A dataclasses serializer for the Django REST Framework.
- Python (3.8+)
- Django (3.2+)
- Django REST Framework (3.11+)
These are the supported Python and package versions. Older versions will probably work as well, but aren't tested.
$ pip install djangorestframework-dataclasses
This package follows semantic versioning. See CHANGELOG for breaking changes and new features, and LICENSE for the complete license (BSD-3-clause).
The package provides the DataclassSerializer serializer, defined in the rest_framework_dataclasses.serializers
namespace.
from rest_framework_dataclasses.serializers import DataclassSerializerThis serializer provides a shortcut that lets you automatically create a Serializer class with fields that
correspond to the fields on a dataclass. In usage, the DataclassSerializer is the same as a regular Serializer
class, except that:
- It will automatically generate fields for you, based on the declaration in the dataclass.
- To make this possible it requires that a
dataclassproperty is specified in theMetasubclass, with as value a dataclass that has type annotations. - It includes default implementations of
.create()and.update().
For example, define a dataclass as follows:
@dataclass
class Person:
name: str
email: str
alive: bool
gender: typing.Literal['male', 'female']
birth_date: typing.Optional[datetime.date]
phone: typing.List[str]
movie_ratings: typing.Dict[str, int]The serializer for this dataclass can now trivially be defined without having to duplicate all fields:
class PersonSerializer(DataclassSerializer):
class Meta:
dataclass = Person
# is equivalent to
class PersonSerializer(Serializer):
name = fields.CharField()
email = fields.CharField()
alive = fields.BooleanField()
gender = fields.ChoiceField(choices=['male', 'female'])
birth_date = fields.DateField(allow_null=True)
phone = fields.ListField(child=fields.CharField())
movie_ratings = fields.DictField(child=fields.IntegerField())You can add extra fields or override default fields by declaring them explicitly on the class, just as you would for a
regular Serializer class. This allows to specify extra field options or change a field type.
class PersonSerializer(Serializer):
email = fields.EmailField()
class Meta:
dataclass = PersonDataclass serializers behave in the same way and can be used in the same places as the built-in serializers from Django
REST Framework: you can retrieve the serialized representation using the .data property, and the deserialized
dataclass instance using the .validated_data property. Furthermore, the save() method is implemented to create
or update an existing dataclass instance. You can find more information on serializer usage in the
Django REST Framework documentation.
Note that this usage pattern is very similar to that of the built-in ModelSerializer. This is intentional, with the
whole API modelled after that of ModelSerializer. Most features and behaviour known from ModelSerializer applies
to dataclass serializers as well.
Currently, automatic field generation is supported for the following types and their subclasses:
str,bool,intandfloat.date,datetime,timeandtimedeltafrom thedatetimepackage.decimal.Decimal(max_digitsanddecimal_placesdefault toNoneand2respectively).uuid.UUIDenum.Enum(mapped to aEnumField)typing.Iterable(includingtyping.Listand PEP 585-style generics such aslist[int]).typing.Mapping(includingtyping.Dictand PEP 585-style generics such asdict[str, int]).typing.Literal(mapped to aChoiceField).typing.Union(mapped to aUnionField, including PEP 604-style unions such asstr | int, see UnionField section below for more information).django.db.Model
The serializer also supports type variables that have an upper bound or are constrained.
The auto-generated serializer fields are configured based on type qualifiers in the dataclass (these can be mixed):
- Fields with a default value (factory) are marked as optional on the serializer (
required=False). This means that these fields don't need to be supplied during deserialization. - Fields marked as nullable through
typing.Optional,typing.Union[X, None]orX | None(PEP 604) are marked as nullable on the serializer (allow_null=True). This means thatNoneis accepted as a valid value during deserialization. - Fields marked as final through
typing.Final(as in PEP 591) are marked as read-only on the serializer (read_only=True).
@dataclass
class Person:
birth_date: typing.Optional[datetime.date]
alive: bool = True
species: typing.Final[str] = 'Human'
# the autogenerated serializer will be equal to
class PersonSerializer(Serializer):
birth_date = fields.DateField(allow_null=True)
alive = fields.BooleanField(required=False)
species = fields.CharField(read_only=True)Besides overriding fields by declaring them explicitly on the serializer, you can also change or override the generated serializer field using metadata on the dataclass field. Currently, two keys are recognized in this dictionary:
serializer_fieldcan be used to replace the auto-generated field with a user-supplied one. Should contain an instance of a field, not a field type.serializer_kwargscan be used to specify arbitrary additional keyword arguments for the generated field. Manually specified arguments will have precedence over generated arguments (so e.g. by supplying{required: True}, a field with a default value can be made required).
