]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/veoh.py
Prepare to release.
[youtubedl] / youtube_dl / extractor / veoh.py
index baa57f3438603e6a5d0d2b406389df4110353ebe..071774a6f79e4d2c974b2fa2abe3e2ea13477d80 100644 (file)
@@ -4,44 +4,127 @@ import re
 import json
 
 from .common import InfoExtractor
+from ..utils import (
+    int_or_none,
+    ExtractorError,
+    sanitized_Request,
+)
 
 
 class VeohIE(InfoExtractor):
-    _VALID_URL = r'http://(?:www\.)?veoh\.com/(?:watch|iphone/#_Watch)/v(?P<id>\d*)'
+    _VALID_URL = r'https?://(?:www\.)?veoh\.com/(?:watch|embed|iphone/#_Watch)/(?P<id>(?:v|e|yapi-)[\da-zA-Z]+)'
 
-    _TEST = {
+    _TESTS = [{
         'url': 'http://www.veoh.com/watch/v56314296nk7Zdmz3',
-        'file': '56314296.mp4',
         'md5': '620e68e6a3cff80086df3348426c9ca3',
         'info_dict': {
+            'id': '56314296',
+            'ext': 'mp4',
             'title': 'Straight Backs Are Stronger',
             'uploader': 'LUMOback',
             'description': 'At LUMOback, we believe straight backs are stronger.  The LUMOback Posture & Movement Sensor:  It gently vibrates when you slouch, inspiring improved posture and mobility.  Use the app to track your data and improve your posture over time. ',
+        },
+    }, {
+        'url': 'http://www.veoh.com/embed/v56314296nk7Zdmz3',
+        'only_matching': True,
+    }, {
+        'url': 'http://www.veoh.com/watch/v27701988pbTc4wzN?h1=Chile+workers+cover+up+to+avoid+skin+damage',
+        'md5': '4a6ff84b87d536a6a71e6aa6c0ad07fa',
+        'info_dict': {
+            'id': '27701988',
+            'ext': 'mp4',
+            'title': 'Chile workers cover up to avoid skin damage',
+            'description': 'md5:2bd151625a60a32822873efc246ba20d',
+            'uploader': 'afp-news',
+            'duration': 123,
+        },
+        'skip': 'This video has been deleted.',
+    }, {
+        'url': 'http://www.veoh.com/watch/v69525809F6Nc4frX',
+        'md5': '4fde7b9e33577bab2f2f8f260e30e979',
+        'note': 'Embedded ooyala video',
+        'info_dict': {
+            'id': '69525809',
+            'ext': 'mp4',
+            'title': 'Doctors Alter Plan For Preteen\'s Weight Loss Surgery',
+            'description': 'md5:f5a11c51f8fb51d2315bca0937526891',
+            'uploader': 'newsy-videos',
+        },
+        'skip': 'This video has been deleted.',
+    }, {
+        'url': 'http://www.veoh.com/watch/e152215AJxZktGS',
+        'only_matching': True,
+    }]
+
+    def _extract_formats(self, source):
+        formats = []
+        link = source.get('aowPermalink')
+        if link:
+            formats.append({
+                'url': link,
+                'ext': 'mp4',
+                'format_id': 'aow',
+            })
+        link = source.get('fullPreviewHashLowPath')
+        if link:
+            formats.append({
+                'url': link,
+                'format_id': 'low',
+            })
+        link = source.get('fullPreviewHashHighPath')
+        if link:
+            formats.append({
+                'url': link,
+                'format_id': 'high',
+            })
+        return formats
+
+    def _extract_video(self, source):
+        return {
+            'id': source.get('videoId'),
+            'title': source.get('title'),
+            'description': source.get('description'),
+            'thumbnail': source.get('highResImage') or source.get('medResImage'),
+            'uploader': source.get('username'),
+            'duration': int_or_none(source.get('length')),
+            'view_count': int_or_none(source.get('views')),
+            'age_limit': 18 if source.get('isMature') == 'true' or source.get('isSexy') == 'true' else 0,
+            'formats': self._extract_formats(source),
         }
-    }
 
     def _real_extract(self, url):
         mobj = re.match(self._VALID_URL, url)
         video_id = mobj.group('id')
+
+        if video_id.startswith('v'):
+            rsp = self._download_xml(
+                r'http://www.veoh.com/api/findByPermalink?permalink=%s' % video_id, video_id, 'Downloading video XML')
+            stat = rsp.get('stat')
+            if stat == 'ok':
+                return self._extract_video(rsp.find('./videoList/video'))
+            elif stat == 'fail':
+                raise ExtractorError(
+                    '%s said: %s' % (self.IE_NAME, rsp.find('./errorList/error').get('errorMessage')), expected=True)
+
         webpage = self._download_webpage(url, video_id)
+        age_limit = 0
+        if 'class="adultwarning-container"' in webpage:
+            self.report_age_confirmation()
+            age_limit = 18
+            request = sanitized_Request(url)
+            request.add_header('Cookie', 'confirmedAdult=true')
+            webpage = self._download_webpage(request, video_id)
 
-        m_youtube = re.search(r'http://www\.youtube\.com/v/(.*?)(\&|")', webpage)
+        m_youtube = re.search(r'http://www\.youtube\.com/v/(.*?)(\&|"|\?)', webpage)
         if m_youtube is not None:
             youtube_id = m_youtube.group(1)
             self.to_screen('%s: detected Youtube video.' % video_id)
             return self.url_result(youtube_id, 'Youtube')
 
-        self.report_extraction(video_id)
-        info = self._search_regex(r'videoDetailsJSON = \'({.*?})\';', webpage, 'info')
-        info = json.loads(info)
-        video_url = info.get('fullPreviewHashHighPath') or info.get('fullPreviewHashLowPath')
+        info = json.loads(
+            self._search_regex(r'videoDetailsJSON = \'({.*?})\';', webpage, 'info').replace('\\\'', '\''))
 
-        return {
-            'id': info['videoId'],
-            'title': info['title'],
-            'url': video_url,
-            'uploader': info['username'],
-            'thumbnail': info.get('highResImage') or info.get('medResImage'),
-            'description': info['description'],
-            'view_count': info['views'],
-        }
+        video = self._extract_video(info)
+        video['age_limit'] = age_limit
+
+        return video