mirror of https://github.com/yt-dlp/yt-dlp.git
[nobelprize] Add new extractor(closes #9999)
This commit is contained in:
parent
ec8705117a
commit
43c53a1700
|
@ -605,6 +605,7 @@ from .ninecninemedia import (
|
||||||
from .ninegag import NineGagIE
|
from .ninegag import NineGagIE
|
||||||
from .ninenow import NineNowIE
|
from .ninenow import NineNowIE
|
||||||
from .nintendo import NintendoIE
|
from .nintendo import NintendoIE
|
||||||
|
from .nobelprize import NobelPrizeIE
|
||||||
from .noco import NocoIE
|
from .noco import NocoIE
|
||||||
from .normalboots import NormalbootsIE
|
from .normalboots import NormalbootsIE
|
||||||
from .nosvideo import NosVideoIE
|
from .nosvideo import NosVideoIE
|
||||||
|
|
|
@ -0,0 +1,62 @@
|
||||||
|
# coding: utf-8
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..utils import (
|
||||||
|
js_to_json,
|
||||||
|
mimetype2ext,
|
||||||
|
determine_ext,
|
||||||
|
update_url_query,
|
||||||
|
get_element_by_attribute,
|
||||||
|
int_or_none,
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
|
class NobelPrizeIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?nobelprize\.org/mediaplayer.*?\bid=(?P<id>\d+)'
|
||||||
|
_TEST = {
|
||||||
|
'url': 'http://www.nobelprize.org/mediaplayer/?id=2636',
|
||||||
|
'md5': '04c81e5714bb36cc4e2232fee1d8157f',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '2636',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Announcement of the 2016 Nobel Prize in Physics',
|
||||||
|
'description': 'md5:05beba57f4f5a4bbd4cf2ef28fcff739',
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
media = self._parse_json(self._search_regex(
|
||||||
|
r'(?s)var\s*config\s*=\s*({.+?});', webpage,
|
||||||
|
'config'), video_id, js_to_json)['media']
|
||||||
|
title = media['title']
|
||||||
|
|
||||||
|
formats = []
|
||||||
|
for source in media.get('source', []):
|
||||||
|
source_src = source.get('src')
|
||||||
|
if not source_src:
|
||||||
|
continue
|
||||||
|
ext = mimetype2ext(source.get('type')) or determine_ext(source_src)
|
||||||
|
if ext == 'm3u8':
|
||||||
|
formats.extend(self._extract_m3u8_formats(
|
||||||
|
source_src, video_id, 'mp4', 'm3u8_native',
|
||||||
|
m3u8_id='hls', fatal=False))
|
||||||
|
elif ext == 'f4m':
|
||||||
|
formats.extend(self._extract_f4m_formats(
|
||||||
|
update_url_query(source_src, {'hdcore': '3.7.0'}),
|
||||||
|
video_id, f4m_id='hds', fatal=False))
|
||||||
|
else:
|
||||||
|
formats.append({
|
||||||
|
'url': source_src,
|
||||||
|
})
|
||||||
|
self._sort_formats(formats)
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': title,
|
||||||
|
'description': get_element_by_attribute('itemprop', 'description', webpage),
|
||||||
|
'duration': int_or_none(media.get('duration')),
|
||||||
|
'formats': formats,
|
||||||
|
}
|
Loading…
Reference in New Issue