%\:\;\=\?\.\,]', '', seriesname_mod)
formatrem_seriesname = re.sub('[\-]', ' ', formatrem_seriesname)
formatrem_seriesname = re.sub('[\/]', ' ', formatrem_seriesname) #not necessary since seriesname in a torrent file won't have /
formatrem_seriesname = re.sub('\s+', ' ', formatrem_seriesname)
if formatrem_seriesname[:1] == ' ': formatrem_seriesname = formatrem_seriesname[1:]
formatrem_torsplit = re.sub('[\'\!\@\#\$\%\:\;\\=\?\.\,]', '', foundname_mod)
formatrem_torsplit = re.sub('[\-]', ' ', formatrem_torsplit) #we replace the - with space so we'll get hits if differnces
formatrem_torsplit = re.sub('[\/]', ' ', formatrem_torsplit) #not necessary since if has a /, should be removed in above line
formatrem_torsplit = re.sub('\s+', ' ', formatrem_torsplit)
logger.fdebug(str(len(formatrem_torsplit)) + ' - formatrem_torsplit : ' + formatrem_torsplit.lower())
logger.fdebug(str(len(formatrem_seriesname)) + ' - formatrem_seriesname :' + formatrem_seriesname.lower())
if formatrem_seriesname.lower() in formatrem_torsplit.lower() or any(x.lower() in formatrem_torsplit.lower() for x in AS_Alt):
logger.fdebug('matched to : ' + torTITLE)
logger.fdebug('matched on series title: ' + seriesname)
titleend = formatrem_torsplit[len(formatrem_seriesname):]
titleend = re.sub('\-', '', titleend) #remove the '-' which is unnecessary
#remove extensions
titleend = re.sub('cbr', '', titleend)
titleend = re.sub('cbz', '', titleend)
titleend = re.sub('none', '', titleend)
logger.fdebug('titleend: ' + titleend)
sptitle = titleend.split()
extra = ''
#the title on 32P has a mix-mash of crap...ignore everything after cbz/cbr to cleanit
#ctitle = torTITLE.find('cbr')
#if ctitle == 0:
# ctitle = torTITLE.find('cbz')
# if ctitle == 0:
# ctitle = torTITLE.find('none')
# if ctitle == 0:
# logger.fdebug('cannot determine title properly - ignoring for now.')
# continue
#cttitle = torTITLE[:ctitle]
# if tor['Site'] == '32P':
# st_pub = rebuiltline.find('(')
# if st_pub < 2 and st_pub != -1:
# st_end = rebuiltline.find(')')
# rebuiltline = rebuiltline[st_end +1:]
tortheinfo.append({
'title': torTITLE, #cttitle,
'link': tor['Link'],
'pubdate': tor['Pubdate'],
'site': tor['Site'],
'length': tor['Size']
})
torinfo['entries'] = tortheinfo
return torinfo
def nzbdbsearch(seriesname, issue, comicid=None, nzbprov=None, searchYear=None, ComicVersion=None, oneoff=False):
myDB = db.DBConnection()
seriesname_alt = None
if any([comicid is None, comicid == 'None', oneoff is True]):
pass
else:
snm = myDB.selectone("SELECT * FROM comics WHERE comicid=?", [comicid]).fetchone()
if snm is None:
logger.info('Invalid ComicID of ' + str(comicid) + '. Aborting search.')
return
else:
seriesname = snm['ComicName']
seriesname_alt = snm['AlternateSearch']
nsearch_seriesname = re.sub('[\'\!\@\#\$\%\:\;\/\\=\?\.\-\s]', '%', seriesname)
formatrem_seriesname = re.sub('[\'\!\@\#\$\%\:\;\/\\=\?\.]', '', seriesname)
nsearch = '%' + nsearch_seriesname + "%"
nresults = myDB.select("SELECT * FROM rssdb WHERE Title like ? AND Site=?", [nsearch, nzbprov])
if nresults is None:
logger.fdebug('nzb search returned no results for ' + seriesname)
if seriesname_alt is None:
logger.fdebug('no nzb Alternate name given. Aborting search.')
