1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-24 05:25:35 +08:00
youtube-dl/youtube_dl/extractor/br.py

135 lines
5.5 KiB
Python
Raw Normal View History

# coding: utf-8
2014-02-23 03:17:26 +08:00
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
int_or_none,
2014-07-28 01:51:38 +08:00
parse_duration,
)
2014-02-23 03:17:26 +08:00
class BRIE(InfoExtractor):
IE_DESC = 'Bayerischer Rundfunk Mediathek'
_VALID_URL = r'https?://(?:www\.)?br\.de/(?:[a-z0-9\-_]+/)+(?P<id>[a-z0-9\-_]+)\.html'
_BASE_URL = 'http://www.br.de'
_TESTS = [
{
2014-06-22 23:33:27 +08:00
'url': 'http://www.br.de/mediathek/video/sendungen/heimatsound/heimatsound-festival-2014-trailer-100.html',
'md5': '93556dd2bcb2948d9259f8670c516d59',
'info_dict': {
2014-06-22 23:33:27 +08:00
'id': '25e279aa-1ffd-40fd-9955-5325bd48a53a',
'ext': 'mp4',
2014-07-28 01:45:46 +08:00
'title': 'Wenn das Traditions-Theater wackelt',
'description': 'Heimatsound-Festival 2014: Wenn das Traditions-Theater wackelt',
2014-07-28 01:51:38 +08:00
'duration': 34,
2014-09-29 11:40:20 +08:00
'uploader': 'BR',
'upload_date': '20140802',
}
},
{
'url': 'http://www.br.de/nachrichten/schaeuble-haushaltsentwurf-bundestag-100.html',
'md5': '3db0df1a9a9cd9fa0c70e6ea8aa8e820',
'info_dict': {
'id': 'c6aae3de-2cf9-43f2-957f-f17fef9afaab',
'ext': 'aac',
'title': '"Keine neuen Schulden im nächsten Jahr"',
'description': 'Haushaltsentwurf: "Keine neuen Schulden im nächsten Jahr"',
2014-07-28 01:51:38 +08:00
'duration': 64,
}
},
{
'url': 'http://www.br.de/radio/bayern1/service/team/videos/team-video-erdelt100.html',
'md5': 'dbab0aef2e047060ea7a21fc1ce1078a',
'info_dict': {
'id': '6ba73750-d405-45d3-861d-1ce8c524e059',
'ext': 'mp4',
'title': 'Umweltbewusster Häuslebauer',
'description': 'Uwe Erdelt: Umweltbewusster Häuslebauer',
2014-07-28 01:51:38 +08:00
'duration': 116,
}
},
{
'url': 'http://www.br.de/fernsehen/br-alpha/sendungen/kant-fuer-anfaenger/kritik-der-reinen-vernunft/kant-kritik-01-metaphysik100.html',
'md5': '23bca295f1650d698f94fc570977dae3',
'info_dict': {
'id': 'd982c9ce-8648-4753-b358-98abb8aec43d',
'ext': 'mp4',
'title': 'Folge 1 - Metaphysik',
'description': 'Kant für Anfänger: Folge 1 - Metaphysik',
2014-07-28 01:51:38 +08:00
'duration': 893,
'uploader': 'Eva Maria Steimle',
'upload_date': '20140117',
}
},
]
def _real_extract(self, url):
2014-09-29 11:40:20 +08:00
display_id = self._match_id(url)
2014-02-23 03:17:26 +08:00
page = self._download_webpage(url, display_id)
xml_url = self._search_regex(
r"return BRavFramework\.register\(BRavFramework\('avPlayer_(?:[a-f0-9-]{36})'\)\.setup\({dataURL:'(/(?:[a-z0-9\-]+/)+[a-z0-9/~_.-]+)'}\)\);", page, 'XMLURL')
xml = self._download_xml(self._BASE_URL + xml_url, None)
medias = []
for xml_media in xml.findall('video') + xml.findall('audio'):
media = {
'id': xml_media.get('externalId'),
'title': xml_media.find('title').text,
2014-07-28 01:51:38 +08:00
'duration': parse_duration(xml_media.find('duration').text),
'formats': self._extract_formats(xml_media.find('assets')),
'thumbnails': self._extract_thumbnails(xml_media.find('teaserImage/variants')),
'description': ' '.join(xml_media.find('shareTitle').text.splitlines()),
'webpage_url': xml_media.find('permalink').text
}
if xml_media.find('author').text:
media['uploader'] = xml_media.find('author').text
if xml_media.find('broadcastDate').text:
media['upload_date'] = ''.join(reversed(xml_media.find('broadcastDate').text.split('.')))
medias.append(media)
if len(medias) > 1:
2014-02-23 03:17:26 +08:00
self._downloader.report_warning(
'found multiple medias; please '
2014-02-23 03:17:26 +08:00
'report this with the video URL to http://yt-dl.org/bug')
if not medias:
raise ExtractorError('No media entries found')
return medias[0]
def _extract_formats(self, assets):
def text_or_none(asset, tag):
elem = asset.find(tag)
return None if elem is None else elem.text
2014-02-23 03:17:26 +08:00
formats = [{
'url': text_or_none(asset, 'downloadUrl'),
'ext': text_or_none(asset, 'mediaType'),
'format_id': asset.get('type'),
'width': int_or_none(text_or_none(asset, 'frameWidth')),
'height': int_or_none(text_or_none(asset, 'frameHeight')),
'tbr': int_or_none(text_or_none(asset, 'bitrateVideo')),
'abr': int_or_none(text_or_none(asset, 'bitrateAudio')),
'vcodec': text_or_none(asset, 'codecVideo'),
'acodec': text_or_none(asset, 'codecAudio'),
'container': text_or_none(asset, 'mediaType'),
'filesize': int_or_none(text_or_none(asset, 'size')),
} for asset in assets.findall('asset')
if asset.find('downloadUrl') is not None]
2014-02-23 03:17:26 +08:00
self._sort_formats(formats)
return formats
def _extract_thumbnails(self, variants):
2014-02-23 03:17:26 +08:00
thumbnails = [{
'url': self._BASE_URL + variant.find('url').text,
'width': int_or_none(variant.find('width').text),
'height': int_or_none(variant.find('height').text),
} for variant in variants.findall('variant')]
thumbnails.sort(key=lambda x: x['width'] * x['height'], reverse=True)
return thumbnails