X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/3477c644417600d9ec8f8d2a44f82da0a4b15eb5..aebf9c8f14c9f5d45f8c6aa523ee135fdfd421b2:/youtube_dl/extractor/dcn.py diff --git a/youtube_dl/extractor/dcn.py b/youtube_dl/extractor/dcn.py index 6f2fea5..15a1c40 100644 --- a/youtube_dl/extractor/dcn.py +++ b/youtube_dl/extractor/dcn.py @@ -1,28 +1,90 @@ # coding: utf-8 from __future__ import unicode_literals +import re +import base64 + from .common import InfoExtractor from ..compat import ( compat_urllib_parse, - compat_urllib_request, + compat_str, ) from ..utils import ( int_or_none, parse_iso8601, + sanitized_Request, + smuggle_url, + unsmuggle_url, ) class DCNIE(InfoExtractor): - _VALID_URL = r'https?://(?:www\.)?dcndigital\.ae/(?:#/)?(?:video/.+|show/\d+/.+?)/(?P\d+)' + _VALID_URL = r'https?://(?:www\.)?dcndigital\.ae/(?:#/)?show/(?P\d+)/[^/]+(?:/(?P\d+)/(?P\d+))?' + + def _real_extract(self, url): + show_id, video_id, season_id = re.match(self._VALID_URL, url).groups() + if video_id and int(video_id) > 0: + return self.url_result( + 'http://www.dcndigital.ae/media/%s' % video_id, 'DCNVideo') + elif season_id and int(season_id) > 0: + return self.url_result(smuggle_url( + 'http://www.dcndigital.ae/program/season/%s' % season_id, + {'show_id': show_id}), 'DCNSeason') + else: + return self.url_result( + 'http://www.dcndigital.ae/program/%s' % show_id, 'DCNSeason') + + +class DCNBaseIE(InfoExtractor): + def _extract_video_info(self, video_data, video_id, is_live): + title = video_data.get('title_en') or video_data['title_ar'] + img = video_data.get('img') + thumbnail = 'http://admin.mangomolo.com/analytics/%s' % img if img else None + duration = int_or_none(video_data.get('duration')) + description = video_data.get('description_en') or video_data.get('description_ar') + timestamp = parse_iso8601(video_data.get('create_time'), ' ') + + return { + 'id': video_id, + 'title': self._live_title(title) if is_live else title, + 'description': description, + 'thumbnail': thumbnail, + 'duration': duration, + 'timestamp': timestamp, + 'is_live': is_live, + } + + def _extract_video_formats(self, webpage, video_id, entry_protocol): + formats = [] + m3u8_url = self._html_search_regex( + r'file\s*:\s*"([^"]+)', webpage, 'm3u8 url', fatal=False) + if m3u8_url: + formats.extend(self._extract_m3u8_formats( + m3u8_url, video_id, 'mp4', entry_protocol, m3u8_id='hls', fatal=None)) + + rtsp_url = self._search_regex( + r']+href="(rtsp://[^"]+)"', webpage, 'rtsp url', fatal=False) + if rtsp_url: + formats.append({ + 'url': rtsp_url, + 'format_id': 'rtsp', + }) + + self._sort_formats(formats) + return formats + + +class DCNVideoIE(DCNBaseIE): + IE_NAME = 'dcn:video' + _VALID_URL = r'https?://(?:www\.)?dcndigital\.ae/(?:#/)?(?:video/[^/]+|media|catchup/[^/]+/[^/]+)/(?P\d+)' _TEST = { - 'url': 'http://www.dcndigital.ae/#/show/199074/%D8%B1%D8%AD%D9%84%D8%A9-%D8%A7%D9%84%D8%B9%D9%85%D8%B1-%D8%A7%D9%84%D8%AD%D9%84%D9%82%D8%A9-1/17375/6887', + 'url': 'http://www.dcndigital.ae/#/video/%D8%B1%D8%AD%D9%84%D8%A9-%D8%A7%D9%84%D8%B9%D9%85%D8%B1-%D8%A7%D9%84%D8%AD%D9%84%D9%82%D8%A9-1/17375', 'info_dict': { 'id': '17375', 'ext': 'mp4', 'title': 'رحلة العمر : الحلقة 1', 'description': 'md5:0156e935d870acb8ef0a66d24070c6d6', - 'thumbnail': 're:^https?://.*\.jpg$', 'duration': 2041, 'timestamp': 1227504126, 'upload_date': '20081124', @@ -36,49 +98,99 @@ class DCNIE(InfoExtractor): def _real_extract(self, url): video_id = self._match_id(url) - request = compat_urllib_request.Request( + request = sanitized_Request( 'http://admin.mangomolo.com/analytics/index.php/plus/video?id=%s' % video_id, headers={'Origin': 'http://www.dcndigital.ae'}) - - video = self._download_json(request, video_id) - title = video.get('title_en') or video['title_ar'] + video_data = self._download_json(request, video_id) + info = self._extract_video_info(video_data, video_id, False) webpage = self._download_webpage( 