X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/a070911bf98c751e4425008f1335cc57c86d30fc..04c6e7e9539d48169931f1911a39ed223d44cf2b:/youtube_dl/extractor/discovery.py?ds=sidebyside diff --git a/youtube_dl/extractor/discovery.py b/youtube_dl/extractor/discovery.py index d3e6675..e0139cc 100644 --- a/youtube_dl/extractor/discovery.py +++ b/youtube_dl/extractor/discovery.py @@ -1,45 +1,118 @@ from __future__ import unicode_literals -from .common import InfoExtractor -from ..utils import ( - parse_iso8601, - int_or_none, -) - - -class DiscoveryIE(InfoExtractor): - _VALID_URL = r'http://www\.discovery\.com\/[a-zA-Z0-9\-]*/[a-zA-Z0-9\-]*/videos/(?P[a-zA-Z0-9_\-]*)(?:\.htm)?' - _TEST = { - 'url': 'http://www.discovery.com/tv-shows/mythbusters/videos/mission-impossible-outtakes.htm', - 'md5': '3c69d77d9b0d82bfd5e5932a60f26504', +import random +import re +import string + +from .discoverygo import DiscoveryGoBaseIE +from ..compat import compat_urllib_parse_unquote +from ..utils import ExtractorError +from ..compat import compat_HTTPError + + +class DiscoveryIE(DiscoveryGoBaseIE): + _VALID_URL = r'''(?x)https?:// + (?P + go\.discovery| + www\. + (?: + investigationdiscovery| + discoverylife| + animalplanet| + ahctv| + destinationamerica| + sciencechannel| + tlc + )| + watch\. + (?: + hgtv| + foodnetwork| + travelchannel| + diynetwork| + cookingchanneltv| + motortrend + ) + )\.com/tv-shows/(?P[^/]+)/(?:video|full-episode)s/(?P[^./?#]+)''' + _TESTS = [{ + 'url': 'https://go.discovery.com/tv-shows/cash-cab/videos/riding-with-matthew-perry', 'info_dict': { - 'id': 'mission-impossible-outtakes', - 'ext': 'flv', - 'title': 'Mission Impossible Outtakes', - 'description': ('Watch Jamie Hyneman and Adam Savage practice being' - ' each other -- to the point of confusing Jamie\'s dog -- and ' - 'don\'t miss Adam moon-walking as Jamie ... behind Jamie\'s' - ' back.'), - 'duration': 156, - 'timestamp': 1303099200, - 'upload_date': '20110418', + 'id': '5a2f35ce6b66d17a5026e29e', + 'ext': 'mp4', + 'title': 'Riding with Matthew Perry', + 'description': 'md5:a34333153e79bc4526019a5129e7f878', + 'duration': 84, }, - } + 'params': { + 'skip_download': True, # requires ffmpeg + } + }, { + 'url': 'https://www.investigationdiscovery.com/tv-shows/final-vision/full-episodes/final-vision', + 'only_matching': True, + }, { + 'url': 'https://go.discovery.com/tv-shows/alaskan-bush-people/videos/follow-your-own-road', + 'only_matching': True, + }, { + # using `show_slug` is important to get the correct video data + 'url': 'https://www.sciencechannel.com/tv-shows/mythbusters-on-science/full-episodes/christmas-special', + 'only_matching': True, + }] + _GEO_COUNTRIES = ['US'] + _GEO_BYPASS = False + _API_BASE_URL = 'https://api.discovery.com/v1/' def _real_extract(self, url): - video_id = self._match_id(url) - webpage = self._download_webpage(url, video_id) - - info = self._parse_json(self._search_regex( - r'(?s)', - webpage, 'video info'), video_id) - - return { - 'id': video_id, - 'title': info['name'], - 'url': info['contentURL'], - 'description': info.get('description'), - 'thumbnail': info.get('thumbnailUrl'), - 'timestamp': parse_iso8601(info.get('uploadDate')), - 'duration': int_or_none(info.get('duration')), - } + site, show_slug, display_id = re.match(self._VALID_URL, url).groups() + + access_token = None + cookies = self._get_cookies(url) + + # prefer Affiliate Auth Token over Anonymous Auth Token + auth_storage_cookie = cookies.get('eosAf') or cookies.get('eosAn') + if auth_storage_cookie and auth_storage_cookie.value: + auth_storage = self._parse_json(compat_urllib_parse_unquote( + compat_urllib_parse_unquote(auth_storage_cookie.value)), + display_id, fatal=False) or {} + access_token = auth_storage.get('a') or auth_storage.get('access_token') + + if not access_token: + access_token = self._download_json( + 'https://%s.com/anonymous' % site, display_id, + 'Downloading token JSON metadata', query={ + 'authRel': 'authorization', + 'client_id': '3020a40c2356a645b4b4', + 'nonce': ''.join([random.choice(string.ascii_letters) for _ in range(32)]), + 'redirectUri': 'https://www.discovery.com/', + })['access_token'] + + headers = self.geo_verification_headers() + headers['Authorization'] = 'Bearer ' + access_token + + try: + video = self._download_json( + self._API_BASE_URL + 'content/videos', + display_id, 'Downloading content JSON metadata', + headers=headers, query={ + 'embed': 'show.name', + 'fields': 'authenticated,description.detailed,duration,episodeNumber,id,name,parental.rating,season.number,show,tags', + 'slug': display_id, + 'show_slug': show_slug, + })[0] + video_id = video['id'] + stream = self._download_json( + self._API_BASE_URL + 'streaming/video/' + video_id, + display_id, 'Downloading streaming JSON metadata', headers=headers) + except ExtractorError as e: + if isinstance(e.cause, compat_HTTPError) and e.cause.code in (401, 403): + e_description = self._parse_json( + e.cause.read().decode(), display_id)['description'] + if 'resource not available for country' in e_description: + self.raise_geo_restricted(countries=self._GEO_COUNTRIES) + if 'Authorized Networks' in e_description: + raise ExtractorError( + 'This video is only available via cable service provider subscription that' + ' is not currently supported. You may want to use --cookies.', expected=True) + raise ExtractorError(e_description) + raise + + return self._extract_video_info(video, stream, display_id)