[swrmediathek] Extract direct links from JSON and add support for audio files
This commit is contained in:
		@@ -4,71 +4,101 @@ from __future__ import unicode_literals
 | 
			
		||||
import re
 | 
			
		||||
 | 
			
		||||
from .common import InfoExtractor
 | 
			
		||||
from ..utils import int_or_none
 | 
			
		||||
from ..utils import parse_duration
 | 
			
		||||
 | 
			
		||||
 | 
			
		||||
class SWRMediathekIE(InfoExtractor):
 | 
			
		||||
    _VALID_URL = r'https?://(?:www\.)?swrmediathek\.de/player\.htm\?show=(?P<videoid>[^?#&]+)'
 | 
			
		||||
    _VALID_URL = r'https?://(?:www\.)?swrmediathek\.de/player\.htm\?show=(?P<id>[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})'
 | 
			
		||||
 | 
			
		||||
    _TESTS = [{
 | 
			
		||||
        'url': 'http://swrmediathek.de/player.htm?show=849790d0-dab8-11e3-a953-0026b975f2e6',
 | 
			
		||||
        'md5': '8c5f6f0172753368547ca8413a7768ac',
 | 
			
		||||
        'info_dict': {
 | 
			
		||||
            'id': '849790d0-dab8-11e3-a953-0026b975f2e6',
 | 
			
		||||
            'ext': 'flv',
 | 
			
		||||
            'ext': 'mp4',
 | 
			
		||||
            'title': 'SWR odysso',
 | 
			
		||||
            'description': 'md5:2012e31baad36162e97ce9eb3f157b8a',
 | 
			
		||||
            'thumbnail': 're:^http:.*\.jpg$',
 | 
			
		||||
        },
 | 
			
		||||
        'params': {
 | 
			
		||||
            'skip_download': True,  # requires rtmpdump
 | 
			
		||||
            'duration': 2602,
 | 
			
		||||
            'upload_date': '20140515',
 | 
			
		||||
            'uploader': 'SWR Fernsehen',
 | 
			
		||||
            'uploader_id': '990030',
 | 
			
		||||
        },
 | 
			
		||||
    }, {
 | 
			
		||||
        'url': 'http://swrmediathek.de/player.htm?show=0e1a8510-ddf2-11e3-9be3-0026b975f2e6',
 | 
			
		||||
        'md5': 'b10ab854f912eecc5a6b55cd6fc1f545',
 | 
			
		||||
        'info_dict': {
 | 
			
		||||
            'id': '0e1a8510-ddf2-11e3-9be3-0026b975f2e6',
 | 
			
		||||
            'ext': 'flv',
 | 
			
		||||
            'ext': 'mp4',
 | 
			
		||||
            'title': 'Nachtcafé - Alltagsdroge Alkohol - zwischen Sektempfang und Komasaufen',
 | 
			
		||||
            'description': 'md5:e0a3adc17e47db2c23aab9ebc36dbee2',
 | 
			
		||||
            'thumbnail': 're:http://.*\.jpg',
 | 
			
		||||
            'duration': 5305,
 | 
			
		||||
            'upload_date': '20140516',
 | 
			
		||||
            'uploader': 'SWR Fernsehen',
 | 
			
		||||
            'uploader_id': '990030',
 | 
			
		||||
        },
 | 
			
		||||
        'params': {
 | 
			
		||||
            'skip_download': True,  # requires rtmpdump
 | 
			
		||||
        },
 | 
			
		||||
    }, {
 | 
			
		||||
        'url': 'http://swrmediathek.de/player.htm?show=bba23e10-cb93-11e3-bf7f-0026b975f2e6',
 | 
			
