GitHub - rnag/dataclass-wizard: Simple, elegant, wizarding tools for interacting with Python's dataclasses.

Simple, elegant wizarding tools for Python’s dataclasses.

Dataclass Wizard is a fast, well-tested serialization library for Python dataclasses with automatic type conversion and support for JSON, YAML, TOML, and environment variables.

Behold, the power of Dataclass Wizard:

>>> from __future__ import annotations
>>> from dataclasses import field
>>> from dataclass_wizard import DataclassWizard
...
>>> class MyClass(DataclassWizard, load_case='AUTO', dump_case='CAMEL'):
...     my_str: str | None
...     is_active_tuple: tuple[bool, ...]
...     list_of_int: list[int] = field(default_factory=list)
...
>>> MyClass.from_json(
...     '{"my_str": 20, "ListOfInt": ["1", "2", 3], "isActiveTuple": [true, false, 1]}'
... )
MyClass(my_str='20', is_active_tuple=(True, False, True), list_of_int=[1, 2, 3])

Note

The example above demonstrates automatic type coercion, key casing transforms, and support for nested dataclasses. DataclassWizard also auto-applies @dataclass to subclasses.

Contents

Why Dataclass Wizard?
Quick Examples
Installation
Wizard Mixins ✨
Supported Types 🧑‍💻
Custom Type Hooks 🔌
JSON Marshalling
No Inheritance Needed
Custom Key Mappings
Mapping Nested JSON Keys
Extending from Meta
Date and Time with Custom Patterns
Recursive Types and Dataclasses with Cyclic References
Dataclasses in Union Types
Supercharged Union Parsing
Conditional Field Skipping
Serialization Options
Environ Magic
Field Properties
Contributing
TODOs
Credits

Why Dataclass Wizard?

Dataclass Wizard helps you turn Python dataclasses into robust, high-performance serialization schemas with minimal effort.

🚀 Fast — code-generated loaders and dumpers
🪶 Lightweight — pure Python, minimal dependencies
🧠 Typed — powered by Python type hints
🧙 Flexible — JSON, YAML, TOML, and environment variables
🧪 Reliable — battle-tested with extensive test coverage

Quick Examples

JSON De/Serialization

from dataclasses import dataclass
from dataclass_wizard import JSONWizard

@dataclass
class Data(JSONWizard):
    value: int

data = Data.from_dict({"value": "123"})
assert data.value == 123
print(data.to_json())

Environment Variables

import os
from dataclass_wizard import EnvWizard

os.environ['DEBUG'] = 'true'

class Config(EnvWizard):
    debug: bool

cfg = Config()
assert cfg.debug is True

EnvWizard

import os
from dataclass_wizard import EnvWizard, Env

os.environ['DEBUG_MODE'] = 'true'


class Config(EnvWizard):
    debug: bool = Env('DEBUG_MODE')


cfg = Config()
assert cfg.debug is True

Installation

Install from PyPI:

pip install dataclass-wizard

Or via conda-forge:

conda install -c conda-forge dataclass-wizard

Dataclass Wizard supports Python 3.9+.

Note

Python 3.6–3.8 users should install the last supported release, dataclass-wizard==0.26.1. See PyPI for historical versions and the Changelog for details.

Wizard Mixins ✨

In addition to JSONWizard, Dataclass Wizard provides a set of focused Mixin classes to simplify common serialization tasks:

🪄 EnvWizard — Load environment variables and .env files into typed schemas, including support for secret directories.
🎩 JSONPyWizard — A helper for JSONWizard that preserves keys as-is (no case transformations).
🔮 JSONListWizard — Extend JSONWizard to convert lists into container objects.
💼 JSONFileWizard — Load and save dataclass instances from local JSON files.

Optional format support:

🌳 TOMLWizard — Map dataclasses to and from TOML.
🧙‍♂️ YAMLWizard — Convert between YAML and dataclasses using PyYAML.

Supported Types 🧑‍💻

Dataclass Wizard supports:

📋 Collections: Handle list, dict, and set effortlessly.
🔢 Typing Generics: Manage Union, Any, and other types from the typing module.
🌟 Advanced Types: Work with Enum, defaultdict, and datetime with ease.

For more info, check out the Supported Types section in the docs for detailed insights into each type and the load/dump process!

Custom Type Hooks 🔌

Dataclass Wizard allows you to extend serialization and deserialization to support custom or unsupported types via type hooks.

