mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-11-14 22:37:59 +09:00
fix #30575
This commit is contained in:
parent
45ae186416
commit
f6c4b9da47
@ -5,7 +5,6 @@ import re
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
extract_attributes,
|
||||
int_or_none,
|
||||
parse_duration,
|
||||
parse_filesize,
|
||||
unified_timestamp,
|
||||
@ -58,39 +57,47 @@ class NewgroundsIE(InfoExtractor):
|
||||
|
||||
webpage = self._download_webpage(url, media_id)
|
||||
|
||||
title = self._html_search_regex(
|
||||
r'<title>([^>]+)</title>', webpage, 'title')
|
||||
paths = url.split('/')
|
||||
if paths[-3] == 'audio':
|
||||
isAudio = True
|
||||
else:
|
||||
isAudio = False
|
||||
|
||||
if isAudio:
|
||||
media_url = self._parse_json(self._search_regex(
|
||||
r'"url"\s*:\s*("[^"]+"),', webpage, ''), media_id)
|
||||
|
||||
uploader = self._html_search_regex(
|
||||
(r'(?s)<h4[^>]*>(.+?)</h4>.*?<em>\s*Author\s*</em>',
|
||||
r'(?:Author|Writer)\s*<a[^>]+>([^<]+)'), webpage, 'uploader',
|
||||
fatal=False)
|
||||
|
||||
formats = [{
|
||||
'url': media_url,
|
||||
'format_id': 'source',
|
||||
'quality': 1,
|
||||
}]
|
||||
|
||||
max_resolution = int_or_none(self._search_regex(
|
||||
r'max_resolution["\']\s*:\s*(\d+)', webpage, 'max resolution',
|
||||
default=None))
|
||||
if max_resolution:
|
||||
url_base = media_url.rpartition('.')[0]
|
||||
for resolution in (360, 720, 1080):
|
||||
if resolution > max_resolution:
|
||||
break
|
||||
else:
|
||||
media_url = f'https://www.newgrounds.com/portal/video/{media_id}'
|
||||
media = self._download_json(media_url, media_id, headers={'X-Requested-With': 'XMLHttpRequest'})
|
||||
|
||||
uploader = media['author']
|
||||
|
||||
sources = media['sources']
|
||||
|
||||
formats = []
|
||||
for source in sources:
|
||||
for i in range(len(sources[source])):
|
||||
formats.append({
|
||||
'url': '%s.%dp.mp4' % (url_base, resolution),
|
||||
'format_id': '%dp' % resolution,
|
||||
'height': resolution,
|
||||
'url': sources[source][i]['src'],
|
||||
'format_id': source,
|
||||
'height': int(source[:-2]) # 1080p -> 1080
|
||||
})
|
||||
|
||||
self._check_formats(formats, media_id)
|
||||
self._sort_formats(formats)
|
||||
|
||||
uploader = self._html_search_regex(
|
||||
(r'(?s)<h4[^>]*>(.+?)</h4>.*?<em>\s*Author\s*</em>',
|
||||
r'(?:Author|Writer)\s*<a[^>]+>([^<]+)'), webpage, 'uploader',
|
||||
fatal=False)
|
||||
title = self._html_search_regex(
|
||||
r'<title>([^>]+)</title>', webpage, 'title')
|
||||
timestamp = unified_timestamp(self._html_search_regex(
|
||||
(r'<dt>\s*Uploaded\s*</dt>\s*<dd>([^<]+</dd>\s*<dd>[^<]+)',
|
||||
r'<dt>\s*Uploaded\s*</dt>\s*<dd>([^<]+)'), webpage, 'timestamp',
|
||||
@ -98,7 +105,6 @@ class NewgroundsIE(InfoExtractor):
|
||||
duration = parse_duration(self._search_regex(
|
||||
r'(?s)<dd>\s*Song\s*</dd>\s*<dd>.+?</dd>\s*<dd>([^<]+)', webpage,
|
||||
'duration', default=None))
|
||||
|
||||
filesize_approx = parse_filesize(self._html_search_regex(
|
||||
r'(?s)<dd>\s*Song\s*</dd>\s*<dd>(.+?)</dd>', webpage, 'filesize',
|
||||
default=None))
|
||||
@ -108,6 +114,9 @@ class NewgroundsIE(InfoExtractor):
|
||||
if '<dd>Song' in webpage:
|
||||
formats[0]['vcodec'] = 'none'
|
||||
|
||||
self._check_formats(formats, media_id)
|
||||
self._sort_formats(formats)
|
||||
|
||||
return {
|
||||
'id': media_id,
|
||||
'title': title,
|
||||
|
Loading…
Reference in New Issue
Block a user