diff --git a/youtube_dl/extractor/__init__.py b/youtube_dl/extractor/__init__.py index 6780f7f993..632e832c72 100644 --- a/youtube_dl/extractor/__init__.py +++ b/youtube_dl/extractor/__init__.py @@ -147,6 +147,7 @@ from .izlesene import IzleseneIE from .jadorecettepub import JadoreCettePubIE from .jeuxvideo import JeuxVideoIE +from .jove import JoveIE from .jukebox import JukeboxIE from .justintv import JustinTVIE from .jpopsukitv import JpopsukiIE diff --git a/youtube_dl/extractor/jove.py b/youtube_dl/extractor/jove.py new file mode 100644 index 0000000000..a7110b6838 --- /dev/null +++ b/youtube_dl/extractor/jove.py @@ -0,0 +1,64 @@ +# coding: utf-8 +from __future__ import unicode_literals + +import re +from datetime import datetime + +from .common import InfoExtractor +from ..utils import determine_ext, ExtractorError + + +class JoveIE(InfoExtractor): + _VALID_URL = r'https?://(?:www\.)?jove\.com/video/(?P[0-9]+)' + _CHAPTERS_URL = 'http://www.jove.com/video-chapters?videoid={video_id:}' + _TEST = { + 'url': 'http://www.jove.com/video/2744/electrode-positioning-montage-transcranial-direct-current', + 'md5': '93723888d82dbd6ba8b3d7d0cd65dd2b', + 'info_dict': { + 'id': '2744', + 'ext': 'mp4', + 'title': 'Electrode Positioning and Montage in Transcranial Direct Current Stimulation', + 'description': 'Transcranial direct current stimulation (tDCS) is an established technique to modulate cortical excitability1,2. It has been ...', + 'thumbnail': 're:^https?://.*\.png$', + 'upload_date': '20110523', + } + } + + def _real_extract(self, url): + mobj = re.match(self._VALID_URL, url) + video_id = mobj.group('id') + + webpage = self._download_webpage(url, video_id) + title = self._html_search_meta('citation_title', webpage, 'title') + thumbnail = self._og_search_thumbnail(webpage) + description = self._html_search_meta( + 'description', webpage, 'description', fatal=False) + publish_date = self._html_search_meta( + 'citation_publication_date', webpage, 'publish date', fatal=False) + if publish_date: + publish_date = datetime.strptime(publish_date, + '%Y/%m/%d').strftime('%Y%m%d') + + # Not the same as video_id. + chapters_id = self._html_search_regex( + r'/video-chapters\?videoid=([0-9]+)', webpage, 'chapters id') + chapters_xml = self._download_xml( + self._CHAPTERS_URL.format(video_id=chapters_id), + video_id, note='Downloading chapter XML', + errnote='Failed to download chapter XML' + ) + video_url = chapters_xml.attrib.get('video') + if not video_url: + raise ExtractorError('Failed to get the video URL') + + ext = determine_ext(video_url) + + return { + 'id': video_id, + 'title': title, + 'url': video_url, + 'ext': ext, + 'thumbnail': thumbnail, + 'description': description, + 'upload_date': publish_date, + }