cloudpickle and cloudpickle_fast differ in the sense that the cloudpickle_fast's CloudPickler only has to reduce objects (the actual saving is taken care of by logic coded in the pickle.Pickler class), whereas the cloudpickle.CloudPickler reduces AND saves objects in one single method (save_{object_type}(self, obj))

Right now, the cloudpickle.CloudPickler reducing parts of the save_{object_type} methods do not use the reduce_{object_type} reducers implemented in cloudpickle_fast.
To avoid the need to write twice any change we make to a reducer, it would be great to refactor most cloudpickle.CloudPickler.save_* methods, to make them look like this

def save_{object_type}(self, obj): rv = reduce_{object_type}(obj) self.save_reduce(*rv, obj=obj)

reduce_{object_type} being used both by cloudpickle and cloudpickle_fast.