+++ /dev/null
-# coding: utf-8
-from __future__ import unicode_literals
-
-from .common import InfoExtractor
-from ..utils import (
- int_or_none,
- unified_strdate,
-)
-
-
-class JpopsukiIE(InfoExtractor):
- IE_NAME = 'jpopsuki.tv'
- _VALID_URL = r'https?://(?:www\.)?jpopsuki\.tv/(?:category/)?video/[^/]+/(?P<id>\S+)'
-
- _TEST = {
- 'url': 'http://www.jpopsuki.tv/video/ayumi-hamasaki---evolution/00be659d23b0b40508169cdee4545771',
- 'md5': '88018c0c1a9b1387940e90ec9e7e198e',
- 'info_dict': {
- 'id': '00be659d23b0b40508169cdee4545771',
- 'ext': 'mp4',
- 'title': 'ayumi hamasaki - evolution',
- 'description': 'Release date: 2001.01.31\r\n浜崎あゆみ - evolution',
- 'thumbnail': 'http://www.jpopsuki.tv/cache/89722c74d2a2ebe58bcac65321c115b2.jpg',
- 'uploader': 'plama_chan',
- 'uploader_id': '404',
- 'upload_date': '20121101'
- }
- }
-
- def _real_extract(self, url):
- video_id = self._match_id(url)
-
- webpage = self._download_webpage(url, video_id)
-
- video_url = 'http://www.jpopsuki.tv' + self._html_search_regex(
- r'<source src="(.*?)" type', webpage, 'video url')
-
- video_title = self._og_search_title(webpage)
- description = self._og_search_description(webpage)
- thumbnail = self._og_search_thumbnail(webpage)
- uploader = self._html_search_regex(
- r'<li>from: <a href="/user/view/user/(.*?)/uid/',
- webpage, 'video uploader', fatal=False)
- uploader_id = self._html_search_regex(
- r'<li>from: <a href="/user/view/user/\S*?/uid/(\d*)',
- webpage, 'video uploader_id', fatal=False)
- upload_date = unified_strdate(self._html_search_regex(
- r'<li>uploaded: (.*?)</li>', webpage, 'video upload_date',
- fatal=False))
- view_count_str = self._html_search_regex(
- r'<li>Hits: ([0-9]+?)</li>', webpage, 'video view_count',
- fatal=False)
- comment_count_str = self._html_search_regex(
- r'<h2>([0-9]+?) comments</h2>', webpage, 'video comment_count',
- fatal=False)
-
- return {
- 'id': video_id,
- 'url': video_url,
- 'title': video_title,
- 'description': description,
- 'thumbnail': thumbnail,
- 'uploader': uploader,
- 'uploader_id': uploader_id,
- 'upload_date': upload_date,
- 'view_count': int_or_none(view_count_str),
- 'comment_count': int_or_none(comment_count_str),
- }