Type hooks define how a value is:

loaded (parsed) from JSON or a dictionary
dumped (serialized) back into JSON-compatible data

This makes it easy to support standard-library types such as ipaddress.IPv4Address as well as user-defined classes.

Type hooks work with or without inheritance, and integrate seamlessly with the v1 code generation engine.

See Type Hooks for details and examples.

JSON Marshalling

JSONWizard (or DataclassWizard which auto-applies @dataclass) 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 JSONWizard as shown below.

Check out a more complete example of using the JSONWizard 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 = {'created_at': '2010-06-10 15:50:00Z',
               'list_of_b': [
                   {'my_status': '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(dump_date_time_as='TIMESTAMP',
         case='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': 1640908800}]}

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 Alias -- which can be considered an alias to dataclasses.field() -- is one approach that can resolve this.

Example below:

from dataclass_wizard import DataclassWizard, Alias


class MyClass(DataclassWizard):
    my_str: str = Alias('myString1')


# 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

Mapping Nested JSON Keys

The dataclass-wizard library allows you to map deeply nested JSON keys to dataclass fields using custom path notation. This is ideal for handling complex or non-standard JSON structures.

You can specify paths to JSON keys with the AliasPath helpers. For example, the deeply nested key data.items.myJSONKey can be mapped to a dataclass field, such as my_str:

from dataclass_wizard import AliasPath, DataclassWizard


class MyData(DataclassWizard):
    my_str: str = AliasPath('data.items.myJSONKey', default="default_value")


input_dict = {'data': {'items': {'myJSONKey': 'Some value'}}}
data_instance = MyData.from_dict(input_dict)
print(data_instance.my_str)  # Output: 'Some value'

Custom Paths for Complex JSON

You can now use custom paths to access nested keys and map them to specific fields, even when keys contain special characters or follow non-standard conventions.

Example with nested and complex keys:

from dataclasses import dataclass
from typing import Annotated
from dataclass_wizard import JSONWizard, AliasPath


@dataclass
class NestedData(JSONWizard):
    my_str: str = AliasPath('data[0].details["key with space"]', default="default_value")
    my_int: Annotated[int, AliasPath('data[0].items[3.14].True')] = 0


input_dict = {
    'data': [
        {
            'details': {'key with space': 'Another value'},
            'items': {3.14: {True: "42"}}
        }
    ]
}

# Deserialize JSON to dataclass
data = NestedData.from_dict(input_dict)
print(data.my_str)  # Output: 'Another value'

# Serialize back to JSON
output_dict = data.to_dict()
print(output_dict)  # {'data': {0: {'details': {'key with space': 'Another value'}, 'items': {3.14: {True: 42}}}}}

# Verify data consistency
assert data == NestedData.from_dict(output_dict)

# Handle empty input gracefully
data = NestedData.from_dict({'data': []})
print(repr(data))  # NestedData(my_str='default_value', my_int=0)

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):
        dump_date_time_as = DateTimeTo.TIMESTAMP
        dump_case = 'SNAKE'

    my_str: str
    my_date: date


data = {'my_str': 'test', 'my_date': '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": 1293667200}

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

Added in v0.28.0

There is now Easier 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='DEBUG')


@dataclass
class Container(JSONWizard):
    class _(JSONWizard.Meta):
        # True to enable Debug mode for additional (more verbose) log output.
        #
        # Pass in a `str` to `int` to set the minimum log level:
        #   logging.getLogger('dataclass_wizard').setLevel('INFO')
        debug = logging.INFO
        # 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.
        on_unknown_key = 'RAISE'

    element: 'MyElement'


@dataclass
class MyElement:
    my_str: str
    my_float: float


d = {
    'element': {
        'my_str': '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.unknown_keys)
    print('JSON object:', e.obj)
    print('Known Fields:', e.fields)
else:
    print('Successfully de-serialized the JSON object.')
    print(repr(c))

See the section on Handling Unknown JSON Keys for more info.

Save or "Catch-All" Unknown JSON Keys

When calling from_dict or from_json, any unknown or extraneous JSON keys that are not mapped to fields in the dataclass are typically ignored or raise an error. However, you can capture these undefined keys in a catch-all field of type CatchAll, allowing you to handle them as needed later.

