# -*- coding: utf-8 -*- ''' Torrenter v2 plugin for XBMC/Kodi Copyright (C) 2012-2015 Vadim Skorba v1 - DiMartino v2 https://forums.tvaddons.ag/addon-releases/29224-torrenter-v2.html 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 re import tempfile import hashlib import os from StringIO import StringIO import zlib import socket import sys proxy = int(sys.modules["__main__"].__settings__.getSetting("proxy")) if proxy == 2: socks_ip = sys.modules["__main__"].__settings__.getSetting("socks_ip") from resources import socks socks.setdefaultproxy(socks.PROXY_TYPE_SOCKS5, socks_ip, 9050) socket.socket = socks.socksocket import urllib import urllib2 import cookielib import xbmcgui import xbmc import Localization from functions import log, debug, showMessage import ssl #ssl._create_default_https_context = ssl._create_unverified_context class SearcherABC: searchIcon = '/icons/video.png' sourceWeight = 1 cookieJar = None timeout_multi=int(sys.modules["__main__"].__settings__.getSetting("timeout")) __plugin__='Empty v 0 0 0' baseurl = 'site.com' socket.setdefaulttimeout(10+(10*int(timeout_multi))) def search(self, keyword): ''' Retrieve keyword from the input and return a list of tuples: filesList.append(( int(weight), int(seeds), str(title), str(link), str(image), )) ''' return def isMagnetLinkSource(self): return 'Should never see this' def getTorrentFile(self, url): return url def sizeConvert(self, sizeBytes): if long(sizeBytes) >= 1024 * 1024 * 1024: size = str(long(sizeBytes) / (1024 * 1024 * 1024)) + 'GB' elif long(sizeBytes) >= 1024 * 1024: size = str(long(sizeBytes) / (1024 * 1024)) + 'MB' elif sizeBytes >= 1024: size = str(long(sizeBytes) / 1024) + 'KB' else: size = str(long(sizeBytes)) + 'B' return size def check_login(self, response=None): return True def login(self): return True def load_cookie(self): cookie=os.path.join(self.tempdir(), self.__class__.__name__+'.txt') self.cookieJar = cookielib.MozillaCookieJar(cookie) try: if os.path.exists(cookie): self.cookieJar.load(ignore_discard=True) except: self.log('[load_cookie]: os.remove(cookie)') os.remove(cookie) self.cookieJar = cookielib.MozillaCookieJar(cookie) def clear_cookie(self, domain): cookie=os.path.join(self.tempdir(),self.__class__.__name__+'.txt') self.cookieJar = cookielib.MozillaCookieJar(cookie) if os.path.exists(cookie): os.remove(cookie) self.log('[clear_cookie]: cookie cleared') def makeRequest(self, url, data={}, headers={}): self.load_cookie() opener = urllib2.build_opener(urllib2.HTTPCookieProcessor(self.cookieJar)) if proxy == 1: try: from resources.proxy import antizapret opener.add_handler(antizapret.AntizapretProxyHandler()) config = antizapret.config() self.debug('[antizapret]: '+str(config["domains"])) self.debug('[antizapret]: '+str(config["server"])) except: showMessage('AntiZapret', Localization.localize('Error')) self.debug('[antizapret]: OFF!') # python ssl Context support - PEP 0466 if 'https:' in url: ssl_context = ssl.create_default_context() ssl_context.check_hostname = False ssl_context.verify_mode = ssl.CERT_NONE log('urllib2.HTTPSHandler(context=ssl_context)') opener.add_handler(urllib2.HTTPSHandler(context=ssl_context)) opener.addheaders = headers if 0 < len(data): encodedData = urllib.urlencode(data) else: encodedData = None try: response = opener.open(url, encodedData) except urllib2.HTTPError as e: if e.code == 404: self.log('[makeRequest]: Not Found! HTTP Error, e.code=' + str(e.code)) return elif e.code in [503]: self.log('[makeRequest]: Denied, HTTP Error, e.code=' + str(e.code)) return else: self.log('[makeRequest]: HTTP Error, e.code=' + str(e.code)) return #self.cookieJar.extract_cookies(response, urllib2) #self.log(response.info().get('Set-Cookie')) if response.info().get('Content-Encoding') == 'gzip': buf = StringIO(response.read()) decomp = zlib.decompressobj(16 + zlib.MAX_WBITS) text = decomp.decompress(buf.getvalue()) else: text = response.read() return text def askCaptcha(self, url): temp_dir = self.tempdir() if isinstance(temp_dir, list): temp_dir = temp_dir[0] urllib.URLopener().retrieve(url, temp_dir + '/captcha.png') window = xbmcgui.Window(xbmcgui.getCurrentWindowId()) if isinstance(temp_dir, list): temp_dir = temp_dir[0] image = xbmcgui.ControlImage(460, 20, 360, 160, temp_dir + '/captcha.png') window.addControl(image) keyboardCaptcha = xbmc.Keyboard('', '[%s] %s' % (self.__plugin__, Localization.localize('Input symbols from CAPTCHA image:'))) keyboardCaptcha.doModal() captchaText = keyboardCaptcha.getText() window.removeControl(image) if not captchaText: return False else: return captchaText htmlCodes = ( ('&', '&'), ('<', '<'), ('>', '>'), ('"', '"'), ('"', '"'), ("'", '''), ("&", '&'), ("'", '''), ("&", '&'),) stripPairs = ( ('

', '\n'), ('

  • ', '\n'), ('
    ', '\n'), ('<.+?>', ' '), ('', ' '), (' ', ' '), ('«', '"'), ('»', '"'), ) def unescape(self, string): for (symbol, code) in self.htmlCodes: string = re.sub(code, symbol, string) return string def stripHtml(self, string): for (html, replacement) in self.stripPairs: string = re.sub(html, replacement, string) return string def md5(self, string): hasher = hashlib.md5() hasher.update(string) return hasher.hexdigest() def tempdir(self): dirname = xbmc.translatePath('special://temp') for subdir in ('xbmcup', 'plugin.video.torrenter'): dirname = os.path.join(dirname, subdir) if not os.path.exists(dirname): os.mkdir(dirname) return dirname def timeout(self, add_seconds=0): seconds=10+(10*int(self.timeout_multi))+int(add_seconds) socket.setdefaulttimeout(int(seconds)) def clean(self, string): specials = ['/', '\\', '-', '[', ']', '(', ')', ','] for symbol in specials: string = string.replace(symbol, ' ') if len(string) > 120: string = string[:120] last_piece = string.split(' ')[-1] string = string[:120 - len(last_piece)].strip() return string def saveTorrentFile(self, url, content): try: temp_dir = tempfile.gettempdir() except: temp_dir = self.tempdir() localFileName = temp_dir + os.path.sep + self.md5(url) + ".torrent" localFile = open(localFileName, 'wb+') localFile.write(content) localFile.close() return localFileName def logout(self): pass def log(self, msg): log('[%s] ' % self.__plugin__ +msg) def debug(self, msg): debug('[%s] ' % self.__plugin__ +msg) def open2(self, url=''): import httplib conn = httplib.HTTPConnection(self.baseurl) conn.request("GET", str(url)) r1 = conn.getresponse() status = str(r1.status) + " " + r1.reason content = r1.read() self.debug('[open2] status:'+str(status)) return content def showMessage(self, heading, message, times=10000): xbmc.executebuiltin('XBMC.Notification("%s", "%s", %s, "%s")' % ( heading.replace('"', "'"), message.replace('"', "'"), times, self.searchIcon)) self.log(str((heading.replace('"', "'"), message.replace('"', "'"), times, self.searchIcon)))