		||||
        'md5': '4382e4ef2c9d7ce6852535fa867a0dd3',
 | 
			
		||||
        'info_dict': {
 | 
			
		||||
            'id': 'bba23e10-cb93-11e3-bf7f-0026b975f2e6',
 | 
			
		||||
            'ext': 'mp3',
 | 
			
		||||
            'title': 'Saša Stanišic: Vor dem Fest',
 | 
			
		||||
            'description': 'md5:5b792387dc3fbb171eb709060654e8c9',
 | 
			
		||||
            'thumbnail': 're:http://.*\.jpg',
 | 
			
		||||
            'duration': 3366,
 | 
			
		||||
            'upload_date': '20140520',
 | 
			
		||||
            'uploader': 'SWR 2',
 | 
			
		||||
            'uploader_id': '284670',
 | 
			
		||||
        }
 | 
			
		||||
    }]
 | 
			
		||||
 | 
			
		||||
    def _real_extract(self, url):
 | 
			
		||||
        mobj = re.match(self._VALID_URL, url)
 | 
			
		||||
        video_id = mobj.group('videoid')
 | 
			
		||||
        video_id = mobj.group('id')
 | 
			
		||||
 | 
			
		||||
        webpage = self._download_webpage(url, video_id)
 | 
			
		||||
        video = self._download_json(
 | 
			
		||||
            'http://swrmediathek.de/AjaxEntry?ekey=%s' % video_id, video_id, 'Downloading video JSON')
 | 
			
		||||
 | 
			
		||||
        smilurl = 'http://swrmediathek.de/rtmpQuals/%s/clips.smil'
 | 
			
		||||
        smildoc = self._download_xml(smilurl % video_id, video_id, 'Downloading SMIL page')
 | 
			
		||||
 | 
			
		||||
        baseurl = smildoc.find('.//meta').attrib['base']
 | 
			
		||||
        attr = video['attr']
 | 
			
		||||
        media_type = attr['entry_etype']
 | 
			
		||||
 | 
			
		||||
        formats = []
 | 
			
		||||
        for video in smildoc.findall('.//video'):
 | 
			
		||||
            vbr = video.attrib.get('system-bitrate')
 | 
			
		||||
            if vbr:
 | 
			
		||||
                vbr = int(vbr) / 1000
 | 
			
		||||
        for entry in video['sub']:
 | 
			
		||||
            if entry['name'] != 'entry_media':
 | 
			
		||||
                continue
 | 
			
		||||
 | 
			
		||||
            formats.append({
 | 
			
		||||
                'format_id': video.attrib['height'] + 'p',
 | 
			
		||||
                'width': int_or_none(video.attrib['width']),
 | 
			
		||||
                'height': int_or_none(video.attrib['height']),
 | 
			
		||||
                'vbr': vbr,
 | 
			
		||||
                'url': baseurl,
 | 
			
		||||
                'play_path': 'mp4:' + video.attrib['src'],
 | 
			
		||||
                'ext': 'flv',
 | 
			
		||||
            })
 | 
			
		||||
            entry_attr = entry['attr']
 | 
			
		||||
            codec = entry_attr['val0']
 | 
			
		||||
            quality = int(entry_attr['val1'])
 | 
			
		||||
 | 
			
		||||
            fmt = {
 | 
			
		||||
                'url': entry_attr['val2'],
 | 
			
		||||
                'quality': quality,
 | 
			
		||||
            }
 | 
			
		||||
 | 
			
		||||
            if media_type == 'Video':
 | 
			
		||||
                fmt.update({
 | 
			
		||||
                    'format_note': ['144p', '288p', '544p'][quality-1],
 | 
			
		||||
                    'vcodec': codec,
 | 
			
		||||
                })
 | 
			
		||||
            elif media_type == 'Audio':
 | 
			
		||||
                fmt.update({
 | 
			
		||||
                    'acodec': codec,
 | 
			
		||||
                })
 | 
			
		||||
            formats.append(fmt)
 | 
			
		||||
 | 
			
		||||
        self._sort_formats(formats)
 | 
			
		||||
 | 
			
		||||
        return {
 | 
			
		||||
            'id': video_id,
 | 
			
		||||
            'title': self._html_search_meta('title', webpage, 'title', fatal=True),
 | 
			
		||||
            'thumbnail': self._search_regex(r'<link rel="image_src".+href="(.+)" />', webpage, 'thumbnail'),
 | 
			
		||||
            'title': attr['entry_title'],
 | 
			
		||||
            'description': attr['entry_descl'],
 | 
			
		||||
            'thumbnail': attr['entry_image_16_9'],
 | 
			
		||||
            'duration': parse_duration(attr['entry_durat']),
 | 
			
		||||
            'upload_date': attr['entry_pdatet'][:-4],
 | 
			
		||||
            'uploader': attr['channel_title'],
 | 
			
		||||
            'uploader_id': attr['channel_idkey'],
 | 
			
		||||
            'formats': formats,
 | 
			
		||||
            'description': self._html_search_meta('description', webpage, 'description'),
 | 
			
		||||
        }
 | 
			
		||||
        }
 | 
			
		||||
		Reference in New Issue
	
	Block a user