diff options
| author | Yen Chi Hsuan <yan12125@gmail.com> | 2015-06-21 23:29:40 +0800 | 
|---|---|---|
| committer | Yen Chi Hsuan <yan12125@gmail.com> | 2015-06-21 23:29:40 +0800 | 
| commit | 3f3308cd75fc068e4d67d00aa7d7892e02ab16e9 (patch) | |
| tree | 26a1ec11069b9963f717ad07be91f117db99e3f4 | |
| parent | 6f96e308d0fa7674ac88e1e80fc602413f9a6b31 (diff) | |
Revert "[sohu] Update extractor"
This reverts commit 32060c6d6b618fa858b2ce43db34d02fd43bc542.
| -rw-r--r-- | youtube_dl/extractor/sohu.py | 44 | 
1 files changed, 25 insertions, 19 deletions
| diff --git a/youtube_dl/extractor/sohu.py b/youtube_dl/extractor/sohu.py index 29bd9ce6f..7644cc02d 100644 --- a/youtube_dl/extractor/sohu.py +++ b/youtube_dl/extractor/sohu.py @@ -8,7 +8,10 @@ from ..compat import (      compat_str,      compat_urllib_request  ) -from ..utils import ExtractorError +from ..utils import ( +    sanitize_url_path_consecutive_slashes, +    ExtractorError, +)  class SohuIE(InfoExtractor): @@ -26,7 +29,7 @@ class SohuIE(InfoExtractor):          'skip': 'On available in China',      }, {          'url': 'http://tv.sohu.com/20150305/n409385080.shtml', -        'md5': 'ac9a5d322b4bf9ae184d53e4711e4f1a', +        'md5': '699060e75cf58858dd47fb9c03c42cfb',          'info_dict': {              'id': '409385080',              'ext': 'mp4', @@ -34,7 +37,7 @@ class SohuIE(InfoExtractor):          }      }, {          'url': 'http://my.tv.sohu.com/us/232799889/78693464.shtml', -        'md5': '49308ff6dafde5ece51137d04aec311e', +        'md5': '9bf34be48f2f4dadcb226c74127e203c',          'info_dict': {              'id': '78693464',              'ext': 'mp4', @@ -48,7 +51,7 @@ class SohuIE(InfoExtractor):              'title': '【神探苍实战秘籍】第13期 战争之影 赫卡里姆',          },          'playlist': [{ -            'md5': '492923eac023ba2f13ff69617c32754a', +            'md5': 'bdbfb8f39924725e6589c146bc1883ad',              'info_dict': {                  'id': '78910339_part1',                  'ext': 'mp4', @@ -56,7 +59,7 @@ class SohuIE(InfoExtractor):                  'title': '【神探苍实战秘籍】第13期 战争之影 赫卡里姆',              }          }, { -            'md5': 'de604848c0e8e9c4a4dde7e1347c0637', +            'md5': '3e1f46aaeb95354fd10e7fca9fc1804e',              'info_dict': {                  'id': '78910339_part2',                  'ext': 'mp4', @@ -64,7 +67,7 @@ class SohuIE(InfoExtractor):                  'title': '【神探苍实战秘籍】第13期 战争之影 赫卡里姆',              }          }, { -            'md5': '93584716ee0657c0b205b8aa3d27aa13', +            'md5': '8407e634175fdac706766481b9443450',              'info_dict': {                  'id': '78910339_part3',                  'ext': 'mp4', @@ -139,21 +142,24 @@ class SohuIE(InfoExtractor):          for i in range(part_count):              formats = []              for format_id, format_data in formats_json.items(): +                allot = format_data['allot'] +                prot = format_data['prot'] +                  data = format_data['data'] +                clips_url = data['clipsURL'] +                su = data['su'] + +                part_str = self._download_webpage( +                    'http://%s/?prot=%s&file=%s&new=%s' % +                    (allot, prot, clips_url[i], su[i]), +                    video_id, +                    'Downloading %s video URL part %d of %d' +                    % (format_id, i + 1, part_count)) + +                part_info = part_str.split('|') -                # URLs starts with http://newflv.sohu.ccgslb.net/ is not usable -                # so retry until got a working URL -                video_url = 'newflv.sohu.ccgslb.net' -                retries = 0 -                while 'newflv.sohu.ccgslb.net' in video_url and retries < 5: -                    download_note = 'Download information from CDN gateway for format ' + format_id -                    if retries > 0: -                        download_note += ' (retry #%d)' % retries -                    retries += 1 -                    cdn_info = self._download_json( -                        'http://data.vod.itc.cn/cdnList?new=' + data['su'][i], -                        video_id, download_note) -                    video_url = cdn_info['url'] +                video_url = sanitize_url_path_consecutive_slashes( +                    '%s%s?key=%s' % (part_info[0], su[i], part_info[3]))                  formats.append({                      'url': video_url, | 
