Search code examples
pythonserializationsimplejson

How to make simplejson serializable class


I have a class defined like this

class A:
    def __init__(self):
        self.item1 = None
    def __repr__(self):
        return str(self.__dict__)

when I do:

>>> import simplejson
>>> myA = A()
>>> simplejson.dumps(myA)
TypeError: {'item1': None} is not JSON serializable

I can't find the reason why.

Do I need to add any particular method to A for simplejson to serialize my class object?


Solution

  • You can't serialize arbitrary objects with simplejson. You need to pass a default and object_hook to dump and load. Here's an example:

    class SerializerRegistry(object):
        def __init__(self):
            self._classes = {}
        def add(self, cls):
            self._classes[cls.__module__, cls.__name__] = cls
            return cls
        def object_hook(self, dct):
            module, cls_name = dct.pop('__type__', (None, None))
            if cls_name is not None:
                return self._classes[module, cls_name].from_dict(dct)
            else:
                return dct
        def default(self, obj):
            dct = obj.to_dict()
            dct['__type__'] = [type(obj).__module__,
                               type(obj).__name__]
            return dct
    
    registry = SerializerRegistry()
    
    @registry.add
    class A(object):
        def __init__(self, item1):
            self.item1 = item1
        def __repr__(self):
            return str(self.__dict__)
        def to_dict(self):
            return dict(item1=self.item1)
        @classmethod
        def from_dict(cls, dct):
            return cls(**dct)
    
    s = json.dumps(A(1), default=registry.default)
    a = json.loads(s, object_hook=registry.object_hook)
    

    This results in this:

    >>> s
    '{"item1": 1, "__type__": ["__main__", "A"]}'
    >>> a
    {'item1': 1}
    

    But what you really need is a function default that creates dictionary from the objects that you wish to serialize, and a function object_hook that returns an object (of the correct type) when it is given a dictionary if a dictionary isn't enough. The best approach is to have methods on the serializable classes that create a dict from the object and that construct it back, and also to have a mapping that recognizes to which class the dictionaries belong.

    You can also add an identifier to the classes to be used as an index for _classes. This way you wouldn't have issues if you have to move a class.