[search engine] Cosmetic update TorrentReactor

This commit is contained in:
DoumanAsh 2015-04-08 07:52:02 +03:00
parent 2fc1487603
commit 7dafb384e9
6 changed files with 12 additions and 10 deletions

Binary file not shown.

Before

Width:  |  Height:  |  Size: 252 B

After

Width:  |  Height:  |  Size: 951 B

View file

@ -1,4 +1,4 @@
#VERSION: 1.35
#VERSION: 1.36
#AUTHORS: Gekko Dam Beer (gekko04@users.sourceforge.net)
#CONTRIBUTORS: Christophe Dumez (chris@qbittorrent.org)
# Bruno Barbieri (brunorex@gmail.com)
@ -32,11 +32,11 @@ from helpers import download_file
import urllib
from HTMLParser import HTMLParser
from httplib import HTTPConnection as http
import re
from re import compile as re_compile
class torrentreactor(object):
url = 'http://www.torrentreactor.net'
name = 'TorrentReactor.Net'
name = 'TorrentReactor'
supported_categories = {'all': '', 'movies': '5', 'tv': '8', 'music': '6', 'games': '3', 'anime': '1', 'software': '2'}
def download_torrent(self, info):
@ -50,6 +50,7 @@ class torrentreactor(object):
self.results = results
self.id = None
self.url = url
self.torrents_matcher = re_compile("/torrents/\d+.*")
self.dispatcher = { 'a' : self.start_a, 'td' : self.start_td }
def handle_starttag(self, tag, attrs):
@ -58,7 +59,7 @@ class torrentreactor(object):
def start_a(self, attr):
params = dict(attr)
if re.match("/torrents/\d+.*", params['href']):
if self.torrents_matcher.match(params['href']):
self.current_item = {}
self.current_item['desc_link'] = self.url+params['href'].strip()
elif 'torrentreactor.net/download.php' in params['href']:

View file

@ -1,5 +1,5 @@
extratorrent: 1.2
torrentreactor: 1.35
torrentreactor: 1.36
mininova: 2.00
piratebay: 2.11
extratorrent: 2.0

Binary file not shown.

Before

Width:  |  Height:  |  Size: 252 B

After

Width:  |  Height:  |  Size: 951 B

View file

@ -1,4 +1,4 @@
#VERSION: 1.35
#VERSION: 1.36
#AUTHORS: Gekko Dam Beer (gekko04@users.sourceforge.net)
#CONTRIBUTORS: Christophe Dumez (chris@qbittorrent.org)
# Bruno Barbieri (brunorex@gmail.com)
@ -32,11 +32,11 @@ from helpers import download_file
from urllib import parse
from html.parser import HTMLParser
from http.client import HTTPConnection as http
import re
from re import compile as re_compile
class torrentreactor(object):
url = 'http://www.torrentreactor.net'
name = 'TorrentReactor.Net'
name = 'TorrentReactor'
supported_categories = {'all': '', 'movies': '5', 'tv': '8', 'music': '6', 'games': '3', 'anime': '1', 'software': '2'}
def download_torrent(self, info):
@ -50,6 +50,7 @@ class torrentreactor(object):
self.results = results
self.id = None
self.url = url
self.torrents_matcher = re_compile("/torrents/\d+.*")
self.dispatcher = { 'a' : self.start_a, 'td' : self.start_td }
def handle_starttag(self, tag, attrs):
@ -58,7 +59,7 @@ class torrentreactor(object):
def start_a(self, attr):
params = dict(attr)
if re.match("/torrents/\d+.*", params['href']):
if self.torrents_matcher.match(params['href']):
self.current_item = {}
self.current_item['desc_link'] = self.url+params['href'].strip()
elif 'torrentreactor.net/download.php' in params['href']:

View file

@ -1,5 +1,5 @@
extratorrent: 1.2
torrentreactor: 1.35
torrentreactor: 1.36
mininova: 2.00
piratebay: 2.11
extratorrent: 2.0