mirror of https://github.com/YGGverse/qBt_SE.git
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
247 lines
8.5 KiB
247 lines
8.5 KiB
# VERSION: 1.0 |
|
# AUTHORS: imDMG [imdmgg@gmail.com] |
|
|
|
# Rutor.is search engine plugin for qBittorrent |
|
|
|
|
|
# gai\"><td>(.+?)</td.+?(/torrent/(\d+).+?)\">(.+?)<\/a.+?right\">(\d+)< |
|
|
|
|
|
import base64 |
|
import json |
|
import logging |
|
import os |
|
import re |
|
import socket |
|
import tempfile |
|
import time |
|
|
|
from concurrent.futures.thread import ThreadPoolExecutor |
|
from html import unescape |
|
from urllib.error import URLError, HTTPError |
|
from urllib.parse import unquote |
|
from urllib.request import build_opener, ProxyHandler |
|
|
|
from novaprinter import prettyPrinter |
|
|
|
# default config |
|
config = { |
|
"version": 2, |
|
"torrentDate": True, |
|
"username": "USERNAME", |
|
"password": "PASSWORD", |
|
"proxy": False, |
|
"proxies": { |
|
"http": "", |
|
"https": "" |
|
}, |
|
"ua": "Mozilla/5.0 (X11; Linux i686; rv:38.0) Gecko/20100101 Firefox/38.0 " |
|
} |
|
|
|
|
|
def path_to(*file): |
|
return os.path.abspath(os.path.join(os.path.dirname(__file__), *file)) |
|
|
|
|
|
def rng(t): |
|
return range(1, -(-t // 100)) |
|
|
|
|
|
PATTERNS = (r'(\d+)\s\(max. 2000\)<table', |
|
r'(?:gai|tum)"><td>(.+?)</td.+?(torrent/(\d+).+?)">(.+?)<\/a.+?' |
|
r'right">([\d\.]+ \w+)</td.+?alt="S"\s/>(.+?)<.+?red">(.+?)<', |
|
'%ssearch/%i/%i/000/0/%s') |
|
|
|
FILENAME = __file__[__file__.rfind('/') + 1:-3] |
|
FILE_J, FILE_C = [path_to(FILENAME + fe) for fe in ['.json', '.cookie']] |
|
|
|
# base64 encoded image |
|
ICON = ("AAABAAEAEBAAAAEAGABoAwAAFgAAACgAAAAQAAAAIAAAAAEAGAAAAAAAAAAAAAAAAAAAAA" |
|
"AAAAAAAAAAAAAAAc4AAMwHNdcQ4vsN3fYS2fUY3fUe3fMj4fkk4fco4PYo5fgk7f5gp8Zu" |
|
"ZZtsa59FIXZEGm4kh74PyeoLGp8NHK4PHrwQHr8VIb8XJL4bJrUcKJ8optEdtPMBGcQAIc" |
|
"XeZAPVYwdA3MQFf8EDAJoFAMEEAM0AANIAAM4AAM0EAL8CAI8bXaEV1/cBHMsGDNTVWAOo" |
|
"dTIU5/ELuOAJM6sEALsIAMoEALkCBbgFALUGAKshgMcvpNUTzOoFQNIFANqxQgBpkmgKue" |
|
"8IT8UUy+8HO7MHPb8Gt+IG3vQHm9YKi84X4foKI7kRl+AWiMwSDYyxjXZAy84HdNYEALcP" |
|
"guYM+vsL6PgGl/wBWN4K1/EF//8LbdQEALgEVc41zMp0YC+t0N0XxPcCIbwGAMkGGOUGUv" |
|
"QKPPUEANsIU9ENvvAJw/ULnekGAr8FJcIUzfRycEZwzuMFnuYEArQCAdYDANYHAMQFAMwG" |
|
"PcwM2vsHU/QKPegLwvYEEckFBrsOt/Y+kYky5/YGgNAGAKkHAc4JMssSoN0GTb0L2/gHYP" |
|
"kCAPkFKOMP0fIHGc0EAKwLgNAq3OMd/P0Al9ACBqQCAMALbOMG+/8E8v0KjugBAO4CAPAG" |
|
"Q9MNyPYEB8QBAKQCe8cW9//T+/09+/8Aqd8GIbIFAMAKbuUG6f8Ht/IFFeEAAMYPqeYMhO" |
|
"EGB6oCgtUY5fuG0tv//vzs+PlQ9fwAw+4CLLoIALgJR+EFU+wEFcweZNAkquMFMrkArOor" |
|
"4fSrxsvWx8n5/fv5+fn3+/iC8fsLzPIAUscEALMDAL8QPtAsetUFWsUHue1r7/vc6evOzM" |
|
