yt-dlp/yt_dlp/extractor/svt.py

489 lines
17 KiB
Python
Raw Permalink Normal View History

import json
2015-02-28 16:25:04 +01:00
import re
from .common import InfoExtractor
from ..utils import (
determine_ext,
dict_get,
int_or_none,
traverse_obj,
try_get,
unified_timestamp,
)
class SVTBaseIE(InfoExtractor):
_GEO_COUNTRIES = ['SE']
2017-02-19 20:25:55 +01:00
def _extract_video(self, video_info, video_id):
is_live = dict_get(video_info, ('live', 'simulcast'), default=False)
m3u8_protocol = 'm3u8' if is_live else 'm3u8_native'
formats = []
subtitles = {}
for vr in video_info['videoReferences']:
2016-09-17 14:25:31 +02:00
player_type = vr.get('playerType') or vr.get('format')
vurl = vr['url']
2015-02-28 16:25:04 +01:00
ext = determine_ext(vurl)
if ext == 'm3u8':
fmts, subs = self._extract_m3u8_formats_and_subtitles(
vurl, video_id,
ext='mp4', entry_protocol=m3u8_protocol,
m3u8_id=player_type, fatal=False)
formats.extend(fmts)
self._merge_subtitles(subs, target=subtitles)
2015-02-28 16:25:04 +01:00
elif ext == 'f4m':
formats.extend(self._extract_f4m_formats(
vurl + '?hdcore=3.3.0', video_id,
2016-03-15 19:33:09 +01:00
f4m_id=player_type, fatal=False))
elif ext == 'mpd':
fmts, subs = self._extract_mpd_formats_and_subtitles(
vurl, video_id, mpd_id=player_type, fatal=False)
formats.extend(fmts)
self._merge_subtitles(subs, target=subtitles)
else:
formats.append({
2016-03-15 19:33:09 +01:00
'format_id': player_type,
'url': vurl,
})
2020-11-21 15:50:42 +01:00
rights = try_get(video_info, lambda x: x['rights'], dict) or {}
if not formats and rights.get('geoBlockedSweden'):
self.raise_geo_restricted(
'This video is only available in Sweden',
countries=self._GEO_COUNTRIES, metadata_available=True)
subtitle_references = dict_get(video_info, ('subtitles', 'subtitleReferences'))
if isinstance(subtitle_references, list):
for sr in subtitle_references:
subtitle_url = sr.get('url')
subtitle_lang = sr.get('language', 'sv')
if subtitle_url:
sub = {
'url': subtitle_url,
}
if determine_ext(subtitle_url) == 'm3u8':
# XXX: no way of testing, is it ever hit?
sub['ext'] = 'vtt'
subtitles.setdefault(subtitle_lang, []).append(sub)
2016-01-12 17:08:55 +01:00
title = video_info.get('title')
series = video_info.get('programTitle')
season_number = int_or_none(video_info.get('season'))
episode = video_info.get('episodeTitle')
episode_number = int_or_none(video_info.get('episodeNumber'))
2020-11-21 15:50:42 +01:00
timestamp = unified_timestamp(rights.get('validFrom'))
duration = int_or_none(dict_get(video_info, ('materialLength', 'contentDuration')))
age_limit = None
adult = dict_get(
video_info, ('inappropriateForChildren', 'blockedForChildren'),
skip_false_values=False)
if adult is not None:
age_limit = 18 if adult else 0
return {
'id': video_id,
'title': title,
'formats': formats,
2016-01-12 17:08:55 +01:00
'subtitles': subtitles,
'duration': duration,
2020-11-21 15:50:42 +01:00
'timestamp': timestamp,
2015-02-28 16:25:04 +01:00
'age_limit': age_limit,
'series': series,
'season_number': season_number,
'episode': episode,
'episode_number': episode_number,
'is_live': is_live,
}
class SVTIE(SVTBaseIE):
_VALID_URL = r'https?://(?:www\.)?svt\.se/wd\?(?:.*?&)?widgetId=(?P<widget_id>\d+)&.*?\barticleId=(?P<id>\d+)'
_EMBED_REGEX = [rf'(?:<iframe src|href)="(?P<url>{_VALID_URL}[^"]*)"']
_TEST = {
'url': 'http://www.svt.se/wd?widgetId=23991&sectionId=541&articleId=2900353&type=embed&contextSectionId=123&autostart=false',