@dataclasses.dataclass
class Person:
email: str = dataclasses.field(metadata={'serializer_field': fields.EmailField()})
age: int = dataclasses.field(metadata={'serializer_kwargs': {'min_value': 0}})
# the autogenerated serializer will be equal to
class PersonSerializer(Serializer):
email = fields.EmailField()
age = fields.IntegerField(min_value=0)To further customize the serializer, the DataclassSerializer accepts the following options in the Meta
subclass. All options have the same behaviour as the identical options in ModelSerializer.
dataclassspecifies the type of dataclass used by the serializer. This is equivalent to themodeloption inModelSerializer.fieldsandexcludecan be used to specify which fields should respectively be included and excluded in the serializer. These cannot both be specified.The
fieldsoption accepts the magic value__all__to specify that all fields on the dataclass should be used. This is also the default value, so it is not mandatory to specify eitherfieldsorexclude.read_only_fieldscan be used to mark a subset of fields as read-only.extra_kwargscan be used to specify arbitrary additional keyword arguments on fields. This can be useful to extend or change the autogenerated field without explicitly declaring the field on the serializer. This option should be a dictionary, mapping field names to a dictionary of keyword arguments.If the autogenerated field is a composite field (a list or dictionary), the arguments are applied to the composite field. To add keyword arguments to the composite field's child field (that is, the field used for the items in the list or dictionary), they should be specified as a nested dictionary under the
child_kwargsname (see Nested dataclasses section below for an example).class PersonSerializer(DataclassSerializer): class Meta: extra_kwargs = { 'height': { 'decimal_places': 1 }, 'movie_ratings': { 'child_kwargs': { 'min_value': 0, 'max_value': 10 } } }
validatorsfunctionality is unchanged.depth(as known fromModelSerializer) is not supported, it will always nest infinitely deep.
Additionally, it is possible to change the default behaviour of the DataclassSerializer by setting one of these
properties on the class:
- The
serializer_field_mappingproperty contains a dictionary that maps types to REST framework serializer classes. You can override or extend this mapping to change the serializer field classes that are used for fields based on their type. This dictionary also accepts dataclasses as keys to change the serializer used for a nested dataclass. - The
serializer_related_fieldproperty is the serializer field class that is used for relations to models. - The
serializer_union_fieldproperty is the serializer field class that is used for union types. - The
serializer_dataclass_fieldproperty is the serializer field class that is used for nested dataclasses. Note that since Python process the class body before it defines the class, this property is implemented using the property decorator to allow it to reference the containing class.
Finally, you can create a subclass that overrides methods of the DataclassSerializer. The field generation is
controlled by the following methods, which are considered a stable part of the API:
- The
build_unknown_field()method is called to create serializer fields for dataclass fields that are not understood. By default this just throws an error, but you can extend this with custom logic to create serializer fields. - The
build_property_field()method is called to create serializer fields for methods. By default this creates a read-only field with the method return value. - The
build_standard_field(),build_relational_field(),build_dataclass_field(),build_union_field(),build_enum_field(),build_literal_field()andbuild_composite_field()methods are used to process respectively fields, nested models, nested dataclasses, union types, enums, literals, and lists or dictionaries. These can be overridden to change the field generation logic.
Note that when creating a subclass of DataclassSerializer, most likely you will want to set the
serializer_dataclass_field property to the subclass, so that any nested dataclasses are serialized using the
subclass as well.
class CustomDataclassSerializer(DataclassSerializer):
@property
def serializer_dataclass_field(self):
return CustomDataclassSerializer
# Implement additional and/or override existing methods hereIf your dataclass has a field that also contains a dataclass instance, the DataclassSerializer will automatically
create another DataclassSerializer for that field, so that its value will be nested. This also works for dataclasses
contained in lists or dictionaries, or even several layers deep.
