Allowing URLs for 13.cl without the /programas prefix
This commit is contained in:
parent
94d5e90b4f
commit
409a16cb72
|
@ -5,22 +5,26 @@ from .common import InfoExtractor
|
|||
|
||||
|
||||
class Canal13clIE(InfoExtractor):
|
||||
_VALID_URL = r'^http://(?:www\.)?13\.cl/programa/'
|
||||
_VALID_URL = r'^http://(?:www\.)?13\.cl/'
|
||||
IE_NAME = 'Canal13cl'
|
||||
|
||||
def _real_extract(self, url):
|
||||
webpage = self._download_webpage(url)
|
||||
webpage = self._download_webpage(url, url)
|
||||
video_id = self._html_search_regex(
|
||||
r'http://streaming.13.cl/(.*)\.mp4',
|
||||
webpage, u'video_id')
|
||||
title = self._html_search_regex(
|
||||
r'articuloTitulo = \'(.*?)\'',
|
||||
r'(articuloTitulo = \"(.*?)\"|(.*?)\|)',
|
||||
webpage, u'title')
|
||||
url = self._html_search_regex(
|
||||
r'articuloVideo = \'(.*?)\'',
|
||||
r'articuloVideo = \"(.*?)\"',
|
||||
webpage, u'url')
|
||||
thumbnail = self._html_search_regex (
|
||||
r'articuloImagen = \'(.*?)\'',
|
||||
r'articuloImagen = \"(.*?)\"',
|
||||
webpage, u'thumbnail')
|
||||
|
||||
return {
|
||||
'video_id': video_id,
|
||||
'url': url,
|
||||
'title': title,
|
||||
'ext': 'mp4',
|
||||
|
|
Loading…
Reference in New Issue