X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/a070911bf98c751e4425008f1335cc57c86d30fc..47d80ec0b18245caeb97018d4c1af18d0b5b972b:/youtube_dl/extractor/veehd.py?ds=inline diff --git a/youtube_dl/extractor/veehd.py b/youtube_dl/extractor/veehd.py index 96353f5..0d8d832 100644 --- a/youtube_dl/extractor/veehd.py +++ b/youtube_dl/extractor/veehd.py @@ -5,6 +5,7 @@ import json from .common import InfoExtractor from ..compat import ( + compat_urllib_parse_unquote, compat_urlparse, ) from ..utils import ( @@ -17,7 +18,9 @@ from ..utils import ( class VeeHDIE(InfoExtractor): _VALID_URL = r'https?://veehd\.com/video/(?P\d+)' - _TEST = { + # Seems VeeHD videos have multiple copies on several servers, all of + # whom have different MD5 checksums, so omit md5 field in all tests + _TESTS = [{ 'url': 'http://veehd.com/video/4639434_Solar-Sinter', 'info_dict': { 'id': '4639434', @@ -26,7 +29,26 @@ class VeeHDIE(InfoExtractor): 'uploader_id': 'VideoEyes', 'description': 'md5:46a840e8692ddbaffb5f81d9885cb457', }, - } + 'skip': 'Video deleted', + }, { + 'url': 'http://veehd.com/video/4905758_Elysian-Fields-Channeling', + 'info_dict': { + 'id': '4905758', + 'ext': 'mp4', + 'title': 'Elysian Fields - Channeling', + 'description': 'md5:360e4e95fdab58aefbea0f2a19e5604b', + 'uploader_id': 'spotted', + } + }, { + 'url': 'http://veehd.com/video/2046729_2012-2009-DivX-Trailer', + 'info_dict': { + 'id': '2046729', + 'ext': 'avi', + 'title': '2012 (2009) DivX Trailer', + 'description': 'md5:75435ee95255e6a9838ac6f6f3a2396b', + 'uploader_id': 'Movie_Trailers', + } + }] def _real_extract(self, url): video_id = self._match_id(url) @@ -48,13 +70,21 @@ class VeeHDIE(InfoExtractor): player_page = self._download_webpage( player_url, video_id, 'Downloading player page') + video_url = None + config_json = self._search_regex( r'value=\'config=({.+?})\'', player_page, 'config json', default=None) if config_json: config = json.loads(config_json) - video_url = compat_urlparse.unquote(config['clip']['url']) - else: + video_url = compat_urllib_parse_unquote(config['clip']['url']) + + if not video_url: + video_url = self._html_search_regex( + r']+type="video/divx"[^>]+src="([^"]+)"', + player_page, 'video url', default=None) + + if not video_url: iframe_src = self._search_regex( r']+src="/?([^"]+)"', player_page, 'iframe url') iframe_url = 'http://veehd.com/%s' % iframe_src @@ -82,7 +112,6 @@ class VeeHDIE(InfoExtractor): 'id': video_id, 'title': title, 'url': video_url, - 'ext': 'mp4', 'uploader_id': uploader_id, 'thumbnail': thumbnail, 'description': description,