mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-01-07 17:16:08 +00:00
[xnxx] Fix extraction (closes #15817)
This commit is contained in:
parent
cc7f6c720e
commit
b4a190fe2a
|
@ -1,19 +1,29 @@
|
|||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..compat import compat_urllib_parse_unquote
|
||||
from ..utils import (
|
||||
determine_ext,
|
||||
int_or_none,
|
||||
NO_DEFAULT,
|
||||
str_to_int,
|
||||
)
|
||||
|
||||
|
||||
class XNXXIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:video|www)\.xnxx\.com/video-?(?P<id>[0-9a-z]+)/'
|
||||
_TESTS = [{
|
||||
'url': 'http://www.xnxx.com/video-55awb78/skyrim_test_video',
|
||||
'md5': 'ef7ecee5af78f8b03dca2cf31341d3a0',
|
||||
'md5': '7583e96c15c0f21e9da3453d9920fbba',
|
||||
'info_dict': {
|
||||
'id': '55awb78',
|
||||
'ext': 'flv',
|
||||
'ext': 'mp4',
|
||||
'title': 'Skyrim Test Video',
|
||||
'thumbnail': r're:^https?://.*\.jpg',
|
||||
'duration': 469,
|
||||
'view_count': int,
|
||||
'age_limit': 18,
|
||||
},
|
||||
}, {
|
||||
|
@ -26,23 +36,49 @@ class XNXXIE(InfoExtractor):
|
|||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
video_url = self._search_regex(r'flv_url=(.*?)&',
|
||||
webpage, 'video URL')
|
||||
video_url = compat_urllib_parse_unquote(video_url)
|
||||
def get(meta, default=NO_DEFAULT, fatal=True):
|
||||
return self._search_regex(
|
||||
r'set%s\s*\(\s*(["\'])(?P<value>(?:(?!\1).)+)\1' % meta,
|
||||
webpage, meta, default=default, fatal=fatal, group='value')
|
||||
|
||||
video_title = self._html_search_regex(r'<title>(.*?)\s+-\s+XNXX.COM',
|
||||
webpage, 'title')
|
||||
title = self._og_search_title(
|
||||
webpage, default=None) or get('VideoTitle')
|
||||
|
||||
video_thumbnail = self._search_regex(r'url_bigthumb=(.*?)&',
|
||||
webpage, 'thumbnail', fatal=False)
|
||||
formats = []
|
||||
for mobj in re.finditer(
|
||||
r'setVideo(?:Url(?P<id>Low|High)|HLS)\s*\(\s*(?P<q>["\'])(?P<url>(?:https?:)?//.+?)(?P=q)', webpage):
|
||||
format_url = mobj.group('url')
|
||||
if determine_ext(format_url) == 'm3u8':
|
||||
formats.extend(self._extract_m3u8_formats(
|
||||
format_url, video_id, 'mp4', entry_protocol='m3u8_native',
|
||||
preference=1, m3u8_id='hls', fatal=False))
|
||||
else:
|
||||
format_id = mobj.group('id')
|
||||
if format_id:
|
||||
format_id = format_id.lower()
|
||||
formats.append({
|
||||
'url': format_url,
|
||||
'format_id': format_id,
|
||||
'quality': -1 if format_id == 'low' else 0,
|
||||
})
|
||||
self._sort_formats(formats)
|
||||
|
||||
thumbnail = self._og_search_thumbnail(webpage, default=None) or get(
|
||||
'ThumbUrl', fatal=False) or get('ThumbUrl169', fatal=False)
|
||||
duration = int_or_none(self._og_search_property('duration', webpage))
|
||||
view_count = str_to_int(self._search_regex(
|
||||
r'id=["\']nb-views-number[^>]+>([\d,.]+)', webpage, 'view count',
|
||||
default=None))
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'url': video_url,
|
||||
'title': video_title,
|
||||
'ext': 'flv',
|
||||
'thumbnail': video_thumbnail,
|
||||
'title': title,
|
||||
'thumbnail': thumbnail,
|
||||
'duration': duration,
|
||||
'view_count': view_count,
|
||||
'age_limit': 18,
|
||||
'formats': formats,
|
||||
}
|
||||
|
|
Loading…
Reference in a new issue