1
0
mirror of https://github.com/moparisthebest/SickRage synced 2024-12-15 12:32:18 -05:00
SickRage/sickbeard/providers/torrentday.py

320 lines
11 KiB
Python
Raw Normal View History

# Author: Mr_Orange <mr_orange@hotmail.it>
#
# This file is part of Sick Beard.
#
# Sick Beard 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.
#
# Sick Beard 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 Sick Beard. If not, see <http://www.gnu.org/licenses/>.
import json
import re
import traceback
import datetime
import urlparse
import sickbeard
import generic
from sickbeard.common import Quality
from sickbeard import logger
from sickbeard import tvcache
from sickbeard import db
from sickbeard import classes
from sickbeard import helpers
from sickbeard import show_name_helpers
from sickbeard.common import Overview
from sickbeard.exceptions import ex
from sickbeard import clients
from lib import requests
from lib.requests import exceptions
from bs4 import BeautifulSoup
from lib.unidecode import unidecode
2014-04-27 10:48:19 -04:00
from sickbeard.helpers import sanitizeSceneName
class TorrentDayProvider(generic.TorrentProvider):
urls = {'base_url': 'http://www.torrentday.com',
'login': 'http://www.torrentday.com/torrents/',
'search': 'http://www.torrentday.com/V3/API/API.php',
'download': 'http://www.torrentday.com/download.php/%s/%s'
}
def __init__(self):
generic.TorrentProvider.__init__(self, "TorrentDay")
self.supportsBacklog = True
self.cache = TorrentDayCache(self)
self.url = self.urls['base_url']
self.cookies = None
self.categories = {'Season': {'c14': 1}, 'Episode': {'c2': 1, 'c26': 1, 'c7': 1, 'c24': 1},
'RSS': {'c2': 1, 'c26': 1, 'c7': 1, 'c24': 1, 'c14': 1}}
def isEnabled(self):
return sickbeard.TORRENTDAY
def imageName(self):
return 'torrentday.png'
def getQuality(self, item):
quality = Quality.sceneQuality(item[0])
return quality
def _doLogin(self):
if any(requests.utils.dict_from_cookiejar(self.session.cookies).values()):
return True
if sickbeard.TORRENTDAY_UID and sickbeard.TORRENTDAY_HASH:
requests.utils.add_dict_to_cookiejar(self.session.cookies, self.cookies)
else:
login_params = {'username': sickbeard.TORRENTDAY_USERNAME,
'password': sickbeard.TORRENTDAY_PASSWORD,
'submit.x': 0,
'submit.y': 0
}
try:
response = self.session.post(self.urls['login'], data=login_params, timeout=30)
except (requests.exceptions.ConnectionError, requests.exceptions.HTTPError), e:
logger.log(u'Unable to connect to ' + self.name + ' provider: ' + ex(e), logger.ERROR)
return False
if re.search('You tried too often', response.text):
logger.log(u'Too many login access for ' + self.name + ', can''t retrive any data', logger.ERROR)
return False
if response.status_code == 401:
logger.log(u'Invalid username or password for ' + self.name + ', Check your settings!', logger.ERROR)
return False
sickbeard.TORRENTDAY_UID = requests.utils.dict_from_cookiejar(self.session.cookies)['uid']
sickbeard.TORRENTDAY_HASH = requests.utils.dict_from_cookiejar(self.session.cookies)['pass']
self.cookies = {'uid': sickbeard.TORRENTDAY_UID,
'pass': sickbeard.TORRENTDAY_HASH
}
return True
def _get_season_search_strings(self, season, episode):
search_string = {'Season': [], 'Episode': []}
for show_name in set(show_name_helpers.allPossibleShowNames(self.show)):
ep_string = show_name + ' S%02d' % int(season) #1) showName SXX
search_string['Season'].append(ep_string)
search_string['Episode'] = self._get_episode_search_strings(season, episode)[0]['Episode']
return [search_string]
def _get_episode_search_strings(self, season, episode, add_string=''):
search_string = {'Episode': []}
if not episode:
return []
if self.show.air_by_date:
for show_name in set(show_name_helpers.allPossibleShowNames(self.show)):
2014-04-27 10:48:19 -04:00
ep_string = sanitizeSceneName(show_name) + ' ' + \
str(episode).replace('-', '|') + '|' + \
episode.strftime('%b')
search_string['Episode'].append(ep_string)
elif self.show.sports:
for show_name in set(show_name_helpers.allPossibleShowNames(self.show)):
ep_string = sanitizeSceneName(show_name) + ' ' + \
str(episode).replace('-', '|') + '|' + \
episode.strftime('%b')
search_string['Episode'].append(ep_string)
else:
for show_name in set(show_name_helpers.allPossibleShowNames(self.show)):
ep_string = show_name_helpers.sanitizeSceneName(show_name) + ' ' + \
