[eyedotv] Add new extractor(closes #9582)

This commit is contained in:
remitamine 2016-05-24 20:10:12 +01:00
parent 688c634b7d
commit 1de32771e1
2 changed files with 65 additions and 0 deletions

View File

@ -231,6 +231,7 @@
from .exfm import ExfmIE from .exfm import ExfmIE
from .expotv import ExpoTVIE from .expotv import ExpoTVIE
from .extremetube import ExtremeTubeIE from .extremetube import ExtremeTubeIE
from .eyedotv import EyedoTVIE
from .facebook import FacebookIE from .facebook import FacebookIE
from .faz import FazIE from .faz import FazIE
from .fc2 import FC2IE from .fc2 import FC2IE

View File

@ -0,0 +1,64 @@
# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
xpath_text,
parse_duration,
ExtractorError,
)
class EyedoTVIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?eyedo\.tv/[^/]+/(?:#!/)?Live/Detail/(?P<id>[0-9]+)'
_TEST = {
'url': 'https://www.eyedo.tv/en-US/#!/Live/Detail/16301',
'md5': 'ba14f17995cdfc20c36ba40e21bf73f7',
'info_dict': {
'id': '16301',
'ext': 'mp4',
'title': 'Journée du conseil scientifique de l\'Afnic 2015',
'description': 'md5:4abe07293b2f73efc6e1c37028d58c98',
'uploader': 'Afnic Live',
'uploader_id': '8023',
}
}
_ROOT_URL = 'http://live.eyedo.net:1935/'
def _real_extract(self, url):
video_id = self._match_id(url)
video_data = self._download_xml('http://eyedo.tv/api/live/GetLive/%s' % video_id, video_id)
def _add_ns(path):
return self._xpath_ns(path, 'http://schemas.datacontract.org/2004/07/EyeDo.Core.Implementation.Web.ViewModels.Api')
title = xpath_text(video_data, _add_ns('Titre'), 'title', True)
state_live_code = xpath_text(video_data, _add_ns('StateLiveCode'), 'title', True)
if state_live_code == 'avenir':
raise ExtractorError(
'%s said: We\'re sorry, but this video is not yet available.' % self.IE_NAME,
expected=True)
is_live = state_live_code == 'live'
m3u8_url = None
# http://eyedo.tv/Content/Html5/Scripts/html5view.js
if is_live:
if xpath_text(video_data, 'Cdn') == 'true':
m3u8_url = 'http://rrr.sz.xlcdn.com/?account=eyedo&file=A%s&type=live&service=wowza&protocol=http&output=playlist.m3u8' % video_id
else:
m3u8_url = self._ROOT_URL + 'w/%s/eyedo_720p/playlist.m3u8' % video_id
else:
m3u8_url = self._ROOT_URL + 'replay-w/%s/mp4:%s.mp4/playlist.m3u8' % (video_id, video_id)
return {
'id': video_id,
'title': title,
'formats': self._extract_m3u8_formats(
m3u8_url, video_id, 'mp4', 'm3u8' if is_live else 'm3u8_native'),
'description': xpath_text(video_data, _add_ns('Description')),
'duration': parse_duration(xpath_text(video_data, _add_ns('Duration'))),
'uploader': xpath_text(video_data, _add_ns('Createur')),
'uploader_id': xpath_text(video_data, _add_ns('CreateurId')),
'chapter': xpath_text(video_data, _add_ns('ChapitreTitre')),
'chapter_id': xpath_text(video_data, _add_ns('ChapitreId')),
}