return "no results"
else:
chkthealt = seriesname_alt.split('##')
if chkthealt == 0:
AS_Alternate = AlternateSearch
for calt in chkthealt:
AS_Alternate = re.sub('##', '', calt)
AS_Alternate = '%' + AS_Alternate + "%"
nresults += myDB.select("SELECT * FROM rssdb WHERE Title like ? AND Site=?", [AS_Alternate, nzbprov])
if nresults is None:
logger.fdebug('nzb alternate name search returned no results.')
return "no results"
nzbtheinfo = []
nzbinfo = {}
if nzbprov == 'experimental':
except_list=['releases', 'gold line', 'distribution', '0-day', '0 day']
if ComicVersion:
ComVersChk = re.sub("[^0-9]", "", ComicVersion)
if ComVersChk == '':
ComVersChk = 0
else:
ComVersChk = 0
else:
ComVersChk = 0
filetype = None
if mylar.CONFIG.PREFERRED_QUALITY == 1: filetype = 'cbr'
elif mylar.CONFIG.PREFERRED_QUALITY == 2: filetype = 'cbz'
for results in nresults:
title = results['Title']
#logger.fdebug("titlesplit: " + str(title.split("\"")))
splitTitle = title.split("\"")
noYear = 'False'
_digits = re.compile('\d')
for subs in splitTitle:
#logger.fdebug(subs)
if len(subs) >= len(seriesname) and not any(d in subs.lower() for d in except_list) and bool(_digits.search(subs)) is True:
if subs.lower().startswith('for'):
# need to filter down alternate names in here at some point...
if seriesname.lower().startswith('for'):
pass
else:
#this is the crap we ignore. Continue
logger.fdebug('this starts with FOR : ' + str(subs) + '. This is not present in the series - ignoring.')
continue
if ComVersChk == 0:
noYear = 'False'
if ComVersChk != 0 and searchYear not in subs:
noYear = 'True'
noYearline = subs
if searchYear in subs and noYear == 'True':
#this would occur on the next check in the line, if year exists and
#the noYear check in the first check came back valid append it
subs = noYearline + ' (' + searchYear + ')'
noYear = 'False'
if noYear == 'False':
if filetype is not None:
if filetype not in subs.lower():
continue
nzbtheinfo.append({
'title': subs,
'link': re.sub('\/release\/', '/download/', results['Link']),
'pubdate': str(results['PubDate']),
'site': str(results['Site']),
'length': str(results['Size'])})
else:
for nzb in nresults:
# no need to parse here, just compile and throw it back ....
nzbtheinfo.append({
'title': nzb['Title'],
'link': nzb['Link'],
'pubdate': nzb['Pubdate'],
'site': nzb['Site'],
'length': nzb['Size']
})
#logger.fdebug("entered info for " + nzb['Title'])
nzbinfo['entries'] = nzbtheinfo
return nzbinfo
def torsend2client(seriesname, issue, seriesyear, linkit, site, pubhash=None):
logger.info('matched on ' + seriesname)
filename = helpers.filesafe(seriesname)
filename = re.sub(' ', '_', filename)
filename += "_" + str(issue) + "_" + str(seriesyear)
if linkit[-7:] != "torrent":
filename += ".torrent"
if any([mylar.USE_UTORRENT, mylar.USE_RTORRENT, mylar.USE_TRANSMISSION, mylar.USE_DELUGE, mylar.USE_QBITTORRENT]):
filepath = os.path.join(mylar.CONFIG.CACHE_DIR, filename)
logger.fdebug('filename for torrent set to : ' + filepath)
elif mylar.USE_WATCHDIR:
if mylar.CONFIG.TORRENT_LOCAL and mylar.CONFIG.LOCAL_WATCHDIR is not None:
filepath = os.path.join(mylar.CONFIG.LOCAL_WATCHDIR, filename)
logger.fdebug('filename for torrent set to : ' + filepath)
elif mylar.CONFIG.TORRENT_SEEDBOX and mylar.CONFIG.SEEDBOX_WATCHDIR is not None:
filepath = os.path.join(mylar.CONFIG.CACHE_DIR, filename)
logger.fdebug('filename for torrent set to : ' + filepath)
else:
logger.error('No Local Watch Directory or Seedbox Watch Directory specified. Set it and try again.')