@dataclass
class House:
address: str
owner: Person
residents: typing.List[Person]
class HouseSerializer(DataclassSerializer):
class Meta:
dataclass = HouseThis will serialize as:
>>> serializer = HouseSerializer(instance=house)
>>> serializer.data
{
'address': 'Main Street 5',
'owner': { 'name': 'Alice' }
'residents': [
{ 'name': 'Alice', 'email': '[email protected]', ... },
{ 'name': 'Bob', 'email': '[email protected]', ... },
{ 'name': 'Charles', 'email': '[email protected]', ... }
]
}This does not give the ability to customize the field generation of the nested dataclasses. If that is needed, you
should declare the serializer to be used for the nested field explicitly. Alternatively, you could use the
extra_kwargs option to provide arguments to fields belonging to the nested dataclasses. Consider the following:
@dataclass
class Transaction:
amount: Decimal
account_number: str
@dataclass
class Company:
sales: List[Transaction]In order to tell DRF to give 2 decimal places to the transaction account number, write the serializer as follows:
class CompanySerializer(DataclassSerializer):
class Meta:
dataclass = Company
extra_kwargs = {
'sales': {
# Arguments here are for the ListField generated for the sales field on Company
'min_length': 1, # requires at least 1 item to be present in the sales list
'child_kwargs': {
# Arguments here are passed to the DataclassSerializer for the Transaction dataclass
'extra_kwargs': {
# Arguments here are the extra arguments for the fields in the Transaction dataclass
'amount': {
'max_digits': 6,
'decimal_places': 2
}
}
}
}
}Likewise, if your dataclass has a field that contains a Django model, the DataclassSerializer will automatically
generate a relational field for you.
class Company(models.Model):
name = models.CharField()
@dataclass
class Person:
name: str
employer: CompanyThis will serialize as:
>>> serializer = PersonSerializer(instance=user)
>>> print(repr(serializer))
PersonSerializer():
name = fields.CharField()
employer = fields.PrimaryKeyRelatedField(queryset=Company.objects.all())
>>> serializer.data
{
"name": "Alice",
"employer": 1
}If you want to nest the model in the serialized representation, you should specify the model serializer to be used by declaring the field explicitly.
If you prefer to use hyperlinks to represent relationships rather than primary keys, in the same package you can find
the HyperlinkedDataclassSerializer class: it generates a HyperlinkedRelatedField instead of a
PrimaryKeyRelatedField.
To handle some types for which DRF does not ship a serializer field, some new serializer field types are shipped in the
rest_framework_dataclasses.fields namespace. These fields can be used independently of the DataclassSerializer
as well.
A subclass of DecimalField that defaults max_digits to None and decimal_places to 2. Used to represent
decimal values which there is no explicit field configured.
A subclass of ChoiceField to represent Python enumerations. The enumeration members can be represented by either their name or value. The member name is used as display name.
Signature: EnumField(enum_class, by_name=False)
enum_class: The enumeration class.by_name: Whether members are represented by their value (False) or name (True).
A subclass of ListField that can return values that aren't of type list, such as set.
Signature: IterableField(container=list)
container: The type of the returned iterable. Must have a constructor that accepts a single parameter of typelist, containing the values for the iterable.
A subclass of DictField that can return values that aren't of type dict, such as collections.OrderedDict.
Signature: MappingField(container=dict)
container: The type of the returned mapping. Must have a constructor that accepts a single parameter of typedict, containing the values for the mapping.
A field that can serialize and deserialize values of multiple types (i.e. values of a union type). The serialized
representation of this field includes an extra discriminator field (by default named type) that indicates the actual
type of the value.
