You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
yt-dlp/yt_dlp/extractor/playsuisse.py

188 lines
6.7 KiB
Python

import json
from .common import InfoExtractor
from ..utils import int_or_none, traverse_obj
class PlaySuisseIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?playsuisse\.ch/(?:watch|detail)/(?:[^#]*[?&]episodeId=)?(?P<id>[0-9]+)'
_TESTS = [
{
# Old URL
'url': 'https://www.playsuisse.ch/watch/763211/0',
'only_matching': True,
},
{
# episode in a series
'url': 'https://www.playsuisse.ch/watch/763182?episodeId=763211',
'md5': '82df2a470b2dfa60c2d33772a8a60cf8',
'info_dict': {
'id': '763211',
'ext': 'mp4',
'title': 'Knochen',
'description': 'md5:8ea7a8076ba000cd9e8bc132fd0afdd8',
'duration': 3344,
'series': 'Wilder',
'season': 'Season 1',
'season_number': 1,
'episode': 'Knochen',
'episode_number': 1,
'thumbnail': 're:https://playsuisse-img.akamaized.net/',
}
}, {
# film
'url': 'https://www.playsuisse.ch/watch/808675',
'md5': '818b94c1d2d7c4beef953f12cb8f3e75',
'info_dict': {
'id': '808675',
'ext': 'mp4',
'title': 'Der Läufer',
'description': 'md5:9f61265c7e6dcc3e046137a792b275fd',
'duration': 5280,
'thumbnail': 're:https://playsuisse-img.akamaized.net/',
}
}, {
# series (treated as a playlist)
'url': 'https://www.playsuisse.ch/detail/1115687',
'info_dict': {
'description': 'md5:e4a2ae29a8895823045b5c3145a02aa3',
'id': '1115687',
'series': 'They all came out to Montreux',
'title': 'They all came out to Montreux',
},
'playlist': [{
'info_dict': {
'description': 'md5:f2462744834b959a31adc6292380cda2',
'duration': 3180,
'episode': 'Folge 1',
'episode_number': 1,
'id': '1112663',
'season': 'Season 1',
'season_number': 1,
'series': 'They all came out to Montreux',
'thumbnail': 're:https://playsuisse-img.akamaized.net/',
'title': 'Folge 1',
'ext': 'mp4'
},
}, {
'info_dict': {
'description': 'md5:9dfd308699fe850d3bce12dc1bad9b27',
'duration': 2935,
'episode': 'Folge 2',
'episode_number': 2,
'id': '1112661',
'season': 'Season 1',
'season_number': 1,
'series': 'They all came out to Montreux',
'thumbnail': 're:https://playsuisse-img.akamaized.net/',
'title': 'Folge 2',
'ext': 'mp4'
},
}, {
'info_dict': {
'description': 'md5:14a93a3356b2492a8f786ab2227ef602',
'duration': 2994,
'episode': 'Folge 3',
'episode_number': 3,
'id': '1112664',
'season': 'Season 1',
'season_number': 1,
'series': 'They all came out to Montreux',
'thumbnail': 're:https://playsuisse-img.akamaized.net/',
'title': 'Folge 3',
'ext': 'mp4'
}
}],
}
]
_GRAPHQL_QUERY = '''
query AssetWatch($assetId: ID!) {
assetV2(id: $assetId) {
...Asset
episodes {
...Asset
}
}
}
fragment Asset on AssetV2 {
id
name
description
duration
episodeNumber
seasonNumber
seriesName
medias {
type
url
}
thumbnail16x9 {
...ImageDetails
}
thumbnail2x3 {
...ImageDetails
}
thumbnail16x9WithTitle {
...ImageDetails
}
thumbnail2x3WithTitle {
...ImageDetails
}
}
fragment ImageDetails on AssetImage {
id
url
}'''
def _get_media_data(self, media_id):
# NOTE In the web app, the "locale" header is used to switch between languages,
# However this doesn't seem to take effect when passing the header here.
response = self._download_json(
'https://4bbepzm4ef.execute-api.eu-central-1.amazonaws.com/prod/graphql',
media_id, data=json.dumps({
'operationName': 'AssetWatch',
'query': self._GRAPHQL_QUERY,
'variables': {'assetId': media_id}
}).encode('utf-8'),
headers={'Content-Type': 'application/json', 'locale': 'de'})
return response['data']['assetV2']
def _real_extract(self, url):
media_id = self._match_id(url)
media_data = self._get_media_data(media_id)
info = self._extract_single(media_data)
if media_data.get('episodes'):
info.update({
'_type': 'playlist',
'entries': map(self._extract_single, media_data['episodes']),
})
return info
def _extract_single(self, media_data):
thumbnails = traverse_obj(media_data, lambda k, _: k.startswith('thumbnail'))
formats, subtitles = [], {}
for media in traverse_obj(media_data, 'medias', default=[]):
if not media.get('url') or media.get('type') != 'HLS':
continue
f, subs = self._extract_m3u8_formats_and_subtitles(
media['url'], media_data['id'], 'mp4', m3u8_id='HLS', fatal=False)
formats.extend(f)
self._merge_subtitles(subs, target=subtitles)
return {
'id': media_data['id'],
'title': media_data.get('name'),
'description': media_data.get('description'),
'thumbnails': thumbnails,
'duration': int_or_none(media_data.get('duration')),
'formats': formats,
'subtitles': subtitles,
'series': media_data.get('seriesName'),
'season_number': int_or_none(media_data.get('seasonNumber')),
'episode': media_data.get('name') if media_data.get('episodeNumber') else None,
'episode_number': int_or_none(media_data.get('episodeNumber')),
}