yt-dlp/yt_dlp/extractor/tf1.py

102 lines
3.8 KiB
Python
Raw Normal View History

import json
2013-06-29 07:18:19 -06:00
from .common import InfoExtractor
from ..utils import (
int_or_none,
parse_iso8601,
try_get,
)
2014-03-29 08:05:15 -06:00
2013-06-29 07:18:19 -06:00
class TF1IE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?tf1\.fr/[^/]+/(?P<program_slug>[^/]+)/videos/(?P<id>[^/?&#]+)\.html'
2015-05-25 23:32:43 -06:00
_TESTS = [{
'url': 'https://www.tf1.fr/tmc/quotidien-avec-yann-barthes/videos/quotidien-premiere-partie-11-juin-2019.html',
2015-01-03 17:05:31 -07:00
'info_dict': {
'id': '13641379',
2015-01-03 17:05:31 -07:00
'ext': 'mp4',
'title': 'md5:f392bc52245dc5ad43771650c96fb620',
'description': 'md5:a02cdb217141fb2d469d6216339b052f',
'upload_date': '20190611',
'timestamp': 1560273989,
'duration': 1738,
'series': 'Quotidien avec Yann Barthès',
'tags': ['intégrale', 'quotidien', 'Replay'],
2015-01-03 17:05:31 -07:00
},
'params': {
# Sometimes wat serves the whole file with the --test option
'skip_download': True,
},
}, {
'url': 'https://www.tf1.fr/tmc/burger-quiz/videos/burger-quiz-du-19-aout-2023-s03-episode-21-85585666.html',
'info_dict': {
'id': '14010600',
'ext': 'mp4',
'title': 'Burger Quiz - S03 EP21 avec Eye Haidara, Anne Depétrini, Jonathan Zaccaï et Pio Marmaï',
'thumbnail': 'https://photos.tf1.fr/1280/720/burger-quiz-11-9adb79-0@1x.jpg',
'description': 'Manu Payet recevra Eye Haidara, Anne Depétrini, Jonathan Zaccaï et Pio Marmaï.',
'upload_date': '20230819',
'timestamp': 1692469471,
'season_number': 3,
'series': 'Burger Quiz',
'episode_number': 21,
'season': 'Season 3',
'tags': 'count:13',
'episode': 'Episode 21',
'duration': 2312,
},
'params': {'skip_download': 'm3u8'},
2015-05-25 23:32:43 -06:00
}, {
'url': 'http://www.tf1.fr/tf1/koh-lanta/videos/replay-koh-lanta-22-mai-2015.html',
'only_matching': True,
2015-12-15 08:31:58 -07:00
}, {
'url': 'http://www.tf1.fr/hd1/documentaire/videos/mylene-farmer-d-une-icone.html',
'only_matching': True,
2015-05-25 23:32:43 -06:00
}]
2013-06-29 07:18:19 -06:00
def _real_extract(self, url):
program_slug, slug = self._match_valid_url(url).groups()
video = self._download_json(
'https://www.tf1.fr/graphql/web', slug, query={
'id': '9b80783950b85247541dd1d851f9cc7fa36574af015621f853ab111a679ce26f',
'variables': json.dumps({
'programSlug': program_slug,
'slug': slug,
}),
})['data']['videoBySlug']
wat_id = video['streamId']
tags = []
for tag in (video.get('tags') or []):
label = tag.get('label')
if not label:
continue
tags.append(label)
decoration = video.get('decoration') or {}
thumbnails = []
for source in (try_get(decoration, lambda x: x['image']['sources'], list) or []):
source_url = source.get('url')
if not source_url:
continue
thumbnails.append({
'url': source_url,
'width': int_or_none(source.get('width')),
})
return {
'_type': 'url_transparent',
'id': wat_id,
'url': 'wat:' + wat_id,
'title': video.get('title'),
'thumbnails': thumbnails,
'description': decoration.get('description'),
'timestamp': parse_iso8601(video.get('date')),
'duration': int_or_none(try_get(video, lambda x: x['publicPlayingInfos']['duration'])),
'tags': tags,
'series': decoration.get('programLabel'),
'season_number': int_or_none(video.get('season')),
'episode_number': int_or_none(video.get('episode')),
}