is it possible to ujson.dumps() python class instance (faster deepcopy)

Question

I am trying to make a fast copy of a class instance. cPickle.loads(cPickle.dumps(),-1) works fine, almost 5 times faster than copy.deepcopy, but I read that ujson is much faster than cPickle. I could not get ujson to work with a custom class, is it possible to do so?

example:

import cPickle as pickle
import ujson

class AClass(object):
    def __init__(self):
        print('init')
        self.v = 10
        self.z = [2,3,4]
        self._zdict = dict(zip(self.z,self.z))

a = AClass()
a
#<__main__.AClass at 0x118b1d390>


# does not work with ujson
ua = ujson.dumps(a)
au = ujson.loads(ua)
au
#{u'v': 10, u'z': [2, 3, 4]}


# but works with pickle
pa = pickle.dumps(a)
ap = pickle.loads(pa)
ap
#<__main__.AClass at 0x117460190>

Answer 1

An idea is to define your own protocole, base of the concept described for pickle. Define a __getstate__ and __setsatte__ instance in your class:

class AClass(object):
    def __init__(self, v, z):
        self.v = v
        self.z = z
        self._zdict = dict(zip(self.z, self.z))

    def __repr__(self):
        return repr({'v': self.v, 'z': self.z, '_zdict': self._zdict})

    def __getstate__(self):
        return {'v': self.v, 'z': self.z}

    def __setstate__(self, state):
        self.__dict__.update(state)
        self._zdict = dict(zip(self.z, self.z))

Then, you can define a save() and a load()function like this:

import importlib
import json
import io

def save(instance, dst_file):
    data = {
        'module': instance.__class__.__module__,
        'class': instance.__class__.__name__,
        'state': instance.__getstate__()}
    json.dump(data, dst_file)


def load(src_file):
    obj = json.load(src_file)
    module_name = obj['module']
    mod = importlib.import_module(module_name)
    cls = getattr(mod, obj['class'])
    instance = cls.__new__(cls)
    instance.__setstate__(obj['state'])
    return instance

Simple usage (using a StringIO here instead of a classic file):

a_class = AClass(10, [2, 3, 4])
my_file = io.StringIO()
save(a_class, my_file)

print(my_file.getvalue())
# -> {"module": "__main__", "class": "AClass", "state": {"v": 10, "z": [2, 3, 4]}}

my_file = io.StringIO(my_file.getvalue())
instance = load(my_file)

print(repr(instance))
# -> {'v': 10, 'z': [2, 3, 4], '_zdict': {2: 2, 3: 3, 4: 4}}

Answer 2

ujson isn't serializing the object; it's just encoding its attribute dict as a JSON object. There isn't enough information there to reproduce the original object in it entirety; the most obvious indication is that nothing in the output of ujson.dumps records what class a was an instance of.

The reason usjon is so much faster than cPickle is that cPickle has to do a lot more.