ExtractorError,
clean_html,
compat_str,
+ float_or_none,
int_or_none,
parse_iso8601,
try_get,
+ urljoin,
)
-class BeamProLiveIE(InfoExtractor):
- IE_NAME = 'Beam:live'
- _VALID_URL = r'https?://(?:\w+\.)?beam\.pro/(?P<id>[^/?#&]+)'
+class BeamProBaseIE(InfoExtractor):
+ _API_BASE = 'https://mixer.com/api/v1'
_RATINGS = {'family': 0, 'teen': 13, '18+': 18}
+
+ def _extract_channel_info(self, chan):
+ user_id = chan.get('userId') or try_get(chan, lambda x: x['user']['id'])
+ return {
+ 'uploader': chan.get('token') or try_get(
+ chan, lambda x: x['user']['username'], compat_str),
+ 'uploader_id': compat_str(user_id) if user_id else None,
+ 'age_limit': self._RATINGS.get(chan.get('audience')),
+ }
+
+
+class BeamProLiveIE(BeamProBaseIE):
+ IE_NAME = 'Mixer:live'
+ _VALID_URL = r'https?://(?:\w+\.)?(?:beam\.pro|mixer\.com)/(?P<id>[^/?#&]+)'
_TEST = {
- 'url': 'http://www.beam.pro/niterhayven',
+ 'url': 'http://mixer.com/niterhayven',
'info_dict': {
'id': '261562',
'ext': 'mp4',
},
}
+ _MANIFEST_URL_TEMPLATE = '%s/channels/%%s/manifest.%%s' % BeamProBaseIE._API_BASE
+
+ @classmethod
+ def suitable(cls, url):
+ return False if BeamProVodIE.suitable(url) else super(BeamProLiveIE, cls).suitable(url)
+
def _real_extract(self, url):
channel_name = self._match_id(url)
chan = self._download_json(
- 'https://beam.pro/api/v1/channels/%s' % channel_name, channel_name)
+ '%s/channels/%s' % (self._API_BASE, channel_name), channel_name)
if chan.get('online') is False:
raise ExtractorError(
channel_id = chan['id']
+ def manifest_url(kind):
+ return self._MANIFEST_URL_TEMPLATE % (channel_id, kind)
+
formats = self._extract_m3u8_formats(
- 'https://beam.pro/api/v1/channels/%s/manifest.m3u8' % channel_id,
- channel_name, ext='mp4', m3u8_id='hls', fatal=False)
+ manifest_url('m3u8'), channel_name, ext='mp4', m3u8_id='hls',
+ fatal=False)
+ formats.extend(self._extract_smil_formats(
+ manifest_url('smil'), channel_name, fatal=False))
self._sort_formats(formats)
- user_id = chan.get('userId') or try_get(chan, lambda x: x['user']['id'])
-
- return {
+ info = {
'id': compat_str(chan.get('id') or channel_name),
'title': self._live_title(chan.get('name') or channel_name),
'description': clean_html(chan.get('description')),
- 'thumbnail': try_get(chan, lambda x: x['thumbnail']['url'], compat_str),
+ 'thumbnail': try_get(
+ chan, lambda x: x['thumbnail']['url'], compat_str),
'timestamp': parse_iso8601(chan.get('updatedAt')),
- 'uploader': chan.get('token') or try_get(
- chan, lambda x: x['user']['username'], compat_str),
- 'uploader_id': compat_str(user_id) if user_id else None,
- 'age_limit': self._RATINGS.get(chan.get('audience')),
'is_live': True,
'view_count': int_or_none(chan.get('viewersTotal')),
'formats': formats,
}
+ info.update(self._extract_channel_info(chan))
+
+ return info
+
+
+class BeamProVodIE(BeamProBaseIE):
+ IE_NAME = 'Mixer:vod'
+ _VALID_URL = r'https?://(?:\w+\.)?(?:beam\.pro|mixer\.com)/[^/?#&]+\?.*?\bvod=(?P<id>\w+)'
+ _TESTS = [{
+ 'url': 'https://mixer.com/willow8714?vod=2259830',
+ 'md5': 'b2431e6e8347dc92ebafb565d368b76b',
+ 'info_dict': {
+ 'id': '2259830',
+ 'ext': 'mp4',
+ 'title': 'willow8714\'s Channel',
+ 'duration': 6828.15,
+ 'thumbnail': r're:https://.*source\.png$',
+ 'timestamp': 1494046474,
+ 'upload_date': '20170506',
+ 'uploader': 'willow8714',
+ 'uploader_id': '6085379',
+ 'age_limit': 13,
+ 'view_count': int,
+ },
+ 'params': {
+ 'skip_download': True,
+ },
+ }, {
+ 'url': 'https://mixer.com/streamer?vod=IxFno1rqC0S_XJ1a2yGgNw',
+ 'only_matching': True,
+ }]
+
+ @staticmethod
+ def _extract_format(vod, vod_type):
+ if not vod.get('baseUrl'):
+ return []
+
+ if vod_type == 'hls':
+ filename, protocol = 'manifest.m3u8', 'm3u8_native'
+ elif vod_type == 'raw':
+ filename, protocol = 'source.mp4', 'https'
+ else:
+ assert False
+
+ data = vod.get('data') if isinstance(vod.get('data'), dict) else {}
+
+ format_id = [vod_type]
+ if isinstance(data.get('Height'), compat_str):
+ format_id.append('%sp' % data['Height'])
+
+ return [{
+ 'url': urljoin(vod['baseUrl'], filename),
+ 'format_id': '-'.join(format_id),
+ 'ext': 'mp4',
+ 'protocol': protocol,
+ 'width': int_or_none(data.get('Width')),
+ 'height': int_or_none(data.get('Height')),
+ 'fps': int_or_none(data.get('Fps')),
+ 'tbr': int_or_none(data.get('Bitrate'), 1000),
+ }]
+
+ def _real_extract(self, url):
+ vod_id = self._match_id(url)
+
+ vod_info = self._download_json(
+ '%s/recordings/%s' % (self._API_BASE, vod_id), vod_id)
+
+ state = vod_info.get('state')
+ if state != 'AVAILABLE':
+ raise ExtractorError(
+ 'VOD %s is not available (state: %s)' % (vod_id, state),
+ expected=True)
+
+ formats = []
+ thumbnail_url = None
+
+ for vod in vod_info['vods']:
+ vod_type = vod.get('format')
+ if vod_type in ('hls', 'raw'):
+ formats.extend(self._extract_format(vod, vod_type))
+ elif vod_type == 'thumbnail':
+ thumbnail_url = urljoin(vod.get('baseUrl'), 'source.png')
+
+ self._sort_formats(formats)
+
+ info = {
+ 'id': vod_id,
+ 'title': vod_info.get('name') or vod_id,
+ 'duration': float_or_none(vod_info.get('duration')),
+ 'thumbnail': thumbnail_url,
+ 'timestamp': parse_iso8601(vod_info.get('createdAt')),
+ 'view_count': int_or_none(vod_info.get('viewsTotal')),
+ 'formats': formats,
+ }
+ info.update(self._extract_channel_info(vod_info.get('channel') or {}))
+
+ return info