1
0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-12-23 08:26:50 +00:00

[ard] Add suppor for plain ARD downloads (Fixes #3546)

This commit is contained in:
Philipp Hagemeister 2014-08-27 02:36:57 +02:00
parent aff216edf4
commit 6d3d3fc083
2 changed files with 63 additions and 2 deletions

View File

@ -9,7 +9,7 @@ from .allocine import AllocineIE
from .aparat import AparatIE from .aparat import AparatIE
from .appletrailers import AppleTrailersIE from .appletrailers import AppleTrailersIE
from .archiveorg import ArchiveOrgIE from .archiveorg import ArchiveOrgIE
from .ard import ARDIE from .ard import ARDIE, ARDMediathekIE
from .arte import ( from .arte import (
ArteTvIE, ArteTvIE,
ArteTVPlus7IE, ArteTVPlus7IE,

View File

@ -10,10 +10,14 @@ from ..utils import (
qualities, qualities,
compat_urllib_parse_urlparse, compat_urllib_parse_urlparse,
compat_urllib_parse, compat_urllib_parse,
int_or_none,
parse_duration,
unified_strdate,
) )
class ARDIE(InfoExtractor): class ARDMediathekIE(InfoExtractor):
IE_NAME = 'ARD:mediathek'
_VALID_URL = r'^https?://(?:(?:www\.)?ardmediathek\.de|mediathek\.daserste\.de)/(?:.*/)(?P<video_id>[0-9]+|[^0-9][^/\?]+)[^/\?]*(?:\?.*)?' _VALID_URL = r'^https?://(?:(?:www\.)?ardmediathek\.de|mediathek\.daserste\.de)/(?:.*/)(?P<video_id>[0-9]+|[^0-9][^/\?]+)[^/\?]*(?:\?.*)?'
_TESTS = [{ _TESTS = [{
@ -128,3 +132,60 @@ class ARDIE(InfoExtractor):
'formats': formats, 'formats': formats,
'thumbnail': thumbnail, 'thumbnail': thumbnail,
} }
class ARDIE(InfoExtractor):
_VALID_URL = '(?P<mainurl>https?://(www\.)?daserste\.de/[^?#]+/videos/(?P<display_id>[^/?#]+)-(?P<id>[0-9]+))\.html'
_TEST = {
'url': 'http://www.daserste.de/information/reportage-dokumentation/dokus/videos/die-story-im-ersten-mission-unter-falscher-flagge-100.html',
'md5': 'd216c3a86493f9322545e045ddc3eb35',
'info_dict': {
'display_id': 'die-story-im-ersten-mission-unter-falscher-flagge',
'id': '100',
'ext': 'mp4',
'duration': 2600,
'title': 'Die Story im Ersten: Mission unter falscher Flagge',
'upload_date': '20140804',
'thumbnail': 're:^https?://.*\.jpg$',
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
display_id = mobj.group('display_id')
player_url = mobj.group('mainurl') + '~playerXml.xml'
doc = self._download_xml(player_url, display_id)
video_node = doc.find('./video')
upload_date = unified_strdate(video_node.find('./broadcastDate').text)
thumbnail = video_node.find('.//teaserImage//variant/url').text
formats = []
for a in video_node.findall('.//asset'):
f = {
'format_id': a.attrib['type'],
'width': int_or_none(a.find('./frameWidth').text),
'height': int_or_none(a.find('./frameHeight').text),
'vbr': int_or_none(a.find('./bitrateVideo').text),
'abr': int_or_none(a.find('./bitrateAudio').text),
'vcodec': a.find('./codecVideo').text,
'tbr': int_or_none(a.find('./totalBitrate').text),
}
if a.find('./serverPrefix').text:
f['url'] = a.find('./serverPrefix').text
f['playpath'] = a.find('./fileName').text
else:
f['url'] = a.find('./fileName').text
formats.append(f)
self._sort_formats(formats)
return {
'id': mobj.group('id'),
'formats': formats,
'display_id': display_id,
'title': video_node.find('./title').text,
'duration': parse_duration(video_node.find('./duration').text),
'upload_date': upload_date,
'thumbnail': thumbnail,
}