mirror of
				https://github.com/yt-dlp/yt-dlp.git
				synced 2025-11-04 08:35:12 +00:00 
			
		
		
		
	
		
			
				
	
	
		
			237 lines
		
	
	
		
			9.6 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
			
		
		
	
	
			237 lines
		
	
	
		
			9.6 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
import json
 | 
						||
import urllib.parse
 | 
						||
 | 
						||
from .common import InfoExtractor
 | 
						||
from ..networking.exceptions import HTTPError
 | 
						||
from ..utils import (
 | 
						||
    ExtractorError,
 | 
						||
    filter_dict,
 | 
						||
    float_or_none,
 | 
						||
    join_nonempty,
 | 
						||
    mimetype2ext,
 | 
						||
    parse_iso8601,
 | 
						||
    unsmuggle_url,
 | 
						||
    update_url_query,
 | 
						||
    url_or_none,
 | 
						||
)
 | 
						||
from ..utils.traversal import traverse_obj
 | 
						||
 | 
						||
 | 
						||
class StreaksBaseIE(InfoExtractor):
 | 
						||
    _API_URL_TEMPLATE = 'https://{}.api.streaks.jp/v1/projects/{}/medias/{}{}'
 | 
						||
    _GEO_BYPASS = False
 | 
						||
    _GEO_COUNTRIES = ['JP']
 | 
						||
 | 
						||
    def _extract_from_streaks_api(self, project_id, media_id, headers=None, query=None, ssai=False):
 | 
						||
        try:
 | 
						||
            response = self._download_json(
 | 
						||
                self._API_URL_TEMPLATE.format('playback', project_id, media_id, ''),
 | 
						||
                media_id, 'Downloading STREAKS playback API JSON', headers={
 | 
						||
                    'Accept': 'application/json',
 | 
						||
                    'Origin': 'https://players.streaks.jp',
 | 
						||
                    **self.geo_verification_headers(),
 | 
						||
                    **(headers or {}),
 | 
						||
                })
 | 
						||
        except ExtractorError as e:
 | 
						||
            if isinstance(e.cause, HTTPError) and e.cause.status in {403, 404}:
 | 
						||
                error = self._parse_json(e.cause.response.read().decode(), media_id, fatal=False)
 | 
						||
                message = traverse_obj(error, ('message', {str}))
 | 
						||
                code = traverse_obj(error, ('code', {str}))
 | 
						||
                if code == 'REQUEST_FAILED':
 | 
						||
                    self.raise_geo_restricted(message, countries=self._GEO_COUNTRIES)
 | 
						||
                elif code == 'MEDIA_NOT_FOUND':
 | 
						||
                    raise ExtractorError(message, expected=True)
 | 
						||
                elif code or message:
 | 
						||
                    raise ExtractorError(join_nonempty(code, message, delim=': '))
 | 
						||
            raise
 | 
						||
 | 
						||
        streaks_id = response['id']
 | 
						||
        live_status = {
 | 
						||
            'clip': 'was_live',
 | 
						||
            'file': 'not_live',
 | 
						||
            'linear': 'is_live',
 | 
						||
            'live': 'is_live',
 | 
						||
        }.get(response.get('type'))
 | 
						||
 | 
						||
        formats, subtitles = [], {}
 | 
						||
        drm_formats = False
 | 
						||
 | 
						||
        for source in traverse_obj(response, ('sources', lambda _, v: v['src'])):
 | 
						||
            if source.get('key_systems'):
 | 
						||
                drm_formats = True
 | 
						||
                continue
 | 
						||
 | 
						||
            src_url = source['src']
 | 
						||
            is_live = live_status == 'is_live'
 | 
						||
            ext = mimetype2ext(source.get('type'))
 | 
						||
            if ext != 'm3u8':
 | 
						||
                self.report_warning(f'Unsupported stream type: {ext}')
 | 
						||
                continue
 | 
						||
 | 
						||
            if is_live and ssai:
 | 
						||
                session_params = traverse_obj(self._download_json(
 | 
						||
                    self._API_URL_TEMPLATE.format('ssai', project_id, streaks_id, '/ssai/session'),
 | 
						||
                    media_id, 'Downloading session parameters',
 | 
						||
                    headers={'Content-Type': 'application/json', 'Accept': 'application/json'},
 | 
						||
                    data=json.dumps({'id': source['id']}).encode(),
 | 
						||
                ), (0, 'query', {urllib.parse.parse_qs}))
 | 
						||
                src_url = update_url_query(src_url, session_params)
