jaweson 1.3.0

A safe, modular, format agnostic, serialiser. Provides support for JSON, MsgPack

JAWESON

(JSON|MsgPack) + Awesome serialisation = JAWESON

JAWESON provides a modular de|serialisation framework. The functions themselves are not dependent on any one framework and can be repurposed to any serialisation format that handles dicts, lists, strings, ints and floats.

JAWESON provides the ability to automatically serialise objects with the only change being an inherited jaweson.Serialisable class.

JAWESON avoids using pickle to avoid potential security issues. Should your pickle store (database, s3, etc) become compromised, your system could be tricked into running malicious code.

Avoiding pickle is not without a cost, and that is the need to provide support for non-serialisation friendly types.

An example of using JAWESON to de|serialise JSON:

from jaweson import json
import numpy as np

a = np.array([1,2,3], dtype=np.float32)
j = json.dumps(a)
print j
>>> {"data": "AACAPwAAAEAAAEBA", "shape": [3], "__type__": "ndarray", "dtype": "<f4"}
b = json.loads(j)
(b == a).all()
>>> True

class Test(json.Serialisable):
    def __init__(self):
        self.a = 1

    def modify(self):
        self.a = 2

a = Test()
a.modify()
j = json.dumps(a)
print j
>>> {"a": 2, "__type__": "serialisable", "__class__": "Test"}
b = json.loads(j)

b.a is 2
>>> True

The same example will work with MsgPack if the import is changed from:

from jaweson import json

To:

from jaweson import msgpack as json

Out-of-the-box Support

JAWESON supports serialisation of the following types out-of-the-box:

• default serialisable types (dict, list, string, int, float, null)

• set

• tuple

• np.ndarray

• np.generic

• jaweson.Serialisable

• JSON support

• MSGPack support

JSON Support

Import the JSON module from Jaweson:

from jaweson import json
json.dumps(set([1,2,3]))
>>> {"data": [1, 2, 3], "__type__": "set"}

MSGPack Support

Import the msgpack module from Jaweson:

from jaweson import msgpack as json
json.dumps(set([1,2,3]))
>>> ��data��__type__�set

Automatic Object Serialisation

JAWESON supports object serialisation through the use of a jaweson.Serialisable base class.

This class provides functionality to:

• Automatically register itself for creation.

• Automatically serialise values.

• Automatically construct and deserialise values.

For 99% of cases, you are not required to make any change to your class, except to inherit from jaweson.Serialisable:

from jaweson import json

class MyObject(json.Serialisable):
    def __init__(self):
        self.a = 1

    def modify(self):
        self.a = 2

a = MyObject()
a.modify()
j = json.dumps(a)
print j
>>> {"a": 2, "__type__": "serialisable", "__class__": "MyObject"}
b = json.loads(j)
b.a is 2
>>> True

Should automatical de|serialisation not work, or require custom de|serialisation, parsing can be overriden through the jaweson.Serialisable to_dict and from_dict class methods:

from jaweson import Serialisable

class MyCustomSerialisableObject(Serialisable):
    @classmethod
    def to_dict(cls, obj):
        data = super(MyCustomSerialisableObject, cls).to_dict(obj)
        data['my_value'] = obj.my_other_value

    @classmethod
    def from_dict(cls, jobj):
        obj = super(MyCustomSerialisableObject, cls).from_dict(jobj)
        obj.my_other_value = jobj['my_value']

White|Blacklisting variables is achieved through the class variables __blacklist and __whitelist:

from jaweson import Serialisable

class BlacklistObject(Serialisable):
    __blacklist = ['dont_serialise']
    __whitelist = ['__im_required']

    def __init__(self):
        self.dont_serialise = 5
        self.__im_required = 10

Sometimes you rename classes or need to / from migrate from another format. Over-riding the serialised class name can be achieved by setting the desired name for the __classname variable of the jaweson.Serialiser class:

class NewClass(jaweson.Serialisable):
    __classname = 'OldClass'

    def __init__(self):
        self.a = 1


a = NewClass()
j = jaweson.dumps(a)
print j
{"a": 1, "__type__": "serialisable", "__class__": "OldClass"}

Custom Serialisers

Support for new seralisers can be added by inheriting from the jaweson.Serialiser class.

