为什么不能使用 class 的成员变量执行多进程？

Question

I tested a code using python 3.6 and python 3.7 by 2 cases.

1 case) using Member variables of class

import multiprocessing as mp
import sys
class Foo:
    def do_multiprocessing(self):
        ctx = mp.get_context('spawn')
        self.process_1 = ctx.Process(target=self.do_stuff1)
        self.process_2 = ctx.Process(target=self.do_stuff2)

        self.process_1.start()
        self.process_2.start()

    def do_stuff1(self):
        print("Doing 1")

    def do_stuff2(self):
        print("Doing 2")

if __name__ == "__main__":
    print("PYTHON VERSION : ", sys.version)
    foo = Foo()
    foo.do_multiprocessing()

Result: When i executed it using python 3.6, working well. However, when i executed it using python 3.7, a error occurred.(TypeError: Can't pickle weakref object)

2 case) using local variables

import multiprocessing as mp
import sys
class Foo:
    def do_multiprocessing(self):
        ctx = mp.get_context('spawn')
        process_1 = ctx.Process(target=self.do_stuff1)
        process_2 = ctx.Process(target=self.do_stuff2)

        process_1.start()
        process_2.start()

    def do_stuff1(self):
        print("Doing 1")

    def do_stuff2(self):
        print("Doing 2")

if __name__ == "__main__":
    print("PYTHON VERSION : ", sys.version)
    foo = Foo()
    foo.do_multiprocessing()

Result: When i executed it using python 3.6 and python 3.7, working well.

Why can't execute multiprocess using member variables of class? And why can execute multiprocess using local variables?

I can't understand why a code execute.

ps) if using mp.get_context('fork'), the code works well using python 3.6 and python 3.7 on case 1.

Answer 1

By passing self.do_stuff2 to process_2 , you're trying to give that process access to the entirety of self (as an implicit argument), which includes self.process_1 . You can see how giving one child process meaningful access to another might be complicated (unless you have the OS replicate the whole (parent) process with 'fork' ).