'md5': '33e9a5d8f646523ce0868ecfb0eed77d',
'info_dict': {
'id': '2900353',
'ext': 'mp4',
'title': 'Stjärnorna skojar till det - under SVT-intervjun',
'duration': 27,
'age_limit': 0,
},
}
def _real_extract(self, url):
mobj = self._match_valid_url(url)
widget_id = mobj.group('widget_id')
article_id = mobj.group('id')
info = self._download_json(
f'http://www.svt.se/wd?widgetId={widget_id}&articleId={article_id}&format=json&type=embed&output=json',
article_id)
info_dict = self._extract_video(info['video'], article_id)
info_dict['title'] = info['context']['title']
return info_dict
2018-04-04 19:28:36 +02:00
class SVTPlayBaseIE(SVTBaseIE):
_SVTPLAY_RE = r'root\s*\[\s*(["\'])_*svtplay\1\s*\]\s*=\s*(?P<json>{.+?})\s*;\s*\n'
class SVTPlayIE(SVTPlayBaseIE):
IE_DESC = 'SVT Play and Öppet arkiv'
_VALID_URL = r'''(?x)
(?:
2020-11-21 15:50:42 +01:00
(?:
svt:|
https?://(?:www\.)?svt\.se/barnkanalen/barnplay/[^/]+/
)
(?P<svt_id>[^/?#&]+)|
https?://(?:www\.)?(?:svtplay|oppetarkiv)\.se/(?:video|klipp|kanaler)/(?P<id>[^/?#&]+)
(?:.*?(?:modalId|id)=(?P<modal_id>[\da-zA-Z-]+))?
)
'''
_TESTS = [{
'url': 'https://www.svtplay.se/video/30479064',
2020-11-21 15:50:42 +01:00
'md5': '2382036fd6f8c994856c323fe51c426e',
'info_dict': {
'id': '8zVbDPA',
'ext': 'mp4',
'title': 'Designdrömmar i Stenungsund',
'timestamp': 1615770000,
'upload_date': '20210315',
'duration': 3519,
2020-11-21 15:50:42 +01:00
'thumbnail': r're:^https?://(?:.*[\.-]jpg|www.svtstatic.se/image/.*)$',
'age_limit': 0,
'subtitles': {
'sv': [{
2020-11-21 15:50:42 +01:00
'ext': 'vtt',
}],
},
},
2020-11-21 15:50:42 +01:00
'params': {
'skip_download': 'm3u8',
},
'skip': 'Episode is no longer available',
}, {
'url': 'https://www.svtplay.se/video/emBxBQj',
'md5': '2382036fd6f8c994856c323fe51c426e',
'info_dict': {
'id': 'eyBd9aj',
'ext': 'mp4',
'title': '1. Farlig kryssning',
'timestamp': 1491019200,
'upload_date': '20170401',
'duration': 2566,
'thumbnail': r're:^https?://(?:.*[\.-]jpg|www.svtstatic.se/image/.*)$',
'age_limit': 0,
'episode': '1. Farlig kryssning',
'series': 'Rederiet',
'subtitles': {
'sv': 'count:3',
},
},
'params': {
'skip_download': 'm3u8',
},
}, {
'url': 'https://www.svtplay.se/video/jz2rYz7/anders-hansen-moter/james-fallon?info=visa',
'info_dict': {
'id': 'jvXAGVb',
'ext': 'mp4',
'title': 'James Fallon',
'timestamp': 1673917200,
'upload_date': '20230117',
'duration': 1081,
'thumbnail': r're:^https?://(?:.*[\.-]jpg|www.svtstatic.se/image/.*)$',
'age_limit': 0,
'episode': 'James Fallon',
'series': 'Anders Hansen möter...',
},
'params': {
'skip_download': 'dash',
2020-11-21 15:50:42 +01:00
},
}, {
'url': 'https://www.svtplay.se/video/30479064/husdrommar/husdrommar-sasong-8-designdrommar-i-stenungsund?modalId=8zVbDPA',
'only_matching': True,
}, {
'url': 'https://www.svtplay.se/video/30684086/rapport/rapport-24-apr-18-00-7?id=e72gVpa',
'only_matching': True,
}, {
# geo restricted to Sweden
'url': 'http://www.oppetarkiv.se/video/5219710/trollflojten',
'only_matching': True,
2016-06-25 19:29:53 +02:00
}, {
'url': 'http://www.svtplay.se/klipp/9023742/stopptid-om-bjorn-borg',
'only_matching': True,
}, {
'url': 'https://www.svtplay.se/kanaler/svt1',
'only_matching': True,
}, {
'url': 'svt:1376446-003A',
'only_matching': True,
}, {
'url': 'svt:14278044',
'only_matching': True,
2020-11-21 15:50:42 +01:00
}, {
'url': 'https://www.svt.se/barnkanalen/barnplay/kar/eWv5MLX/',
'only_matching': True,
}, {
'url': 'svt:eWv5MLX',
'only_matching': True,
}]
def _extract_by_video_id(self, video_id, webpage=None):
