1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-27 02:44:33 +01:00
youtube-dl/youtube_dl/extractor/rtlnow.py

156 lines
5.9 KiB
Python
Raw Normal View History

# encoding: utf-8
2014-01-25 14:18:54 +01:00
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
2014-02-26 11:36:16 +01:00
clean_html,
unified_strdate,
int_or_none,
)
class RTLnowIE(InfoExtractor):
"""Information Extractor for RTL NOW, RTL2 NOW, RTL NITRO, SUPER RTL NOW, VOX NOW and n-tv NOW"""
2014-02-26 11:36:16 +01:00
_VALID_URL = r'''(?x)
(?:https?://)?
(?P<url>
(?P<domain>
rtl-now\.rtl\.de|
rtl2now\.rtl2\.de|
(?:www\.)?voxnow\.de|
(?:www\.)?rtlnitronow\.de|
(?:www\.)?superrtlnow\.de|
(?:www\.)?n-tvnow\.de)
/+[a-zA-Z0-9-]+/[a-zA-Z0-9-]+\.php\?
(?:container_id|film_id)=(?P<video_id>[0-9]+)&
player=1(?:&season=[0-9]+)?(?:&.*)?
)'''
_TESTS = [
{
'url': 'http://rtl-now.rtl.de/ahornallee/folge-1.php?film_id=90419&player=1&season=1',
'info_dict': {
'id': '90419',
'ext': 'flv',
'title': 'Ahornallee - Folge 1 - Der Einzug',
'description': 'md5:ce843b6b5901d9a7f7d04d1bbcdb12de',
'upload_date': '20070416',
'duration': 1685,
},
'params': {
'skip_download': True,
},
'skip': 'Only works from Germany',
2013-08-19 08:57:36 +02:00
},
2014-02-26 11:36:16 +01:00
{
'url': 'http://rtl2now.rtl2.de/aerger-im-revier/episode-15-teil-1.php?film_id=69756&player=1&season=2&index=5',
'info_dict': {
'id': '69756',
'ext': 'flv',
'title': 'Ärger im Revier - Ein junger Ladendieb, ein handfester Streit u.a.',
'description': 'md5:3fb247005ed21a935ffc82b7dfa70cf0',
'thumbnail': 'http://autoimg.static-fra.de/rtl2now/219850/1500x1500/image2.jpg',
'upload_date': '20120519',
'duration': 1245,
},
'params': {
'skip_download': True,
},
'skip': 'Only works from Germany',
2013-08-24 13:22:28 +02:00
},
2014-02-26 11:36:16 +01:00
{
'url': 'http://www.voxnow.de/voxtours/suedafrika-reporter-ii.php?film_id=13883&player=1&season=17',
'info_dict': {
'id': '13883',
'ext': 'flv',
'title': 'Voxtours - Südafrika-Reporter II',
'description': 'md5:de7f8d56be6fd4fed10f10f57786db00',
'upload_date': '20090627',
'duration': 1800,
},
'params': {
'skip_download': True,
},
2013-08-24 13:22:28 +02:00
},
2014-02-26 11:36:16 +01:00
{
'url': 'http://superrtlnow.de/medicopter-117/angst.php?film_id=99205&player=1',
'info_dict': {
'id': '99205',
'ext': 'flv',
'title': 'Medicopter 117 - Angst!',
'description': 'md5:895b1df01639b5f61a04fc305a5cb94d',
'thumbnail': 'http://autoimg.static-fra.de/superrtlnow/287529/1500x1500/image2.jpg',
'upload_date': '20080928',
'duration': 2691,
},
'params': {
'skip_download': True,
},
},
2014-02-26 11:36:16 +01:00
{
2014-02-26 11:43:56 +01:00
'url': 'http://www.n-tvnow.de/deluxe-alles-was-spass-macht/thema-ua-luxushotel-fuer-vierbeiner.php?container_id=153819&player=1&season=0',
2014-07-23 01:49:25 +02:00
'only_matching': True,
},
2014-02-26 11:36:16 +01:00
]
2014-01-25 14:18:54 +01:00
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
2014-02-26 11:36:16 +01:00
video_page_url = 'http://%s/' % mobj.group('domain')
2014-01-25 14:18:54 +01:00
video_id = mobj.group('video_id')
2014-02-26 11:36:16 +01:00
webpage = self._download_webpage('http://' + mobj.group('url'), video_id)
mobj = re.search(r'(?s)<div style="margin-left: 20px; font-size: 13px;">(.*?)<div id="playerteaser">', webpage)
if mobj:
raise ExtractorError(clean_html(mobj.group(1)), expected=True)
title = self._og_search_title(webpage)
description = self._og_search_description(webpage)
thumbnail = self._og_search_thumbnail(webpage, default=None)
2014-02-26 11:36:16 +01:00
upload_date = unified_strdate(self._html_search_meta('uploadDate', webpage, 'upload date'))
2014-02-26 11:49:49 +01:00
mobj = re.search(r'<meta itemprop="duration" content="PT(?P<seconds>\d+)S" />', webpage)
duration = int(mobj.group('seconds')) if mobj else None
2014-01-25 14:18:54 +01:00
playerdata_url = self._html_search_regex(
2014-02-26 11:36:16 +01:00
r"'playerdata': '(?P<playerdata_url>[^']+)'", webpage, 'playerdata_url')
2014-02-26 11:36:16 +01:00
playerdata = self._download_xml(playerdata_url, video_id, 'Downloading player data XML')
2014-02-26 11:36:16 +01:00
videoinfo = playerdata.find('./playlist/videoinfo')
2014-02-26 11:49:49 +01:00
2014-02-26 11:36:16 +01:00
formats = []
for filename in videoinfo.findall('filename'):
mobj = re.search(r'(?P<url>rtmpe://(?:[^/]+/){2})(?P<play_path>.+)', filename.text)
if mobj:
fmt = {
'url': mobj.group('url'),
'play_path': 'mp4:' + mobj.group('play_path'),
'page_url': video_page_url,
'player_url': video_page_url + 'includes/vodplayer.swf',
}
else:
fmt = {
'url': filename.text,
}
fmt.update({
'width': int_or_none(filename.get('width')),
'height': int_or_none(filename.get('height')),
'vbr': int_or_none(filename.get('bitrate')),
'ext': 'flv',
})
formats.append(fmt)
2014-01-25 14:18:54 +01:00
return {
'id': video_id,
2014-02-26 11:36:16 +01:00
'title': title,
'description': description,
'thumbnail': thumbnail,
'upload_date': upload_date,
'duration': duration,
'formats': formats,
}