aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor')
-rw-r--r--youtube_dl/extractor/aftonbladet.py11
1 files changed, 6 insertions, 5 deletions
diff --git a/youtube_dl/extractor/aftonbladet.py b/youtube_dl/extractor/aftonbladet.py
index a117502bc..e0518cf26 100644
--- a/youtube_dl/extractor/aftonbladet.py
+++ b/youtube_dl/extractor/aftonbladet.py
@@ -6,11 +6,11 @@ from ..utils import int_or_none
class AftonbladetIE(InfoExtractor):
- _VALID_URL = r'http://tv\.aftonbladet\.se/webbtv.+?(?P<id>article[0-9]+)\.ab(?:$|[?#])'
+ _VALID_URL = r'http://tv\.aftonbladet\.se/abtv/articles/(?P<id>[0-9]+)'
_TEST = {
- 'url': 'http://tv.aftonbladet.se/webbtv/nyheter/vetenskap/rymden/article36015.ab',
+ 'url': 'http://tv.aftonbladet.se/abtv/articles/36015',
'info_dict': {
- 'id': 'article36015',
+ 'id': '36015',
'ext': 'mp4',
'title': 'Vulkanutbrott i rymden - nu släpper NASA bilderna',
'description': 'Jupiters måne mest aktiv av alla himlakroppar',
@@ -25,8 +25,9 @@ class AftonbladetIE(InfoExtractor):
# find internal video meta data
meta_url = 'http://aftonbladet-play.drlib.aptoma.no/video/%s.json'
- internal_meta_id = self._html_search_regex(
- r'data-aptomaId="([\w\d]+)"', webpage, 'internal_meta_id')
+ player_config = self._parse_json(self._html_search_regex(
+ r'data-player-config="([^"]+)"', webpage, 'player config'), video_id)
+ internal_meta_id = player_config['videoId']
internal_meta_url = meta_url % internal_meta_id
internal_meta_json = self._download_json(
internal_meta_url, video_id, 'Downloading video meta data')