summaryrefslogtreecommitdiffstats
path: root/module
diff options
context:
space:
mode:
Diffstat (limited to 'module')
-rw-r--r--module/plugins/hoster/XvidstageCom.py92
1 files changed, 7 insertions, 85 deletions
diff --git a/module/plugins/hoster/XvidstageCom.py b/module/plugins/hoster/XvidstageCom.py
index 4962c05af..32c5fe0c3 100644
--- a/module/plugins/hoster/XvidstageCom.py
+++ b/module/plugins/hoster/XvidstageCom.py
@@ -17,98 +17,20 @@
@author: 4Christopher
"""
-import re
-import HTMLParser
+from module.plugins.hoster.XFileSharingPro import XFileSharingPro, create_getInfo
-from module.plugins.Hoster import Hoster
-from module.network.RequestFactory import getURL
-
-
-def setup(self):
- self.wantReconnect = False
- self.resumeDownload = True
- self.multiDL = True
-
-
-def getInfo(urls):
- result = []
-
- for url in urls:
- result.append(parseFileInfo(url, getInfoMode=True))
- yield result
-
-
-def parseFileInfo(url, getInfoMode=False):
- html = getURL(url)
- info = {"name": url, "size": 0, "status": 3}
- try:
- info['name'] = re.search(r'(?:Filename|Dateiname):</b></td><td nowrap[^>]*?>(.*?)<', html).group(1)
- info['size'] = re.search(r'(?:Size|Größe):</b></td><td>.*? <small>\((\d+?) bytes\)', html).group(1)
- except: ## The file is offline
- info['status'] = 1
- else:
- info['status'] = 2
-
- if getInfoMode:
- return info['name'], info['size'], info['status'], url
- else:
- return info['name'], info['size'], info['status'], html
-
-
-class XvidstageCom(Hoster):
+class XvidstageCom(XFileSharingPro):
__name__ = 'XvidstageCom'
- __version__ = '0.4'
+ __version__ = '0.5'
__pattern__ = r'http://(?:www.)?xvidstage.com/(?P<id>[0-9A-Za-z]+)'
__type__ = 'hoster'
__description__ = """A Plugin that allows you to download files from http://xvidstage.com"""
__author_name__ = ('4Christopher')
__author_mail__ = ('4Christopher@gmx.de')
+ HOSTER_NAME = "xvidstage.com"
- def process(self, pyfile):
- pyfile.name, pyfile.size, pyfile.status, self.html = parseFileInfo(pyfile.url)
- self.logDebug('Name: %s' % pyfile.name)
- if pyfile.status == 1: ## offline
- self.offline()
- self.id = re.search(self.__pattern__, pyfile.url).group('id')
-
- wait_sec = int(re.search(r'countdown_str">.+?>(\d+?)<', self.html).group(1))
- self.setWait(wait_sec, reconnect=False)
- self.logDebug('Waiting %d seconds before submitting the captcha' % wait_sec)
- self.wait()
-
- rand = re.search(r'<input type="hidden" name="rand" value="(.*?)">', self.html).group(1)
- self.logDebug('rand: %s, id: %s' % (rand, self.id))
- self.html = self.req.load(pyfile.url,
- post={'op': 'download2', 'id': self.id, 'rand': rand, 'code': self.get_captcha()})
- file_url = re.search(r'<a href="(?P<url>.*?)">(?P=url)</a>', self.html).group('url')
- try:
- hours_file_available = int(
- re.search(r'This direct link will be available for your IP next (?P<hours>\d+?) hours',
- self.html).group('hours'))
- self.logDebug(
- 'You have %d hours to download this file with your current IP address.' % hours_file_available)
- except:
- self.logDebug('Failed')
- self.logDebug('Download file: %s' % file_url)
- self.download(file_url)
- check = self.checkDownload({'empty': re.compile(r'^$')})
-
- if check == 'empty':
- self.logInfo('Downloaded File was empty')
- # self.retry()
-
- def get_captcha(self):
- ## <span style='position:absolute;padding-left:7px;padding-top:6px;'>&#49; …
- cap_chars = {}
- for pad_left, char in re.findall(r"position:absolute;padding-left:(\d+?)px;.*?;'>(.*?)<", self.html):
- cap_chars[int(pad_left)] = char
-
- h = HTMLParser.HTMLParser()
- ## Sorting after padding-left
- captcha = ''
- for pad_left in sorted(cap_chars):
- captcha += h.unescape(cap_chars[pad_left])
+ def setup(self):
+ self.resumeDownload = self.multiDL = self.premium
- self.logDebug('The captcha is: %s' % captcha)
- return captcha
+getInfo = create_getInfo(XvidstageCom)