|
|
@ -85,9 +85,9 @@ class FranceTVBaseInfoExtractor(InfoExtractor): |
|
|
|
|
|
|
|
subtitles = {} |
|
|
|
subtitles_list = [{ |
|
|
|
'url': subtitle['url'], |
|
|
|
'ext': subtitle.get('format'), |
|
|
|
} for subtitle in info.get('subtitles', []) if subtitle.get('url')] |
|
|
|
'url': subformat['url'], |
|
|
|
'ext': subformat.get('format'), |
|
|
|
} for subformat in info.get('subtitles', []) if subformat.get('url')] |
|
|
|
if subtitles_list: |
|
|
|
subtitles['fr'] = subtitles_list |
|
|
|
|
|
|
@ -118,7 +118,7 @@ class PluzzIE(FranceTVBaseInfoExtractor): |
|
|
|
'id_video', webpage, 'video id', default=None) |
|
|
|
if not video_id: |
|
|
|
video_id = self._search_regex( |
|
|
|
r'data-diffusion=["\'](\d+)', webpage, 'video id') |
|
|
|
r'data-diffusion=["\'](\d+)', webpage, 'video id') |
|
|
|
|
|
|
|
return self._extract_video(video_id, 'Pluzz') |
|
|
|
|
|
|
|