#!/usr/bin/env python
# -*- coding: utf-8 -*-
import re
from module.plugins.Crypter import Crypter
class HoerbuchIn(Crypter):
__name__ = "HoerbuchIn"
__type__ = "container"
__pattern__ = r"http://(www\.)?hoerbuch\.in/(blog\.php\?id=|download_(.*)\.html)"
__version__ = "0.4"
__description__ = """Hoerbuch.in Container Plugin"""
__author_name__ = ("spoob")
__author_mail__ = ("spoob@pyload.org")
def __init__(self, parent):
Crypter.__init__(self, parent)
self.parent = parent
self.html = None
def download_html(self):
url = self.parent.url
self.html = self.req.load(url)
def file_exists(self):
""" returns True or False
"""
self.download_html()
if re.search(r"Download", self.html) is not None:
return True
return False
def proceed(self, url, location):
temp_links = []
download_container = ("Download", "Mirror #1", "Mirror #2", "Mirror #3")
for container in download_container:
download_content = re.search("
" + container + ":(.*?)
", self.html).group(1)
tmp = re.findall('Part \d+', download_content)
if tmp == []: continue
for link in tmp:
link_html = self.req.load(link, cookies=True)
temp_links.append(re.search('