|
|
|
# VERSION: 2.0
|
|
|
|
# AUTHORS: imDMG [imdmgg@gmail.com]
|
|
|
|
|
|
|
|
# NoNaMe-Club search engine plugin for qBittorrent
|
|
|
|
|
|
|
|
import json
|
|
|
|
import logging
|
|
|
|
import math
|
|
|
|
import os
|
|
|
|
import re
|
|
|
|
import tempfile
|
|
|
|
import threading
|
|
|
|
import time
|
|
|
|
|
|
|
|
from urllib.request import build_opener, HTTPCookieProcessor, ProxyHandler
|
|
|
|
from urllib.parse import urlencode # , parse_qs
|
|
|
|
from urllib.error import URLError, HTTPError
|
|
|
|
from http.cookiejar import Cookie, CookieJar
|
|
|
|
from novaprinter import prettyPrinter
|
|
|
|
|
|
|
|
# setup logging into qBittorrent/logs
|
|
|
|
logging.basicConfig(level=logging.INFO,
|
|
|
|
format='%(asctime)s %(name)-12s %(levelname)-8s %(message)s',
|
|
|
|
datefmt='%m-%d %H:%M',
|
|
|
|
filename=os.path.abspath(os.path.join(os.path.dirname(__file__), '../../logs', 'nnmclub.log')),
|
|
|
|
filemode='w')
|
|
|
|
|
|
|
|
# benchmark
|
|
|
|
start_time = time.time()
|
|
|
|
|
|
|
|
|
|
|
|
class nnmclub(object):
|
|
|
|
name = 'NoNaMe-Club'
|
|
|
|
url = 'https://nnm-club.me/forum/'
|
|
|
|
supported_categories = {'all': '-1',
|
|
|
|
'movies': '14',
|
|
|
|
'tv': '27',
|
|
|
|
'music': '16',
|
|
|
|
'games': '17',
|
|
|
|
'anime': '24',
|
|
|
|
'software': '21'}
|
|
|
|
|
|
|
|
# getting config from kinozal.json
|
|
|
|
config = None
|
|
|
|
try:
|
|
|
|
# try to load user data from file
|
|
|
|
with open(os.path.abspath(os.path.join(os.path.dirname(__file__), 'nnmclub.json'))) as f:
|
|
|
|
config: dict = json.load(f)
|
|
|
|
except OSError as e:
|
|
|
|
# file not found
|
|
|
|
logging.error(e)
|
|
|
|
raise e
|
|
|
|
|
|
|
|
def __init__(self):
|
|
|
|
# establish connection
|
|
|
|
#
|
|
|
|
# make cookie
|
|
|
|
cj = CookieJar()
|
|
|
|
# if we wanna use https we mast add ssl=enable_ssl to cookie
|
|
|
|
c = Cookie(0, 'ssl', "enable_ssl", None, False, '.nnm-club.me',
|
|
|
|
True, False, '/', True, False, None, 'ParserCookie', None, None, None)
|
|
|
|
cj.set_cookie(c)
|
|
|
|
self.session = build_opener(HTTPCookieProcessor(cj))
|
|
|
|
|
|
|
|
# avoid endless waiting
|
|
|
|
self.blocked = False
|
|
|
|
|
|
|
|
# add proxy handler if needed
|
|
|
|
if self.config['proxy'] and any(self.config['proxies'].keys()):
|
|
|
|
self.session.add_handler(ProxyHandler(self.config['proxies']))
|
|
|
|
|
|
|
|
# change user-agent
|
|
|
|
self.session.addheaders.pop()
|
|
|
|
self.session.addheaders.append(('User-Agent', self.config['ua']))
|
|
|
|
|
|
|
|
response = self._catch_error_request(self.url + 'login.php')
|
|
|
|
if not self.blocked:
|
|
|
|
code = re.search(r'code"\svalue="(.+?)"', response.read().decode('cp1251'))[1]
|
|
|
|
form_data = {"username": self.config['username'],
|
|
|
|
"password": self.config['password'],
|
|
|
|
"autologin": "on",
|
|
|
|
"code": code,
|
|
|
|
"login": "Вход"}
|
|
|
|
# so we first encode keys to cp1251 then do default decode whole string
|
|
|
|
data_encoded = urlencode({k: v.encode('cp1251') for k, v in form_data.items()}).encode()
|
|
|
|
|
|
|
|
self._catch_error_request(self.url + 'login.php', data_encoded)
|
|
|
|
|
|
|
|
if 'phpbb2mysql_4_sid' not in [cookie.name for cookie in cj]:
|
|
|
|
logging.warning("we not authorized, please check your credentials")
|
|
|
|
else:
|
|
|
|
logging.info('We successfully authorized')
|
|
|
|
|
|
|
|
def draw(self, html: str):
|
|
|
|
torrents = re.findall(r'd\stopic.+?href="(.+?)".+?<b>(.+?)</b>.+?href="(d.+?)"'
