]> Raphaël G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/porn91.py
Prepate to upload.
[youtubedl] / youtube_dl / extractor / porn91.py
index 72d1b2718692d902b16efa303f9a14dd7d456f1f..24c3600fe3224347a415226f0cfa73f38eaeca51 100644 (file)
@@ -1,7 +1,6 @@
-# encoding: utf-8
+# coding: utf-8
 from __future__ import unicode_literals
 
 from __future__ import unicode_literals
 
-from ..compat import compat_urllib_parse
 from .common import InfoExtractor
 from ..utils import (
     parse_duration,
 from .common import InfoExtractor
 from ..utils import (
     parse_duration,
@@ -16,20 +15,22 @@ class Porn91IE(InfoExtractor):
 
     _TEST = {
         'url': 'http://91porn.com/view_video.php?viewkey=7e42283b4f5ab36da134',
 
     _TEST = {
         'url': 'http://91porn.com/view_video.php?viewkey=7e42283b4f5ab36da134',
-        'md5': '6df8f6d028bc8b14f5dbd73af742fb20',
+        'md5': '7fcdb5349354f40d41689bd0fa8db05a',
         'info_dict': {
             'id': '7e42283b4f5ab36da134',
             'title': '18岁大一漂亮学妹,水嫩性感,再爽一次!',
             'ext': 'mp4',
             'duration': 431,
         'info_dict': {
             'id': '7e42283b4f5ab36da134',
             'title': '18岁大一漂亮学妹,水嫩性感,再爽一次!',
             'ext': 'mp4',
             'duration': 431,
+            'age_limit': 18,
         }
     }
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
         }
     }
 
     def _real_extract(self, url):
         video_id = self._match_id(url)
-        url = 'http://91porn.com/view_video.php?viewkey=%s' % video_id
         self._set_cookie('91porn.com', 'language', 'cn_CN')
         self._set_cookie('91porn.com', 'language', 'cn_CN')
-        webpage = self._download_webpage(url, video_id, 'get HTML content')
+
+        webpage = self._download_webpage(
+            'http://91porn.com/view_video.php?viewkey=%s' % video_id, video_id)
 
         if '作为游客,你每天只可观看10个视频' in webpage:
             raise ExtractorError('91 Porn says: Daily limit 10 videos exceeded', expected=True)
 
         if '作为游客,你每天只可观看10个视频' in webpage:
             raise ExtractorError('91 Porn says: Daily limit 10 videos exceeded', expected=True)
@@ -38,23 +39,7 @@ class Porn91IE(InfoExtractor):
             r'<div id="viewvideo-title">([^<]+)</div>', webpage, 'title')
         title = title.replace('\n', '')
 
             r'<div id="viewvideo-title">([^<]+)</div>', webpage, 'title')
         title = title.replace('\n', '')
 
-        # get real url
-        file_id = self._search_regex(
-            r'so.addVariable\(\'file\',\'(\d+)\'', webpage, 'file id')
-        sec_code = self._search_regex(
-            r'so.addVariable\(\'seccode\',\'([^\']+)\'', webpage, 'sec code')
-        max_vid = self._search_regex(
-            r'so.addVariable\(\'max_vid\',\'(\d+)\'', webpage, 'max vid')
-        url_params = compat_urllib_parse.urlencode({
-            'VID': file_id,
-            'mp4': '1',
-            'seccode': sec_code,
-            'max_vid': max_vid,
-        })
-        info_cn = self._download_webpage(
-            'http://91porn.com/getfile.php?' + url_params, video_id,
-            'get real video url')
-        video_url = self._search_regex(r'file=([^&]+)&', info_cn, 'url')
+        info_dict = self._parse_html5_media_entries(url, webpage, video_id)[0]
 
         duration = parse_duration(self._search_regex(
             r'时长:\s*</span>\s*(\d+:\d+)', webpage, 'duration', fatal=False))
 
         duration = parse_duration(self._search_regex(
             r'时长:\s*</span>\s*(\d+:\d+)', webpage, 'duration', fatal=False))
@@ -62,10 +47,12 @@ class Porn91IE(InfoExtractor):
         comment_count = int_or_none(self._search_regex(
             r'留言:\s*</span>\s*(\d+)', webpage, 'comment count', fatal=False))
 
         comment_count = int_or_none(self._search_regex(
             r'留言:\s*</span>\s*(\d+)', webpage, 'comment count', fatal=False))
 
-        return {
+        info_dict.update({
             'id': video_id,
             'title': title,
             'id': video_id,
             'title': title,
-            'url': video_url,
             'duration': duration,
             'comment_count': comment_count,
             'duration': duration,
             'comment_count': comment_count,
-        }
+            'age_limit': self._rta_search(webpage),
+        })
+
+        return info_dict