注意,本项目仅支持Python3,不提供backporting到Python2
后来我才搞清楚pcs并不是指百度网盘,想过改成baidupanapi但是由于历史原因就算了。
http://baidupcsapi.readthedocs.org/
- 获取分享链接中文件列表接口
- 保存分享到用户网盘接口
百度网盘现在不开放PCS API,所以目前只能用百度自己的API
应该是可以了,参考 (https://github.com/mozillazg/baidu-pcs-python-sdk/wiki/%E5%A6%82%E4%BD%95%E8%8E%B7%E5%8F%96-Access-Token-%E5%92%8C-Refresh-Token%EF%BC%9F)
一个基于此api的 fuse(测试环境ubuntu12.04) http://github.com/ly0/baidu-fuse
web版百度网盘(可以用来开放资源,测试环境ubuntu14.04) https://github.com/ly0/web.baidupan
-
Free software: MIT license
-
Python version: 2.7
-
require: requests>=2.0.0, requests_toolbelt>=0.1.2
-
删掉了一些原作者在pan.baidu.com上没有的api,修改了大部分api和相关的程序部分,为了区分原作者的程序,在pypi上发布为baidupcsapi
To install baidupcsapi, simply:
$ pip3 install baidupcsapi
>>> from baidupcsapi import PCS
>>> pcs = PCS('username','password')
>>> print(pcs.quota().content)
>>> print(pcs.list_files('/').content)
>>> headers = {'Range': 'bytes=0-99'}
>>> pcs = PCS('username','password')
>>> pcs.download('/test_sdk/test.txt', headers=headers)
回调函数参数要求 有size和progress两个参数名, size:文件总字节数 progress:当前传输完成字节数
import progressbar
from baidupcsapi import PCS
class ProgressBar():
def __init__(self):
self.first_call = True
def __call__(self, *args, **kwargs):
if self.first_call:
self.widgets = [progressbar.Percentage(), ' ', progressbar.Bar(marker=progressbar.RotatingMarker('>')),
' ', progressbar.ETA()]
self.pbar = progressbar.ProgressBar(widgets=self.widgets, maxval=kwargs['size']).start()
self.first_call = False
if kwargs['size'] <= kwargs['progress']:
self.pbar.finish()
else:
self.pbar.update(kwargs['progress'])
pcs = PCS('username','password')
test_file = open('bigfile.pdf','rb').read()
ret = pcs.upload('/',test_file,'bigfile.pdf',callback=ProgressBar())
将大文件切成一个个块,分批上传 注意upload系列的函数都可以指定callback参数
#coding: utf-8
import os,json,sys,tempfile
from baidupcsapi import PCS
pcs = PCS('username','password')
chinksize = 1024*1024*16
fid = 1
md5list = []
tmpdir = tempfile.mkdtemp('bdpcs')
with open(sys.argv[1],'rb') as infile:
while 1:
data = infile.read(chinksize)
if len(data) == 0: break
smallfile = os.path.join(tmpdir, 'tmp%d' %fid)
with open(smallfile, 'wb') as f:
f.write(data)
print('chunk%d size %d' %(fid, len(data)))
fid += 1
print('start uploading...')
ret = pcs.upload_tmpfile(open(smallfile, 'rb'))
md5list.append(json.loads(ret.content)['md5'])
print('md5: %s' %(md5list[-1]))
os.remove(smallfile)
os.rmdir(tmpdir)
ret = pcs.upload_superfile('/'+os.path.basename(sys.argv[1]), md5list)
print ret.content
python upload.py huge_file