mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-17 13:52:36 +00:00
Merge branch 'sprhawk-weibo'
This commit is contained in:
commit
4df1098c3f
|
@ -1298,6 +1298,10 @@
|
|||
WebOfStoriesIE,
|
||||
WebOfStoriesPlaylistIE,
|
||||
)
|
||||
from .weibo import (
|
||||
WeiboIE,
|
||||
WeiboMobileIE
|
||||
)
|
||||
from .weiqitv import WeiqiTVIE
|
||||
from .wimp import WimpIE
|
||||
from .wistia import WistiaIE
|
||||
|
|
140
youtube_dl/extractor/weibo.py
Normal file
140
youtube_dl/extractor/weibo.py
Normal file
|
@ -0,0 +1,140 @@
|
|||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
from .common import InfoExtractor
|
||||
|
||||
import json
|
||||
import random
|
||||
import re
|
||||
|
||||
from ..compat import (
|
||||
compat_parse_qs,
|
||||
compat_str,
|
||||
)
|
||||
from ..utils import (
|
||||
js_to_json,
|
||||
strip_jsonp,
|
||||
urlencode_postdata,
|
||||
)
|
||||
|
||||
|
||||
class WeiboIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://weibo\.com/[0-9]+/(?P<id>[a-zA-Z0-9]+)'
|
||||
_TEST = {
|
||||
'url': 'https://weibo.com/6275294458/Fp6RGfbff?type=comment',
|
||||
'info_dict': {
|
||||
'id': 'Fp6RGfbff',
|
||||
'ext': 'mp4',
|
||||
'title': 'You should have servants to massage you,... 来自Hosico_猫 - 微博',
|
||||
}
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
# to get Referer url for genvisitor
|
||||
webpage, urlh = self._download_webpage_handle(url, video_id)
|
||||
|
||||
visitor_url = urlh.geturl()
|
||||
|
||||
if 'passport.weibo.com' in visitor_url:
|
||||
# first visit
|
||||
visitor_data = self._download_json(
|
||||
'https://passport.weibo.com/visitor/genvisitor', video_id,
|
||||
note='Generating first-visit data',
|
||||
transform_source=strip_jsonp,
|
||||
headers={'Referer': visitor_url},
|
||||
data=urlencode_postdata({
|
||||
'cb': 'gen_callback',
|
||||
'fp': json.dumps({
|
||||
'os': '2',
|
||||
'browser': 'Gecko57,0,0,0',
|
||||
'fonts': 'undefined',
|
||||
'screenInfo': '1440*900*24',
|
||||
'plugins': '',
|
||||
}),
|
||||
}))
|
||||
|
||||
tid = visitor_data['data']['tid']
|
||||
cnfd = '%03d' % visitor_data['data']['confidence']
|
||||
|
||||
self._download_webpage(
|
||||
'https://passport.weibo.com/visitor/visitor', video_id,
|
||||
note='Running first-visit callback',
|
||||
query={
|
||||
'a': 'incarnate',
|
||||
't': tid,
|
||||
'w': 2,
|
||||
'c': cnfd,
|
||||
'cb': 'cross_domain',
|
||||
'from': 'weibo',
|
||||
'_rand': random.random(),
|
||||
})
|
||||
|
||||
webpage = self._download_webpage(
|
||||
url, video_id, note='Revisiting webpage')
|
||||
|
||||
title = self._html_search_regex(
|
||||
r'<title>(.+?)</title>', webpage, 'title')
|
||||
|
||||
video_formats = compat_parse_qs(self._search_regex(
|
||||
r'video-sources=\\\"(.+?)\"', webpage, 'video_sources'))
|
||||
|
||||
formats = []
|
||||
supported_resolutions = (480, 720)
|
||||
for res in supported_resolutions:
|
||||
vid_urls = video_formats.get(compat_str(res))
|
||||
if not vid_urls or not isinstance(vid_urls, list):
|
||||
continue
|
||||
|
||||
vid_url = vid_urls[0]
|
||||
formats.append({
|
||||
'url': vid_url,
|
||||
'height': res,
|
||||
})
|
||||
|
||||
self._sort_formats(formats)
|
||||
|
||||
uploader = self._og_search_property(
|
||||
'nick-name', webpage, 'uploader', default=None)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'uploader': uploader,
|
||||
'formats': formats
|
||||
}
|
||||
|
||||
|
||||
class WeiboMobileIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://m\.weibo\.cn/status/(?P<id>[0-9]+)(\?.+)?'
|
||||
_TEST = {
|
||||
'url': 'https://m.weibo.cn/status/4189191225395228?wm=3333_2001&sourcetype=weixin&featurecode=newtitle&from=singlemessage&isappinstalled=0',
|
||||
'info_dict': {
|
||||
'id': '4189191225395228',
|
||||
'ext': 'mp4',
|
||||
'title': '午睡当然是要甜甜蜜蜜的啦',
|
||||
'uploader': '柴犬柴犬'
|
||||
}
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
# to get Referer url for genvisitor
|
||||
webpage = self._download_webpage(url, video_id, note='visit the page')
|
||||
|
||||
weibo_info = self._parse_json(self._search_regex(
|
||||
r'var\s+\$render_data\s*=\s*\[({.*})\]\[0\]\s*\|\|\s*{};',
|
||||
webpage, 'js_code', flags=re.DOTALL),
|
||||
video_id, transform_source=js_to_json)
|
||||
|
||||
status_data = weibo_info.get('status', {})
|
||||
page_info = status_data.get('page_info')
|
||||
title = status_data['status_title']
|
||||
uploader = status_data.get('user', {}).get('screen_name')
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'uploader': uploader,
|
||||
'url': page_info['media_info']['stream_url']
|
||||
}
|
Loading…
Reference in a new issue