mirror of
https://github.com/yt-dlp/yt-dlp
synced 2024-12-29 22:24:25 +01:00
[udemy] Extract formats from view_html (Closes #8979)
This commit is contained in:
parent
6dee688e6d
commit
efcba804f6
1 changed files with 37 additions and 4 deletions
|
@ -1,5 +1,7 @@
|
||||||
from __future__ import unicode_literals
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import re
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..compat import (
|
from ..compat import (
|
||||||
compat_HTTPError,
|
compat_HTTPError,
|
||||||
|
@ -8,6 +10,8 @@ from ..compat import (
|
||||||
compat_urlparse,
|
compat_urlparse,
|
||||||
)
|
)
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
|
determine_ext,
|
||||||
|
extract_attributes,
|
||||||
ExtractorError,
|
ExtractorError,
|
||||||
float_or_none,
|
float_or_none,
|
||||||
int_or_none,
|
int_or_none,
|
||||||
|
@ -73,11 +77,8 @@ class UdemyIE(InfoExtractor):
|
||||||
return self._download_json(
|
return self._download_json(
|
||||||
'https://www.udemy.com/api-2.0/users/me/subscribed-courses/%s/lectures/%s?%s' % (
|
'https://www.udemy.com/api-2.0/users/me/subscribed-courses/%s/lectures/%s?%s' % (
|
||||||
course_id, lecture_id, compat_urllib_parse_urlencode({
|
course_id, lecture_id, compat_urllib_parse_urlencode({
|
||||||
'video_only': '',
|
'fields[lecture]': 'title,description,view_html,asset',
|
||||||
'auto_play': '',
|
|
||||||
'fields[lecture]': 'title,description,asset',
|
|
||||||
'fields[asset]': 'asset_type,stream_url,thumbnail_url,download_urls,data',
|
'fields[asset]': 'asset_type,stream_url,thumbnail_url,download_urls,data',
|
||||||
'instructorPreviewMode': 'False',
|
|
||||||
})),
|
})),
|
||||||
lecture_id, 'Downloading lecture JSON')
|
lecture_id, 'Downloading lecture JSON')
|
||||||
|
|
||||||
|
@ -246,6 +247,38 @@ class UdemyIE(InfoExtractor):
|
||||||
f['format_id'] = '%sp' % format_id
|
f['format_id'] = '%sp' % format_id
|
||||||
formats.append(f)
|
formats.append(f)
|
||||||
|
|
||||||
|
view_html = lecture.get('view_html')
|
||||||
|
if view_html:
|
||||||
|
view_html_urls = set()
|
||||||
|
for source in re.findall(r'<source[^>]+>', view_html):
|
||||||
|
attributes = extract_attributes(source)
|
||||||
|
src = attributes.get('src')
|
||||||
|
if not src:
|
||||||
|
continue
|
||||||
|
res = attributes.get('data-res')
|
||||||
|
height = int_or_none(res)
|
||||||
|
if src in view_html_urls:
|
||||||
|
continue
|
||||||
|
view_html_urls.add(src)
|
||||||
|
if attributes.get('type') == 'application/x-mpegURL' or determine_ext(src) == 'm3u8':
|
||||||
|
m3u8_formats = self._extract_m3u8_formats(
|
||||||
|
src, video_id, 'mp4', entry_protocol='m3u8_native',
|
||||||
|
m3u8_id='hls', fatal=False)
|
||||||
|
for f in m3u8_formats:
|
||||||
|
m = re.search(r'/hls_(?P<height>\d{3,4})_(?P<tbr>\d{2,})/', f['url'])
|
||||||
|
if m:
|
||||||
|
if not f.get('height'):
|
||||||
|
f['height'] = int(m.group('height'))
|
||||||
|
if not f.get('tbr'):
|
||||||
|
f['tbr'] = int(m.group('tbr'))
|
||||||
|
formats.extend(m3u8_formats)
|
||||||
|
else:
|
||||||
|
formats.append({
|
||||||
|
'url': src,
|
||||||
|
'format_id': '%dp' % height if height else None,
|
||||||
|
'height': height,
|
||||||
|
})
|
||||||
|
|
||||||
self._sort_formats(formats)
|
self._sort_formats(formats)
|
||||||
|
|
||||||
return {
|
return {
|
||||||
|
|
Loading…
Reference in a new issue