114 lines
3.9 KiB
Python
114 lines
3.9 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import json
|
|
import xml.etree.ElementTree as ET
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
int_or_none,
|
|
ExtractorError,
|
|
)
|
|
|
|
|
|
class BiliBiliIE(InfoExtractor):
|
|
_VALID_URL = r'http://www\.bilibili\.(?:tv|com)/video/av(?P<id>[0-9]+)/'
|
|
|
|
_TESTS = [{
|
|
'url': 'http://www.bilibili.tv/video/av1074402/',
|
|
'md5': '2c301e4dab317596e837c3e7633e7d86',
|
|
'info_dict': {
|
|
'id': '1554319',
|
|
'ext': 'flv',
|
|
'title': '【金坷垃】金泡沫',
|
|
'duration': 308313,
|
|
'upload_date': '20140420',
|
|
'thumbnail': 're:^https?://.+\.jpg',
|
|
'description': 'md5:ce18c2a2d2193f0df2917d270f2e5923',
|
|
'timestamp': 1397983878,
|
|
'uploader': '菊子桑',
|
|
},
|
|
}, {
|
|
'url': 'http://www.bilibili.com/video/av1041170/',
|
|
'info_dict': {
|
|
'id': '1041170',
|
|
'title': '【BD1080P】刀语【诸神&异域】',
|
|
},
|
|
'playlist_count': 12,
|
|
}]
|
|
|
|
def _extract_video_info(self, cid, view_data, page_num=1, num_pages=1):
|
|
title = view_data['title']
|
|
|
|
page = self._download_webpage(
|
|
'http://interface.bilibili.com/v_cdn_play?appkey=8e9fc618fbd41e28&cid=%s' % cid,
|
|
cid,
|
|
'Downloading page %d/%d' % (page_num, num_pages)
|
|
)
|
|
try:
|
|
err_info = json.loads(page)
|
|
raise ExtractorError(
|
|
'BiliBili said: ' + err_info['error_text'], expected=True)
|
|
except ValueError:
|
|
pass
|
|
|
|
doc = ET.fromstring(page)
|
|
durls = doc.findall('./durl')
|
|
|
|
entries = []
|
|
|
|
for durl in durls:
|
|
formats = []
|
|
backup_url = durl.find('./backup_url')
|
|
if backup_url is not None:
|
|
formats.append({'url': backup_url.find('./url').text})
|
|
size = durl.find('./filesize|./size')
|
|
formats.append({
|
|
'url': durl.find('./url').text,
|
|
'filesize': int_or_none(size.text) if size else None,
|
|
'ext': 'flv',
|
|
})
|
|
entries.append({
|
|
'id': '%s_part%s' % (cid, durl.find('./order').text),
|
|
'title': title,
|
|
'duration': int_or_none(durl.find('./length').text) // 1000,
|
|
'formats': formats,
|
|
})
|
|
|
|
info = {
|
|
'id': cid,
|
|
'title': title,
|
|
'description': view_data.get('description'),
|
|
'thumbnail': view_data.get('pic'),
|
|
'uploader': view_data.get('author'),
|
|
'timestamp': int_or_none(view_data.get('created')),
|
|
'view_count': view_data.get('play'),
|
|
'duration': int_or_none(doc.find('./timelength').text),
|
|
}
|
|
|
|
if len(entries) == 1:
|
|
entries[0].update(info)
|
|
return entries[0]
|
|
else:
|
|
info.update({
|
|
'_type': 'multi_video',
|
|
'entries': entries,
|
|
})
|
|
return info
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
view_data = self._download_json('http://api.bilibili.com/view?type=json&appkey=8e9fc618fbd41e28&id=%s' % video_id, video_id)
|
|
|
|
num_pages = int_or_none(view_data['pages'])
|
|
if num_pages > 1:
|
|
play_list_title = view_data['title']
|
|
page_list = self._download_json('http://www.bilibili.com/widget/getPageList?aid=%s' % video_id, video_id, 'Downloading page list metadata')
|
|
entries = []
|
|
for page in page_list:
|
|
view_data['title'] = page['pagename']
|
|
entries.append(self._extract_video_info(str(page['cid']), view_data, page['page'], num_pages))
|
|
return self.playlist_result(entries, video_id, play_list_title, view_data.get('description'))
|
|
else:
|
|
return self._extract_video_info(str(view_data['cid']), view_data)
|