"fFx8n5/fvy+fj89vb/9/e+9/o44/oNi9kBD54CFKQJg9Qu4vu09vr/+ff89fTIz8rFx8n5" |
|
"/fvy+fj59vb49vf/+fbh+vtk6vw1rN03suFn6vnl/f3/+fn49vj18/TIz8rFx8n5/fvy+f" |
|
"j59vb39vf39/f//P3w+fme6/ak8Prv+fj//f369/r39vj18/TIz8rFx8ngBwAA4AMAAMAD" |
|
"AADAAwAAwAMAAMABAACAAQAAgAEAAAAAAAAAAAAAgAEAAMADAADgBwAA+B8AAPw/AAD" |
|
"+fwAA") |
|
|
|
# setup logging |
|
logging.basicConfig( |
|
format="%(asctime)s %(name)-12s %(levelname)-8s %(message)s", |
|
datefmt="%m-%d %H:%M") |
|
logger = logging.getLogger(__name__) |
|
logger.setLevel(logging.DEBUG) |
|
|
|
try: |
|
# try to load user data from file |
|
with open(FILE_J, 'r+') as f: |
|
config = json.load(f) |
|
logger.debug("Config is loaded.") |
|
except OSError as e: |
|
logger.error(e) |
|
# if file doesn't exist, we'll create it |
|
with open(FILE_J, 'w') as f: |
|
f.write(json.dumps(config, indent=4, sort_keys=False)) |
|
# also write/rewrite ico file |
|
with open(path_to(FILENAME + '.ico'), 'wb') as f: |
|
f.write(base64.b64decode(ICON)) |
|
logger.debug("Write files.") |
|
|
|
|
|
class rutor: |
|
name = 'Rutor' |
|
url = 'http://rutor.info/' |
|
url_dl = url.replace("//", "//d.") |
|
supported_categories = {'all': 0, |
|
'movies': 1, |
|
'tv': 6, |
|
'music': 2, |
|
'games': 8, |
|
'anime': 10, |
|
'software': 9, |
|
'pictures': 3, |
|
'books': 11} |
|
|
|
def __init__(self): |
|
# error message |
|
self.error = None |
|
|
|
# establish connection |
|
self.session = build_opener() |
|
|
|
# add proxy handler if needed |
|
if config['proxy']: |
|
if any(config['proxies'].values()): |
|
self.session.add_handler(ProxyHandler(config['proxies'])) |
|
logger.debug("Proxy is set!") |
|
else: |
|
self.error = "Proxy enabled, but not set!" |
|
|
|
# change user-agent |
|
self.session.addheaders.pop() |
|
self.session.addheaders.append(('User-Agent', config['ua'])) |
|
|
|
def search(self, what, cat='all'): |
|
if self.error: |
|
self.pretty_error(what) |
|
return |
|
query = PATTERNS[2] % (self.url, 0, self.supported_categories[cat], |
|
what.replace(" ", "+")) |
|
|
|
# make first request (maybe it enough) |
|
t0, total = time.time(), self.searching(query, True) |
|
if self.error: |
|
self.pretty_error(what) |
|
return |
|
# do async requests |
|
if total > 100: |
|
query = query.replace('h/0', 'h/%i') |
|
qrs = [query % x for x in rng(total)] |
|
with ThreadPoolExecutor(len(qrs)) as executor: |
|
executor.map(self.searching, qrs, timeout=30) |
|
|
|
logger.debug(f"--- {time.time() - t0} seconds ---") |
|
logger.info(f"Found torrents: {total}") |
|
|
|
def download_torrent(self, url: str): |
|
# Download url |
|
response = self._catch_error_request(url) |
|
if self.error: |
|
self.pretty_error(url) |
