You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
yt-dlc/yt_dlp/extractor/porn91.py

64 lines
2.1 KiB
Python

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
parse_duration,
int_or_none,
ExtractorError,
)
class Porn91IE(InfoExtractor):
IE_NAME = '91porn'
_VALID_URL = r'(?:https?://)(?:www\.|)91porn\.com/.+?\?viewkey=(?P<id>[\w\d]+)'
_TEST = {
'url': 'http://91porn.com/view_video.php?viewkey=7e42283b4f5ab36da134',
'md5': '7fcdb5349354f40d41689bd0fa8db05a',
'info_dict': {
'id': '7e42283b4f5ab36da134',
'title': '18岁大一漂亮学妹水嫩性感再爽一次',
'ext': 'mp4',
'duration': 431,
'age_limit': 18,
}
}
def _real_extract(self, url):
video_id = self._match_id(url)
self._set_cookie('91porn.com', 'language', 'cn_CN')
webpage = self._download_webpage(
'http://91porn.com/view_video.php?viewkey=%s' % video_id, video_id)
if '作为游客你每天只可观看10个视频' in webpage:
raise ExtractorError('91 Porn says: Daily limit 10 videos exceeded', expected=True)
title = self._search_regex(
r'<div id="viewvideo-title">([^<]+)</div>', webpage, 'title')
title = title.replace('\n', '')
video_link_url = self._search_regex(
r'<textarea[^>]+id=["\']fm-video_link[^>]+>([^<]+)</textarea>',
webpage, 'video link')
videopage = self._download_webpage(video_link_url, video_id)
info_dict = self._parse_html5_media_entries(url, videopage, video_id)[0]
duration = parse_duration(self._search_regex(
r'时长:\s*</span>\s*(\d+:\d+)', webpage, 'duration', fatal=False))
comment_count = int_or_none(self._search_regex(
r'留言:\s*</span>\s*(\d+)', webpage, 'comment count', fatal=False))
info_dict.update({
'id': video_id,
'title': title,
'duration': duration,
'comment_count': comment_count,
'age_limit': self._rta_search(webpage),
})
return info_dict