Dataclasses serializer

A dataclasses serializer for the Django REST Framework.

Requirements

• Python (3.7+)
• Django (2.0+)
• Django REST Framework (3.9+)

These are the supported Python and package versions. Older versions will probably work as well, but haven't been tested by the author.

Installation

$ 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).

Basic usage

The package provides the DataclassSerializer serializer, defined in the rest_framework_dataclasses.serializers namespace.

from rest_framework_dataclasses.serializers import DataclassSerializer

This 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 dataclass property is specified in the Meta subclass, 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(required=False, 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 = Person

Dataclass 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.

Customize field generation

To customize the generated fields, the DataclassSerializer accepts the following options in the Meta subclass. All options have the same behaviour as the identical options in ModelSerializer.

• dataclass specifies the type of dataclass used by the serializer. This is equivalent to the model option in ModelSerializer.

• fields and exclude can be used to specify which fields should respectively be included and excluded in the serializer. These cannot both be specified.

  The fields option 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 either fields or exclude.

• read_only_fields can be used to mark a subset of fields as read-only.

• extra_kwargs can 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_kwargs name (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 } }
          }

• validators functionality is unchanged.

• depth (as known from ModelSerializer) is not supported, it will always nest infinitely deep.

Additionally,

• Fields marked final (as in PEP 591) are made read-only.

Nesting

Nested dataclasses

If 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 = House

This will serialize as:

>>> serializer = HouseSerializer(instance=house)
>>> serializer.data
{
    'address': 'Main Street 5',
    'owner': { 'name': 'Alice' }
    'residents': [
        { 'name': 'Alice', 'email': 'alice@example.org', ... },
        { 'name': 'Bob', 'email': 'bob@example.org', ... },
        { 'name': 'Charles', 'email': 'charles@example.org', ... }
    ]
}

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
                        }
                    }
                }
            }
        }

Nesting models

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: Company

This 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.

New serializer field types

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.

DefaultDecimalField

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.

EnumField

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).

Advanced usage

• 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 fields option in the Meta class.

• If you don't need to customize the generated fields, DataclassSerializer can also be used directly without creating a subclass. In that case, the dataclass should be specified using the dataclass constructor parameter:

  serializer = DataclassSerializer(data=request.data, dataclass=Person)

• Partial updates are supported by setting the partial argument to True. 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() or update() methods, the dataclass instance passed in the validated_data argument will have the special rest_framework.fields.empty value 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 these empty markers and replace them with the default value by calling the parent update() or create() 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_data property on the serializer has these empty markers stripped as well, and replaced with the default values for not-provided fields. Note that this means you cannot access validated_data on the serializer for partial updates where no data has been provided for fields without a default value, an Exception will be thrown.

Field mappings

So far, field generation is supported for the following types and their subclasses:

• str, bool, int and float.
• date, datetime, time and timedelta from the datetime package.
• decimal.Decimal (max_digits and decimal_places default to None and 2 respectively).
• uuid.UUID
• enum.Enum (mapped to a EnumField)
• typing.Iterable (including typing.List and PEP 585-style generics such as list[int]).
• typing.Mapping (including typing.Dict and PEP 585-style generics such as dict[str, int]).
• typing.Literal (mapped to a ChoiceField).
• django.db.Model

The serializer also supports type variables that have an upper bound or are constrained. Type unions are not supported yet.

For advanced users, the DataclassSerializer also exposes an API that you can override in order to alter how serializer fields are generated:

• The serializer_field_mapping property 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 nested dataclass.
• The serializer_related_field property is the serializer field class that is used for relations to models.
• The serializer_dataclass_field property is the serializer field class that is used for nested dataclasses. If you subclass DataclassSerializer to customize behaviour, you probably want to change this property to use the subclass as well. 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.
• 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_enum_field(), build_literal_field() and build_composite_field() methods are used to process respectively fields, nested models, nested dataclasses, enums, literals, and lists or dictionaries. These can be overridden to change the field generation logic.

Note that until version 1.0 these API's are not declared stable yet and might change if necessary.