X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/a4f82408d9a3d921d9c2af7e4d757f44737dc7ea..b61cf51bd505290c3e7e60c68d235a6202c81a4d:/youtube_dl/extractor/wdr.py diff --git a/youtube_dl/extractor/wdr.py b/youtube_dl/extractor/wdr.py index 54d37da..621de1e 100644 --- a/youtube_dl/extractor/wdr.py +++ b/youtube_dl/extractor/wdr.py @@ -1,119 +1,242 @@ -# -*- coding: utf-8 -*- +# coding: utf-8 from __future__ import unicode_literals import re from .common import InfoExtractor from ..utils import ( - compat_parse_qs, - compat_urlparse, determine_ext, + ExtractorError, + js_to_json, + strip_jsonp, unified_strdate, + update_url_query, + urlhandle_detect_ext, ) -class WDRIE(InfoExtractor): - _PLAYER_REGEX = '-(?:video|audio)player(?:_size-[LMS])?' - _VALID_URL = r'(?Phttps?://www\d?\.(?:wdr\d?|funkhauseuropa)\.de/)(?P.+?)(?P%s)?\.html' % _PLAYER_REGEX +class WDRBaseIE(InfoExtractor): + def _extract_wdr_video(self, webpage, display_id): + # for wdr.de the data-extension is in a tag with the class "mediaLink" + # for wdr.de radio players, in a tag with the class "wdrrPlayerPlayBtn" + # for wdrmaus, in a tag with the class "videoButton" (previously a link + # to the page in a multiline "videoLink"-tag) + json_metadata = self._html_search_regex( + r'''(?sx)class= + (?: + (["\'])(?:mediaLink|wdrrPlayerPlayBtn|videoButton)\b.*?\1[^>]+| + (["\'])videoLink\b.*?\2[\s]*>\n[^\n]* + )data-extension=(["\'])(?P(?:(?!\3).)+)\3 + ''', + webpage, 'media link', default=None, group='data') + + if not json_metadata: + return + + media_link_obj = self._parse_json(json_metadata, display_id, + transform_source=js_to_json) + jsonp_url = media_link_obj['mediaObj']['url'] + + metadata = self._download_json( + jsonp_url, display_id, transform_source=strip_jsonp) + + metadata_tracker_data = metadata['trackerData'] + metadata_media_resource = metadata['mediaResource'] + + formats = [] + + # check if the metadata contains a direct URL to a file + for kind, media_resource in metadata_media_resource.items(): + if kind not in ('dflt', 'alt'): + continue + + for tag_name, medium_url in media_resource.items(): + if tag_name not in ('videoURL', 'audioURL'): + continue + + ext = determine_ext(medium_url) + if ext == 'm3u8': + formats.extend(self._extract_m3u8_formats( + medium_url, display_id, 'mp4', 'm3u8_native', + m3u8_id='hls')) + elif ext == 'f4m': + manifest_url = update_url_query( + medium_url, {'hdcore': '3.2.0', 'plugin': 'aasp-3.2.0.77.18'}) + formats.extend(self._extract_f4m_formats( + manifest_url, display_id, f4m_id='hds', fatal=False)) + elif ext == 'smil': + formats.extend(self._extract_smil_formats( + medium_url, 'stream', fatal=False)) + else: + a_format = { + 'url': medium_url + } + if ext == 'unknown_video': + urlh = self._request_webpage( + medium_url, display_id, note='Determining extension') + ext = urlhandle_detect_ext(urlh) + a_format['ext'] = ext + formats.append(a_format) + + self._sort_formats(formats) + + subtitles = {} + caption_url = metadata_media_resource.get('captionURL') + if caption_url: + subtitles['de'] = [{ + 'url': caption_url, + 'ext': 'ttml', + }] + + title = metadata_tracker_data['trackerClipTitle'] + + return { + 'id': metadata_tracker_data.get('trackerClipId', display_id), + 'display_id': display_id, + 'title': title, + 'alt_title': metadata_tracker_data.get('trackerClipSubcategory'), + 'formats': formats, + 'subtitles': subtitles, + 'upload_date': unified_strdate(metadata_tracker_data.get('trackerClipAirTime')), + } + + +class WDRIE(WDRBaseIE): + _CURRENT_MAUS_URL = r'https?://(?:www\.)wdrmaus.de/(?:[^/]+/){1,2}[^/?