]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/rentv.py
debian/patches: Remove patch cherry-picked from upstream.
[youtubedl] / youtube_dl / extractor / rentv.py
index d338b3a933cf10fb50621ea4d785d4e83324466c..7c8909d95377b38f5e9bbc3a204337f1f8b540a6 100644 (file)
@@ -3,6 +3,11 @@ from __future__ import unicode_literals
 
 from .common import InfoExtractor
 from ..compat import compat_str
 
 from .common import InfoExtractor
 from ..compat import compat_str
+from ..utils import (
+    determine_ext,
+    int_or_none,
+    url_or_none,
+)
 
 
 class RENTVIE(InfoExtractor):
 
 
 class RENTVIE(InfoExtractor):
@@ -13,7 +18,9 @@ class RENTVIE(InfoExtractor):
         'info_dict': {
             'id': '118577',
             'ext': 'mp4',
         'info_dict': {
             'id': '118577',
             'ext': 'mp4',
-            'title': 'Документальный спецпроект: "Промывка мозгов. Технологии XXI века"'
+            'title': 'Документальный спецпроект: "Промывка мозгов. Технологии XXI века"',
+            'timestamp': 1472230800,
+            'upload_date': '20160826',
         }
     }, {
         'url': 'http://ren.tv/player/118577',
         }
     }, {
         'url': 'http://ren.tv/player/118577',
@@ -26,9 +33,33 @@ class RENTVIE(InfoExtractor):
     def _real_extract(self, url):
         video_id = self._match_id(url)
         webpage = self._download_webpage('http://ren.tv/player/' + video_id, video_id)
     def _real_extract(self, url):
         video_id = self._match_id(url)
         webpage = self._download_webpage('http://ren.tv/player/' + video_id, video_id)
-        jw_config = self._parse_json(self._search_regex(
-            r'config\s*=\s*({.+});', webpage, 'jw config'), video_id)
-        return self._parse_jwplayer_data(jw_config, video_id, m3u8_id='hls')
+        config = self._parse_json(self._search_regex(
+            r'config\s*=\s*({.+})\s*;', webpage, 'config'), video_id)
+        title = config['title']
+        formats = []
+        for video in config['src']:
+            src = url_or_none(video.get('src'))
+            if not src:
+                continue
+            ext = determine_ext(src)
+            if ext == 'm3u8':
+                formats.extend(self._extract_m3u8_formats(
+                    src, video_id, 'mp4', entry_protocol='m3u8_native',
+                    m3u8_id='hls', fatal=False))
+            else:
+                formats.append({
+                    'url': src,
+                })
+        self._sort_formats(formats)
+        return {
+            'id': video_id,
+            'title': title,
+            'description': config.get('description'),
+            'thumbnail': config.get('image'),
+            'duration': int_or_none(config.get('duration')),
+            'timestamp': int_or_none(config.get('date')),
+            'formats': formats,
+        }
 
 
 class RENTVArticleIE(InfoExtractor):
 
 
 class RENTVArticleIE(InfoExtractor):