[urplay] Add new extractor(closes #9332)

This commit is contained in:
Remita Amine 2016-06-29 20:19:31 +01:00
parent 06a96da15b
commit e496fa50cd
2 changed files with 68 additions and 0 deletions

View file

@ -894,6 +894,7 @@ from .udn import UDNEmbedIE
from .digiteka import DigitekaIE
from .unistra import UnistraIE
from .urort import UrortIE
from .urplay import URPlayIE
from .usatoday import USATodayIE
from .ustream import UstreamIE, UstreamChannelIE
from .ustudio import (

View file

@ -0,0 +1,67 @@
# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
class URPlayIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?urplay\.se/program/(?P<id>[0-9]+)'
_TEST = {
'url': 'http://urplay.se/program/190031-tripp-trapp-trad-sovkudde',
'md5': '15ca67b63fd8fb320ac2bcd854bad7b6',
'info_dict': {
'id': '190031',
'ext': 'mp4',
'title': 'Tripp, Trapp, Träd : Sovkudde',
'description': 'md5:b86bffdae04a7e9379d1d7e5947df1d1',
}
}
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
urplayer_data = self._parse_json(self._search_regex(
r'urPlayer.init\(({.+?})\);', webpage, 'urplayer data'), video_id)
host = self._download_json('http://streaming-loadbalancer.ur.se/loadbalancer.json', video_id)['redirect']
formats = []
for quality_attr, quality, preference in (('', 'sd', 0), ('_hd', 'hd', 1)):
file_rtmp = urplayer_data.get('file_rtmp' + quality_attr)
if file_rtmp:
formats.append({
'url': 'rtmp://%s/urplay/mp4:%s' % (host, file_rtmp),
'format_id': quality + '-rtmp',
'ext': 'flv',
'preference': preference,
})
file_http = urplayer_data.get('file_http' + quality_attr) or urplayer_data.get('file_http_sub' + quality_attr)
if file_http:
file_http_base_url = 'http://%s/%s' % (host, file_http)
formats.extend(self._extract_f4m_formats(
file_http_base_url + 'manifest.f4m', video_id,
preference, '%s-hds' % quality, fatal=False))
formats.extend(self._extract_m3u8_formats(
file_http_base_url + 'playlist.m3u8', video_id, 'mp4',
'm3u8_native', preference, '%s-hls' % quality, fatal=False))
self._sort_formats(formats)
subtitles = {}
for subtitle in urplayer_data.get('subtitles', []):
subtitle_url = subtitle.get('file')
kind = subtitle.get('kind')
if subtitle_url or kind and kind != 'captions':
continue
subtitles.setdefault(subtitle.get('label', 'Svenska'), []).append({
'url': subtitle_url,
})
return {
'id': video_id,
'title': urplayer_data['title'],
'description': self._og_search_description(webpage),
'thumbnail': urplayer_data.get('image'),
'series': urplayer_data.get('series_title'),
'subtitles': subtitles,
'formats': formats,
}