Dataclass Wizard

Full documentation is available at Read The Docs. (Installation)

https://img.shields.io/pypi/v/dataclass-wizard.svg https://img.shields.io/conda/vn/conda-forge/dataclass-wizard.svg https://img.shields.io/pypi/pyversions/dataclass-wizard.svg https://github.com/rnag/dataclass-wizard/actions/workflows/dev.yml/badge.svg Documentation Status Updates

This library provides a set of simple, yet elegant wizarding tools for interacting with the Python dataclasses module.

The primary use is as a fast serialization framework that enables dataclass instances to be converted to/from JSON; this works well in particular with a nested dataclass model.


Behold, the power of the Dataclass Wizard:

>>> from __future__ import annotations
>>> from dataclasses import dataclass, field
>>> from dataclass_wizard import JSONWizard
...
>>> @dataclass
... class MyClass(JSONWizard):
...     my_str: str | None
...     is_active_tuple: tuple[bool, ...]
...     list_of_int: list[int] = field(default_factory=list)
...
>>> string = """
... {
...   "my_str": 20,
...   "ListOfInt": ["1", "2", 3],
...   "isActiveTuple": ["true", false, 1]
... }
... """
...
>>> instance = MyClass.from_json(string)
>>> instance
MyClass(my_str='20', is_active_tuple=(True, False, True), list_of_int=[1, 2, 3])
>>> instance.to_json()
'{"myStr": "20", "isActiveTuple": [true, false, true], "listOfInt": [1, 2, 3]}'
>>> instance == MyClass.from_dict(instance.to_dict())
True

Installation

The Dataclass Wizard library is available on PyPI, and can be installed with pip:

$ pip install dataclass-wizard

Alternatively, this library is available on conda under the conda-forge channel:

$ conda install dataclass-wizard -c conda-forge

The dataclass-wizard library officially supports Python 3.6 or higher.

Features

Here are the supported features that dataclass-wizard currently provides:

  • JSON/YAML (de)serialization: marshal dataclasses to/from JSON, YAML, and Python dict objects.

  • Field properties: support for using properties with default values in dataclass instances.

  • JSON to Dataclass generation: construct a dataclass schema with a JSON file or string input.

Wizard Mixins

In addition to the JSONWizard, here are a few extra Mixin classes that might prove quite convenient to use.

  • JSONListWizard – Extends JSONWizard to return Container – instead of list – objects where possible.

  • JSONFileWizard – Makes it easier to convert dataclass instances from/to JSON files on a local drive.

  • YAMLWizard – Provides support to convert dataclass instances to/from YAML, using the default PyYAML parser.

Supported Types

The Dataclass Wizard library provides inherent support for standard Python collections such as list, dict and set, as well as most Generics from the typing module, such as Union and Any. Other commonly used types such as Enum, defaultdict, and date and time objects such as datetime are also natively supported.

For a complete list of the supported Python types, including info on the load/dump process for special types, check out the Supported Types section in the docs.

Usage and Examples

Using the built-in JSON marshalling support for dataclasses:

Note: The following example should work in Python 3.7+ with the included __future__ import.

from __future__ import annotations  # This can be removed in Python 3.10+

from dataclasses import dataclass, field
from datetime import date
from enum import Enum

from dataclass_wizard import JSONWizard


@dataclass
class Data(JSONWizard):

    class _(JSONWizard.Meta):
        # Sets the target key transform to use for serialization;
        # defaults to `camelCase` if not specified.
        key_transform_with_dump = 'LISP'

    a_sample_bool: bool
    values: list[Inner] = field(default_factory=list)


@dataclass
class Inner:
    vehicle: Car | None
    my_dates: dict[int, date]


class Car(Enum):
    SEDAN = 'BMW Coupe'
    SUV = 'Toyota 4Runner'
    JEEP = 'Jeep Cherokee'


def main():

    my_dict = {
        'values': [
            {
                'vehicle': 'Toyota 4Runner',
                'My-Dates': {'123': '2023-01-31'}
            },
            {
                'vehicle': None,
                'my_dates': {}
            }
        ],
        'aSampleBool': 'TRUE'
    }

    # De-serialize (a JSON string or dictionary data) into a `Data` instance.
    data = Data.from_dict(my_dict)

    print(repr(data))
    # > Data(a_sample_bool=True, values=[Inner(vehicle=<Car.SUV: 'Toyota 4Runner'>, ...)])

    # assert enums values are as expected
    assert data.values[0].vehicle is Car.SUV

    print(data.to_json(indent=2))
    # {
    #   "a-sample-bool": true,
    #   "values": [
    #     {
    #       "vehicle": "Toyota 4Runner",
    #       "my-dates": {
    #         "123": "2023-01-31"
    #       },
    #   ...