|
|
|
|
r'.+?/u>\s(.+?)<.+?b>(\d+)</.+?b>(\d+)<', html, re.S)
|
|
|
|
|
|
|
|
for tor in torrents:
|
|
|
|
torrent = {"engine_url": self.url,
|
|
|
|
"desc_link": self.url + tor[0],
|
|
|
|
"name": tor[1],
|
|
|
|
"link": self.url + tor[2],
|
|
|
|
"size": tor[3].replace(',', '.'),
|
|
|
|
"seeds": tor[4],
|
|
|
|
"leech": tor[5]}
|
|
|
|
|
|
|
|
prettyPrinter(torrent)
|
|
|
|
del torrents
|
|
|
|
# return len(torrents)
|
|
|
|
|
|
|
|
def download_torrent(self, url):
|
|
|
|
if self.blocked:
|
|
|
|
return
|
|
|
|
# Create a torrent file
|
|
|
|
file, path = tempfile.mkstemp('.torrent')
|
|
|
|
file = os.fdopen(file, "wb")
|
|
|
|
|
|
|
|
# Download url
|
|
|
|
response = self._catch_error_request(url)
|
|
|
|
|
|
|
|
# Write it to a file
|
|
|
|
file.write(response.read())
|
|
|
|
file.close()
|
|
|
|
|
|
|
|
# return file path
|
|
|
|
logging.debug(path + " " + url)
|
|
|
|
print(path + " " + url)
|
|
|
|
|
|
|
|
def searching(self, query, first=False):
|
|
|
|
response = self._catch_error_request(query)
|
|
|
|
page = response.read().decode('cp1251')
|
|
|
|
self.draw(page)
|
|
|
|
total = int(re.search(r'\(max:\s(\d{1,3})\)', page)[1]) if first else -1
|
|
|
|
|
|
|
|
return total
|
|
|
|
|
|
|
|
def search(self, what, cat='all'):
|
|
|
|
if self.blocked:
|
|
|
|
return
|
|
|
|
c = self.supported_categories[cat]
|
|
|
|
query = '{}tracker.php?nm={}&{}'.format(self.url, what.replace(" ", "+"), "f=-1" if c == '-1' else "c=" + c)
|
|
|
|
|
|
|
|
# make first request (maybe it enough)
|
|
|
|
total = self.searching(query, True)
|
|
|
|
# do async requests
|
|
|
|
if total > 50:
|
|
|
|
tasks = []
|
|
|
|
for x in range(1, math.ceil(total / 50)):
|
|
|
|
task = threading.Thread(target=self.searching, args=(query + "&start={}".format(x * 50),))
|
|
|
|
tasks.append(task)
|
|
|
|
task.start()
|
|
|
|
|
|
|
|
# wait slower request in stack
|
|
|
|
for task in tasks:
|
|
|
|
task.join()
|
|
|
|
del tasks
|
|
|
|
|
|
|
|
logging.debug("--- {} seconds ---".format(time.time() - start_time))
|
|
|
|
logging.info("Found torrents: {}".format(total))
|
|
|
|
|
|
|
|
def _catch_error_request(self, url='', data=None):
|
|
|
|
url = url if url else self.url
|
|
|
|
|
|
|
|
try:
|
|
|
|
response = self.session.open(url, data)
|
|
|
|
# Only continue if response status is OK.
|
|
|
|
if response.getcode() != 200:
|
|
|
|
logging.error('Unable connect')
|
|
|
|
raise HTTPError(response.geturl(), response.getcode(),
|
|
|
|
"HTTP request to {} failed with status: {}".format(url, response.getcode()),
|
|
|
|
response.info(), None)
|
|
|
|
except (URLError, HTTPError) as e:
|
|
|
|
logging.error(e)
|
|
|
|
raise e
|
|
|
|
|
|
|
|
# checking that tracker is'nt blocked
|
|
|
|
self.blocked = False
|
|
|
|
if self.url not in response.geturl():
|
|
|
|
logging.warning("{} is blocked. Try proxy or another proxy".format(self.url))
|
|
|
|
self.blocked = True
|
|
|
|
|
|
|
|
return response
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
nnmclub_se = nnmclub()
|
|
|
|
nnmclub_se.search('supernatural')
|