#]+\.php5' + _PAGE_REGEX = r'/(?:mediathek/)?[^/]+/(?P[^/]+)/(?P.+)\.html' + _VALID_URL = r'(?Phttps?://(?:www\d\.)?wdr\d?\.de)' + _PAGE_REGEX + '|' + _CURRENT_MAUS_URL _TESTS = [ { - 'url': 'http://www1.wdr.de/mediathek/video/sendungen/servicezeit/videoservicezeit560-videoplayer_size-L.html', + 'url': 'http://www1.wdr.de/mediathek/video/sendungen/doku-am-freitag/video-geheimnis-aachener-dom-100.html', + # HDS download, MD5 is unstable 'info_dict': { - 'id': 'mdb-362427', + 'id': 'mdb-1058683', 'ext': 'flv', - 'title': 'Servicezeit', - 'description': 'md5:c8f43e5e815eeb54d0b96df2fba906cb', - 'upload_date': '20140310', + 'display_id': 'doku-am-freitag/video-geheimnis-aachener-dom-100', + 'title': 'Geheimnis Aachener Dom', + 'alt_title': 'Doku am Freitag', + 'upload_date': '20160304', + 'description': 'md5:87be8ff14d8dfd7a7ee46f0299b52318', + 'is_live': False, + 'subtitles': {'de': [{ + 'url': 'http://ondemand-ww.wdr.de/medp/fsk0/105/1058683/1058683_12220974.xml', + 'ext': 'ttml', + }]}, }, - 'params': { - 'skip_download': True, + }, + { + 'url': 'http://www1.wdr.de/mediathek/audio/wdr3/wdr3-gespraech-am-samstag/audio-schriftstellerin-juli-zeh-100.html', + 'md5': 'f4c1f96d01cf285240f53ea4309663d8', + 'info_dict': { + 'id': 'mdb-1072000', + 'ext': 'mp3', + 'display_id': 'wdr3-gespraech-am-samstag/audio-schriftstellerin-juli-zeh-100', + 'title': 'Schriftstellerin Juli Zeh', + 'alt_title': 'WDR 3 Gespräch am Samstag', + 'upload_date': '20160312', + 'description': 'md5:e127d320bc2b1f149be697ce044a3dd7', + 'is_live': False, + 'subtitles': {} }, }, { - 'url': 'http://www1.wdr.de/themen/av/videomargaspiegelisttot101-videoplayer.html', + 'url': 'http://www1.wdr.de/mediathek/video/live/index.html', 'info_dict': { - 'id': 'mdb-363194', - 'ext': 'flv', - 'title': 'Marga Spiegel ist tot', - 'description': 'md5:2309992a6716c347891c045be50992e4', - 'upload_date': '20140311', + 'id': 'mdb-103364', + 'ext': 'mp4', + 'display_id': 'index', + 'title': r're:^WDR Fernsehen im Livestream [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$', + 'alt_title': 'WDR Fernsehen Live', + 'upload_date': None, + 'description': 'md5:ae2ff888510623bf8d4b115f95a9b7c9', + 'is_live': True, + 'subtitles': {} }, 'params': { - 'skip_download': True, + 'skip_download': True, # m3u8 download }, }, { - 'url': 'http://www1.wdr.de/themen/kultur/audioerlebtegeschichtenmargaspiegel100-audioplayer.html', - 'md5': '83e9e8fefad36f357278759870805898', + 'url': 'http://www1.wdr.de/mediathek/video/sendungen/aktuelle-stunde/aktuelle-stunde-120.html', + 'playlist_mincount': 8, 'info_dict': { - 'id': 'mdb-194332', - 'ext': 'mp3', - 'title': 'Erlebte Geschichten: Marga Spiegel (29.11.2009)', - 'description': 'md5:2309992a6716c347891c045be50992e4', - 'upload_date': '20091129', + 'id': 