yt-dlc/youtube_dl/extractor/reverbnation.py

45 lines
1.4 KiB
Python
Raw Normal View History

from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import str_or_none
class ReverbNationIE(InfoExtractor):
_VALID_URL = r'^https?://(?:www\.)?reverbnation\.com/.*?/song/(?P<id>\d+).*?$'
_TESTS = [{
'url': 'http://www.reverbnation.com/alkilados/song/16965047-mona-lisa',
'md5': '3da12ebca28c67c111a7f8b262d3f7a7',
'info_dict': {
2016-02-14 09:37:17 +00:00
'id': '16965047',
'ext': 'mp3',
'title': 'MONA LISA',
'uploader': 'ALKILADOS',
'uploader_id': '216429',
'thumbnail': 're:^https://gp1\.wac\.edgecastcdn\.net/.*?\.jpg$'
},
}]
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
song_id = mobj.group('id')
api_res = self._download_json(
'https://api.reverbnation.com/song/%s' % song_id,
song_id,
note='Downloading information of song %s' % song_id
)
return {
'id': song_id,
'title': api_res.get('name'),
'url': api_res.get('url'),
'uploader': api_res.get('artist', {}).get('name'),
'uploader_id': str_or_none(api_res.get('artist', {}).get('id')),
'thumbnail': self._proto_relative_url(
api_res.get('image', api_res.get('thumbnail'))),
'ext': 'mp3',
'vcodec': 'none',
}