from __future__ import unicode_literals from .common import InfoExtractor from ..compat import compat_str from ..utils import ( int_or_none, parse_duration, ) class MwaveIE(InfoExtractor): _VALID_URL = r'https?://mwave\.interest\.me/mnettv/videodetail\.m\?searchVideoDetailVO\.clip_id=(?P[0-9]+)' _TEST = { 'url': 'http://mwave.interest.me/mnettv/videodetail.m?searchVideoDetailVO.clip_id=168859', # md5 is unstable 'info_dict': { 'id': '168859', 'ext': 'flv', 'title': '[M COUNTDOWN] SISTAR - SHAKE IT', 'thumbnail': 're:^https?://.*\.jpg$', 'uploader': 'M COUNTDOWN', 'duration': 206, 'view_count': int, } } def _real_extract(self, url): video_id = self._match_id(url) vod_info = self._download_json( 'http://mwave.interest.me/onair/vod_info.m?vodtype=CL§orid=&endinfo=Y&id=%s' % video_id, video_id, 'Download vod JSON') formats = [] for num, cdn_info in enumerate(vod_info['cdn']): stream_url = cdn_info.get('url') if not stream_url: continue stream_name = cdn_info.get('name') or compat_str(num) f4m_stream = self._download_json( stream_url, video_id, 'Download %s stream JSON' % stream_name) f4m_url = f4m_stream.get('fileurl') if not f4m_url: continue formats.extend( self._extract_f4m_formats(f4m_url + '&hdcore=3.0.3', video_id, f4m_id=stream_name)) self._sort_formats(formats) return { 'id': video_id, 'title': vod_info['title'], 'thumbnail': vod_info.get('cover'), 'uploader': vod_info.get('program_title'), 'duration': parse_duration(vod_info.get('time')), 'view_count': int_or_none(vod_info.get('hit')), 'formats': formats, } class MwaveMeetGreetIE(InfoExtractor): _VALID_URL = r'https?://mwave\.interest\.me/meetgreet/view/(?P[0-9]+)' _TEST = { 'url': 'http://mwave.interest.me/meetgreet/view/256', 'info_dict': { 'id': '173294', 'ext': 'flv', 'title': '[MEET&GREET] Park BoRam', 'thumbnail': 're:^https?://.*\.jpg$', 'uploader': 'Mwave', 'duration': 3634, 'view_count': int, } } def _real_extract(self, url): video_id = self._match_id(url) webpage = self._download_webpage(url, video_id) clip_id = self._html_search_regex(r'