| 
					
				 | 
			
			
				@@ -3,7 +3,7 @@ from __future__ import unicode_literals 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				  
			 | 
		
	
		
			
				 | 
				 | 
			
			
				 import re 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				  
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-from .common import InfoExtractor 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+from .adobepass import AdobePassIE 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				 from ..compat import compat_str 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				 from ..utils import ( 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				     xpath_text, 
			 | 
		
	
	
		
			
				| 
					
				 | 
			
			
				@@ -16,11 +16,11 @@ from ..utils import ( 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				 ) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				  
			 | 
		
	
		
			
				 | 
				 | 
			
			
				  
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-class TurnerBaseIE(InfoExtractor): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+class TurnerBaseIE(AdobePassIE): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				     def _extract_timestamp(self, video_data): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				         return int_or_none(xpath_attr(video_data, 'dateCreated', 'uts')) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				  
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-    def _extract_cvp_info(self, data_src, video_id, path_data={}): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+    def _extract_cvp_info(self, data_src, video_id, path_data={}, ap_data={}): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				         video_data = self._download_xml(data_src, video_id) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				         video_id = video_data.attrib['id'] 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				         title = xpath_text(video_data, 'headline', fatal=True) 
			 | 
		
	
	
		
			
				| 
					
				 | 
			
			
				@@ -70,11 +70,14 @@ class TurnerBaseIE(InfoExtractor): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                 secure_path = self._search_regex(r'https?://[^/]+(.+/)', video_url, 'secure path') + '*' 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                 token = tokens.get(secure_path) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                 if not token: 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                    query = { 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                        'path': secure_path, 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                        'videoId': content_id, 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                    } 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                    if ap_data.get('auth_required'): 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                        query['accessToken'] = self._extract_mvpd_auth(ap_data['url'], video_id, ap_data['site_name'], ap_data['site_name']) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                     auth = self._download_xml( 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-                        secure_path_data['tokenizer_src'], video_id, query={ 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-                            'path': secure_path, 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-                            'videoId': content_id, 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				-                        }) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				+                        secure_path_data['tokenizer_src'], video_id, query=query) 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                     error_msg = xpath_text(auth, 'error/msg') 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                     if error_msg: 
			 | 
		
	
		
			
				 | 
				 | 
			
			
				                         raise ExtractorError(error_msg, expected=True) 
			 |