1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-22 14:12:52 +08:00
youtube-dl/youtube_dl/extractor/rai.py
2015-04-10 22:44:41 +06:00

164 lines
6.4 KiB
Python

from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import (
compat_urllib_parse,
)
from ..utils import (
parse_duration,
unified_strdate,
)
class RaiIE(InfoExtractor):
_VALID_URL = r'(?P<url>(?P<host>http://(?:.+?\.)?(?:rai\.it|rai\.tv|rainews\.it))/dl/.+?-(?P<id>[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})(?:-.+?)?\.html)'
_TESTS = [
{
'url': 'http://www.rai.tv/dl/RaiTV/programmi/media/ContentItem-cb27157f-9dd0-4aee-b788-b1f67643a391.html',
'md5': 'c064c0b2d09c278fb293116ef5d0a32d',
'info_dict': {
'id': 'cb27157f-9dd0-4aee-b788-b1f67643a391',
'ext': 'mp4',
'title': 'Report del 07/04/2014',
'description': 'md5:f27c544694cacb46a078db84ec35d2d9',
'upload_date': '20140407',
'duration': 6160,
}
},
{
'url': 'http://www.raisport.rai.it/dl/raiSport/media/rassegna-stampa-04a9f4bd-b563-40cf-82a6-aad3529cb4a9.html',
'md5': '8bb9c151924ce241b74dd52ef29ceafa',
'info_dict': {
'id': '04a9f4bd-b563-40cf-82a6-aad3529cb4a9',
'ext': 'mp4',
'title': 'TG PRIMO TEMPO',
'description': '',
'upload_date': '20140612',
'duration': 1758,
},
'skip': 'Error 404',
},
{
'url': 'http://www.rainews.it/dl/rainews/media/state-of-the-net-Antonella-La-Carpia-regole-virali-7aafdea9-0e5d-49d5-88a6-7e65da67ae13.html',
'md5': '35cf7c229f22eeef43e48b5cf923bef0',
'info_dict': {
'id': '7aafdea9-0e5d-49d5-88a6-7e65da67ae13',
'ext': 'mp4',
'title': 'State of the Net, Antonella La Carpia: regole virali',
'description': 'md5:b0ba04a324126903e3da7763272ae63c',
'upload_date': '20140613',
},
'skip': 'Error 404',
},
{
'url': 'http://www.rai.tv/dl/RaiTV/programmi/media/ContentItem-b4a49761-e0cc-4b14-8736-2729f6f73132-tg2.html',
'md5': '35694f062977fe6619943f08ed935730',
'info_dict': {
'id': 'b4a49761-e0cc-4b14-8736-2729f6f73132',
'ext': 'mp4',
'title': 'Alluvione in Sardegna e dissesto idrogeologico',
'description': 'Edizione delle ore 20:30 ',
}
},
{
'url': 'http://www.ilcandidato.rai.it/dl/ray/media/Il-Candidato---Primo-episodio-Le-Primarie-28e5525a-b495-45e8-a7c3-bc48ba45d2b6.html',
'md5': '02b64456f7cc09f96ff14e7dd489017e',
'info_dict': {
'id': '28e5525a-b495-45e8-a7c3-bc48ba45d2b6',
'ext': 'flv',
'title': 'Il Candidato - Primo episodio: "Le Primarie"',
'description': 'Primo appuntamento con "Il candidato" con Filippo Timi, alias Piero Zucca presidente!',
'uploader': 'RaiTre',
}
}
]
def _extract_relinker_url(self, webpage):
return self._proto_relative_url(self._search_regex(
[r'name="videourl" content="([^"]+)"', r'var\s+videoURL(?:_MP4)?\s*=\s*"([^"]+)"'],
webpage, 'relinker url', default=None))
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
host = mobj.group('host')
webpage = self._download_webpage(url, video_id)
relinker_url = self._extract_relinker_url(webpage)
if not relinker_url:
iframe_path = self._search_regex(
r'<iframe[^>]+src="/?(dl/[^"]+\?iframe\b[^"]*)"',
webpage, 'iframe')
webpage = self._download_webpage(
'%s/%s' % (host, iframe_path), video_id)
relinker_url = self._extract_relinker_url(webpage)
relinker = self._download_json(
'%s&output=47' % relinker_url, video_id)
media_url = relinker['video'][0]
ct = relinker.get('ct')
if ct == 'f4m':
formats = self._extract_f4m_formats(
media_url + '&hdcore=3.7.0&plugin=aasp-3.7.0.39.44', video_id)
else:
formats = [{
'url': media_url,
'format_id': ct,
}]
json_link = self._html_search_meta(
'jsonlink', webpage, 'JSON link', default=None)
if json_link:
media = self._download_json(
host + json_link, video_id, 'Downloading video JSON')
title = media.get('name')
description = media.get('desc')
thumbnail = media.get('image_300') or media.get('image_medium') or media.get('image')
duration = parse_duration(media.get('length'))
uploader = media.get('author')
upload_date = unified_strdate(media.get('date'))
else:
title = (self._search_regex(
r'var\s+videoTitolo\s*=\s*"(.+?)";',
webpage, 'title', default=None) or self._og_search_title(webpage)).replace('\\"', '"')
description = self._og_search_description(webpage)
thumbnail = self._og_search_thumbnail(webpage)
duration = None
uploader = self._html_search_meta('Editore', webpage, 'uploader')
upload_date = unified_strdate(self._html_search_meta(
'item-date', webpage, 'upload date', default=None))
subtitles = self.extract_subtitles(video_id, webpage)
return {
'id': video_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'uploader': uploader,
'upload_date': upload_date,
'duration': duration,
'formats': formats,
'subtitles': subtitles,
}
def _get_subtitles(self, video_id, webpage):
subtitles = {}
m = re.search(r'<meta name="closedcaption" content="(?P<captions>[^"]+)"', webpage)
if m:
captions = m.group('captions')
STL_EXT = '.stl'
SRT_EXT = '.srt'
if captions.endswith(STL_EXT):
captions = captions[:-len(STL_EXT)] + SRT_EXT
subtitles['it'] = [{
'ext': 'srt',
'url': 'http://www.rai.tv%s' % compat_urllib_parse.quote(captions),
}]
return subtitles