For example, suppose you have the following dictionary:

dump_dict = {
    "endpoint": "some_api_endpoint",
    "data": {"foo": 1, "bar": "2"},
    "undefined_field_name": [1, 2, 3]
}

You can save the undefined keys in a catch-all field and process them later. Simply define a field of type CatchAll in your dataclass. This field will act as a dictionary to store any unmapped keys and their values. If there are no undefined keys, the field will default to an empty dictionary.

from typing import Any
from dataclass_wizard import DataclassWizard
from dataclass_wizard.models import CatchAll


class UnknownAPIDump(DataclassWizard):
    endpoint: str
    data: dict[str, Any]
    unknown_things: CatchAll

dump_dict = {
    "endpoint": "some_api_endpoint",
    "data": {"foo": 1, "bar": "2"},
    "undefined_field_name": [1, 2, 3]
}

dump = UnknownAPIDump.from_dict(dump_dict)
print(f'{dump!r}')
# > UnknownAPIDump(endpoint='some_api_endpoint', data={'foo': 1, 'bar': '2'},
#       unknown_things={'undefined_field_name': [1, 2, 3]})

print(dump.to_dict())
# > {'endpoint': 'some_api_endpoint', 'data': {'foo': 1, 'bar': '2'}, 'undefined_field_name': [1, 2, 3]}

Note

When using a "catch-all" field, it is strongly recommended to define exactly one field of type CatchAll in the dataclass.
LetterCase transformations do not apply to keys stored in the CatchAll field; the keys remain as they are provided.
If you specify a default (or a default factory) for the CatchAll field, such as unknown_things: CatchAll = None, the default value will be used instead of an empty dictionary when no undefined parameters are present.
The CatchAll functionality is guaranteed only when using from_dict or from_json. Currently, unknown keyword arguments passed to __init__ will not be written to a CatchAll field.

Date and Time with Custom Patterns

Tip

As of v1.x, Dataclass Wizard now supports timezone-aware and UTC datetime and time patterns, as well as multiple pattern strings (i.e. multiple custom formats) for greater flexibility in pattern matching. These features are not available in the v0.* versions.

The new features include:

Timezone-aware datetime and time patterns.
UTC datetime and time patterns.
Multiple custom formats for a single field, providing more control over pattern matching.

For more details and examples on how to use these new features, refer to the Documentation for Patterned Date and Time.

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

Internally, these annotations use datetime.strptime with the specified format and the fromisoformat() method for ISO-8601 formatted strings. All date and time values are still serialized to ISO format strings by default. For more information, refer to the Patterned Date and Time section in the documentation.

Here is an example demonstrating how to use these annotations:

from dataclasses import dataclass
from datetime import time, datetime
from typing import Annotated

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


@dataclass
class MyClass:
    # Custom format for date (Month-Year)
    date_field: DatePattern['%m-%Y']
    # Custom format for datetime (Month/Day/Year Hour.Minute.Second)
    dt_field: Annotated[datetime, Pattern('%m/%d/%y %H.%M.%S')]
    # Custom format for time (Hour:Minute)
    time_field1: TimePattern['%H:%M']
    # Custom format for a list of times (12-hour format with AM/PM)
    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 to date, time, or datetime objects, 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']}

# The patterned date/times can be de-serialized back after serialization, which will be faster than
# re-parsing the custom patterns!
assert class_obj == fromdict(MyClass, asdict(class_obj))

Recursive Types and Dataclasses with Cyclic References

Prior to version 0.27.0, dataclasses with cyclic references or self-referential structures were not supported. This limitation is shown in the following toy example:

from dataclasses import dataclass

@dataclass
class A:
    a: 'A | None' = None

a = A(a=A(a=A(a=A())))

This was a longstanding issue, but in v1.x, Dataclass Wizard now supports recursive dataclasses OOTB, including cyclic references.

The example below demonstrates recursive dataclasses with cyclic dependencies, following the pattern A -> B -> A -> B. For more details, see the Cyclic or "Recursive" Dataclasses section in the documentation.

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

from dataclasses import dataclass
from dataclass_wizard import JSONWizard

@dataclass
class A(JSONWizard):
    b: 'B | None' = None


@dataclass
class B:
    a: A | None = None

# Confirm that `from_dict` with a recursive, self-referential
# input `dict` works as expected.
a = A.from_dict({'b': {'a': {'b': {'a': None}}}})

assert a == A(b=B(a=A(b=B())))

Starting with version 1.x, recursive types are supported out of the box (OOTB), removing the need for any Meta settings.

This makes working with recursive dataclasses even easier and more streamlined. In addition, recursive types are now supported for the following Python type constructs:

NamedTuple
TypedDict
Union
Literal
Nested dataclasses
Type aliases (introduced in Python 3.12+)

Example Usage

Recursive types allow handling complex nested data structures, such as deeply nested JSON objects or lists. With v0.34.0 of Dataclass Wizard, de/serializing these structures becomes seamless and more intuitive.