    # True
    assert data == data.from_json(data.to_json())

if __name__ == '__main__':
    main()

… and with the property_wizard, which provides support for field properties with default values in dataclasses:

from __future__ import annotations  # This can be removed in Python 3.10+

from dataclasses import dataclass, field
from typing_extensions import Annotated

from dataclass_wizard import property_wizard


@dataclass
class Vehicle(metaclass=property_wizard):
    # Note: The example below uses the default value from the `field` extra in
    # the `Annotated` definition; if `wheels` were annotated as `int | str`,
    # it would default to 0, because `int` appears as the first type argument.
    #
    # Any right-hand value assigned to `wheels` is ignored as it is simply
    # re-declared by the property; here it is simply omitted for brevity.
    wheels: Annotated[int | str, field(default=4)]

    # This is a shorthand version of the above; here an IDE suggests
    # `_wheels` as a keyword argument to the constructor method, though
    # it will actually be named as `wheels`.
    # _wheels: int | str = 4

    @property
    def wheels(self) -> int:
        return self._wheels

    @wheels.setter
    def wheels(self, wheels: int | str):
        self._wheels = int(wheels)


if __name__ == '__main__':
    v = Vehicle()
    print(v)
    # prints:
    #   Vehicle(wheels=4)

    v = Vehicle(wheels=3)
    print(v)

    v = Vehicle('6')
    print(v)

    assert v.wheels == 6, 'The constructor should use our setter method'

    # Confirm that we go through our setter method
    v.wheels = '123'
    assert v.wheels == 123

… or generate a dataclass schema for JSON input, via the wiz-cli tool:

$ echo '{"myFloat": "1.23", "Products": [{"created_at": "2021-11-17"}]}' | wiz gs - my_file

# Contents of my_file.py
from dataclasses import dataclass
from datetime import date
from typing import List, Union

from dataclass_wizard import JSONWizard


@dataclass
class Data(JSONWizard):
    """
    Data dataclass

    """
    my_float: Union[float, str]
    products: List['Product']


@dataclass
class Product:
    """
    Product dataclass

    """
    created_at: date

JSON Marshalling

JSONSerializable (aliased to JSONWizard) is a Mixin class which provides the following helper methods that are useful for serializing (and loading) a dataclass instance to/from JSON, as defined by the AbstractJSONWizard interface.

Method

Example

Description

from_json

item = Product.from_json(string)

Converts a JSON string to an instance of the dataclass, or a list of the dataclass instances.

from_list

list_of_item = Product.from_list(l)

Converts a Python list object to a list of the dataclass instances.

from_dict

item = Product.from_dict(d)

Converts a Python dict object to an instance of the dataclass.

to_dict

d = item.to_dict()

Converts the dataclass instance to a Python dict object that is JSON serializable.

to_json

string = item.to_json()

Converts the dataclass instance to a JSON string representation.

list_to_json

string = Product.list_to_json(list_of_item)

Converts a list of dataclass instances to a JSON string representation.

Additionally, it adds a default __str__ method to subclasses, which will pretty print the JSON representation of an object; this is quite useful for debugging purposes. Whenever you invoke print(obj) or str(obj), for example, it’ll call this method which will format the dataclass object as a prettified JSON string. If you prefer a __str__ method to not be added, you can pass in str=False when extending from the Mixin class as mentioned here.

Note that the __repr__ method, which is implemented by the dataclass decorator, is also available. To invoke the Python object representation of the dataclass instance, you can instead use repr(obj) or f'{obj!r}'.

To mark a dataclass as being JSON serializable (and de-serializable), simply sub-class from JSONSerializable as shown below. You can also extend from the aliased name JSONWizard, if you prefer to use that instead.

Check out a more complete example of using the JSONSerializable Mixin class.

No Inheritance Needed

It is important to note that the main purpose of sub-classing from JSONWizard Mixin class is to provide helper methods like from_dict and to_dict, which makes it much more convenient and easier to load or dump your data class from and to JSON.

That is, it’s meant to complement the usage of the dataclass decorator, rather than to serve as a drop-in replacement for data classes, or to provide type validation for example; there are already excellent libraries like pydantic that provide these features if so desired.

However, there may be use cases where we prefer to do away with the class inheritance model introduced by the Mixin class. In the interests of convenience and also so that data classes can be used as is, the Dataclass Wizard library provides the helper functions fromlist and fromdict for de-serialization, and asdict for serialization. These functions also work recursively, so there is full support for nested dataclasses – just as with the class inheritance approach.

