[extractor/screencastify] Add extractor (#5604)
Closes #5603 Authored by: bashonly
This commit is contained in:
parent
7ff2fafe47
commit
27c0f899c8
2 changed files with 53 additions and 0 deletions
|
@ -1603,6 +1603,7 @@ from .savefrom import SaveFromIE
|
||||||
from .sbs import SBSIE
|
from .sbs import SBSIE
|
||||||
from .screen9 import Screen9IE
|
from .screen9 import Screen9IE
|
||||||
from .screencast import ScreencastIE
|
from .screencast import ScreencastIE
|
||||||
|
from .screencastify import ScreencastifyIE
|
||||||
from .screencastomatic import ScreencastOMaticIE
|
from .screencastomatic import ScreencastOMaticIE
|
||||||
from .scrippsnetworks import (
|
from .scrippsnetworks import (
|
||||||
ScrippsNetworksWatchIE,
|
ScrippsNetworksWatchIE,
|
||||||
|
|
52
yt_dlp/extractor/screencastify.py
Normal file
52
yt_dlp/extractor/screencastify.py
Normal file
|
@ -0,0 +1,52 @@
|
||||||
|
import urllib.parse
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..utils import traverse_obj, update_url_query
|
||||||
|
|
||||||
|
|
||||||
|
class ScreencastifyIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://watch\.screencastify\.com/v/(?P<id>[^/?#]+)'
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://watch.screencastify.com/v/sYVkZip3quLKhHw4Ybk8',
|
||||||
|
'info_dict': {
|
||||||
|
'id': 'sYVkZip3quLKhHw4Ybk8',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Inserting and Aligning the Case Top and Bottom',
|
||||||
|
'description': '',
|
||||||
|
'uploader': 'Paul Gunn',
|
||||||
|
'extra_param_to_segment_url': str,
|
||||||
|
},
|
||||||
|
'params': {
|
||||||
|
'skip_download': 'm3u8',
|
||||||
|
},
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
info = self._download_json(
|
||||||
|
f'https://umbrella.svc.screencastify.com/api/umbrellaService/watch/{video_id}', video_id)
|
||||||
|
|
||||||
|
query_string = traverse_obj(info, ('manifest', 'auth', 'query'))
|
||||||
|
query = urllib.parse.parse_qs(query_string)
|
||||||
|
formats = []
|
||||||
|
dash_manifest_url = traverse_obj(info, ('manifest', 'url'))
|
||||||
|
if dash_manifest_url:
|
||||||
|
formats.extend(
|
||||||
|
self._extract_mpd_formats(
|
||||||
|
dash_manifest_url, video_id, mpd_id='dash', query=query, fatal=False))
|
||||||
|
hls_manifest_url = traverse_obj(info, ('manifest', 'hlsUrl'))
|
||||||
|
if hls_manifest_url:
|
||||||
|
formats.extend(
|
||||||
|
self._extract_m3u8_formats(
|
||||||
|
hls_manifest_url, video_id, ext='mp4', m3u8_id='hls', query=query, fatal=False))
|
||||||
|
for f in formats:
|
||||||
|
f['url'] = update_url_query(f['url'], query)
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': info.get('title'),
|
||||||
|
'description': info.get('description'),
|
||||||
|
'uploader': info.get('userName'),
|
||||||
|
'formats': formats,
|
||||||
|
'extra_param_to_segment_url': query_string,
|
||||||
|
}
|
Loading…
Reference in a new issue