]> git.bitcoin.ninja Git - youtube-dl/commitdiff
[mgoon] Add new extractor
authorNaglis Jonaitis <njonaitis@gmail.com>
Sun, 21 Sep 2014 00:26:12 +0000 (03:26 +0300)
committerNaglis Jonaitis <njonaitis@gmail.com>
Sun, 21 Sep 2014 00:26:12 +0000 (03:26 +0300)
youtube_dl/extractor/__init__.py
youtube_dl/extractor/mgoon.py [new file with mode: 0644]

index 625666acbe5c98edeab7278cb4ac31de217a8dfb..fb546eeae2534a53fb50422f05ed1237c6f32e98 100644 (file)
@@ -199,6 +199,7 @@ from .malemotion import MalemotionIE
 from .mdr import MDRIE
 from .metacafe import MetacafeIE
 from .metacritic import MetacriticIE
+from .mgoon import MgoonIE
 from .ministrygrid import MinistryGridIE
 from .mit import TechTVMITIE, MITIE, OCWMITIE
 from .mitele import MiTeleIE
diff --git a/youtube_dl/extractor/mgoon.py b/youtube_dl/extractor/mgoon.py
new file mode 100644 (file)
index 0000000..94bc87b
--- /dev/null
@@ -0,0 +1,87 @@
+# coding: utf-8
+from __future__ import unicode_literals
+
+import re
+
+from .common import InfoExtractor
+from ..utils import (
+    ExtractorError,
+    qualities,
+    unified_strdate,
+)
+
+
+class MgoonIE(InfoExtractor):
+    _VALID_URL = r'''(?x)https?://(?:www\.)?
+    (?:(:?m\.)?mgoon\.com/(?:ch/(?:.+)/v|play/view)|
+        video\.mgoon\.com)/(?P<id>[0-9]+)'''
+    _API_URL = 'http://mpos.mgoon.com/player/video?id={0:}'
+    _TESTS = [
+        {
+            'url': 'http://m.mgoon.com/ch/hi6618/v/5582148',
+            'md5': 'dd46bb66ab35cf6d51cc812fd82da79d',
+            'info_dict': {
+                'id': '5582148',
+                'uploader_id': 'hi6618',
+                'duration': 240.419,
+                'upload_date': '20131220',
+                'ext': 'mp4',
+                'title': 'md5:543aa4c27a4931d371c3f433e8cebebc',
+                'thumbnail': 're:^https?://.*\.jpg$',
+            }
+        },
+        {
+            'url': 'http://www.mgoon.com/play/view/5582148',
+            'only_matching': True,
+        },
+        {
+            'url': 'http://video.mgoon.com/5582148',
+            'only_matching': True,
+        },
+    ]
+
+    def _real_extract(self, url):
+        mobj = re.match(self._VALID_URL, url)
+        video_id = mobj.group('id')
+
+        data = self._download_json(self._API_URL.format(video_id), video_id)
+
+        if data.get('errorInfo', {}).get('code') != 'NONE':
+            raise ExtractorError('%s encountered an error: %s' % (
+                self.IE_NAME, data['errorInfo']['message']), expected=True)
+
+        v_info = data['videoInfo']
+        title = v_info.get('v_title')
+        thumbnail = v_info.get('v_thumbnail')
+        duration = v_info.get('v_duration')
+        upload_date = unified_strdate(v_info.get('v_reg_date'))
+        uploader_id = data.get('userInfo', {}).get('u_alias')
+        if duration:
+            duration /= 1000.0
+
+        age_limit = None
+        if data.get('accessInfo', {}).get('code') == 'VIDEO_STATUS_ADULT':
+            age_limit = 18
+
+        formats = []
+        get_quality = qualities(['360p', '480p', '720p', '1080p'])
+        for fmt in data['videoFiles']:
+            formats.append({
+                'format_id': fmt['label'],
+                'quality': get_quality(fmt['label']),
+                'url': fmt['url'],
+                'ext': fmt['format'],
+
+            })
+        self._sort_formats(formats)
+
+        return {
+            'id': video_id,
+            'title': title,
+            'formats': formats,
+            'thumbnail': thumbnail,
+            'duration': duration,
+            'upload_date': upload_date,
+            'uploader_id': uploader_id,
+            'age_limit': age_limit,
+        }