X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/9f2b33881274af98a9145c533a1d295fad71521a..4eb7d8923b3bad26edab01df0fd2650b0563ea8b:/youtube_dl/extractor/hotstar.py diff --git a/youtube_dl/extractor/hotstar.py b/youtube_dl/extractor/hotstar.py index f05d765..3a7a66a 100644 --- a/youtube_dl/extractor/hotstar.py +++ b/youtube_dl/extractor/hotstar.py @@ -34,11 +34,9 @@ class HotStarIE(InfoExtractor): 'only_matching': True, }] - _GET_CONTENT_TEMPLATE = 'http://account.hotstar.com/AVS/besc?action=GetAggregatedContentDetails&channel=PCTV&contentId=%s' - _GET_CDN_TEMPLATE = 'http://getcdn.hotstar.com/AVS/besc?action=GetCDN&asJson=Y&channel=%s&id=%s&type=%s' - - def _download_json(self, url_or_request, video_id, note='Downloading JSON metadata', fatal=True): - json_data = super(HotStarIE, self)._download_json(url_or_request, video_id, note, fatal=fatal) + def _download_json(self, url_or_request, video_id, note='Downloading JSON metadata', fatal=True, query=None): + json_data = super(HotStarIE, self)._download_json( + url_or_request, video_id, note, fatal=fatal, query=query) if json_data['resultCode'] != 'OK': if fatal: raise ExtractorError(json_data['errorDescription']) @@ -48,20 +46,37 @@ class HotStarIE(InfoExtractor): def _real_extract(self, url): video_id = self._match_id(url) video_data = self._download_json( - self._GET_CONTENT_TEMPLATE % video_id, - video_id)['contentInfo'][0] + 'http://account.hotstar.com/AVS/besc', video_id, query={ + 'action': 'GetAggregatedContentDetails', + 'channel': 'PCTV', + 'contentId': video_id, + })['contentInfo'][0] + title = video_data['episodeTitle'] + + if video_data.get('encrypted') == 'Y': + raise ExtractorError('This video is DRM protected.', expected=True) formats = [] - # PCTV for extracting f4m manifest - for f in ('TABLET',): + for f in ('JIO',): format_data = self._download_json( - self._GET_CDN_TEMPLATE % (f, video_id, 'VOD'), - video_id, 'Downloading %s JSON metadata' % f, fatal=False) + 'http://getcdn.hotstar.com/AVS/besc', + video_id, 'Downloading %s JSON metadata' % f, + fatal=False, query={ + 'action': 'GetCDN', + 'asJson': 'Y', + 'channel': f, + 'id': video_id, + 'type': 'VOD', + }) if format_data: - format_url = format_data['src'] + format_url = format_data.get('src') + if not format_url: + continue ext = determine_ext(format_url) if ext == 'm3u8': - formats.extend(self._extract_m3u8_formats(format_url, video_id, 'mp4', m3u8_id='hls', fatal=False)) + formats.extend(self._extract_m3u8_formats( + format_url, video_id, 'mp4', + m3u8_id='hls', fatal=False)) elif ext == 'f4m': # produce broken files continue @@ -75,9 +90,12 @@ class HotStarIE(InfoExtractor): return { 'id': video_id, - 'title': video_data['episodeTitle'], + 'title': title, 'description': video_data.get('description'), 'duration': int_or_none(video_data.get('duration')), 'timestamp': int_or_none(video_data.get('broadcastDate')), 'formats': formats, + 'episode': title, + 'episode_number': int_or_none(video_data.get('episodeNumber')), + 'series': video_data.get('contentTitle'), }