youtube-dl/youtube_dl/extractor/arte.py
Jaime Marquínez Ferrándiz 75c9481224 ArteTvIE: rewrite the extract process to support the new site (fixes #875)
The video can be downloaded with rtmp or http, but the best quality format seems to always use rtmp.
Deleted the old methods.
2013-06-27 00:09:51 +02:00

86 lines
3.1 KiB
Python

import re
import json
from .common import InfoExtractor
from ..utils import (
# This is used by the not implemented extractLiveStream method
compat_urllib_parse,
ExtractorError,
unified_strdate,
)
class ArteTvIE(InfoExtractor):
_VALID_URL = r'(?:http://)?www\.arte.tv/guide/(?:fr|de)/(?:(?:sendungen|emissions)/)?(?P<id>.*?)/(?P<name>.*?)(\?.*)?'
_LIVE_URL = r'index-[0-9]+\.html$'
IE_NAME = u'arte.tv'
# TODO implement Live Stream
# def extractLiveStream(self, url):
# video_lang = url.split('/')[-4]
# info = self.grep_webpage(
# url,
# r'src="(.*?/videothek_js.*?\.js)',
# 0,
# [
# (1, 'url', u'Invalid URL: %s' % url)
# ]
# )
# http_host = url.split('/')[2]
# next_url = 'http://%s%s' % (http_host, compat_urllib_parse.unquote(info.get('url')))
# info = self.grep_webpage(
# next_url,
# r'(s_artestras_scst_geoFRDE_' + video_lang + '.*?)\'.*?' +
# '(http://.*?\.swf).*?' +
# '(rtmp://.*?)\'',
# re.DOTALL,
# [
# (1, 'path', u'could not extract video path: %s' % url),
# (2, 'player', u'could not extract video player: %s' % url),
# (3, 'url', u'could not extract video url: %s' % url)
# ]
# )
# video_url = u'%s/%s' % (info.get('url'), info.get('path'))
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
name = mobj.group('name')
# This is not a real id, it can be for example AJT for the news
# http://www.arte.tv/guide/fr/emissions/AJT/arte-journal
video_id = mobj.group('id')
if re.search(self._LIVE_URL, video_id) is not None:
raise ExtractorError(u'Arte live streams are not yet supported, sorry')
# self.extractLiveStream(url)
# return
webpage = self._download_webpage(url, video_id)
json_url = self._html_search_regex(r'arte_vp_url="(.*?)"', webpage, 'json url')
json_info = self._download_webpage(json_url, video_id, 'Downloading info json')
self.report_extraction(video_id)
info = json.loads(json_info)
player_info = info['videoJsonPlayer']
info_dict = {'id': player_info['VID'],
'title': player_info['VTI'],
'description': player_info['VDE'],
'upload_date': unified_strdate(player_info['VDA'].split(' ')[0]),
'thumbnail': player_info['programImage'],
}
formats = player_info['VSR'].values()
# We order the formats by quality
formats = sorted(formats, key=lambda f: int(f['height']))
# Pick the best quality
format_info = formats[-1]
if format_info['mediaType'] == u'rtmp':
info_dict['url'] = format_info['streamer']
info_dict['play_path'] = 'mp4:' + format_info['url']
info_dict['ext'] = 'mp4'
else:
info_dict['url'] = format_info['url']
info_dict['ext'] = 'mp4'
return info_dict