return "fail"
cf_cookievalue = None
if site == '32P':
url = 'https://32pag.es/torrents.php'
if mylar.CONFIG.VERIFY_32P == 1 or mylar.CONFIG.VERIFY_32P == True:
verify = True
else:
verify = False
logger.fdebug('[32P] Verify SSL set to : ' + str(verify))
if mylar.CONFIG.MODE_32P == 0:
if mylar.KEYS_32P is None or mylar.CONFIG.PASSKEY_32P is None:
logger.warn('[32P] Unable to retrieve keys from provided RSS Feed. Make sure you have provided a CURRENT RSS Feed from 32P')
mylar.KEYS_32P = helpers.parse_32pfeed(mylar.FEED_32P)
if mylar.KEYS_32P is None or mylar.KEYS_32P == '':
return "fail"
else:
logger.fdebug('[32P-AUTHENTICATION] 32P (Legacy) Authentication Successful. Re-establishing keys.')
mylar.AUTHKEY_32P = mylar.KEYS_32P['authkey']
else:
logger.fdebug('[32P-AUTHENTICATION] 32P (Legacy) Authentication already done. Attempting to use existing keys.')
mylar.AUTHKEY_32P = mylar.KEYS_32P['authkey']
else:
if any([mylar.CONFIG.USERNAME_32P is None, mylar.CONFIG.USERNAME_32P == '', mylar.CONFIG.PASSWORD_32P is None, mylar.CONFIG.PASSWORD_32P == '']):
logger.error('[RSS] Unable to sign-on to 32P to validate settings and initiate download sequence. Please enter/check your username password in the configuration.')
return "fail"
elif mylar.CONFIG.PASSKEY_32P is None or mylar.AUTHKEY_32P is None or mylar.KEYS_32P is None:
logger.fdebug('[32P-AUTHENTICATION] 32P (Auth Mode) Authentication enabled. Keys have not been established yet, attempting to gather.')
feed32p = auth32p.info32p(reauthenticate=True)
feedinfo = feed32p.authenticate()
if feedinfo == "disable":
mylar.CONFIG.ENABLE_32P = 0
#mylar.config_write()
return "fail"
if mylar.CONFIG.PASSKEY_32P is None or mylar.AUTHKEY_32P is None or mylar.KEYS_32P is None:
logger.error('[RSS] Unable to sign-on to 32P to validate settings and initiate download sequence. Please enter/check your username password in the configuration.')
return "fail"
else:
logger.fdebug('[32P-AUTHENTICATION] 32P (Auth Mode) Authentication already done. Attempting to use existing keys.')
payload = {'action': 'download',
'torrent_pass': mylar.CONFIG.PASSKEY_32P,
'authkey': mylar.AUTHKEY_32P,
'id': linkit}
headers = None #{'Accept-encoding': 'gzip',
# 'User-Agent': str(mylar.USER_AGENT)}
elif site == 'TPSE':
pass
#linkit should be the magnet link since it's TPSE
#url = linkit
#url = helpers.torrent_create('TPSE', linkit)
#if url.startswith('https'):
# tpse_referrer = 'https://torrentproject.se/'
#else:
# tpse_referrer = 'http://torrentproject.se/'
#try:
# scraper = cfscrape.create_scraper()
# cf_cookievalue, cf_user_agent = scraper.get_tokens(url)
# headers = {'Accept-encoding': 'gzip',
# 'User-Agent': cf_user_agent}
#except Exception, e:
# return "fail"
#logger.fdebug('Grabbing torrent from url:' + str(url))
#payload = None
#verify = False
elif site == 'DEM':
url = helpers.torrent_create('DEM', linkit)
if url.startswith('https'):
dem_referrer = mylar.DEMURL + 'files/download/'
else:
dem_referrer = 'http' + mylar.DEMURL[5:] + 'files/download/'
headers = {'Accept-encoding': 'gzip',
'User-Agent': str(mylar.USER_AGENT),
'Referer': dem_referrer}
logger.fdebug('Grabbing torrent from url:' + str(url))
payload = None
verify = False
elif site == 'WWT':
url = helpers.torrent_create('WWT', linkit)
if url.startswith('https'):
wwt_referrer = mylar.WWTURL
else:
wwt_referrer = 'http' + mylar.WWTURL[5:]
headers = {'Accept-encoding': 'gzip',
'User-Agent': str(mylar.USER_AGENT),
'Referer': wwt_referrer}
logger.fdebug('Grabbing torrent [id:' + str(linkit) + '] from url:' + str(url))
payload = {'id': linkit}
verify = False
else:
headers = {'Accept-encoding': 'gzip',
'User-Agent': str(mylar.USER_AGENT)}
url = linkit
payload = None
verify = False
if site != 'TPSE':
if not verify:
#32P throws back an insecure warning because it can't validate against the CA. The below suppresses the message just for 32P instead of being displayed.