data = self._download_json(
f'https://api.svt.se/videoplayer-api/video/{video_id}',
video_id, headers=self.geo_verification_headers())
info_dict = self._extract_video(data, video_id)
if not info_dict.get('title'):
title = dict_get(info_dict, ('episode', 'series'))
if not title and webpage:
title = re.sub(
r'\s*\|\s*.+?$', '', self._og_search_title(webpage))
if not title:
title = video_id
info_dict['title'] = title
return info_dict
def _real_extract(self, url):
mobj = self._match_valid_url(url)
video_id = mobj.group('id')
svt_id = mobj.group('svt_id') or mobj.group('modal_id')
if svt_id:
return self._extract_by_video_id(svt_id)
webpage = self._download_webpage(url, video_id)
data = self._parse_json(
self._search_regex(
2018-04-04 19:28:36 +02:00
self._SVTPLAY_RE, webpage, 'embedded data', default='{}',
group='json'),
video_id, fatal=False)
thumbnail = self._og_search_thumbnail(webpage)
if data:
video_info = try_get(
data, lambda x: x['context']['dispatcher']['stores']['VideoTitlePageStore']['data']['video'],
dict)
if video_info:
info_dict = self._extract_video(video_info, video_id)
info_dict.update({
'title': data['context']['dispatcher']['stores']['MetaStore']['title'],
'thumbnail': thumbnail,
})
return info_dict
svt_id = try_get(
data, lambda x: x['statistics']['dataLake']['content']['id'],
str)
if not svt_id:
nextjs_data = self._search_nextjs_data(webpage, video_id, fatal=False)
svt_id = traverse_obj(nextjs_data, (
'props', 'urqlState', ..., 'data', {json.loads}, 'detailsPageByPath',
'video', 'svtId', {str}), get_all=False)
if not svt_id:
svt_id = self._search_regex(
(r'<video[^>]+data-video-id=["\']([\da-zA-Z-]+)',
r'<[^>]+\bdata-rt=["\']top-area-play-button["\'][^>]+\bhref=["\'][^"\']*video/[\w-]+/[^"\']*\b(?:modalId|id)=([\w-]+)'),
webpage, 'video id')
2020-11-21 15:50:42 +01:00
info_dict = self._extract_by_video_id(svt_id, webpage)
info_dict['thumbnail'] = thumbnail
return info_dict
2018-04-04 19:28:36 +02:00
class SVTSeriesIE(SVTPlayBaseIE):
_VALID_URL = r'https?://(?:www\.)?svtplay\.se/(?P<id>[^/?&#]+)(?:.+?\btab=(?P<season_slug>[^&#]+))?'
_TESTS = [{
'url': 'https://www.svtplay.se/rederiet',
'info_dict': {
'id': '14445680',
'title': 'Rederiet',
'description': 'md5:d9fdfff17f5d8f73468176ecd2836039',
},
'playlist_mincount': 318,
}, {
'url': 'https://www.svtplay.se/rederiet?tab=season-2-14445680',
'info_dict': {
'id': 'season-2-14445680',
'title': 'Rederiet - Säsong 2',
'description': 'md5:d9fdfff17f5d8f73468176ecd2836039',
},
'playlist_mincount': 12,
}]
@classmethod
def suitable(cls, url):
return False if SVTIE.suitable(url) or SVTPlayIE.suitable(url) else super().suitable(url)
def _real_extract(self, url):
series_slug, season_id = self._match_valid_url(url).groups()
series = self._download_json(
'https://api.svt.se/contento/graphql', series_slug,
'Downloading series page', query={
'query': '''{
listablesBySlug(slugs: ["%s"]) {
associatedContent(include: [productionPeriod, season]) {
items {
item {
... on Episode {
videoSvtId
}
}
}
id
name
}
id
longDescription
name
shortDescription
}
}''' % series_slug, # noqa: UP031
})['data']['listablesBySlug'][0]
season_name = None
entries = []
for season in series['associatedContent']:
if not isinstance(season, dict):
continue
if season_id:
if season.get('id') != season_id:
continue
season_name = season.get('name')
items = season.get('items')
if not isinstance(items, list):
continue
for item in items:
video = item.get('item') or {}
