[go90] Add new extractor(closes #10127)
authorRemita Amine <remitamine@gmail.com>
Thu, 13 Apr 2017 11:27:28 +0000 (12:27 +0100)
committerRemita Amine <remitamine@gmail.com>
Thu, 13 Apr 2017 11:28:22 +0000 (12:28 +0100)
youtube_dl/extractor/extractors.py
youtube_dl/extractor/go90.py [new file with mode: 0644]

index 45591f2a7586b015fff0c40dd9f0697cdc8d9da2..1671090f47caab4f8a02b05166645942aabafd1d 100644 (file)
@@ -384,6 +384,7 @@ from .globo import (
     GloboArticleIE,
 )
 from .go import GoIE
+from .go90 import Go90IE
 from .godtube import GodTubeIE
 from .godtv import GodTVIE
 from .golem import GolemIE
diff --git a/youtube_dl/extractor/go90.py b/youtube_dl/extractor/go90.py
new file mode 100644 (file)
index 0000000..3550eca
--- /dev/null
@@ -0,0 +1,92 @@
+# coding: utf-8
+from __future__ import unicode_literals
+
+import re
+
+from .common import InfoExtractor
+from ..utils import (
+    int_or_none,
+    parse_iso8601,
+)
+
+
+class Go90IE(InfoExtractor):
+    _VALID_URL = r'https?://(?:www\.)?go90\.com/videos/(?P<id>[0-9a-zA-Z]+)'
+    _TEST = {
+        'url': 'https://www.go90.com/videos/84BUqjLpf9D',
+        'md5': 'efa7670dbbbf21a7b07b360652b24a32',
+        'info_dict': {
+            'id': '84BUqjLpf9D',
+            'ext': 'mp4',
+            'title': 'Inside The Utah Coalition Against Pornography Convention',
+            'description': 'VICE\'s Karley Sciortino meets with activists who discuss the state\'s strong anti-porn stance. Then, VICE Sports explains NFL contracts.',
+            'timestamp': 1491868800,
+            'upload_date': '20170411',
+        }
+    }
+
+    def _real_extract(self, url):
+        video_id = self._match_id(url)
+        video_data = self._download_json(
+            'https://www.go90.com/api/view/items/' + video_id,
+            video_id, headers={
+                'Content-Type': 'application/json; charset=utf-8',
+            }, data=b'{"client":"web","device_type":"pc"}')
+        title = video_data['title']
+        main_video_asset = video_data['main_video_asset']
+
+        thumbnails = []
+        formats = []
+        for asset in video_data.get('assets'):
+            if asset.get('id') == main_video_asset:
+                for source in asset.get('sources', []):
+                    source_location = source.get('location')
+                    if not source_location:
+                        continue
+                    source_type = source.get('type')
+                    if source_type == 'hls':
+                        m3u8_formats = self._extract_m3u8_formats(
+                            source_location, video_id, 'mp4',
+                            'm3u8_native', m3u8_id='hls', fatal=False)
+                        for f in m3u8_formats:
+                            mobj = re.search(r'/hls-(\d+)-(\d+)K', f['url'])
+                            if mobj:
+                                height, tbr = mobj.groups()
+                                height = int_or_none(height)
+                                f.update({
+                                    'height': f.get('height') or height,
+                                    'width': f.get('width') or int_or_none(height / 9.0 * 16.0 if height else None),
+                                    'tbr': f.get('tbr') or int_or_none(tbr),
+                                })
+                        formats.extend(m3u8_formats)
+                    elif source_type == 'dash':
+                        formats.extend(self._extract_mpd_formats(
+                            source_location, video_id, mpd_id='dash', fatal=False))
+                    else:
+                        formats.append({
+                            'format_id': source.get('name'),
+                            'url': source_location,
+                            'width': int_or_none(source.get('width')),
+                            'height': int_or_none(source.get('height')),
+                            'tbr': int_or_none(source.get('bitrate')),
+                        })
+            elif asset.get('type') == 'image':
+                asset_location = asset.get('location')
+                if not asset_location:
+                    continue
+                thumbnails.append({
+                    'url': asset_location,
+                    'width': int_or_none(asset.get('width')),
+                    'height': int_or_none(asset.get('height')),
+                })
+        self._sort_formats(formats)
+
+        return {
+            'id': video_id,
+            'title': title,
+            'formats': formats,
+            'thumbnails': thumbnails,
+            'description': video_data.get('short_description'),
+            'like_count': int_or_none(video_data.get('like_count')),
+            'timestamp': parse_iso8601(video_data.get('released_at')),
+        }