@dataclass
class A:
a: str
@dataclass
class B:
b: int
@dataclass
class Response:
obj: A | B
class ResponseSerializer(DataclassSerializer):
class Meta:
dataclass = Response>>> response = Response(obj=A('hello'))
>>> serializer = ResponseSerializer(instance=response)
>>> serializer.data
{
'obj': {'type': 'A', 'a': 'hello'}
}
>>> deserializer = ResponseSerializer(data={'obj': {'type': 'B', 'b': 42}})
>>> deserializer.is_valid()
True
>>> deserializer.validated_data
Response(obj=B(b=42))The name of the discriminator field can be changed by setting the discriminator_field_name keyword argument for the
field:
@dataclass
class Response:
obj: A | B = dataclasses.field(metadata={'serializer_kwargs': {'discriminator_field_name': 'a_or_b'}})
# or:
class ResponseSerializer(DataclassSerializer):
class Meta:
dataclass = Response
extra_kwargs = {
'obj': {'discriminator_field_name': 'a_or_b'}
}Unions containing a type that does not serialize to a mapping (e.g. an integer or string) can be serialized by enabling
nesting with the nest_value keyword argument:
@dataclass
class Response:
amount: int | float
class ResponseSerializer(DataclassSerializer):
class Meta:
dataclass = Response
extra_kwargs = {
'amount': {'nest_value': True}
}>>> response = Response(amount=42)
>>> serializer = ResponseSerializer(instance=response)
>>> serializer.data
{
'amount': {'type': 'int', 'value': 42}
}Signature: UnionField(child_fields, nest_value=False, discriminator_field_name=None, value_field_name=None).
child_fields: A dictionary mapping the individual types to the serializer field to be used for them.nest_value: Whether the value should be put under a key (True), or merged directly into the serialized representation of this field (False). This is disabled by default, and should usually only be set toTrueif any of the union member types is a primitive.discriminator_field_name: Name of the discriminator field, defaults totype.value_field_name: Name of the field under which values are nested ifnest_valueis used defaults tovalue.
The values used in the discriminator field can be changed by subclassing UnionField and overriding the
get_discriminator(self, type) method. The lone argument to this method is one of the member types of union (a key
from the child_fields parameter), and it should return the appropriate string to be used in the discriminator field
for values of this type.
The output of methods or properties on the dataclass can be included as a (read-only) field in the serialized state by adding their name to the
fieldsoption in theMetaclass.If you don't need to customize the generated fields,
DataclassSerializercan also be used directly without creating a subclass. In that case, the dataclass should be specified using thedataclassconstructor parameter:serializer = DataclassSerializer(data=request.data, dataclass=Person)
Partial updates are supported by setting the
partialargument toTrue. Nested dataclasses will also be partially updated, but nested fields and dictionaries will be replaced in full with the supplied value:@dataclass class Company: name: str location: Optional[str] = None @dataclass class Person: name: str current_employer: Company past_employers: List[Company] alice = Person(name='Alice', current_employer=Company('Acme Corp.', 'New York City'), past_employers=[Company('PSF', 'Delaware'), Company('Ministry of Silly Walks', 'London')]) data = {'current_employer': {'location': 'Los Angeles'}, 'past_employers': [{'name': 'OsCorp', 'location': 'NYC'}]} >>> serializer = PersonSerializer(partial=True, instance=alice, data=data) >>> print(serializer.save()) Person(name='Alice', current_employer=Company('Acme Corp.', 'Los Angeles'), past_employers=[Company(name='OsCorp', location='NYC')])
If you override the
create()orupdate()methods, the dataclass instance passed in thevalidated_dataargument will have the specialrest_framework.fields.emptyvalue for any fields for which no data was provided. This is required to distinguish between not-provided fields and fields with the default value, as needed for (both regular and partial) updates. You can get rid of theseemptymarkers and replace them with the default value by calling the parentupdate()orcreate()methods - this is the only thing they do.class CompanySerializer(DataclassSerializer): def create(self, validated_data): instance = super(CompanySerializer, self).create(validated_data) # if no value is provided for location, these will both hold assert validated_data.location == rest_framework.fields.empty assert instance.location is None # None is the default value of Company.location (see previous example)
The
validated_dataproperty on the serializer has theseemptymarkers stripped as well, and replaced with the default values for not-provided fields. Note that this means you cannot accessvalidated_dataon the serializer for partial updates where no data has been provided for fields without a default value, an Exception will be thrown.
Starting from version 0.21.2, drf-spectacular natively supports DataclassSerializer. For previous versions, you
can include the extension in your project manually. You don't need to configure it, but you do need to import the
module that contains the extension.
When using a type checker such as mypy, please ensure that the djangorestframework-stubs package is installed. The
type hints for this library depend on the type hints for DRF being available to validate successfully, and might
otherwise generate some seemingly bizarre mypy errors.
The DataclassSerializer class is generic, and must be parameterized with the dataclass type to have correct types on
its properties and methods:
class PersonSerializer(DataclassSerializer[Person]):
class Meta:
dataclass = Person