- # coding: utf-8
- from __future__ import unicode_literals
-
- import re
-
- from .jwplatform import JWPlatformBaseIE
- from ..utils import remove_end
-
-
- class ThisAVIE(JWPlatformBaseIE):
- _VALID_URL = r'https?://(?:www\.)?thisav\.com/video/(?P<id>[0-9]+)/.*'
- _TESTS = [{
- 'url': 'http://www.thisav.com/video/47734/%98%26sup1%3B%83%9E%83%82---just-fit.html',
- 'md5': '0480f1ef3932d901f0e0e719f188f19b',
- 'info_dict': {
- 'id': '47734',
- 'ext': 'flv',
- 'title': '高樹マリア - Just fit',
- 'uploader': 'dj7970',
- 'uploader_id': 'dj7970'
- }
- }, {
- 'url': 'http://www.thisav.com/video/242352/nerdy-18yo-big-ass-tattoos-and-glasses.html',
- 'md5': 'ba90c076bd0f80203679e5b60bf523ee',
- 'info_dict': {
- 'id': '242352',
- 'ext': 'mp4',
- 'title': 'Nerdy 18yo Big Ass Tattoos and Glasses',
- 'uploader': 'cybersluts',
- 'uploader_id': 'cybersluts',
- },
- }]
-
- def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
-
- video_id = mobj.group('id')
- webpage = self._download_webpage(url, video_id)
- title = remove_end(self._html_search_regex(
- r'<title>([^<]+)</title>', webpage, 'title'),
- ' - 視頻 - ThisAV.com-世界第一中文成人娛樂網站')
- video_url = self._html_search_regex(
- r"addVariable\('file','([^']+)'\);", webpage, 'video url', default=None)
- if video_url:
- info_dict = {
- 'formats': [{
- 'url': video_url,
- }],
- }
- else:
- info_dict = self._extract_jwplayer_data(
- webpage, video_id, require_title=False)
- uploader = self._html_search_regex(
- r': <a href="http://www.thisav.com/user/[0-9]+/(?:[^"]+)">([^<]+)</a>',
- webpage, 'uploader name', fatal=False)
- uploader_id = self._html_search_regex(
- r': <a href="http://www.thisav.com/user/[0-9]+/([^"]+)">(?:[^<]+)</a>',
- webpage, 'uploader id', fatal=False)
-
- info_dict.update({
- 'id': video_id,
- 'uploader': uploader,
- 'uploader_id': uploader_id,
- 'title': title,
- })
-
- return info_dict
|