Customizing encoding and decoding

Both the encoder and decoder can be customized to support a wider range of types.

On the encoder side, this is accomplished by passing a callback as the default constructor argument. This callback will receive an object that the encoder could not serialize on its own. The callback should then return a value that the encoder can serialize on its own, although the return value is allowed to contain objects that also require the encoder to use the callback, as long as it won’t result in an infinite loop.

On the decoder side, you have two options: tag_hook and object_hook. The former is called by the decoder to process any semantic tags that have no predefined decoders. The latter is called for any newly decoded dict objects, and is mostly useful for implementing a JSON compatible custom type serialization scheme. Unless your requirements restrict you to JSON compatible types only, it is recommended to use tag_hook for this purpose.

Using the CBOR tags for custom types

The most common way to use default is to call encode() to add a custom tag in the data stream, with the payload as the value:

class Point:
    def __init__(self, x, y):
        self.x = x
        self.y = y

def default_encoder(encoder, value):
    # Tag number 4000 was chosen arbitrarily
    encoder.encode(CBORTag(4000, [value.x, value.y]))

The corresponding tag_hook would be:

def tag_hook(decoder, tag, shareable_index=None):
    if tag.tag != 4000:
        return tag

    # tag.value is now the [x, y] list we serialized before
    return Point(*tag.value)

Using dicts to carry custom types

The same could be done with object_hook, except less efficiently:

def default_encoder(encoder, value):
    encoder.encode(dict(typename='Point', x=value.x, y=value.y))

def object_hook(decoder, value):
    if value.get('typename') != 'Point':
        return value

    return Point(value['x'], value['y'])

You should make sure that whatever way you decide to use for telling apart your “specially marked” dicts from arbitrary data dicts won’t mistake on for the other.

Value sharing with custom types

In order to properly encode and decode cyclic references with custom types, some special care has to be taken. Suppose you have a custom type as below, where every child object contains a reference to its parent and the parent contains a list of children:

from cbor2 import dumps, loads, shareable_encoder, CBORTag

class MyType:
    def __init__(self, parent=None):
        self.parent = parent
        self.children = []
        if parent:

This would not normally be serializable, as it would lead to an endless loop (in the worst case) and raise some exception (in the best case). Now, enter CBOR’s extension tags 28 and 29. These tags make it possible to add special markers into the data stream which can be later referenced and substituted with the object marked earlier.

To do this, in default hooks used with the encoder you will need to use the shareable_encoder() decorator on your default hook function. It will automatically automatically add the object to the shared values registry on the encoder and prevent it from being serialized twice (instead writing a reference to the data stream):

def default_encoder(encoder, value):
    # The state has to be serialized separately so that the decoder would have a chance to
    # create an empty instance before the shared value references are decoded
    serialized_state = encoder.encode_to_bytes(value.__dict__)
    encoder.encode(CBORTag(3000, serialized_state))

On the decoder side, you will need to initialize an empty instance for shared value lookup before the object’s state (which may contain references to it) is decoded. This is done with the set_shareable() method:

def tag_hook(decoder, tag, shareable_index=None):
    # Return all other tags as-is
    if tag.tag != 3000:
        return tag

    # Create a raw instance before initializing its state to make it possible for cyclic
    # references to work
    instance = MyType.__new__(MyType)
    decoder.set_shareable(shareable_index, instance)

    # Separately decode the state of the new object and then apply it
    state = decoder.decode_from_bytes(tag.value)
    return instance

You could then verify that the cyclic references have been restored after deserialization:

parent = MyType()
child1 = MyType(parent)
child2 = MyType(parent)
serialized = dumps(parent, default=default_encoder, value_sharing=True)

new_parent = loads(serialized, tag_hook=tag_hook)
assert new_parent.children[0].parent is new_parent
assert new_parent.children[1].parent is new_parent

Decoding Tagged items as keys

Since the CBOR specification allows any type to be used as a key in the mapping type, the decoder provides a flag that indicates it is expecting an immutable (and by implication hashable) type. If your custom class cannot be used this way you can raise an exception if this flag is set:

def tag_hook(decoder, tag, shareable_index=None):
    if tag.tag != 3000:
        return tag

    if decoder.immutable:
        raise CBORDecodeException('MyType cannot be used as a key or set member')

    return MyType(*tag.value)

An example where the data could be used as a dict key:

from collections import namedtuple

Pair = namedtuple('Pair', 'first second')

def tag_hook(decoder, tag, shareable_index=None):
    if tag.tag != 4000:
        return tag

    return Pair(*tag.value)

The object_hook can check for the immutable flag in the same way.