content_id = video.get('videoSvtId')
if not content_id or not isinstance(content_id, str):
continue
entries.append(self.url_result(
'svt:' + content_id, SVTPlayIE.ie_key(), content_id))
title = series.get('name')
season_name = season_name or season_id
if title and season_name:
title = f'{title} - {season_name}'
elif season_id:
title = season_id
return self.playlist_result(
entries, season_id or series.get('id'), title,
dict_get(series, ('longDescription', 'shortDescription')))
class SVTPageIE(SVTBaseIE):
_VALID_URL = r'https?://(?:www\.)?svt\.se/(?:[^/?#]+/)*(?P<id>[^/?&#]+)'
_TESTS = [{
'url': 'https://www.svt.se/nyheter/lokalt/skane/viktor-18-forlorade-armar-och-ben-i-sepsis-vill-ateruppta-karaten-och-bli-svetsare',
'info_dict': {
'title': 'Viktor, 18, förlorade armar och ben i sepsis vill återuppta karaten och bli svetsare',
'id': 'viktor-18-forlorade-armar-och-ben-i-sepsis-vill-ateruppta-karaten-och-bli-svetsare',
},
'playlist_count': 2,
}, {
'url': 'https://www.svt.se/nyheter/lokalt/skane/forsvarsmakten-om-trafikkaoset-pa-e22-kunde-inte-varit-dar-snabbare',
'info_dict': {
'id': 'jXvk42E',
'title': 'Försvarsmakten om trafikkaoset på E22: Kunde inte varit där snabbare',
'ext': 'mp4',
'duration': 80,
'age_limit': 0,
'timestamp': 1704370009,
'episode': 'Försvarsmakten om trafikkaoset på E22: Kunde inte varit där snabbare',
'series': 'Lokala Nyheter Skåne',
'upload_date': '20240104',
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.svt.se/nyheter/svtforum/2023-tungt-ar-for-svensk-media',
'info_dict': {
'title': '2023 tungt år för svensk media',
'id': 'ewqAZv4',
'ext': 'mp4',
'duration': 3074,
'age_limit': 0,
'series': '',
'timestamp': 1702980479,
'upload_date': '20231219',
'episode': 'Mediestudier',
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.svt.se/sport/ishockey/bakom-masken-lehners-kamp-mot-mental-ohalsa',
'info_dict': {
'id': '25298267',
'title': 'Bakom masken Lehners kamp mot mental ohälsa',
},
'playlist_count': 4,
'skip': 'Video is gone',
}, {
'url': 'https://www.svt.se/nyheter/utrikes/svenska-andrea-ar-en-mil-fran-branderna-i-kalifornien',
'info_dict': {
'id': '24243746',
'title': 'Svenska Andrea redo att fly sitt hem i Kalifornien',
},
'playlist_count': 2,
'skip': 'Video is gone',
}, {
# only programTitle
'url': 'http://www.svt.se/sport/ishockey/jagr-tacklar-giroux-under-intervjun',
'info_dict': {
'id': '8439V2K',
'ext': 'mp4',
'title': 'Stjärnorna skojar till det - under SVT-intervjun',
'duration': 27,
'age_limit': 0,
},
'skip': 'Video is gone',
}, {
'url': 'https://www.svt.se/nyheter/lokalt/vast/svt-testar-tar-nagon-upp-skrapet-1',
'only_matching': True,
}, {
'url': 'https://www.svt.se/vader/manadskronikor/maj2018',
'only_matching': True,
}]
@classmethod
def suitable(cls, url):
return False if SVTIE.suitable(url) or SVTPlayIE.suitable(url) else super().suitable(url)
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
title = self._og_search_title(webpage)
urql_state = self._search_json(
r'window\.svt\.(?:nyh\.)?urqlState\s*=', webpage, 'json data', display_id)
data = traverse_obj(urql_state, (..., 'data', {str}, {json.loads}), get_all=False) or {}
def entries():
for video_id in set(traverse_obj(data, (
'page', (('topMedia', 'svtId'), ('body', ..., 'video', 'svtId')), {str},
))):
info = self._extract_video(
self._download_json(f'https://api.svt.se/video/{video_id}', video_id), video_id)
info['title'] = title
yield info
return self.playlist_result(entries(), display_id, title)