|
|
@ -0,0 +1,30 @@ |
|
|
|
# coding: utf-8 |
|
|
|
from __future__ import unicode_literals |
|
|
|
|
|
|
|
from .common import InfoExtractor |
|
|
|
|
|
|
|
|
|
|
|
class NRLTVIE(InfoExtractor): |
|
|
|
_VALID_URL = r'https?://(?:www\.)?nrl\.com/tv(/[^/]+)*/(?P<id>[^/?&#]+)' |
|
|
|
_TEST = { |
|
|
|
'url': 'https://www.nrl.com/tv/news/match-highlights-titans-v-knights-862805/', |
|
|
|
'info_dict': { |
|
|
|
'id': 'YyNnFuaDE6kPJqlDhG4CGQ_w89mKTau4', |
|
|
|
'ext': 'mp4', |
|
|
|
'title': 'Match Highlights: Titans v Knights', |
|
|
|
}, |
|
|
|
'params': { |
|
|
|
# m3u8 download |
|
|
|
'skip_download': True, |
|
|
|
'format': 'bestvideo', |
|
|
|
}, |
|
|
|
} |
|
|
|
|
|
|
|
def _real_extract(self, url): |
|
|
|
display_id = self._match_id(url) |
|
|
|
webpage = self._download_webpage(url, display_id) |
|
|
|
q_data = self._parse_json(self._search_regex( |
|
|
|
r"(?s)q-data='({.+?})'", webpage, 'player data'), display_id) |
|
|
|
ooyala_id = q_data['videoId'] |
|
|
|
return self.url_result( |
|
|
|
'ooyala:' + ooyala_id, 'Ooyala', ooyala_id, q_data.get('title')) |