X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/d4ff594119bc679aa175947eb59a97bee8f966f4..0fada1afdee4b9d13610291943d8b2d9ea7283a2:/youtube_dl/extractor/viddler.py diff --git a/youtube_dl/extractor/viddler.py b/youtube_dl/extractor/viddler.py index 8516a29..67808e7 100644 --- a/youtube_dl/extractor/viddler.py +++ b/youtube_dl/extractor/viddler.py @@ -1,12 +1,14 @@ from __future__ import unicode_literals from .common import InfoExtractor +from ..compat import ( + compat_urllib_parse_urlencode, + compat_urlparse, +) from ..utils import ( float_or_none, int_or_none, -) -from ..compat import ( - compat_urllib_request + sanitized_Request, ) @@ -14,27 +16,27 @@ class ViddlerIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?viddler\.com/(?:v|embed|player)/(?P[a-z0-9]+)' _TESTS = [{ 'url': 'http://www.viddler.com/v/43903784', - 'md5': 'ae43ad7cb59431ce043f0ff7fa13cbf4', + 'md5': '9eee21161d2c7f5b39690c3e325fab2f', 'info_dict': { 'id': '43903784', - 'ext': 'mp4', + 'ext': 'mov', 'title': 'Video Made Easy', 'description': 'md5:6a697ebd844ff3093bd2e82c37b409cd', 'uploader': 'viddler', 'timestamp': 1335371429, 'upload_date': '20120425', 'duration': 100.89, - 'thumbnail': 're:^https?://.*\.jpg$', + 'thumbnail': r're:^https?://.*\.jpg$', 'view_count': int, 'comment_count': int, 'categories': ['video content', 'high quality video', 'video made easy', 'how to produce video with limited resources', 'viddler'], } }, { 'url': 'http://www.viddler.com/v/4d03aad9/', - 'md5': 'faa71fbf70c0bee7ab93076fd007f4b0', + 'md5': 'f12c5a7fa839c47a79363bfdf69404fb', 'info_dict': { 'id': '4d03aad9', - 'ext': 'mp4', + 'ext': 'ts', 'title': 'WALL-TO-GORTAT', 'upload_date': '20150126', 'uploader': 'deadspin', @@ -44,10 +46,10 @@ class ViddlerIE(InfoExtractor): } }, { 'url': 'http://www.viddler.com/player/221ebbbd/0/', - 'md5': '0defa2bd0ea613d14a6e9bd1db6be326', + 'md5': '740511f61d3d1bb71dc14a0fe01a1c10', 'info_dict': { 'id': '221ebbbd', - 'ext': 'mp4', + 'ext': 'mov', 'title': 'LETeens-Grammar-snack-third-conditional', 'description': ' ', 'upload_date': '20140929', @@ -56,16 +58,42 @@ class ViddlerIE(InfoExtractor): 'view_count': int, 'comment_count': int, } + }, { + # secret protected + 'url': 'http://www.viddler.com/v/890c0985?secret=34051570', + 'info_dict': { + 'id': '890c0985', + 'ext': 'mp4', + 'title': 'Complete Property Training - Traineeships', + 'description': ' ', + 'upload_date': '20130606', + 'uploader': 'TiffanyBowtell', + 'timestamp': 1370496993, + 'view_count': int, + 'comment_count': int, + }, + 'params': { + 'skip_download': True, + }, }] def _real_extract(self, url): video_id = self._match_id(url) - json_url = ( - 'http://api.viddler.com/api/v2/viddler.videos.getPlaybackDetails.json?video_id=%s&key=v0vhrt7bg2xq1vyxhkct' % - video_id) + query = { + 'video_id': video_id, + 'key': 'v0vhrt7bg2xq1vyxhkct', + } + + qs = compat_urlparse.parse_qs(compat_urlparse.urlparse(url).query) + secret = qs.get('secret', [None])[0] + if secret: + query['secret'] = secret + headers = {'Referer': 'http://static.cdn-ec.viddler.com/js/arpeggio/v2/embed.html'} - request = compat_urllib_request.Request(json_url, None, headers) + request = sanitized_Request( + 'http://api.viddler.com/api/v2/viddler.videos.getPlaybackDetails.json?%s' + % compat_urllib_parse_urlencode(query), None, headers) data = self._download_json(request, video_id)['video'] formats = []