# coding: utf-8 from __future__ import unicode_literals from .common import InfoExtractor from ..utils import ( ExtractorError, int_or_none, float_or_none, mimetype2ext, parse_iso8601, remove_end, update_url_query, ) class DRTVIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?dr\.dk/(?:tv/se|nyheder|radio/ondemand)/(?:[^/]+/)*(?P[\da-z-]+)(?:[/#?]|$)' _GEO_BYPASS = False _GEO_COUNTRIES = ['DK'] IE_NAME = 'drtv' _TESTS = [{ 'url': 'https://www.dr.dk/tv/se/boern/ultra/klassen-ultra/klassen-darlig-taber-10', 'md5': '25e659cccc9a2ed956110a299fdf5983', 'info_dict': { 'id': 'klassen-darlig-taber-10', 'ext': 'mp4', 'title': 'Klassen - Dårlig taber (10)', 'description': 'md5:815fe1b7fa656ed80580f31e8b3c79aa', 'timestamp': 1471991907, 'upload_date': '20160823', 'duration': 606.84, }, 'params': { 'skip_download': True, }, }, { 'url': 'https://www.dr.dk/nyheder/indland/live-christianias-rydning-af-pusher-street-er-i-gang', 'md5': '2c37175c718155930f939ef59952474a', 'info_dict': { 'id': 'christiania-pusher-street-ryddes-drdkrjpo', 'ext': 'mp4', 'title': 'LIVE Christianias rydning af Pusher Street er i gang', 'description': '- Det er det fedeste, der er sket i 20 år, fortæller christianit til DR Nyheder.', 'timestamp': 1472800279, 'upload_date': '20160902', 'duration': 131.4, }, }, { 'url': 'https://www.dr.dk/tv/se/historien-om-danmark/-/historien-om-danmark-stenalder', 'md5': '7c8ca12e6c3d3e3edd59ba5a9b7ca10a', 'info_dict': { 'id': 'historien-om-danmark-stenalder', 'ext': 'mp4', 'title': 'Historien om Danmark: Stenalder (1)', 'description': 'Én fascinerende historie om tusindvis af år, hvor vores land bliver skabt ud af is og vand, og hvor de første danskere ankommer til vores egn. Det bliver en rejse ind i urtiden og det liv, som urtidsjægerne har levet i skovene og ved havet og helt frem til bondestenalderen. Gennem skeletfund afslører eksperter, hvordan vores forfædre har set ud i stenalderen og hvorfor stenaldermennesket byggede de imponerende jættestuer, som ligger overalt i det danske.', 'timestamp': 1490401996, 'upload_date': '20170325', 'duration': 3502.04, }, }] def _real_extract(self, url): video_id = self._match_id(url) webpage = self._download_webpage(url, video_id) if '>Programmet er ikke længere tilgængeligt' in webpage: raise ExtractorError( 'Video %s is not available' % video_id, expected=True) video_id = self._search_regex( (r'data-(?:material-identifier|episode-slug)="([^"]+)"', r'data-resource="[^>"]+mu/programcard/expanded/([^"]+)"'), webpage, 'video id') programcard = self._download_json( 'http://www.dr.dk/mu/programcard/expanded/%s' % video_id, video_id, 'Downloading video JSON') data = programcard['Data'][0] title = remove_end(self._og_search_title( webpage, default=None), ' | TV | DR') or data['Title'] description = self._og_search_description( webpage, default=None) or data.get('Description') timestamp = parse_iso8601(data.get('CreatedTime')) thumbnail = None duration = None restricted_to_denmark = False formats = [] subtitles = {} for asset in data['Assets']: kind = asset.get('Kind') if kind == 'Image': thumbnail = asset.get('Uri') preference = 0 sign_language = asset.get('Target') == 'SignLanguage' if kind in ('VideoResource', 'AudioResource'): duration = float_or_none(asset.get('DurationInMilliseconds'), 1000) restricted_to_denmark = asset.get('RestrictedToDenmark') spoken_subtitles = asset.get('Target') == 'SpokenSubtitles' for link in asset.get('Links', []): uri = link.get('Uri') if not uri: continue target = link.get('Target') format_id = target or '' if spoken_subtitles: preference = -1 format_id += '-spoken-subtitles' if sign_language: preference = -1 format_id += "-sign-language" if target == 'HDS': f4m_formats = self._extract_f4m_formats( uri + '?hdcore=3.3.0&plugin=aasp-3.3.0.99.43', video_id, preference, f4m_id=format_id) if kind == 'AudioResource': for f in f4m_formats: f['vcodec'] = 'none' formats.extend(f4m_formats) elif target == 'HLS': formats.extend(self._extract_m3u8_formats( uri, video_id, 'mp4', entry_protocol='m3u8_native', preference=preference, m3u8_id=format_id)) else: bitrate = link.get('Bitrate') if bitrate: format_id += '-%s' % bitrate formats.append({ 'url': uri, 'format_id': format_id, 'tbr': int_or_none(bitrate), 'ext': link.get('FileFormat'), 'vcodec': 'none' if kind == 'AudioResource' else None, }) subtitles_list = asset.get('SubtitlesList') if isinstance(subtitles_list, list): LANGS = { 'Danish': 'da', } for subs in subtitles_list: if not subs.get('Uri'): continue lang = subs.get('Language') or 'da' subtitles.setdefault(LANGS.get(lang, lang), []).append({ 'url': subs['Uri'], 'ext': mimetype2ext(subs.get('MimeType')) or 'vtt' }) if not formats and restricted_to_denmark: self.raise_geo_restricted( 'Unfortunately, DR is not allowed to show this program outside Denmark.', countries=self._GEO_COUNTRIES) self._sort_formats(formats) return { 'id': video_id, 'title': title, 'description': description, 'thumbnail': thumbnail, 'timestamp': timestamp, 'duration': duration, 'formats': formats, 'subtitles': subtitles, } class DRTVLiveIE(InfoExtractor): IE_NAME = 'drtv:live' _VALID_URL = r'https?://(?:www\.)?dr\.dk/(?:tv|TV)/live/(?P[\da-z-]+)' _GEO_COUNTRIES = ['DK'] _TEST = { 'url': 'https://www.dr.dk/tv/live/dr1', 'info_dict': { 'id': 'dr1', 'ext': 'mp4', 'title': 're:^DR1 [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$', }, 'params': { # m3u8 download 'skip_download': True, }, } def _real_extract(self, url): channel_id = self._match_id(url) channel_data = self._download_json( 'https://www.dr.dk/mu-online/api/1.0/channel/' + channel_id, channel_id) title = self._live_title(channel_data['Title']) formats = [] for streaming_server in channel_data.get('StreamingServers', []): server = streaming_server.get('Server') if not server: continue link_type = streaming_server.get('LinkType') for quality in streaming_server.get('Qualities', []): for stream in quality.get('Streams', []): stream_path = stream.get('Stream') if not stream_path: continue stream_url = update_url_query( '%s/%s' % (server, stream_path), {'b': ''}) if link_type == 'HLS': formats.extend(self._extract_m3u8_formats( stream_url, channel_id, 'mp4', m3u8_id=link_type, fatal=False, live=True)) elif link_type == 'HDS': formats.extend(self._extract_f4m_formats(update_url_query( '%s/%s' % (server, stream_path), {'hdcore': '3.7.0'}), channel_id, f4m_id=link_type, fatal=False)) self._sort_formats(formats) return { 'id': channel_id, 'title': title, 'thumbnail': channel_data.get('PrimaryImageUri'), 'formats': formats, 'is_live': True, }