aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/lynda.py
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor/lynda.py')
-rw-r--r--youtube_dl/extractor/lynda.py72
1 files changed, 62 insertions, 10 deletions
diff --git a/youtube_dl/extractor/lynda.py b/youtube_dl/extractor/lynda.py
index 299873ecc..f4dcfd93f 100644
--- a/youtube_dl/extractor/lynda.py
+++ b/youtube_dl/extractor/lynda.py
@@ -94,12 +94,12 @@ class LyndaBaseIE(InfoExtractor):
class LyndaIE(LyndaBaseIE):
IE_NAME = 'lynda'
IE_DESC = 'lynda.com videos'
- _VALID_URL = r'https?://(?:www\.)?lynda\.com/(?:[^/]+/[^/]+/\d+|player/embed)/(?P<id>\d+)'
+ _VALID_URL = r'https?://(?:www\.)?lynda\.com/(?:[^/]+/[^/]+/(?P<course_id>\d+)|player/embed)/(?P<id>\d+)'
_TIMECODE_REGEX = r'\[(?P<timecode>\d+:\d+:\d+[\.,]\d+)\]'
_TESTS = [{
- 'url': 'http://www.lynda.com/Bootstrap-tutorials/Using-exercise-files/110885/114408-4.html',
+ 'url': 'https://www.lynda.com/Bootstrap-tutorials/Using-exercise-files/110885/114408-4.html',
# md5 is unstable
'info_dict': {
'id': '114408',
@@ -112,19 +112,71 @@ class LyndaIE(LyndaBaseIE):
'only_matching': True,
}]
+ def _raise_unavailable(self, video_id):
+ self.raise_login_required(
+ 'Video %s is only available for members' % video_id)
+
def _real_extract(self, url):
- video_id = self._match_id(url)
+ mobj = re.match(self._VALID_URL, url)
+ video_id = mobj.group('id')
+ course_id = mobj.group('course_id')
+
+ query = {
+ 'videoId': video_id,
+ 'type': 'video',
+ }
video = self._download_json(
- 'http://www.lynda.com/ajax/player?videoId=%s&type=video' % video_id,
- video_id, 'Downloading video JSON')
+ 'https://www.lynda.com/ajax/player', video_id,
+ 'Downloading video JSON', fatal=False, query=query)
+
+ # Fallback scenario
+ if not video:
+ query['courseId'] = course_id
+
+ play = self._download_json(
+ 'https://www.lynda.com/ajax/course/%s/%s/play'
+ % (course_id, video_id), video_id, 'Downloading play JSON')
+
+ if not play:
+ self._raise_unavailable(video_id)
+
+ formats = []
+ for formats_dict in play:
+ urls = formats_dict.get('urls')
+ if not isinstance(urls, dict):
+ continue
+ cdn = formats_dict.get('name')
+ for format_id, format_url in urls.items():
+ if not format_url:
+ continue
+ formats.append({
+ 'url': format_url,
+ 'format_id': '%s-%s' % (cdn, format_id) if cdn else format_id,
+ 'height': int_or_none(format_id),
+ })
+ self._sort_formats(formats)
+
+ conviva = self._download_json(
+ 'https://www.lynda.com/ajax/player/conviva', video_id,
+ 'Downloading conviva JSON', query=query)
+
+ return {
+ 'id': video_id,
+ 'title': conviva['VideoTitle'],
+ 'description': conviva.get('VideoDescription'),
+ 'release_year': int_or_none(conviva.get('ReleaseYear')),
+ 'duration': int_or_none(conviva.get('Duration')),
+ 'creator': conviva.get('Author'),
+ 'formats': formats,
+ }
if 'Status' in video:
raise ExtractorError(
'lynda returned error: %s' % video['Message'], expected=True)
if video.get('HasAccess') is False:
- self.raise_login_required('Video %s is only available for members' % video_id)
+ self._raise_unavailable(video_id)
video_id = compat_str(video.get('ID') or video_id)
duration = int_or_none(video.get('DurationInSeconds'))
@@ -148,7 +200,7 @@ class LyndaIE(LyndaBaseIE):
for prioritized_stream_id, prioritized_stream in prioritized_streams.items():
formats.extend([{
'url': video_url,
- 'width': int_or_none(format_id),
+ 'height': int_or_none(format_id),
'format_id': '%s-%s' % (prioritized_stream_id, format_id),
} for format_id, video_url in prioritized_stream.items()])
@@ -187,7 +239,7 @@ class LyndaIE(LyndaBaseIE):
return srt
def _get_subtitles(self, video_id):
- url = 'http://www.lynda.com/ajax/player?videoId=%s&type=transcript' % video_id
+ url = 'https://www.lynda.com/ajax/player?videoId=%s&type=transcript' % video_id
subs = self._download_json(url, None, False)
if subs:
return {'en': [{'ext': 'srt', 'data': self._fix_subtitles(subs)}]}
@@ -209,7 +261,7 @@ class LyndaCourseIE(LyndaBaseIE):
course_id = mobj.group('courseid')
course = self._download_json(
- 'http://www.lynda.com/ajax/player?courseId=%s&type=course' % course_id,
+ 'https://www.lynda.com/ajax/player?courseId=%s&type=course' % course_id,
course_id, 'Downloading course JSON')
if course.get('Status') == 'NotFound':
@@ -231,7 +283,7 @@ class LyndaCourseIE(LyndaBaseIE):
if video_id:
entries.append({
'_type': 'url_transparent',
- 'url': 'http://www.lynda.com/%s/%s-4.html' % (course_path, video_id),
+ 'url': 'https://www.lynda.com/%s/%s-4.html' % (course_path, video_id),
'ie_key': LyndaIE.ie_key(),
'chapter': chapter.get('Title'),
'chapter_number': int_or_none(chapter.get('ChapterIndex')),