|
|
|
@ -6,6 +6,7 @@ from .common import InfoExtractor
|
|
|
|
|
from ..utils import (
|
|
|
|
|
clean_html,
|
|
|
|
|
determine_ext,
|
|
|
|
|
ExtractorError,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@ -29,14 +30,15 @@ class FKTVIE(InfoExtractor):
|
|
|
|
|
webpage = self._download_webpage('http://fernsehkritik.tv/folge-%s/play' % episode, episode)
|
|
|
|
|
title = clean_html(self._html_search_regex('<h3>([^<]+?)</h3>', webpage, 'title'))
|
|
|
|
|
matches = re.search(r'(?s)<video[^>]*poster="([^"]+)"[^>]*>(.*?)</video>', webpage)
|
|
|
|
|
if matches:
|
|
|
|
|
poster, sources = matches.groups()
|
|
|
|
|
urls = re.findall(r'(?s)<source[^>]*src="([^"]+)"[^>]*>', sources)
|
|
|
|
|
if sources:
|
|
|
|
|
formats = [{'url': url, 'format_id': determine_ext(url)} for url in urls]
|
|
|
|
|
return {
|
|
|
|
|
'id': episode,
|
|
|
|
|
'title': title,
|
|
|
|
|
'formats': formats,
|
|
|
|
|
'thumbnail': poster,
|
|
|
|
|
}
|
|
|
|
|
if matches is None:
|
|
|
|
|
raise ExtractorError('Unable to extract the video')
|
|
|
|
|
|
|
|
|
|
poster, sources = matches.groups()
|
|
|
|
|
urls = re.findall(r'(?s)<source[^>]*src="([^"]+)"[^>]*>', sources)
|
|
|
|
|
formats = [{'url': url, 'format_id': determine_ext(url)} for url in urls]
|
|
|
|
|
return {
|
|
|
|
|
'id': episode,
|
|
|
|
|
'title': title,
|
|
|
|
|
'formats': formats,
|
|
|
|
|
'thumbnail': poster,
|
|
|
|
|
}
|
|
|
|
|