 | 
						||
 | 
						||
            fmts, subs = self._extract_m3u8_formats_and_subtitles(
 | 
						||
                src_url, media_id, 'mp4', m3u8_id='hls', fatal=False, live=is_live, query=query)
 | 
						||
            formats.extend(fmts)
 | 
						||
            self._merge_subtitles(subs, target=subtitles)
 | 
						||
 | 
						||
        if not formats and drm_formats:
 | 
						||
            self.report_drm(media_id)
 | 
						||
        self._remove_duplicate_formats(formats)
 | 
						||
 | 
						||
        for subs in traverse_obj(response, (
 | 
						||
            'tracks', lambda _, v: v['kind'] in ('captions', 'subtitles') and url_or_none(v['src']),
 | 
						||
        )):
 | 
						||
            lang = traverse_obj(subs, ('srclang', {str.lower})) or 'ja'
 | 
						||
            subtitles.setdefault(lang, []).append({'url': subs['src']})
 | 
						||
 | 
						||
        return {
 | 
						||
            'id': streaks_id,
 | 
						||
            'display_id': media_id,
 | 
						||
            'formats': formats,
 | 
						||
            'live_status': live_status,
 | 
						||
            'subtitles': subtitles,
 | 
						||
            'uploader_id': project_id,
 | 
						||
            **traverse_obj(response, {
 | 
						||
                'title': ('name', {str}),
 | 
						||
                'description': ('description', {str}, filter),
 | 
						||
                'duration': ('duration', {float_or_none}),
 | 
						||
                'modified_timestamp': ('updated_at', {parse_iso8601}),
 | 
						||
                'tags': ('tags', ..., {str}),
 | 
						||
                'thumbnails': (('poster', 'thumbnail'), 'src', {'url': {url_or_none}}),
 | 
						||
                'timestamp': ('created_at', {parse_iso8601}),
 | 
						||
            }),
 | 
						||
        }
 | 
						||
 | 
						||
 | 
						||
class StreaksIE(StreaksBaseIE):
 | 
						||
    _VALID_URL = [
 | 
						||
        r'https?://players\.streaks\.jp/(?P<project_id>[\w-]+)/[\da-f]+/index\.html\?(?:[^#]+&)?m=(?P<id>(?:ref:)?[\w-]+)',
 | 
						||
        r'https?://playback\.api\.streaks\.jp/v1/projects/(?P<project_id>[\w-]+)/medias/(?P<id>(?:ref:)?[\w-]+)',
 | 
						||
    ]
 | 
						||
    _EMBED_REGEX = [rf'<iframe\s+[^>]*\bsrc\s*=\s*["\'](?P<url>{_VALID_URL[0]})']
 | 
						||
    _TESTS = [{
 | 
						||
        'url': 'https://players.streaks.jp/tipness/08155cd19dc14c12bebefb69b92eafcc/index.html?m=dbdf2df35b4d483ebaeeaeb38c594647',
 | 
						||
        'info_dict': {
 | 
						||
            'id': 'dbdf2df35b4d483ebaeeaeb38c594647',
 | 
						||
            'ext': 'mp4',
 | 
						||
            'title': '3shunenCM_edit.mp4',
 | 
						||
            'display_id': 'dbdf2df35b4d483ebaeeaeb38c594647',
 | 
						||
            'duration': 47.533,
 | 
						||
            'live_status': 'not_live',
 | 
						||
            'modified_date': '20230726',
 | 
						||
            'modified_timestamp': 1690356180,
 | 
						||
            'timestamp': 1690355996,
 | 
						||
            'upload_date': '20230726',
 | 
						||
            'uploader_id': 'tipness',
 | 
						||
        },
 | 
						||
    }, {
 | 
						||
        'url': 'https://players.streaks.jp/ktv-web/0298e8964c164ab384c07ef6e08c444b/index.html?m=ref:mycoffeetime_250317',
 | 
						||
        'info_dict': {
 | 
						||
            'id': 'dccdc079e3fd41f88b0c8435e2d453ab',
 | 
						||
            'ext': 'mp4',
 | 
						||
            'title': 'わたしの珈琲時間_250317',
 | 
						||
            'display_id': 'ref:mycoffeetime_250317',
 | 
						||
            'duration': 122.99,
 | 
						||
            'live_status': 'not_live',
 | 
						||
            'modified_date': '20250310',
 | 
						||
            'modified_timestamp': 1741586302,
 | 
						||
            'thumbnail': r're:https?://.+\.jpg',
 | 
						||
            'timestamp': 1741585839,
 | 
						||
            'upload_date': '20250310',
 | 
						||
            'uploader_id': 'ktv-web',
 | 
						||
        },
 | 
						||
    }, {
 | 
						||
        'url': 'https://playback.api.streaks.jp/v1/projects/ktv-web/medias/b5411938e1e5435dac71edf829dd4813',
 | 
						||
        'info_dict': {
 | 
						||
            'id': 'b5411938e1e5435dac71edf829dd4813',
 | 
						||
            'ext': 'mp4',
 | 
						||
