mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-16 17:15:59 +00:00
54 lines
2 KiB
Python
54 lines
2 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import unified_strdate
|
|
|
|
|
|
class DctpTvIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?dctp\.tv/(#/)?filme/(?P<id>.+?)/$'
|
|
_TEST = {
|
|
'url': 'http://www.dctp.tv/filme/videoinstallation-fuer-eine-kaufhausfassade/',
|
|
'md5': '174dd4a8a6225cf5655952f969cfbe24',
|
|
'info_dict': {
|
|
'id': '95eaa4f33dad413aa17b4ee613cccc6c',
|
|
'display_id': 'videoinstallation-fuer-eine-kaufhausfassade',
|
|
'ext': 'mp4',
|
|
'title': 'Videoinstallation für eine Kaufhausfassade',
|
|
'description': 'Kurzfilm',
|
|
'upload_date': '20110407',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
object_id = self._html_search_meta('DC.identifier', webpage)
|
|
|
|
servers_json = self._download_json(
|
|
'http://www.dctp.tv/elastic_streaming_client/get_streaming_server/',
|
|
video_id, note='Downloading server list')
|
|
server = servers_json[0]['server']
|
|
m3u8_path = self._search_regex(
|
|
r'\'([^\'"]+/playlist\.m3u8)"', webpage, 'm3u8 path')
|
|
formats = self._extract_m3u8_formats(
|
|
'http://%s%s' % (server, m3u8_path), video_id, ext='mp4',
|
|
entry_protocol='m3u8_native')
|
|
|
|
title = self._og_search_title(webpage)
|
|
description = self._html_search_meta('DC.description', webpage)
|
|
upload_date = unified_strdate(
|
|
self._html_search_meta('DC.date.created', webpage))
|
|
thumbnail = self._og_search_thumbnail(webpage)
|
|
|
|
return {
|
|
'id': object_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'display_id': video_id,
|
|
'description': description,
|
|
'upload_date': upload_date,
|
|
'thumbnail': thumbnail,
|
|
}
|