#disable SSL warnings - too many 'warning' messages about invalid certificates
try:
from requests.packages.urllib3 import disable_warnings
disable_warnings()
except ImportError:
#this is probably not necessary and redudant, but leaving in for the time being.
from requests.packages.urllib3.exceptions import InsecureRequestWarning
requests.packages.urllib3.disable_warnings()
try:
from urllib3.exceptions import InsecureRequestWarning
urllib3.disable_warnings()
except ImportError:
logger.warn('[EPIC FAILURE] Cannot load the requests module')
return "fail"
try:
logger.info('url: %s' % url)
logger.info('payload: %s' % payload)
scraper = cfscrape.create_scraper()
if cf_cookievalue:
r = scraper.get(url, params=payload, cookies=cf_cookievalue, verify=verify, stream=True, headers=headers)
else:
r = scraper.get(url, params=payload, verify=verify, stream=True, headers=headers)
#r = requests.get(url, params=payload, verify=verify, stream=True, headers=headers)
except Exception, e:
logger.warn('Error fetching data from %s (%s): %s' % (site, url, e))
if site == '32P':
logger.info('[TOR2CLIENT-32P] Retrying with 32P')
if mylar.CONFIG.MODE_32P == 1:
logger.info('[TOR2CLIENT-32P] Attempting to re-authenticate against 32P and poll new keys as required.')
feed32p = auth32p.info32p(reauthenticate=True)
feedinfo = feed32p.authenticate()
if feedinfo == "disable":
mylar.CONFIG.ENABLE_32P = 0
#mylar.config_write()
return "fail"
logger.debug('[TOR2CLIENT-32P] Creating CF Scraper')
scraper = cfscrape.create_scraper()
logger.debug('[TOR2CLIENT-32P] payload: %s \n verify %s \n headers %s \n', payload, verify, headers)
try:
r = scraper.get(url, params=payload, verify=verify, allow_redirects=True)
except Exception, e:
logger.warn('[TOR2CLIENT-32P] Unable to GET %s (%s): %s' % (site, url, e))
return "fail"
else:
logger.warn('[TOR2CLIENT-32P] Unable to authenticate using existing RSS Feed given. Make sure that you have provided a CURRENT feed from 32P')
return "fail"
else:
return "fail"
if any([site == 'TPSE', site == 'DEM', site == 'WWT']) and any([str(r.status_code) == '403', str(r.status_code) == '404', str(r.status_code) == '503']):
if str(r.status_code) != '503':
logger.warn('Unable to download from ' + site + ' [' + str(r.status_code) + ']')
#retry with the alternate torrent link.
url = helpers.torrent_create(site, linkit, True)
logger.fdebug('Trying alternate url: ' + str(url))
try:
r = requests.get(url, params=payload, verify=verify, stream=True, headers=headers)
except Exception, e:
return "fail"
else:
logger.warn('Cloudflare protection online for ' + site + '. Attempting to bypass...')