Classes are automatically registered with the jaweson serialiser when parsed.

The following code is for the built-in Python type serialiser:

from jaweson import Serialiser

class PythonTypeSerialiser(Serialiser):
    python_types = (set, tuple, complex)
    serialised_types = ('set', 'tuple', 'complex')

    def to_dict(self, obj):
        if isinstance(obj, set):
            return {
                '__type__': 'set',
                'data': list(obj),
            }
        if isinstance(obj, tuple):
            return {
                '__type__': 'tuple',
                'data': list(obj)
            }
        if isinstance(obj, complex):
            return {
                '__type__': 'complex',
                'data': obj.__repr__()
            }

        return super(PythonTypeSerialiser, self).to_dict(obj)

    def from_dict(self, jobj):
        if jobj.get('__type__') == 'set':
            return set(jobj['data'])
        if jobj.get('__type__') == 'tuple':
            return tuple(jobj['data'])
        if jobj.get('__type__') == 'complex':
            return complex(jobj['data'])

        return super(PythonTypeSerialiser, self).from_dict(jobj)

Gotchas

Jawson expects unique class names

Having multiple classes with the same name defined will cause the de-serialiser to become confused and fail.

To get around this, assign a string value to the __classname property of the Serialisable class:

class Duplicate(jaweson.Serialisable):
    __classname = 'AnotherDuplicate'

    def __init__(self):
        self.a = 1
        self.classname = "I'm really Duplicate"


class AnotherDuplicate(jaweson.Serialisable):
    __classname = 'Duplicate'

    def __init__(self):
        self.b = 2
        self.classname = "I'm really AnotherDuplicate"

a = Duplicate()
b = AnotherDuplicate()
assert a.a is 1
assert b.b is 2
ja = jaweson.dumps(a)
jb = jaweson.dumps(b)
print ja
{"a": 1, "classname": "I'm really Duplicate", "__type__": "serialisable", "__class__": "AnotherDuplicate"}
print jb
{"b": 2, "classname": "I'm really AnotherDuplicate", "__type__": "serialisable", "__class__": "Duplicate"}

Serialisable does not serialise any variables with ‘__’ in its name

To avoid serialising internal data-structures, Serialisable derived objects will not serialise variables with __ in them by default. This can be over-ridden by defining specific variables in the __whitelist list:

class Obj(jaweson.Serialisable):
    __whitelist = ['__a']
    def __init__(self):
        self.__a = 1

Constructors are not called

If you define a constant in a class constructor:

class Obj(jaweson.Serialisable):
    def __init__(self):
        self.a = 1

And later change the value:

class Obj(jaweson.Serialisable):
    def __init__(self):
        self.a = 2

Variables serialised before the change will still have the value a=1.

To over-come this, over-ride the from_dict method to force the variable to the new value:

class Obj(jaweson.Serialisable):
    @classmethod
    def from_dict(cls, jobj):
        obj = super(Obj, cls).from_dict(jobj)
        obj.a = 2
        return obj

    def __init__(self):
        self.a = 2

JSON does not support dict keys of type int

JSON imposes a limitation that dictionary keys must be strings:

import msgpack
msgpack.loads(msgpack.dumps({1:1,2:2}))
>>> {1: 1, 2: 2}
import json
json.loads(json.dumps({1:1,2:2}))
>>> {u'1': 1, u'2': 2}

Data format

JAWESON stores complex objects in the following structure:

{
    '__type__': '<type name>',
    <other fields>
}

JAWESON implements the following serialisation formats.

numpy.ndarray:

{
    '__type__': 'ndarray',
    'data': '<base 64 encoded data>',
    'dtype': '<numpy dtype>',
    'shape': [<shape>,],
}

numpy.generic:

{
    '__type__': 'ndarray',
    'data': '<base 64 encoded data>',
    'dtype': '<numpy dtype>',
}

set:

{
    '__type__': 'set',
    'data': [<set>],
}

tuple:

{
    '__type__': 'tuple',
    'data': [<tuple>],
}

complex:

{
    '__type__': 'complex',
    'data': '<base 64 encoded data>',
}

jaweson.Serialisable:

{
    '__type__': 'serialisable',
    '__class__': '<class name>',

}

TODO

• datetime serialisation

Authors

• Adam Griffiths