|
return |
|
|
|
# Create a torrent file |
|
file, path = tempfile.mkstemp('.torrent') |
|
with os.fdopen(file, "wb") as fd: |
|
# Write it to a file |
|
fd.write(response.read()) |
|
|
|
# return file path |
|
logger.debug(path + " " + url) |
|
print(path + " " + url) |
|
|
|
def searching(self, query, first=False): |
|
response = self._catch_error_request(query) |
|
if not response: |
|
return None |
|
page = response.read().decode() |
|
self.draw(page) |
|
|
|
return int(re.search(PATTERNS[0], page)[1]) if first else -1 |
|
|
|
def draw(self, html: str): |
|
torrents = re.findall(PATTERNS[1], html, re.S) |
|
for tor in torrents: |
|
torrent_date = "" |
|
if config['torrentDate']: |
|
# replace names month |
|
table = {'Янв': '01', 'Фев': '02', 'Мар': '03', 'Апр': '04', |
|
'Май': '05', 'Июн': '06', 'Июл': '07', 'Авг': '08', |
|
'Сен': '09', 'Окт': '10', 'Ноя': '11', 'Дек': '12'} |
|
td = tor[0].split(" ") |
|
td[1] = table[td[1]] |
|
ct = " ".join(td) |
|
ct = time.strftime("%y.%m.%d", time.strptime(ct, "%d %m %y")) |
|
torrent_date = f'[{ct}] ' |
|
|
|
prettyPrinter({ |
|
"engine_url": self.url, |
|
"desc_link": self.url + tor[1], |
|
"name": torrent_date + unescape(tor[3]), |
|
"link": self.url_dl + "download/" + tor[2], |
|
"size": unescape(tor[4]), |
|
"seeds": unescape(tor[5]), |
|
"leech": unescape(tor[6]) |
|
}) |
|
del torrents |
|
|
|
def _catch_error_request(self, url='', data=None, retrieve=False): |
|
url = url or self.url |
|
|
|
try: |
|
response = self.session.open(url, data, 5) |
|
# checking that tracker is'nt blocked |
|
if not response.geturl().startswith((self.url, self.url_dl)): |
|
raise URLError(f"{self.url} is blocked. Try another proxy.") |
|
except (socket.error, socket.timeout) as err: |
|
if not retrieve: |
|
return self._catch_error_request(url, data, True) |
|
logger.error(err) |
|
self.error = f"{self.url} is not response! Maybe it is blocked." |
|
if "no host given" in err.args: |
|
self.error = "Proxy is bad, try another!" |
|
except (URLError, HTTPError) as err: |
|
logger.error(err.reason) |
|
self.error = err.reason |
|
if hasattr(err, 'code'): |
|
self.error = f"Request to {url} failed with status: {err.code}" |
|
else: |
|
return response |
|
|
|
return None |
|
|
|
def pretty_error(self, what): |
|
prettyPrinter({"engine_url": self.url, |
|
"desc_link": "https://github.com/imDMG/qBt_SE", |
|
"name": f"[{unquote(what)}][Error]: {self.error}", |
|
"link": self.url + "error", |
|
"size": "1 TB", # lol |
|
"seeds": 100, |
|
"leech": 100}) |
|
|
|
self.error = None |
|
|
|
|
|
if __name__ == "__main__": |
|
engine = rutor() |
|
engine.search('doctor')
|
|
|