try:
scraper = cfscrape.create_scraper()
cf_cookievalue, cf_user_agent = cfscrape.get_cookie_string(url)
headers = {'Accept-encoding': 'gzip',
'User-Agent': cf_user_agent}
r = scraper.get(url, verify=verify, cookies=cf_cookievalue, stream=True, headers=headers)
except Exception, e:
return "fail"
if str(r.status_code) != '200':
logger.warn('Unable to download torrent from ' + site + ' [Status Code returned: ' + str(r.status_code) + ']')
logger.info('content: %s' % r.content)
return "fail"
if any([site == 'TPSE', site == 'DEM', site == 'WWT']):
if r.headers.get('Content-Encoding') == 'gzip':
buf = StringIO(r.content)
f = gzip.GzipFile(fileobj=buf)
with open(filepath, 'wb') as f:
for chunk in r.iter_content(chunk_size=1024):
if chunk: # filter out keep-alive new chunks
f.write(chunk)
f.flush()
logger.fdebug('[' + site + '] Saved torrent file to : ' + filepath)
else:
#tpse is magnet links only...
filepath = linkit
if mylar.USE_UTORRENT:
uTC = utorrent.utorrentclient()
if site == 'TPSE':
ti = uTC.addurl(linkit)
else:
ti = uTC.addfile(filepath, filename)
if ti == 'fail':
return ti
else:
#if ti is value, it will return the hash
torrent_info = []
torrent_info['hash'] = ti
torrent_info['clientmode'] = 'utorrent'
torrent_info['link'] = linkit
return torrent_info
elif mylar.USE_RTORRENT:
import test
rp = test.RTorrent()
torrent_info = rp.main(filepath=filepath)
if torrent_info:
torrent_info['clientmode'] = 'rtorrent'
torrent_info['link'] = linkit
return torrent_info
else:
return 'fail'
elif mylar.USE_TRANSMISSION:
try:
rpc = transmission.TorrentClient()
if not rpc.connect(mylar.CONFIG.TRANSMISSION_HOST, mylar.CONFIG.TRANSMISSION_USERNAME, mylar.CONFIG.TRANSMISSION_PASSWORD):
return "fail"
torrent_info = rpc.load_torrent(filepath)
if torrent_info:
torrent_info['clientmode'] = 'transmission'
torrent_info['link'] = linkit
return torrent_info
else:
return "fail"
except Exception as e:
logger.error(e)
return "fail"
elif mylar.USE_DELUGE:
try:
dc = deluge.TorrentClient()
if not dc.connect(mylar.CONFIG.DELUGE_HOST, mylar.CONFIG.DELUGE_USERNAME, mylar.CONFIG.DELUGE_PASSWORD):
logger.info('Not connected to Deluge!')
return "fail"
else:
logger.info('Connected to Deluge! Will try to add torrent now!')
torrent_info = dc.load_torrent(filepath)
if torrent_info:
torrent_info['clientmode'] = 'deluge'
torrent_info['link'] = linkit
return torrent_info
else:
return "fail"
logger.info('Unable to connect to Deluge!')
except Exception as e:
logger.error(e)
return "fail"
elif mylar.USE_QBITTORRENT:
try:
qc = qbittorrent.TorrentClient()
if not qc.connect(mylar.CONFIG.QBITTORRENT_HOST, mylar.CONFIG.QBITTORRENT_USERNAME, mylar.CONFIG.QBITTORRENT_PASSWORD):
logger.info('Not connected to qBittorrent - Make sure the Web UI is enabled and the port is correct!')
return "fail"
else:
logger.info('Connected to qBittorrent! Will try to add torrent now!')
torrent_info = qc.load_torrent(filepath)
if torrent_info['status'] is True:
torrent_info['clientmode'] = 'qbittorrent'
torrent_info['link'] = linkit
return torrent_info
else:
logger.info('Unable to add torrent to qBittorrent')
return "fail"
except Exception as e:
logger.error(e)
return "fail"
elif mylar.USE_WATCHDIR:
if mylar.CONFIG.TORRENT_LOCAL:
if site == 'TPSE':
torrent_info = {'hash': pubhash}
else:
#get the hash so it doesn't mess up...
torrent_info = helpers.get_the_hash(filepath)
torrent_info['clientmode'] = 'watchdir'
torrent_info['link'] = linkit
torrent_info['filepath'] = filepath
return torrent_info
else:
tssh = ftpsshup.putfile(filepath, filename)
return tssh
if __name__ == '__main__':
#torrents(sys.argv[1])
#torrentdbsearch(sys.argv[1], sys.argv[2], sys.argv[3])
nzbs(provider=sys.argv[1])