'http://admin.mangomolo.com/analytics/index.php/customers/embed/video?' + compat_urllib_parse.urlencode({ - 'id': video['id'], - 'user_id': video['user_id'], - 'signature': video['signature'], + 'id': video_data['id'], + 'user_id': video_data['user_id'], + 'signature': video_data['signature'], 'countries': 'Q0M=', 'filter': 'DENY', }), video_id) + info['formats'] = self._extract_video_formats(webpage, video_id, 'm3u8_native') + return info - m3u8_url = self._html_search_regex(r'file:\s*"([^"]+)', webpage, 'm3u8 url') - formats = self._extract_m3u8_formats( - m3u8_url, video_id, 'mp4', entry_protocol='m3u8_native', m3u8_id='hls') - rtsp_url = self._search_regex( - r']+href="(rtsp://[^"]+)"', webpage, 'rtsp url', fatal=False) - if rtsp_url: - formats.append({ - 'url': rtsp_url, - 'format_id': 'rtsp', +class DCNLiveIE(DCNBaseIE): + IE_NAME = 'dcn:live' + _VALID_URL = r'https?://(?:www\.)?dcndigital\.ae/(?:#/)?live/(?P\d+)' + + def _real_extract(self, url): + channel_id = self._match_id(url) + + request = sanitized_Request( + 'http://admin.mangomolo.com/analytics/index.php/plus/getchanneldetails?channel_id=%s' % channel_id, + headers={'Origin': 'http://www.dcndigital.ae'}) + + channel_data = self._download_json(request, channel_id) + info = self._extract_video_info(channel_data, channel_id, True) + + webpage = self._download_webpage( + 'http://admin.mangomolo.com/analytics/index.php/customers/embed/index?' + + compat_urllib_parse.urlencode({ + 'id': base64.b64encode(channel_data['user_id'].encode()).decode(), + 'channelid': base64.b64encode(channel_data['id'].encode()).decode(), + 'signature': channel_data['signature'], + 'countries': 'Q0M=', + 'filter': 'DENY', + }), channel_id) + info['formats'] = self._extract_video_formats(webpage, channel_id, 'm3u8') + return info + + +class DCNSeasonIE(InfoExtractor): + IE_NAME = 'dcn:season' + _VALID_URL = r'https?://(?:www\.)?dcndigital\.ae/(?:#/)?program/(?:(?P\d+)|season/(?P\d+))' + _TEST = { + 'url': 'http://dcndigital.ae/#/program/205024/%D9%85%D8%AD%D8%A7%D8%B6%D8%B1%D8%A7%D8%AA-%D8%A7%D9%84%D8%B4%D9%8A%D8%AE-%D8%A7%D9%84%D8%B4%D8%B9%D8%B1%D8%A7%D9%88%D9%8A', + 'info_dict': + { + 'id': '7910', + 'title': 'محاضرات الشيخ الشعراوي', + }, + 'playlist_mincount': 27, + } + + def _real_extract(self, url): + url, smuggled_data = unsmuggle_url(url, {}) + show_id, season_id = re.match(self._VALID_URL, url).groups() + + data = {} + if season_id: + data['season'] = season_id + show_id = smuggled_data.get('show_id') + if show_id is None: + request = sanitized_Request( + 'http://admin.mangomolo.com/analytics/index.php/plus/season_info?id=%s' % season_id, + headers={'Origin': 'http://www.dcndigital.ae'}) + season = self._download_json(request, season_id) + show_id = season['id'] + data['show_id'] = show_id + request = sanitized_Request( + 'http://admin.mangomolo.com/analytics/index.php/plus/show', + compat_urllib_parse.urlencode(data), + { + 'Origin': 'http://www.dcndigital.ae', + 'Content-Type': 'application/x-www-form-urlencoded' }) - self._sort_formats(formats) + show = self._download_json(request, show_id) + if not season_id: + season_id = show['default_season'] + for season in show['seasons']: + if season['id'] == season_id: + title = season.get('title_en') or season['title_ar'] - img = video.get('img') - thumbnail = 'http://admin.mangomolo.com/analytics/%s' % img if img else None - duration = int_or_none(video.get('duration')) - description = video.get('description_en') or video.get('description_ar') - timestamp = parse_iso8601(video.get('create_time') or video.get('update_time'), ' ') + entries = [] + for video in show['videos']: + video_id = compat_str(video['id']) + entries.append(self.url_result( + 'http://www.dcndigital.ae/media/%s' % video_id, 'DCNVideo', video_id)) - return { - 'id': video_id, - 'title': title, - 'description': description, - 'thumbnail': thumbnail, - 'duration': duration, - 'timestamp': timestamp, - 'formats': formats, - } + return self.playlist_result(entries, season_id, title)