Here is an example to demonstrate the usage of these helper functions:

Note

As of v0.18.0, the Meta config for the main dataclass will cascade down and be merged with the Meta config (if specified) of each nested dataclass. To disable this behavior, you can pass in recursive=False to the Meta config.

from __future__ import annotations

from dataclasses import dataclass, field
from datetime import datetime, date

from dataclass_wizard import fromdict, asdict, DumpMeta


@dataclass
class A:
    created_at: datetime
    list_of_b: list[B] = field(default_factory=list)


@dataclass
class B:
    my_status: int | str
    my_date: date | None = None


source_dict = {'createdAt': '2010-06-10 15:50:00Z',
               'List-Of-B': [
                   {'MyStatus': '200', 'my_date': '2021-12-31'}
               ]}

# De-serialize the JSON dictionary object into an `A` instance.
a = fromdict(A, source_dict)

print(repr(a))
# A(created_at=datetime.datetime(2010, 6, 10, 15, 50, tzinfo=datetime.timezone.utc),
#   list_of_b=[B(my_status='200', my_date=datetime.date(2021, 12, 31))])

# Set an optional dump config for the main dataclass, for example one which
# converts converts date and datetime objects to a unix timestamp (as an int)
#
# Note that `recursive=True` is the default, so this Meta config will be
# merged with the Meta config (if specified) of each nested dataclass.
DumpMeta(marshal_date_time_as='TIMESTAMP',
         key_transform='SNAKE',
         # Finally, apply the Meta config to the main dataclass.
         ).bind_to(A)

# Serialize the `A` instance to a Python dict object.
json_dict = asdict(a)

expected_dict = {'created_at': 1276185000, 'list_of_b': [{'my_status': '200', 'my_date': 1640926800}]}

print(json_dict)
# Assert that we get the expected dictionary object.
assert json_dict == expected_dict

Custom Key Mappings

If you ever find the need to add a custom mapping of a JSON key to a dataclass field (or vice versa), the helper function json_field – which can be considered an alias to dataclasses.field() – is one approach that can resolve this.

Example below:

from dataclasses import dataclass

from dataclass_wizard import JSONSerializable, json_field


@dataclass
class MyClass(JSONSerializable):

    my_str: str = json_field('myString1', all=True)


# De-serialize a dictionary object with the newly mapped JSON key.
d = {'myString1': 'Testing'}
c = MyClass.from_dict(d)

print(repr(c))
# prints:
#   MyClass(my_str='Testing')

# Assert we get the same dictionary object when serializing the instance.
assert c.to_dict() == d

Extending from Meta

Looking to change how date and datetime objects are serialized to JSON? Or prefer that field names appear in snake case when a dataclass instance is serialized?

The inner Meta class allows easy configuration of such settings, as shown below; and as a nice bonus, IDEs should be able to assist with code completion along the way.

Note

As of v0.18.0, the Meta config for the main dataclass will cascade down and be merged with the Meta config (if specified) of each nested dataclass. To disable this behavior, you can pass in recursive=False to the Meta config.

from dataclasses import dataclass
from datetime import date

from dataclass_wizard import JSONWizard
from dataclass_wizard.enums import DateTimeTo


@dataclass
class MyClass(JSONWizard):

    class _(JSONWizard.Meta):
        marshal_date_time_as = DateTimeTo.TIMESTAMP
        key_transform_with_dump = 'SNAKE'

    my_str: str
    my_date: date


data = {'my_str': 'test', 'myDATE': '2010-12-30'}

c = MyClass.from_dict(data)

print(repr(c))
# prints:
#   MyClass(my_str='test', my_date=datetime.date(2010, 12, 30))

string = c.to_json()
print(string)
# prints:
#   {"my_str": "test", "my_date": 1293685200}

Other Uses for Meta

Here are a few additional use cases for the inner Meta class. Note that a full list of available settings can be found in the Meta section in the docs.

Debug Mode

Enables additional (more verbose) log output. For example, a message can be logged whenever an unknown JSON key is encountered when from_dict or from_json is called.

This also results in more helpful error messages during the JSON load (de-serialization) process, such as when values are an invalid type – i.e. they don’t match the annotation for the field. This can be particularly useful for debugging purposes.

Note

There is a minor performance impact when DEBUG mode is enabled; for that reason, I would personally advise against enabling this in a production environment.

Handle Unknown JSON Keys

The default behavior is to ignore any unknown or extraneous JSON keys that are encountered when from_dict or from_json is called, and emit a “warning” which is visible when debug mode is enabled (and logging is properly configured). An unknown key is one that does not have a known mapping to a dataclass field.

