Merge branch 'eroprofile' of https://github.com/peugeot/youtube-dl into peugeot-eroprofile
This commit is contained in:
		
						commit
						c2500434c3
					
				| @ -108,6 +108,7 @@ from .elpais import ElPaisIE | |||||||
| from .empflix import EMPFlixIE | from .empflix import EMPFlixIE | ||||||
| from .engadget import EngadgetIE | from .engadget import EngadgetIE | ||||||
| from .eporner import EpornerIE | from .eporner import EpornerIE | ||||||
|  | from .eroprofile import EroProfileIE | ||||||
| from .escapist import EscapistIE | from .escapist import EscapistIE | ||||||
| from .everyonesmixtape import EveryonesMixtapeIE | from .everyonesmixtape import EveryonesMixtapeIE | ||||||
| from .exfm import ExfmIE | from .exfm import ExfmIE | ||||||
|  | |||||||
							
								
								
									
										53
									
								
								youtube_dl/extractor/eroprofile.py
									
									
									
									
									
										Normal file
									
								
							
							
						
						
									
										53
									
								
								youtube_dl/extractor/eroprofile.py
									
									
									
									
									
										Normal file
									
								
							| @ -0,0 +1,53 @@ | |||||||
|  | from __future__ import unicode_literals | ||||||
|  | 
 | ||||||
|  | import re | ||||||
|  | 
 | ||||||
|  | from .common import InfoExtractor | ||||||
|  | 
 | ||||||
|  | class EroProfileIE(InfoExtractor): | ||||||
|  |     _VALID_URL = r'https?://(?:www\.)?eroprofile\.com/m/videos/view/(?P<display_id>[^/]+)' | ||||||
|  |     _TEST = { | ||||||
|  |         'url': 'http://www.eroprofile.com/m/videos/view/sexy-babe-softcore', | ||||||
|  |         'md5': 'c26f351332edf23e1ea28ce9ec9de32f', | ||||||
|  |         'info_dict': { | ||||||
|  |             'id': '3733775', | ||||||
|  |             'ext': 'm4v', | ||||||
|  |             'title': 'HD MOVIES - sexy babe softcore', | ||||||
|  |             'description': 'md5:831ee50526c333eb4f6c1e58d382c295', | ||||||
|  |             'categories': list,  # NSFW | ||||||
|  |             'thumbnail': 're:https?://.*\.jpg', | ||||||
|  |             'age_limit': 18, | ||||||
|  |         } | ||||||
|  |     } | ||||||
|  | 
 | ||||||
|  |     def _real_extract(self, url): | ||||||
|  |         webpage = self._download_webpage(url, 'main') | ||||||
|  | 
 | ||||||
|  |         video_id = self._html_search_regex(r'glbUpdViews\s*\(\'\d*\',\'(\d+)\'', webpage, 'id') | ||||||
|  | 
 | ||||||
|  |         video_url = self._html_search_regex(r'<source src="([^"]+)', webpage, 'video_url') | ||||||
|  | 
 | ||||||
|  |         title = self._html_search_regex( | ||||||
|  |             r'<title>([^<]+)\s*-\s*EroProfile</title>', webpage, 'title') | ||||||
|  | 
 | ||||||
|  |         description = self._html_search_meta('description', webpage, 'description', fatal=False) | ||||||
|  | 
 | ||||||
|  |         thumbnail = self._html_search_regex( | ||||||
|  |           r'onclick="showVideoPlayer\(\)"><img src="([^"]+)', webpage, 'thumbnail', fatal=False) | ||||||
|  | 
 | ||||||
|  |         categories_str = self._html_search_meta( | ||||||
|  |             'keywords', webpage, 'categories', fatal=False) | ||||||
|  |         categories = ( | ||||||
|  |             None if categories_str is None | ||||||
|  |             else categories_str.split(',')) | ||||||
|  | 
 | ||||||
|  |         return { | ||||||
|  |             'id': video_id, | ||||||
|  |             'url': video_url, | ||||||
|  |             'title': title, | ||||||
|  |             'ext': 'm4v', | ||||||
|  |             'description': description, | ||||||
|  |             'thumbnail': thumbnail, | ||||||
|  |             'categories': categories, | ||||||
|  |             'age_limit': 18, | ||||||
|  |         } | ||||||
		Loading…
	
	
			
			x
			
			
		
	
		Reference in New Issue
	
	Block a user