mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-17 01:25:11 +00:00
Merge pull request #65 from TobiX/fix-netzkino
[Netzkino] Only use video id to find metadata
This commit is contained in:
commit
4f51913680
1 changed files with 24 additions and 23 deletions
|
@ -13,17 +13,16 @@
|
|||
|
||||
|
||||
class NetzkinoIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?netzkino\.de/\#!/(?P<category>[^/]+)/(?P<id>[^/]+)'
|
||||
_VALID_URL = r'https?://(?:www\.)?netzkino\.de/\#!/[^/]+/(?P<id>[^/]+)'
|
||||
|
||||
_TEST = {
|
||||
'url': 'http://www.netzkino.de/#!/scifikino/rakete-zum-mond',
|
||||
_TESTS = [{
|
||||
'url': 'https://www.netzkino.de/#!/scifikino/rakete-zum-mond',
|
||||
'md5': '92a3f8b76f8d7220acce5377ea5d4873',
|
||||
'info_dict': {
|
||||
'id': 'rakete-zum-mond',
|
||||
'ext': 'mp4',
|
||||
'title': 'Rakete zum Mond (Endstation Mond, Destination Moon)',
|
||||
'comments': 'mincount:3',
|
||||
'description': 'md5:1eddeacc7e62d5a25a2d1a7290c64a28',
|
||||
'title': 'Rakete zum Mond \u2013 Jules Verne',
|
||||
'description': 'md5:f0a8024479618ddbfa450ff48ffa6c60',
|
||||
'upload_date': '20120813',
|
||||
'thumbnail': r're:https?://.*\.jpg$',
|
||||
'timestamp': 1344858571,
|
||||
|
@ -32,17 +31,30 @@ class NetzkinoIE(InfoExtractor):
|
|||
'params': {
|
||||
'skip_download': 'Download only works from Germany',
|
||||
}
|
||||
}
|
||||
}, {
|
||||
'url': 'https://www.netzkino.de/#!/filme/dr-jekyll-mrs-hyde-2',
|
||||
'md5': 'c7728b2dadd04ff6727814847a51ef03',
|
||||
'info_dict': {
|
||||
'id': 'dr-jekyll-mrs-hyde-2',
|
||||
'ext': 'mp4',
|
||||
'title': 'Dr. Jekyll & Mrs. Hyde 2',
|
||||
'description': 'md5:c2e9626ebd02de0a794b95407045d186',
|
||||
'upload_date': '20190130',
|
||||
'thumbnail': r're:https?://.*\.jpg$',
|
||||
'timestamp': 1548849437,
|
||||
'age_limit': 18,
|
||||
},
|
||||
'params': {
|
||||
'skip_download': 'Download only works from Germany',
|
||||
}
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
category_id = mobj.group('category')
|
||||
video_id = mobj.group('id')
|
||||
|
||||
api_url = 'http://api.netzkino.de.simplecache.net/capi-2.0a/categories/%s.json?d=www' % category_id
|
||||
api_info = self._download_json(api_url, video_id)
|
||||
info = next(
|
||||
p for p in api_info['posts'] if p['slug'] == video_id)
|
||||
api_url = 'https://api.netzkino.de.simplecache.net/capi-2.0a/movies/%s.json?d=www' % video_id
|
||||
info = self._download_json(api_url, video_id)
|
||||
custom_fields = info['custom_fields']
|
||||
|
||||
production_js = self._download_webpage(
|
||||
|
@ -67,23 +79,12 @@ def _real_extract(self, url):
|
|||
} for key, tpl in templates.items()]
|
||||
self._sort_formats(formats)
|
||||
|
||||
comments = [{
|
||||
'timestamp': parse_iso8601(c.get('date'), delimiter=' '),
|
||||
'id': c['id'],
|
||||
'author': c['name'],
|
||||
'html': c['content'],
|
||||
'parent': 'root' if c.get('parent', 0) == 0 else c['parent'],
|
||||
} for c in info.get('comments', [])]
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'formats': formats,
|
||||
'comments': comments,
|
||||
'title': info['title'],
|
||||
'age_limit': int_or_none(custom_fields.get('FSK')[0]),
|
||||
'timestamp': parse_iso8601(info.get('date'), delimiter=' '),
|
||||
'description': clean_html(info.get('content')),
|
||||
'thumbnail': info.get('thumbnail'),
|
||||
'playlist_title': api_info.get('title'),
|
||||
'playlist_id': category_id,
|
||||
}
|
||||
|
|
Loading…
Reference in a new issue