71 lines
		
	
	
		
			2.2 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
		
		
			
		
	
	
			71 lines
		
	
	
		
			2.2 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
|  | from __future__ import unicode_literals | ||
|  | 
 | ||
|  | import json | ||
|  | 
 | ||
|  | from .common import InfoExtractor | ||
|  | from ..utils import ( | ||
|  |     parse_iso8601, | ||
|  |     compat_urllib_parse, | ||
|  |     parse_age_limit, | ||
|  |     int_or_none, | ||
|  | ) | ||
|  | 
 | ||
|  | 
 | ||
|  | class OpenFilmIE(InfoExtractor): | ||
|  |     _VALID_URL = r'http://(?:www\.)openfilm\.com/videos/(?P<id>.+)' | ||
|  |     _TEST = { | ||
|  |         'url': 'http://www.openfilm.com/videos/human-resources-remastered', | ||
|  |         'md5': '42bcd88c2f3ec13b65edf0f8ad1cac37', | ||
|  |         'info_dict': { | ||
|  |             'id': '32736', | ||
|  |             'display_id': 'human-resources-remastered', | ||
|  |             'ext': 'mp4', | ||
|  |             'title': 'Human Resources (Remastered)', | ||
|  |             'description': 'Social Engineering in the 20th Century.', | ||
|  |             'thumbnail': 're:^https?://.*\.jpg$', | ||
|  |             'duration': 7164, | ||
|  |             'timestamp': 1334756988, | ||
|  |             'upload_date': '20120418', | ||
|  |             'uploader_id': '41117', | ||
|  |             'view_count': int, | ||
|  |             'age_limit': 0, | ||
|  |         }, | ||
|  |     } | ||
|  | 
 | ||
|  |     def _real_extract(self, url): | ||
|  |         display_id = self._match_id(url) | ||
|  | 
 | ||
|  |         webpage = self._download_webpage(url, display_id) | ||
|  | 
 | ||
|  |         player = compat_urllib_parse.unquote_plus( | ||
|  |             self._og_search_video_url(webpage)) | ||
|  | 
 | ||
|  |         video = json.loads(self._search_regex( | ||
|  |             r'\bp=({.+?})(?:&|$)', player, 'video JSON')) | ||
|  | 
 | ||
|  |         video_url = '%s1.mp4' % video['location'] | ||
|  |         video_id = video.get('video_id') | ||
|  |         display_id = video.get('alias') or display_id | ||
|  |         title = video.get('title') | ||
|  |         description = video.get('description') | ||
|  |         thumbnail = video.get('main_thumb') | ||
|  |         duration = int_or_none(video.get('duration')) | ||
|  |         timestamp = parse_iso8601(video.get('dt_published'), ' ') | ||
|  |         uploader_id = video.get('user_id') | ||
|  |         view_count = int_or_none(video.get('views_count')) | ||
|  |         age_limit = parse_age_limit(video.get('age_limit')) | ||
|  | 
 | ||
|  |         return { | ||
|  |             'id': video_id, | ||
|  |             'display_id': display_id, | ||
|  |             'url': video_url, | ||
|  |             'title': title, | ||
|  |             'description': description, | ||
|  |             'thumbnail': thumbnail, | ||
|  |             'duration': duration, | ||
|  |             'timestamp': timestamp, | ||
|  |             'uploader_id': uploader_id, | ||
|  |             'view_count': view_count, | ||
|  |             'age_limit': age_limit, | ||
|  |         } |