aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/odatv.py
diff options
context:
space:
mode:
authorskacurt <skacurt@outlook.com>2016-04-23 04:20:04 +0300
committerSergey M․ <dstftw@gmail.com>2016-07-20 21:42:05 +0700
commit740d7c49c251464b25bea5146e054390ada6b997 (patch)
treed36af02048aa83e777ec4ff53ae878b574d58831 /youtube_dl/extractor/odatv.py
parent4e51ec5f579465d5b64ecfef3614ac5c726fca1f (diff)
[odatv] Add extractor
Diffstat (limited to 'youtube_dl/extractor/odatv.py')
-rw-r--r--youtube_dl/extractor/odatv.py51
1 files changed, 51 insertions, 0 deletions
diff --git a/youtube_dl/extractor/odatv.py b/youtube_dl/extractor/odatv.py
new file mode 100644
index 000000000..dbf96aefe
--- /dev/null
+++ b/youtube_dl/extractor/odatv.py
@@ -0,0 +1,51 @@
+# coding: utf-8
+
+from __future__ import unicode_literals
+from .common import InfoExtractor
+from ..utils import (
+ ExtractorError,
+ remove_start
+)
+import re
+
+
+class OdaTVIE(InfoExtractor):
+ _VALID_URL = r'^https?://(?:www\.)?odatv\.com/(?:mob|vid)_video\.php\?id=(?P<id>[^&]*)'
+ _TESTS = [{
+ 'url': 'http://odatv.com/vid_video.php?id=8E388',
+ 'md5': 'dc61d052f205c9bf2da3545691485154',
+ 'info_dict': {
+ 'id': '8E388',
+ 'ext': 'mp4',
+ 'title': 'md5:69654805a16a16cf9ec9d055e079831c'
+ }
+ }, {
+ 'url': 'http://odatv.com/mob_video.php?id=8E388',
+ 'md5': 'dc61d052f205c9bf2da3545691485154',
+ 'info_dict': {
+ 'id': '8E388',
+ 'ext': 'mp4',
+ 'title': 'md5:69654805a16a16cf9ec9d055e079831c'
+ }
+ }, {
+ 'url': 'http://odatv.com/mob_video.php?id=8E900',
+ 'md5': '',
+ 'info_dict': {
+ 'id': '8E900',
+ 'ext': 'mp4',
+ 'title': 'not found check'
+ }
+ }]
+
+ def _real_extract(self, url):
+ video_id = self._match_id(url)
+ webpage = self._download_webpage(url, video_id)
+ if 'NO VIDEO!' in webpage:
+ raise ExtractorError('Video %s does not exist' % video_id, expected=True)
+
+ return {
+ 'id': video_id,
+ 'title': remove_start(self._og_search_title(webpage), 'Video: '),
+ 'thumbnail': self._og_search_thumbnail(webpage),
+ 'url': self._html_search_regex(r"(http.+?video_%s\.mp4)" % re.escape(video_id), webpage, 'url', flags=re.IGNORECASE)
+ }