            'title': 'KANTELE_SYUSEi_0630',
 | 
						||
            'display_id': 'b5411938e1e5435dac71edf829dd4813',
 | 
						||
            'live_status': 'not_live',
 | 
						||
            'modified_date': '20250122',
 | 
						||
            'modified_timestamp': 1737522999,
 | 
						||
            'thumbnail': r're:https?://.+\.jpg',
 | 
						||
            'timestamp': 1735205137,
 | 
						||
            'upload_date': '20241226',
 | 
						||
            'uploader_id': 'ktv-web',
 | 
						||
        },
 | 
						||
    }, {
 | 
						||
        # TVer Olympics: website already down, but api remains accessible
 | 
						||
        'url': 'https://playback.api.streaks.jp/v1/projects/tver-olympic/medias/ref:sp_240806_1748_dvr',
 | 
						||
        'info_dict': {
 | 
						||
            'id': 'c10f7345adb648cf804d7578ab93b2e3',
 | 
						||
            'ext': 'mp4',
 | 
						||
            'title': 'サッカー 男子 準決勝_dvr',
 | 
						||
            'display_id': 'ref:sp_240806_1748_dvr',
 | 
						||
            'duration': 12960.0,
 | 
						||
            'live_status': 'was_live',
 | 
						||
            'modified_date': '20240805',
 | 
						||
            'modified_timestamp': 1722896263,
 | 
						||
            'timestamp': 1722777618,
 | 
						||
            'upload_date': '20240804',
 | 
						||
            'uploader_id': 'tver-olympic',
 | 
						||
        },
 | 
						||
    }, {
 | 
						||
        # TBS FREE: 24-hour stream
 | 
						||
        'url': 'https://playback.api.streaks.jp/v1/projects/tbs/medias/ref:simul-02',
 | 
						||
        'info_dict': {
 | 
						||
            'id': 'c4e83a7b48f4409a96adacec674b4e22',
 | 
						||
            'ext': 'mp4',
 | 
						||
            'title': str,
 | 
						||
            'display_id': 'ref:simul-02',
 | 
						||
            'live_status': 'is_live',
 | 
						||
            'modified_date': '20241031',
 | 
						||
            'modified_timestamp': 1730339858,
 | 
						||
            'timestamp': 1705466840,
 | 
						||
            'upload_date': '20240117',
 | 
						||
            'uploader_id': 'tbs',
 | 
						||
        },
 | 
						||
    }, {
 | 
						||
        # DRM protected
 | 
						||
        'url': 'https://players.streaks.jp/sp-jbc/a12d7ee0f40c49d6a0a2bff520639677/index.html?m=5f89c62f37ee4a68be8e6e3b1396c7d8',
 | 
						||
        'only_matching': True,
 | 
						||
    }]
 | 
						||
    _WEBPAGE_TESTS = [{
 | 
						||
        'url': 'https://event.play.jp/playnext2023/',
 | 
						||
        'info_dict': {
 | 
						||
            'id': '2d975178293140dc8074a7fc536a7604',
 | 
						||
            'ext': 'mp4',
 | 
						||
            'title': 'PLAY NEXTキームービー(本番)',
 | 
						||
            'uploader_id': 'play',
 | 
						||
            'duration': 17.05,
 | 
						||
            'thumbnail': r're:https?://.+\.jpg',
 | 
						||
            'timestamp': 1668387517,
 | 
						||
            'upload_date': '20221114',
 | 
						||
            'modified_timestamp': 1739411523,
 | 
						||
            'modified_date': '20250213',
 | 
						||
            'live_status': 'not_live',
 | 
						||
        },
 | 
						||
    }, {
 | 
						||
        'url': 'https://wowshop.jp/Page/special/cooking_goods/?bid=wowshop&srsltid=AfmBOor_phUNoPEE_UCPiGGSCMrJE5T2US397smvsbrSdLqUxwON0el4',
 | 
						||
        'playlist_mincount': 2,
 | 
						||
        'info_dict': {
 | 
						||
            'id': '?bid=wowshop&srsltid=AfmBOor_phUNoPEE_UCPiGGSCMrJE5T2US397smvsbrSdLqUxwON0el4',
 | 
						||
            'title': 'ワンランク上の料理道具でとびきりの“おいしい”を食卓へ|wowshop',
 | 
						||
            'description': 'md5:914b5cb8624fc69274c7fb7b2342958f',
 | 
						||
            'age_limit': 0,
 | 
						||
            'thumbnail': 'https://wowshop.jp/Page/special/cooking_goods/images/ogp.jpg',
 | 
						||
        },
 | 
						||
    }]
 | 
						||
 | 
						||
    def _real_extract(self, url):
 | 
						||
        url, smuggled_data = unsmuggle_url(url, {})
 | 
						||
        project_id, media_id = self._match_valid_url(url).group('project_id', 'id')
 | 
						||
 | 
						||
        return self._extract_from_streaks_api(
 | 
						||
            project_id, media_id, headers=filter_dict({
 | 
						||
                'X-Streaks-Api-Key': smuggled_data.get('api_key'),
 | 
						||
            }))
 |