1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-24 10:22:53 +08:00
youtube-dl/youtube_dl/extractor/ard.py

84 lines
3.0 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
determine_ext,
ExtractorError,
)
class ARDIE(InfoExtractor):
_VALID_URL = r'^https?://(?:(?:www\.)?ardmediathek\.de|mediathek\.daserste\.de)/(?:.*/)(?P<video_id>[^/\?]+)(?:\?.*)?'
2013-06-28 02:46:46 +08:00
_TEST = {
'url': 'http://www.ardmediathek.de/das-erste/guenther-jauch/edward-snowden-im-interview-held-oder-verraeter?documentId=19288786',
'file': '19288786.mp4',
'md5': '515bf47ce209fb3f5a61b7aad364634c',
'info_dict': {
'title': 'Edward Snowden im Interview - Held oder Verräter?',
'description': 'Edward Snowden hat alles aufs Spiel gesetzt, um die weltweite \xdcberwachung durch die Geheimdienste zu enttarnen. Nun stellt sich der ehemalige NSA-Mitarbeiter erstmals weltweit in einem TV-Interview den Fragen eines NDR-Journalisten. Die Sendung vom Sonntagabend.',
'thumbnail': 'http://www.ardmediathek.de/ard/servlet/contentblob/19/28/87/90/19288790/bild/2250037',
2013-06-28 02:46:46 +08:00
},
'skip': 'Blocked outside of Germany',
2013-06-28 02:46:46 +08:00
}
def _real_extract(self, url):
# determine video id from url
m = re.match(self._VALID_URL, url)
numid = re.search(r'documentId=([0-9]+)', url)
if numid:
video_id = numid.group(1)
else:
video_id = m.group('video_id')
webpage = self._download_webpage(url, video_id)
title = self._html_search_regex(
r'<h1(?:\s+class="boxTopHeadline")?>(.*?)</h1>', webpage, 'title')
description = self._html_search_meta(
'dcterms.abstract', webpage, 'description')
thumbnail = self._og_search_thumbnail(webpage)
streams = [
mo.groupdict()
for mo in re.finditer(
r'mediaCollection\.addMediaStream\((?P<media_type>\d+), (?P<quality>\d+), "(?P<rtmp_url>[^"]*)", "(?P<video_url>[^"]*)", "[^"]*"\)', webpage)]
if not streams:
if '"fsk"' in webpage:
raise ExtractorError('This video is only available after 20:00')
formats = []
for s in streams:
format = {
'quality': int(s['quality']),
}
if s.get('rtmp_url'):
format['protocol'] = 'rtmp'
format['url'] = s['rtmp_url']
format['playpath'] = s['video_url']
else:
format['url'] = s['video_url']
quality_name = self._search_regex(
r'[,.]([a-zA-Z0-9_-]+),?\.mp4', format['url'],
'quality name', default='NA')
format['format_id'] = '%s-%s-%s-%s' % (
determine_ext(format['url']), quality_name, s['media_type'],
s['quality'])
formats.append(format)
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'description': description,
'formats': formats,
'thumbnail': thumbnail,
}