However, we can also raise an error in such cases if desired. The below example demonstrates a use case where we want to raise an error when an unknown JSON key is encountered in the load (de-serialization) process.

import logging
from dataclasses import dataclass

from dataclass_wizard import JSONWizard
from dataclass_wizard.errors import UnknownJSONKey


# Sets up application logging if we haven't already done so
logging.basicConfig(level='INFO')


@dataclass
class Container(JSONWizard):

    class _(JSONWizard.Meta):
        # True to enable Debug mode for additional (more verbose) log output.
        debug_enabled = True
        # True to raise an class:`UnknownJSONKey` when an unmapped JSON key is
        # encountered when `from_dict` or `from_json` is called. Note that by
        # default, this is also recursively applied to any nested dataclasses.
        raise_on_unknown_json_key = True

    element: 'MyElement'


@dataclass
class MyElement:
    my_str: str
    my_float: float


d = {
    'element': {
        'myStr': 'string',
        'my_float': '1.23',
        # Notice how this key is not mapped to a known dataclass field!
        'my_bool': 'Testing'
    }
}

# Try to de-serialize the dictionary object into a `MyClass` object.
try:
    c = Container.from_dict(d)
except UnknownJSONKey as e:
    print('Received error:', type(e).__name__)
    print('Class:', e.class_name)
    print('Unknown JSON key:', e.json_key)
    print('JSON object:', e.obj)
    print('Known Fields:', e.fields)
else:
    print('Successfully de-serialized the JSON object.')
    print(repr(c))

Date and Time with Custom Patterns

As of v0.20.0, date and time strings in a custom format can be de-serialized using the DatePattern, TimePattern, and DateTimePattern type annotations, representing patterned date, time, and datetime objects respectively.

This will internally call datetime.strptime with the format specified in the annotation, and also use the fromisoformat() method in case the date string is in ISO-8601 format. All dates and times will continue to be serialized as ISO format strings by default. For more info, check out the Patterned Date and Time section in the docs.

A brief example of the intended usage is shown below:

from dataclasses import dataclass
from datetime import time, datetime
from typing import List
# Note: in Python 3.9+, you can import this from `typing` instead
from typing_extensions import Annotated

from dataclass_wizard import fromdict, asdict, DatePattern, TimePattern, Pattern


@dataclass
class MyClass:
    date_field: DatePattern['%m-%Y']
    dt_field: Annotated[datetime, Pattern('%m/%d/%y %H.%M.%S')]
    time_field1: TimePattern['%H:%M']
    time_field2: Annotated[List[time], Pattern('%I:%M %p')]


data = {'date_field': '12-2022',
        'time_field1': '15:20',
        'dt_field': '1/02/23 02.03.52',
        'time_field2': ['1:20 PM', '12:30 am']}

class_obj = fromdict(MyClass, data)

# All annotated fields de-serialize as just date, time, or datetime, as shown.
print(class_obj)
# MyClass(date_field=datetime.date(2022, 12, 1), dt_field=datetime.datetime(2023, 1, 2, 2, 3, 52),
#         time_field1=datetime.time(15, 20), time_field2=[datetime.time(13, 20), datetime.time(0, 30)])

# All date/time fields are serialized as ISO-8601 format strings by default.
print(asdict(class_obj))
# {'dateField': '2022-12-01', 'dtField': '2023-01-02T02:03:52',
#  'timeField1': '15:20:00', 'timeField2': ['13:20:00', '00:30:00']}

# But, the patterned date/times can still be de-serialized back after
# serialization. In fact, it'll be faster than parsing the custom patterns!
assert class_obj == fromdict(MyClass, asdict(class_obj))

Dataclasses in Union Types

The dataclass-wizard library fully supports declaring dataclass models in Union types as field annotations, such as list[Wizard | Archer | Barbarian].

As of v0.19.0, there is added support to auto-generate tags for a dataclass model – based on the class name – as well as to specify a custom tag key that will be present in the JSON object, which defaults to a special __tag__ key otherwise. These two options are controlled by the auto_assign_tags and tag_key attributes (respectively) in the Meta config.

To illustrate a specific example, a JSON object such as {"oneOf": {"type": "A", ...}, ...} will now automatically map to a dataclass instance A, provided that the tag_key is correctly set to “type”, and the field one_of is annotated as a Union type in the A | B syntax.