'aktuelle-stunde/aktuelle-stunde-120', }, }, { - 'url': 'http://www.funkhauseuropa.de/av/audioflaviacoelhoamaramar100-audioplayer.html', - 'md5': '99a1443ff29af19f6c52cf6f4dc1f4aa', + 'url': 'http://www.wdrmaus.de/aktuelle-sendung/index.php5', 'info_dict': { - 'id': 'mdb-478135', - 'ext': 'mp3', - 'title': 'Flavia Coelho: Amar é Amar', - 'description': 'md5:7b29e97e10dfb6e265238b32fa35b23a', - 'upload_date': '20140717', + 'id': 'mdb-1323501', + 'ext': 'mp4', + 'upload_date': 're:^[0-9]{8}$', + 'title': 're:^Die Sendung mit der Maus vom [0-9.]{10}$', + 'description': 'Die Seite mit der Maus -', }, + 'skip': 'The id changes from week to week because of the new episode' }, + { + 'url': 'http://www.wdrmaus.de/filme/sachgeschichten/achterbahn.php5', + 'md5': '803138901f6368ee497b4d195bb164f2', + 'info_dict': { + 'id': 'mdb-186083', + 'ext': 'mp4', + 'upload_date': '20130919', + 'title': 'Sachgeschichte - Achterbahn ', + 'description': 'Die Seite mit der Maus -', + }, + }, + { + 'url': 'http://www1.wdr.de/radio/player/radioplayer116~_layout-popupVersion.html', + # Live stream, MD5 unstable + 'info_dict': { + 'id': 'mdb-869971', + 'ext': 'flv', + 'title': 'COSMO Livestream', + 'description': 'md5:2309992a6716c347891c045be50992e4', + 'upload_date': '20160101', + }, + } ] def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) - page_url = mobj.group('url') - page_id = mobj.group('id') + url_type = mobj.group('type') + page_url = mobj.group('page_url') + display_id = mobj.group('display_id') + webpage = self._download_webpage(url, display_id) - webpage = self._download_webpage(url, page_id) + info_dict = self._extract_wdr_video(webpage, display_id) - if mobj.group('player') is None: + if not info_dict: entries = [ - self.url_result(page_url + href, 'WDR') - for href in re.findall(r']+data-extension=' % self._PAGE_REGEX, + webpage) ] - return self.playlist_result(entries, page_id) - - flashvars = compat_parse_qs( - self._html_search_regex(r'

Sendedatum:\s*([0-9\.]+)

', - webpage, 'air date') - title_str = self._html_search_regex( - r'

(.*?)

', webpage, 'title') - title = '%s - %s' % (title_date, title_str) - upload_date = unified_strdate( - self._html_search_meta('dc.date', webpage)) - - fields = compat_parse_qs(param_code) - video_url = fields['firstVideo'][0] - thumbnail = compat_urlparse.urljoin(url, fields['startPicture'][0]) - - formats = [{ - 'format_id': 'rtmp', - 'url': video_url, - }] - - jscode = self._download_webpage( - 'http://www.wdrmaus.de/codebase/js/extended-medien.min.js', - video_id, fatal=False, - note='Downloading URL translation table', - errnote='Could not download URL translation table') - if jscode: - for m in re.finditer( - r"stream:\s*'dslSrc=(?P[^']+)',\s*download:\s*'(?P
[^']+)'\s*\}", - jscode): - if video_url.startswith(m.group('stream')): - http_url = video_url.replace( - m.group('stream'), m.group('dl')) - formats.append({ - 'format_id': 'http', - 'url': http_url, - }) - break - - self._sort_formats(formats) - - return { - 'id': video_id, - 'title': title, - 'formats': formats, - 'thumbnail': thumbnail, - 'upload_date': upload_date, + 'http_headers': { + 'User-Agent': 'mobile', + }, } - -# TODO test _1 \ No newline at end of file