X-Git-Url: http://git.bitcoin.ninja/index.cgi?a=blobdiff_plain;f=youtube_dl%2Fextractor%2Flynda.py;h=1b6f5091d36eecb0f241b95e38739d4b967d6c34;hb=603fc4e0ea472c7c2a78ff201d69686a9e3fe1f2;hp=da94eab561b91d6b70675911e432b5750d5d5b04;hpb=ec85ded83cbfa652ba94cb080aab52d8b270212a;p=youtube-dl diff --git a/youtube_dl/extractor/lynda.py b/youtube_dl/extractor/lynda.py index da94eab56..1b6f5091d 100644 --- a/youtube_dl/extractor/lynda.py +++ b/youtube_dl/extractor/lynda.py @@ -94,7 +94,7 @@ class LyndaBaseIE(InfoExtractor): class LyndaIE(LyndaBaseIE): IE_NAME = 'lynda' IE_DESC = 'lynda.com videos' - _VALID_URL = r'https?://(?:www\.)?lynda\.com/(?:[^/]+/[^/]+/(?P\d+)|player/embed)/(?P\d+)' + _VALID_URL = r'https?://(?:www\.)?(?:lynda\.com|educourse\.ga)/(?:[^/]+/[^/]+/(?P\d+)|player/embed)/(?P\d+)' _TIMECODE_REGEX = r'\[(?P\d+:\d+:\d+[\.,]\d+)\]' @@ -110,6 +110,9 @@ class LyndaIE(LyndaBaseIE): }, { 'url': 'https://www.lynda.com/player/embed/133770?tr=foo=1;bar=g;fizz=rt&fs=0', 'only_matching': True, + }, { + 'url': 'https://educourse.ga/Bootstrap-tutorials/Using-exercise-files/110885/114408-4.html', + 'only_matching': True, }] def _raise_unavailable(self, video_id): @@ -253,16 +256,31 @@ class LyndaCourseIE(LyndaBaseIE): # Course link equals to welcome/introduction video link of same course # We will recognize it as course link - _VALID_URL = r'https?://(?:www|m)\.lynda\.com/(?P[^/]+/[^/]+/(?P\d+))-\d\.html' + _VALID_URL = r'https?://(?:www|m)\.(?:lynda\.com|educourse\.ga)/(?P[^/]+/[^/]+/(?P\d+))-\d\.html' def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) course_path = mobj.group('coursepath') course_id = mobj.group('courseid') + item_template = 'https://www.lynda.com/%s/%%s-4.html' % course_path + course = self._download_json( 'https://www.lynda.com/ajax/player?courseId=%s&type=course' % course_id, - course_id, 'Downloading course JSON') + course_id, 'Downloading course JSON', fatal=False) + + if not course: + webpage = self._download_webpage(url, course_id) + entries = [ + self.url_result( + item_template % video_id, ie=LyndaIE.ie_key(), + video_id=video_id) + for video_id in re.findall( + r'data-video-id=["\'](\d+)', webpage)] + return self.playlist_result( + entries, course_id, + self._og_search_title(webpage, fatal=False), + self._og_search_description(webpage)) if course.get('Status') == 'NotFound': raise ExtractorError( @@ -283,7 +301,7 @@ class LyndaCourseIE(LyndaBaseIE): if video_id: entries.append({ '_type': 'url_transparent', - 'url': 'https://www.lynda.com/%s/%s-4.html' % (course_path, video_id), + 'url': item_template % video_id, 'ie_key': LyndaIE.ie_key(), 'chapter': chapter.get('Title'), 'chapter_number': int_or_none(chapter.get('ChapterIndex')),