youtube-dl/youtube_dl/extractor/skillshare.py

51 lines
2.1 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from .brightcove import BrightcoveNewIE
from ..utils import (
try_get,
compat_str
)
class SkillshareClassIE(InfoExtractor):
IE_NAME = 'skillshare:class'
_VALID_URL = r'https?://(?:www\.)?skillshare\.com/classes/[^/]+/(?P<id>[0-9]+)'
_TEST = {
'url': 'https://www.skillshare.com/classes/SEO-Today-Strategies-to-Earn-Trust-Rank-High-and-Stand-Out/423483018',
'only_matching': True,
}
def _real_extract(self, url):
class_id = self._match_id(url)
class_json_data = self._parse_json(self._search_regex(
r'(?P<json_line>{.+pageData.+});\n',
self._download_webpage(url, class_id), 'class_json_data'), class_id)
account_id = class_json_data['pageData']['videoPlayerData']['brightcoveAccountId']
lessons = class_json_data['pageData']['videoPlayerData']['units'][0]['sessions']
entries = []
for lesson in lessons:
lesson_id = lesson.get('id')
lesson_info_api_response = self._download_json(
"https://www.skillshare.com/sessions/%s/video" % lesson_id,
lesson_id)
if 'video_hashed_id' not in lesson_info_api_response:
break
video_hashed_id = self._search_regex(
r'(\d+)', lesson_info_api_response.get('video_hashed_id'),
'video_hashed_id')
entry = {
# the brightcove extractor extracts the title and id
'_type': 'url_transparent',
'ie_key': BrightcoveNewIE.ie_key(),
'url': 'https://players.brightcove.net/%s/default_default/index.html?videoId=%s' % (account_id, video_hashed_id),
}
entries.append(entry)
return self.playlist_result(
entries, class_id, try_get(
class_json_data, lambda x: x['pageData']['headerData']['title'],
compat_str),
try_get(
class_json_data, lambda x: x['pageData']['sectionData']['description'],
compat_str))