#f = open("yesterday2",‘r+‘,encoding="utf-8") #文件句柄 读写
#f = open("yesterday2",‘w+‘,encoding="utf-8") #文件句柄 写读
#f = open("yesterday2",‘a+‘,encoding="utf-8") #文件句柄 追加读写
f = open("yesterday2",‘wb‘) #文件句柄 二进制文件
函数传参:
#*args:接受N个位置参数,转换成元组形式
# def test(*args):
# print(args)
#
# test(1,2,3,4,5,5)
# test(*[1,2,4,5,5])# args=tuple([1,2,3,4,5])
# def test1(x,*args):
# print(x)
# print(args)
#
# test1(1,2,3,4,5,6,7)
#**kwargs:接受N个关键字参数,转换成字典的方式
# def test2(**kwargs):
# print(kwargs)
# print(kwargs[‘name‘])
# print(kwargs[‘age‘])
# print(kwargs[‘sex‘])
#
# test2(name=‘alex‘,age=8,sex=‘F‘)
# test2(**{‘name‘:‘alex‘,‘age‘:8})
# def test3(name,**kwargs):
# print(name)
# print(kwargs)
#
# test3(‘alex‘,age=18,sex=‘m‘)
# def test4(name,age=18,**kwargs):
# print(name)
# print(age)
# print(kwargs)
#
# test4(‘alex‘,age=34,sex=‘m‘,hobby=‘tesla‘)
def test4(name,age=18,*args,**kwargs):
print(name)
print(age)
print(args)
print(kwargs)
logger("TEST4")
def logger(source):
print("from %s" % source)
test4(‘alex‘,age=34,sex=‘m‘,hobby=‘tesla‘)