sickbeard.config.naming_ep_type[2] % {'seasonnumber': season,
'episodenumber': episode}
search_string['Episode'].append(re.sub('\s+', ' ', ep_string))
return [search_string]
2014-04-05 07:34:49 -04:00
def _doSearch(self, search_params, show=None, age=None):
results = []
items = {'Season': [], 'Episode': [], 'RSS': []}
freeleech = '&free=on' if sickbeard.TORRENTDAY_FREELEECH else ''
if not self.session:
self._doLogin()
if not self._doLogin():
return []
for mode in search_params.keys():
for search_string in search_params[mode]:
logger.log(u"Search string: " + search_string, logger.DEBUG)
search_string = '+'.join(search_string.split())
post_data = dict({'/browse.php?': None, 'cata': 'yes', 'jxt': 8, 'jxw': 'b', 'search': search_string},
**self.categories[mode])
if sickbeard.TORRENTDAY_FREELEECH:
post_data.update({'free': 'on'})
data = self.session.post(self.urls['search'], data=post_data).json()
try:
torrents = data.get('Fs', [])[0].get('Cn', {}).get('torrents', [])
except:
continue
for torrent in torrents:
title = re.sub(r"\[.*\=.*\].*\[/.*\]", "", torrent['name'])
url = self.urls['download'] % ( torrent['id'], torrent['fname'] )
seeders = int(torrent['seed'])
leechers = int(torrent['leech'])
if mode != 'RSS' and seeders == 0:
continue
if not title or not url:
continue
item = title, url, seeders, leechers
items[mode].append(item)
results += items[mode]
return results
def _get_title_and_url(self, item):
title, url = item[0], item[1]
if url:
url = str(url).replace('&amp;', '&')
return (title, url)
2014-04-22 04:02:43 -04:00
def getURL(self, url, post_data=None, headers=None):
if not self.session:
self._doLogin()
try:
# Remove double-slashes from url
parsed = list(urlparse.urlparse(url))
parsed[2] = re.sub("/{2,}", "/", parsed[2]) # replace two or more / with one
url = urlparse.urlunparse(parsed)
response = self.session.get(url, verify=False)
except (requests.exceptions.ConnectionError, requests.exceptions.HTTPError), e:
logger.log(u"Error loading " + self.name + " URL: " + ex(e), logger.ERROR)
return None
if response.status_code != 200:
logger.log(self.name + u" page requested with url " + url + " returned status code is " + str(
response.status_code) + ': ' + clients.http_error_code[response.status_code], logger.WARNING)
return None
return response.content
def findPropers(self, search_date=datetime.datetime.today()):
results = []
sqlResults = db.DBConnection().select(
'SELECT s.show_name, e.showid, e.season, e.episode, e.status, e.airdate FROM tv_episodes AS e' +
' INNER JOIN tv_shows AS s ON (e.showid = s.indexer_id)' +
' WHERE e.airdate >= ' + str(search_date.toordinal()) +
' AND (e.status IN (' + ','.join([str(x) for x in Quality.DOWNLOADED]) + ')' +
' OR (e.status IN (' + ','.join([str(x) for x in Quality.SNATCHED]) + ')))'
)
if not sqlResults:
return []
for sqlshow in sqlResults:
self.show = curshow = helpers.findCertainShow(sickbeard.showList, int(sqlshow["showid"]))
curEp = curshow.getEpisode(int(sqlshow["season"]), int(sqlshow["episode"]))
season = curEp.scene_season
episode = curEp.scene_episode
if curshow.air_by_date or curshow.sports:
episode = curEp.airdate
searchString = self._get_episode_search_strings(curEp.scene_season, curEp.airdate if curshow.air_by_date or curshow.sports else curEp.scene_episode, add_string='PROPER|REPACK')
for item in self._doSearch(searchString[0], show=curshow):
title, url = self._get_title_and_url(item)
results.append(classes.Proper(title, url, datetime.datetime.today()))
return results
class TorrentDayCache(tvcache.TVCache):
def __init__(self, provider):
tvcache.TVCache.__init__(self, provider)
# Only poll IPTorrents every 10 minutes max
self.minTime = 10
def updateCache(self):
if not self.shouldUpdate():
return
search_params = {'RSS': ['']}
rss_results = self.provider._doSearch(search_params)
if rss_results:
self.setLastUpdate()
else:
return []
logger.log(u"Clearing " + self.provider.name + " cache and updating with new information")
self._clearCache()
cl = []
for result in rss_results:
item = (result[0], result[1])
ci = self._parseItem(item)
if ci is not None:
cl.append(ci)
if len(cl) > 0:
myDB = self._getDB()
myDB.mass_action(cl)
def _parseItem(self, item):
(title, url) = item
if not title or not url:
return None
logger.log(u"Adding item to cache: " + title, logger.DEBUG)
return self._addCacheEntry(title, url)
provider = TorrentDayProvider()