You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
youtube-dl/youtube_dl/extractor/rtlnl.py

134 lines
5.2 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
int_or_none,
parse_duration,
)
class RtlNlIE(InfoExtractor):
IE_NAME = 'rtl.nl'
IE_DESC = 'rtl.nl and rtlxl.nl'
_VALID_URL = r'''(?x)
https?://(?:www\.)?
(?:
rtlxl\.nl/\#!/[^/]+/|
rtl\.nl/system/videoplayer/(?:[^/]+/)+(?:video_)?embed\.html\b.+?\buuid=
)
(?P<id>[0-9a-f-]+)'''
_TESTS = [{
'url': 'http://www.rtlxl.nl/#!/rtl-nieuws-132237/6e4203a6-0a5e-3596-8424-c599a59e0677',
'md5': 'cc16baa36a6c169391f0764fa6b16654',
'info_dict': {
'id': '6e4203a6-0a5e-3596-8424-c599a59e0677',
'ext': 'mp4',
'title': 'RTL Nieuws - Laat',
'description': 'md5:6b61f66510c8889923b11f2778c72dc5',
'timestamp': 1408051800,
'upload_date': '20140814',
'duration': 576.880,
},
}, {
'url': 'http://www.rtl.nl/system/videoplayer/derden/rtlnieuws/video_embed.html#uuid=84ae5571-ac25-4225-ae0c-ef8d9efb2aed/autoplay=false',
'md5': 'dea7474214af1271d91ef332fb8be7ea',
'info_dict': {
'id': '84ae5571-ac25-4225-ae0c-ef8d9efb2aed',
'ext': 'mp4',
'timestamp': 1424039400,
'title': 'RTL Nieuws - Nieuwe beelden Kopenhagen: chaos direct na aanslag',
'thumbnail': 're:^https?://screenshots\.rtl\.nl/system/thumb/sz=[0-9]+x[0-9]+/uuid=84ae5571-ac25-4225-ae0c-ef8d9efb2aed$',
'upload_date': '20150215',
'description': 'Er zijn nieuwe beelden vrijgegeven die vlak na de aanslag in Kopenhagen zijn gemaakt. Op de video is goed te zien hoe omstanders zich bekommeren om één van de slachtoffers, terwijl de eerste agenten ter plaatse komen.',
}
}, {
# empty synopsis and missing episodes (see https://github.com/rg3/youtube-dl/issues/6275)
'url': 'http://www.rtl.nl/system/videoplayer/derden/rtlnieuws/video_embed.html#uuid=f536aac0-1dc3-4314-920e-3bd1c5b3811a/autoplay=false',
'info_dict': {
'id': 'f536aac0-1dc3-4314-920e-3bd1c5b3811a',
'ext': 'mp4',
'title': 'RTL Nieuws - Meer beelden van overval juwelier',
'thumbnail': 're:^https?://screenshots\.rtl\.nl/system/thumb/sz=[0-9]+x[0-9]+/uuid=f536aac0-1dc3-4314-920e-3bd1c5b3811a$',
'timestamp': 1437233400,
'upload_date': '20150718',
'duration': 30.474,
},
'params': {
'skip_download': True,
},
}, {
# encrypted m3u8 streams, georestricted
'url': 'http://www.rtlxl.nl/#!/afl-2-257632/52a74543-c504-4cde-8aa8-ec66fe8d68a7',
'only_matching': True,
}, {
'url': 'http://www.rtl.nl/system/videoplayer/derden/embed.html#!/uuid=bb0353b0-d6a4-1dad-90e9-18fe75b8d1f0',
'only_matching': True,
}]
def _real_extract(self, url):
uuid = self._match_id(url)
info = self._download_json(
'http://www.rtl.nl/system/s4m/vfd/version=2/uuid=%s/fmt=adaptive/' % uuid,
uuid)
material = info['material'][0]
title = info['abstracts'][0]['name']
subtitle = material.get('title')
if subtitle:
title += ' - %s' % subtitle
description = material.get('synopsis')
9 years ago
meta = info.get('meta', {})
# Use unencrypted m3u8 streams (See https://github.com/rg3/youtube-dl/issues/4118)
# NB: nowadays, recent ffmpeg and avconv can handle these encrypted streams, so
# this adaptive -> flash workaround is not required in general, but it also
# allows bypassing georestriction therefore is retained for now.
videopath = material['videopath']
9 years ago
m3u8_url = meta.get('videohost', 'http://manifest.us.rtl.nl') + videopath
formats = self._extract_m3u8_formats(m3u8_url, uuid, ext='mp4')
video_urlpart = videopath.split('/adaptive/')[1][:-5]
PG_URL_TEMPLATE = 'http://pg.us.rtl.nl/rtlxl/network/%s/progressive/%s.mp4'
formats.extend([
{
'url': PG_URL_TEMPLATE % ('a2m', video_urlpart),
'format_id': 'pg-sd',
},
{
'url': PG_URL_TEMPLATE % ('a3m', video_urlpart),
'format_id': 'pg-hd',
'quality': 0,
}
])
self._sort_formats(formats)
thumbnails = []
9 years ago
for p in ('poster_base_url', '"thumb_base_url"'):
if not meta.get(p):
continue
thumbnails.append({
'url': self._proto_relative_url(meta[p] + uuid),
'width': int_or_none(self._search_regex(
r'/sz=([0-9]+)', meta[p], 'thumbnail width', fatal=False)),
'height': int_or_none(self._search_regex(
r'/sz=[0-9]+x([0-9]+)',
meta[p], 'thumbnail height', fatal=False))
})
return {
'id': uuid,
'title': title,
'formats': formats,
'timestamp': material['original_date'],
'description': description,
'duration': parse_duration(material.get('duration')),
'thumbnails': thumbnails,
}