]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/redtube.py
Merge tag 'upstream/2017.03.07'
[youtubedl] / youtube_dl / extractor / redtube.py
index c2254ae8abdca2ab9dde2388fb2182b056ffd0e2..c367a6ae74f3a7b63dd50f035a2d380f76dc3719 100644 (file)
@@ -1,46 +1,95 @@
+from __future__ import unicode_literals
+
 import re
 
 from .common import InfoExtractor
+from ..utils import (
+    ExtractorError,
+    int_or_none,
+    str_to_int,
+    unified_strdate,
+)
 
 
 class RedTubeIE(InfoExtractor):
-    _VALID_URL = r'(?:http://)?(?:www\.)?redtube\.com/(?P<id>[0-9]+)'
-    _TEST = {
-        u'url': u'http://www.redtube.com/66418',
-        u'file': u'66418.mp4',
-        # md5 varies from time to time, as in
-        # https://travis-ci.org/rg3/youtube-dl/jobs/14052463#L295
-        #u'md5': u'7b8c22b5e7098a3e1c09709df1126d2d',
-        u'info_dict': {
-            u"title": u"Sucked on a toilet",
-            u"age_limit": 18,
+    _VALID_URL = r'https?://(?:(?:www\.)?redtube\.com/|embed\.redtube\.com/\?.*?\bid=)(?P<id>[0-9]+)'
+    _TESTS = [{
+        'url': 'http://www.redtube.com/66418',
+        'md5': '7b8c22b5e7098a3e1c09709df1126d2d',
+        'info_dict': {
+            'id': '66418',
+            'ext': 'mp4',
+            'title': 'Sucked on a toilet',
+            'upload_date': '20120831',
+            'duration': 596,
+            'view_count': int,
+            'age_limit': 18,
         }
-    }
+    }, {
+        'url': 'http://embed.redtube.com/?bgcolor=000000&id=1443286',
+        'only_matching': True,
+    }]
+
+    @staticmethod
+    def _extract_urls(webpage):
+        return re.findall(
+            r'<iframe[^>]+?src=["\'](?P<url>(?:https?:)?//embed\.redtube\.com/\?.*?\bid=\d+)',
+            webpage)
 
     def _real_extract(self, url):
-        mobj = re.match(self._VALID_URL, url)
+        video_id = self._match_id(url)
+        webpage = self._download_webpage(
+            'http://www.redtube.com/%s' % video_id, video_id)
 
-        video_id = mobj.group('id')
-        video_extension = 'mp4'
-        webpage = self._download_webpage(url, video_id)
+        if any(s in webpage for s in ['video-deleted-info', '>This video has been removed']):
+            raise ExtractorError('Video %s has been removed' % video_id, expected=True)
 
-        self.report_extraction(video_id)
+        title = self._html_search_regex(
+            (r'<h1 class="videoTitle[^"]*">(?P<title>.+?)</h1>',
+             r'videoTitle\s*:\s*(["\'])(?P<title>)\1'),
+            webpage, 'title', group='title')
 
-        video_url = self._html_search_regex(
-            r'<source src="(.+?)" type="video/mp4">', webpage, u'video URL')
+        formats = []
+        sources = self._parse_json(
+            self._search_regex(
+                r'sources\s*:\s*({.+?})', webpage, 'source', default='{}'),
+            video_id, fatal=False)
+        if sources and isinstance(sources, dict):
+            for format_id, format_url in sources.items():
+                if format_url:
+                    formats.append({
+                        'url': format_url,
+                        'format_id': format_id,
+                        'height': int_or_none(format_id),
+                    })
+        else:
+            video_url = self._html_search_regex(
+                r'<source src="(.+?)" type="video/mp4">', webpage, 'video URL')
+            formats.append({'url': video_url})
+        self._sort_formats(formats)
 
-        video_title = self._html_search_regex(
-            r'<h1 class="videoTitle[^"]*">(.+?)</h1>',
-            webpage, u'title')
+        thumbnail = self._og_search_thumbnail(webpage)
+        upload_date = unified_strdate(self._search_regex(
+            r'<span[^>]+class="added-time"[^>]*>ADDED ([^<]+)<',
+            webpage, 'upload date', fatal=False))
+        duration = int_or_none(self._search_regex(
+            r'videoDuration\s*:\s*(\d+)', webpage, 'duration', fatal=False))
+        view_count = str_to_int(self._search_regex(
+            r'<span[^>]*>VIEWS</span></td>\s*<td>([\d,.]+)',
+            webpage, 'view count', fatal=False))
 
         # No self-labeling, but they describe themselves as
         # "Home of Videos Porno"
         age_limit = 18
 
         return {
-            'id':        video_id,
-            'url':       video_url,
-            'ext':       video_extension,
-            'title':     video_title,
+            'id': video_id,
+            'ext': 'mp4',
+            'title': title,
+            'thumbnail': thumbnail,
+            'upload_date': upload_date,
+            'duration': duration,
+            'view_count': view_count,
             'age_limit': age_limit,
+            'formats': formats,
         }