Let’s start out with an example, which aims to demonstrate the simplest usage of dataclasses in Union types. For more info, check out the Dataclasses in Union Types section in the docs.

from __future__ import annotations

from dataclasses import dataclass

from dataclass_wizard import JSONWizard


@dataclass
class Container(JSONWizard):

    class Meta(JSONWizard.Meta):
        tag_key = 'type'
        auto_assign_tags = True

    objects: list[A | B | C]


@dataclass
class A:
    my_int: int
    my_bool: bool = False


@dataclass
class B:
    my_int: int
    my_bool: bool = True


@dataclass
class C:
    my_str: str


data = {
    'objects': [
        {'type': 'A', 'my_int': 42},
        {'type': 'C', 'my_str': 'hello world'},
        {'type': 'B', 'my_int': 123},
        {'type': 'A', 'my_int': 321, 'myBool': True}
    ]
}


c = Container.from_dict(data)
print(f'{c!r}')

# True
assert c == Container(objects=[A(my_int=42, my_bool=False),
                               C(my_str='hello world'),
                               B(my_int=123, my_bool=True),
                               A(my_int=321, my_bool=True)])


print(c.to_dict())
# prints the following on a single line:
# {'objects': [{'myInt': 42, 'myBool': False, 'type': 'A'},
#              {'myStr': 'hello world', 'type': 'C'},
#              {'myInt': 123, 'myBool': True, 'type': 'B'},
#              {'myInt': 321, 'myBool': True, 'type': 'A'}]}

# True
assert c == c.from_json(c.to_json())

Serialization Options

The following parameters can be used to fine-tune and control how the serialization of a dataclass instance to a Python dict object or JSON string is handled.

Skip Defaults

A common use case is skipping fields with default values - based on the default or default_factory argument to dataclasses.field - in the serialization process.

The attribute skip_defaults in the inner Meta class can be enabled, to exclude such field values from serialization.The to_dict method (or the asdict helper function) can also be passed an skip_defaults argument, which should have the same result. An example of both these approaches is shown below.

from collections import defaultdict
from dataclasses import field, dataclass
from typing import DefaultDict, List

from dataclass_wizard import JSONWizard


@dataclass
class MyClass(JSONWizard):

    class _(JSONWizard.Meta):
        skip_defaults = True

    my_str: str
    other_str: str = 'any value'
    optional_str: str = None
    my_list: List[str] = field(default_factory=list)
    my_dict: DefaultDict[str, List[float]] = field(
        default_factory=lambda: defaultdict(list))


print('-- Load (Deserialize)')
c = MyClass('abc')
print(f'Instance: {c!r}')

print('-- Dump (Serialize)')
string = c.to_json()
print(string)

assert string == '{"myStr": "abc"}'

print('-- Dump (with `skip_defaults=False`)')
print(c.to_dict(skip_defaults=False))

Exclude Fields

You can also exclude specific dataclass fields (and their values) from the serialization process. There are two approaches that can be used for this purpose:

  • The argument dump=False can be passed in to the json_key and json_field helper functions. Note that this is a more permanent option, as opposed to the one below.

  • The to_dict method (or the asdict helper function ) can be passed an exclude argument, containing a list of one or more dataclass field names to exclude from the serialization process.

Additionally, here is an example to demonstrate usage of both these approaches:

from dataclasses import dataclass
from typing import Annotated

from dataclass_wizard import JSONWizard, json_key, json_field


@dataclass
class MyClass(JSONWizard):

    my_str: str
    my_int: int
    other_str: Annotated[str, json_key('AnotherStr', dump=False)]
    my_bool: bool = json_field('TestBool', dump=False)


data = {'MyStr': 'my string',
        'myInt': 1,
        'AnotherStr': 'testing 123',
        'TestBool': True}

print('-- From Dict')
c = MyClass.from_dict(data)
print(f'Instance: {c!r}')

# dynamically exclude the `my_int` field from serialization
additional_exclude = ('my_int',)

print('-- To Dict')
out_dict = c.to_dict(exclude=additional_exclude)
print(out_dict)

assert out_dict == {'myStr': 'my string'}

Field Properties

The Python dataclasses library has some key limitations with how it currently handles properties and default values.

The dataclass-wizard package natively provides support for using field properties with default values in dataclasses. The main use case here is to assign an initial value to the field property, if one is not explicitly passed in via the constructor method.

To use it, simply import the property_wizard helper function, and add it as a metaclass on any dataclass where you would benefit from using field properties with default values. The metaclass also pairs well with the JSONSerializable mixin class.

For more examples and important how-to’s on properties with default values, refer to the Using Field Properties section in the documentation.

Contributing

Contributions are welcome! Open a pull request to fix a bug, or open an issue to discuss a new feature or change.

Check out the Contributing section in the docs for more info.

TODOs

All feature ideas or suggestions for future consideration, have been currently added as milestones in the project’s GitHub repo.

Credits

This package was created with Cookiecutter and the rnag/cookiecutter-pypackage project template.