1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-12-27 14:03:02 +08:00
youtube-dl/youtube_dl/extractor/srgssr.py
2016-02-02 23:09:50 +01:00

156 lines
6.3 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
parse_iso8601,
qualities,
)
class SRGSSRIE(InfoExtractor):
_VALID_URL = r'(?:https?://tp\.srgssr\.ch/p(?:/[^/]+)+\?urn=urn|srgssr):(?P<bu>srf|rts|rsi|rtr|swi):(?:[^:]+:)?(?P<type>video|audio):(?P<id>[0-9a-f\-]{36}|\d+)'
_ERRORS = {
'AGERATING12': 'To protect children under the age of 12, this video is only available between 8 p.m. and 6 a.m.',
'AGERATING18': 'To protect children under the age of 18, this video is only available between 11 p.m. and 5 a.m.',
# 'ENDDATE': 'For legal reasons, this video was only available for a specified period of time.',
'GEOBLOCK': 'For legal reasons, this video is only available in Switzerland.',
'LEGAL': 'The video cannot be transmitted for legal reasons.',
'STARTDATE': 'This video is not yet available. Please try again later.',
}
def get_media_data(self, bu, media_type, media_id):
media_data = self._download_json(
'http://il.srgssr.ch/integrationlayer/1.0/ue/%s/%s/play/%s.json' % (bu, media_type, media_id),
media_id)[media_type.capitalize()]
if media_data.get('block') and media_data['block'] in self._ERRORS:
raise ExtractorError('%s said: %s' % (
self.IE_NAME, self._ERRORS[media_data['block']]), expected=True)
return media_data
def _real_extract(self, url):
bu, media_type, media_id = re.match(self._VALID_URL, url).groups()
if bu == 'rts':
return self.url_result('rts:%s' % media_id, 'RTS')
media_data = self.get_media_data(bu, media_type, media_id)
metadata = media_data['AssetMetadatas']['AssetMetadata'][0]
title = metadata['title']
description = metadata.get('description')
created_date = media_data.get('createdDate') or metadata.get('createdDate')
timestamp = parse_iso8601(created_date)
thumbnails = [{
'id': image.get('id'),
'url': image['url'],
} for image in media_data.get('Image', {}).get('ImageRepresentations', {}).get('ImageRepresentation', [])]
preference = qualities(['LQ', 'MQ', 'SD', 'HQ', 'HD'])
formats = []
for source in media_data.get('Playlists', {}).get('Playlist', []) + media_data.get('Downloads', {}).get('Download', []):
protocol = source.get('@protocol')
for asset in source['url']:
asset_url = asset['text']
quality = asset['@quality']
format_id = '%s-%s' % (protocol, quality)
if protocol == 'HTTP-HDS':
formats.extend(self._extract_f4m_formats(
asset_url + '?hdcore=3.4.0', media_id,
f4m_id=format_id, fatal=False))
elif protocol == 'HTTP-HLS':
formats.extend(self._extract_m3u8_formats(
asset_url, media_id, 'mp4', 'm3u8_native',
m3u8_id=format_id, fatal=False))
else:
formats.append({
'format_id': format_id,
'url': asset_url,
'preference': preference(quality),
'ext': 'flv' if protocol == 'RTMP' else None,
})
self._sort_formats(formats)
return {
'id': media_id,
'title': title,
'description': description,
'timestamp': timestamp,
'thumbnails': thumbnails,
'formats': formats,
}
class SRGSSRPlayIE(InfoExtractor):
IE_DESC = 'srf.ch, rts.ch, rsi.ch, rtr.ch and swissinfo.ch play sites'
_VALID_URL = r'https?://(?:(?:www|play)\.)?(?P<bu>srf|rts|rsi|rtr|swissinfo)\.ch/play/(?:tv|radio)/[^/]+/(?P<type>video|audio)/[^?]+\?id=(?P<id>[0-9a-f\-]{36}|\d+)'
_TESTS = [{
'url': 'http://www.srf.ch/play/tv/10vor10/video/snowden-beantragt-asyl-in-russland?id=28e1a57d-5b76-4399-8ab3-9097f071e6c5',
'md5': '4cd93523723beff51bb4bee974ee238d',
'info_dict': {
'id': '28e1a57d-5b76-4399-8ab3-9097f071e6c5',
'ext': 'm4v',
'upload_date': '20130701',
'title': 'Snowden beantragt Asyl in Russland',
'timestamp': 1372713995,
}
}, {
# No Speichern (Save) button
'url': 'http://www.srf.ch/play/tv/top-gear/video/jaguar-xk120-shadow-und-tornado-dampflokomotive?id=677f5829-e473-4823-ac83-a1087fe97faa',
'md5': '0a274ce38fda48c53c01890651985bc6',
'info_dict': {
'id': '677f5829-e473-4823-ac83-a1087fe97faa',
'ext': 'flv',
'upload_date': '20130710',
'title': 'Jaguar XK120, Shadow und Tornado-Dampflokomotive',
'description': 'md5:88604432b60d5a38787f152dec89cd56',
'timestamp': 1373493600,
},
}, {
'url': 'http://www.rtr.ch/play/radio/actualitad/audio/saira-tujetsch-tuttina-cuntinuar-cun-sedrun-muster-turissem?id=63cb0778-27f8-49af-9284-8c7a8c6d15fc',
'info_dict': {
'id': '63cb0778-27f8-49af-9284-8c7a8c6d15fc',
'ext': 'mp3',
'upload_date': '20151013',
'title': 'Saira: Tujetsch - tuttina cuntinuar cun Sedrun Mustér Turissem',
'timestamp': 1444750398,
},
'params': {
# rtmp download
'skip_download': True,
},
}, {
'url': 'http://www.rts.ch/play/tv/-/video/le-19h30?id=6348260',
'md5': '67a2a9ae4e8e62a68d0e9820cc9782df',
'info_dict': {
'id': '6348260',
'display_id': '6348260',
'ext': 'mp4',
'duration': 1796,
'title': 'Le 19h30',
'description': '',
'uploader': '19h30',
'upload_date': '20141201',
'timestamp': 1417458600,
'thumbnail': 're:^https?://.*\.image',
'view_count': int,
},
'params': {
# m3u8 download
'skip_download': True,
}
}]
def _real_extract(self, url):
bu, media_type, media_id = re.match(self._VALID_URL, url).groups()
# other info can be extracted from url + '&layout=json'
return self.url_result('srgssr:%s:%s:%s' % (bu[:3], media_type, media_id), 'SRGSSR')