Arreglado serieslan
This commit is contained in:
@@ -14,10 +14,7 @@ from channels import autoplay
|
||||
|
||||
IDIOMAS = {'latino': 'Latino'}
|
||||
list_language = IDIOMAS.values()
|
||||
list_servers = ['openload',
|
||||
'okru',
|
||||
'netutv',
|
||||
'rapidvideo'
|
||||
list_servers = ['openload'
|
||||
]
|
||||
list_quality = ['default']
|
||||
|
||||
@@ -49,7 +46,11 @@ def lista(item):
|
||||
patron = '<a href="([^"]+)" '
|
||||
patron += 'class="link">.+?<img src="([^"]+)".*?'
|
||||
patron += 'title="([^"]+)">'
|
||||
|
||||
if item.url==host:
|
||||
a=1
|
||||
else:
|
||||
num=(item.url).split('-')
|
||||
a=int(num[1])
|
||||
matches = scrapertools.find_multiple_matches(data, patron)
|
||||
|
||||
# Paginacion
|
||||
@@ -57,17 +58,28 @@ def lista(item):
|
||||
min = item.page * num_items_x_pagina
|
||||
min=min-item.page
|
||||
max = min + num_items_x_pagina - 1
|
||||
|
||||
b=0
|
||||
for link, img, name in matches[min:max]:
|
||||
title = name
|
||||
b=b+1
|
||||
if " y " in name:
|
||||
title=name.replace(" y "," & ")
|
||||
else:
|
||||
title = name
|
||||
url = host + link
|
||||
scrapedthumbnail = host + img
|
||||
context1=[renumbertools.context(item), autoplay.context]
|
||||
itemlist.append(item.clone(title=title, url=url, action="episodios", thumbnail=scrapedthumbnail, show=title,
|
||||
context=context1))
|
||||
|
||||
itemlist.append(
|
||||
Item(channel=item.channel, title="Página Siguiente >>", url=item.url, action="lista", page=item.page + 1))
|
||||
logger.info("gasdfsa "+str(b))
|
||||
if b<29:
|
||||
a=a+1
|
||||
url="https://serieslan.com/pag-"+str(a)
|
||||
if b>10:
|
||||
itemlist.append(
|
||||
Item(channel=item.channel, title="Página Siguiente >>", url=url, action="lista", page=0))
|
||||
else:
|
||||
itemlist.append(
|
||||
Item(channel=item.channel, title="Página Siguiente >>", url=item.url, action="lista", page=item.page + 1))
|
||||
|
||||
tmdb.set_infoLabels(itemlist)
|
||||
return itemlist
|
||||
@@ -93,6 +105,10 @@ def episodios(item):
|
||||
|
||||
title = ""
|
||||
pat = "/"
|
||||
if "Mike, Lu & Og"==item.title:
|
||||
pat="&/"
|
||||
if "KND" in item.title:
|
||||
pat="-"
|
||||
# varios episodios en un enlace
|
||||
if len(name.split(pat)) > 1:
|
||||
i = 0
|
||||
@@ -134,7 +150,7 @@ def findvideos(item):
|
||||
itemlist = []
|
||||
|
||||
url_server = "https://openload.co/embed/%s/"
|
||||
url_api_get_key = "https://serieslan.com/ide.php?i=%s&k=%s"
|
||||
url_api_get_key = "https://serieslan.com/idx.php?i=%s&k=%s"
|
||||
|
||||
def txc(key, _str):
|
||||
s = range(256)
|
||||
@@ -157,7 +173,7 @@ def findvideos(item):
|
||||
return res
|
||||
|
||||
data = httptools.downloadpage(item.url).data
|
||||
pattern = '<div id="video" idv="([^"]*)" ide="([^"]*)" ids="[^"]*" class="video">'
|
||||
pattern = "<script type=.+?>.+?\['(.+?)','(.+?)','.+?'\]"
|
||||
idv, ide = scrapertools.find_single_match(data, pattern)
|
||||
thumbnail = scrapertools.find_single_match(data,
|
||||
'<div id="tab-1" class="tab-content current">.+?<img src="([^"]*)">')
|
||||
|
||||
Reference in New Issue
Block a user