Files
addon/servers/vidoza.py
2020-09-19 16:32:37 +02:00

48 lines
1.6 KiB
Python

# -*- coding: utf-8 -*-
from core import httptools
from core import jsontools
from core import scrapertools
from platformcode import logger, config
def test_video_exists(page_url):
logger.info("(page_url='%s')" % page_url)
global data
data = httptools.downloadpage(page_url).data
if "Page not found" in data or "File was deleted" in data:
return False, config.get_localized_string(70449) % "vidoza"
elif "processing" in data:
return False, config.get_localized_string(70449) % "Vidoza"
return True, ""
def get_video_url(page_url, premium=False, user="", password="", video_password=""):
logger.info("(page_url='%s')" % page_url)
global data
video_urls = []
s = scrapertools.find_single_match(data, r'sourcesCode\s*:\s*(\[\{.*?\}\])')
s = s.replace('src:', '"src":').replace('file:', '"file":').replace('type:', '"type":').replace('label:', '"label":').replace('res:', '"res":')
try:
data = jsontools.load(s)
for enlace in data:
if 'src' in enlace or 'file' in enlace:
url = enlace['src'] if 'src' in enlace else enlace['file']
tit = ''
if 'label' in enlace: tit += ' [%s]' % enlace['label']
if 'res' in enlace: tit += ' [%s]' % enlace['res']
if tit == '' and 'type' in enlace: tit = enlace['type']
if tit == '': tit = '.mp4'
video_urls.append(["%s [Vidoza]" % tit, url])
except:
logger.debug('No se detecta json %s' % s)
pass
video_urls.reverse()
return video_urls