]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/extractor/rtvnh.py
Imported Upstream version 2015.11.10
[youtubedl] / youtube_dl / extractor / rtvnh.py
diff --git a/youtube_dl/extractor/rtvnh.py b/youtube_dl/extractor/rtvnh.py
new file mode 100644 (file)
index 0000000..7c9d4b0
--- /dev/null
@@ -0,0 +1,47 @@
+# coding: utf-8
+from __future__ import unicode_literals
+
+from .common import InfoExtractor
+from ..utils import ExtractorError
+
+
+class RTVNHIE(InfoExtractor):
+    _VALID_URL = r'https?://(?:www\.)?rtvnh\.nl/video/(?P<id>[0-9]+)'
+    _TEST = {
+        'url': 'http://www.rtvnh.nl/video/131946',
+        'md5': '6e1d0ab079e2a00b6161442d3ceacfc1',
+        'info_dict': {
+            'id': '131946',
+            'ext': 'mp4',
+            'title': 'Grote zoektocht in zee bij Zandvoort naar vermiste vrouw',
+            'thumbnail': 're:^https?:.*\.jpg$'
+        }
+    }
+
+    def _real_extract(self, url):
+        video_id = self._match_id(url)
+
+        meta = self._parse_json(self._download_webpage(
+            'http://www.rtvnh.nl/video/json?m=' + video_id, video_id), video_id)
+
+        status = meta.get('status')
+        if status != 200:
+            raise ExtractorError(
+                '%s returned error code %d' % (self.IE_NAME, status), expected=True)
+
+        formats = self._extract_smil_formats(
+            'http://www.rtvnh.nl/video/smil?m=' + video_id, video_id, fatal=False)
+
+        for item in meta['source']['fb']:
+            if item.get('type') == 'hls':
+                formats.extend(self._extract_m3u8_formats(
+                    item['file'], video_id, ext='mp4', entry_protocol='m3u8_native'))
+            elif item.get('type') == '':
+                formats.append({'url': item['file']})
+
+        return {
+            'id': video_id,
+            'title': meta['title'].strip(),
+            'thumbnail': meta.get('image'),
+            'formats': formats
+        }