PyTorch optimizer __getstate__ implementation

1

This is a somewhat advanced question, but I am going through the source code for the optimizer, and I found something interesting about the pickle-specific function __getstate__:

    def __getstate__(self):
        return {
            'defaults': self.defaults,
            'state': self.state,
            'param_groups': self.param_groups,
        }

    def __setstate__(self, state):
        self.__dict__.update(state)

My limited understanding of pickling/serialization is that __getstate__ is going to just return the state of the class in whatever implemented form, often by just returning a copy of the dictionary of attributes self.__dict__. Here, it is returning a defined dictionary of the attributes self.defaults, self.state, and self.param_groups. But as far as I can tell, these are the only attributes of the optimizer class. So why can’t __getstate__ just return self.__dict__.copy()?