Recursive `Union`

from dataclasses import dataclass
from dataclass_wizard import DataclassWizard

# For Python 3.9, use this `Union` approach:
from typing_extensions import TypeAlias

JSON: TypeAlias = 'str | int | float | bool | dict[str, JSON] | list[JSON] | None'


# For Python 3.10 and above, use this simpler approach:
# JSON = str | int | float | bool | dict[str, 'JSON'] | list['JSON'] | None

# For Python 3.12+, you can use the `type` statement:
# type JSON = str | int | float | bool | dict[str, JSON] | list[JSON] | None

class MyTestClass(DataclassWizard):
    name: str
    meta: str
    msg: JSON


x = MyTestClass.from_dict(
    {
        "name": "name",
        "meta": "meta",
        "msg": [{"x": {"x": [{"x": ["x", 1, 1.0, True, None]}]}}],
    }
)
assert x == MyTestClass(
    name="name",
    meta="meta",
    msg=[{"x": {"x": [{"x": ["x", 1, 1.0, True, None]}]}}],
)

Note

The type statement in Python 3.12+ simplifies type alias definitions by avoiding string annotations for recursive references.

Recursive `Union` with Nested `dataclasses`

from dataclasses import dataclass, field
from dataclass_wizard import JSONWizard

@dataclass
class A(JSONWizard):
    value: int
    nested: 'B'
    next: 'A | None' = None


@dataclass
class B:
    items: list[A] = field(default_factory=list)


x = A.from_dict(
    {
        "value": 1,
        "next": {"value": 2, "next": None, "nested": {}},
        "nested": {"items": [{"value": 3, "nested": {}}]},
    }
)
assert x == A(
    value=1,
    next=A(value=2, next=None, nested=B(items=[])),
    nested=B(items=[A(value=3, nested=B())]),
)

Note

Nested dataclasses are particularly useful for representing hierarchical structures, such as trees or graphs, in a readable and maintainable way.

Official References

For more information, see:

These examples illustrate the power of recursive types in simplifying complex data structures while leveraging the functionality of dataclass-wizard.

Dataclasses in `Union` Types

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

Starting from v0.19.0, the library introduces two key features: - Auto-generated tags for dataclass models (based on class names). - A customizable tag key (default: __tag__) that identifies the model in JSON.

These options are controlled by the auto_assign_tags and tag_key attributes in the Meta config.

For example, if a JSON object looks like {"type": "A", ...}, you can set tag_key = "type" to automatically deserialize it into the appropriate class, like A.

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(repr(c))

# Output:
# 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())

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

Supercharged `Union` Parsing

What about untagged dataclasses in Union types or | syntax? With the major release V1, dataclass-wizard supercharges Union parsing, making it intuitive and flexible, even without tags.

This is especially useful for collections like list[Wizard] or when tags (discriminators) are not feasible.

from __future__ import annotations  # Remove in Python 3.10+

from dataclasses import dataclass
from typing import Literal

from dataclass_wizard import DataclassWizard


class MyClass(DataclassWizard):
    class _(DataclassWizard.Meta):
        unsafe_parse_dataclass_in_union = True

    literal_or_float: Literal['Auto'] | float
    entry: int | MoreDetails
    collection: list[MoreDetails | int]

@dataclass
class MoreDetails:
    arg: str

# OK: Union types work seamlessly
c = MyClass.from_dict({
    "literal_or_float": 1.23,
    "entry": 123,
    "collection": [{"arg": "test"}]
})
print(repr(c))
# > MyClass(literal_or_float=1.23, entry=123, collection=[MoreDetails(arg='test')])

# OK: Handles primitive and dataclass parsing
c = MyClass.from_dict({
    "literal_or_float": "Auto",
    "entry": {"arg": "example"},
    "collection": [123]
})
print(repr(c))
# > MyClass(literal_or_float='Auto', entry=MoreDetails(arg='example'), collection=[123])

Conditional Field Skipping

Added in v0.30.0

Dataclass Wizard introduces conditional skipping to omit fields during JSON serialization based on user-defined conditions. This feature works seamlessly with:

Global rules via Meta settings.
Per-field controls using SkipIf() annotations.
Field wrappers for maximum flexibility.

Quick Examples

Globally Skip Fields Matching a Condition

Define a global skip rule using Meta.skip_if:

from dataclass_wizard import DataclassWizard
from dataclass_wizard.conditions import IS_NOT


class Example(DataclassWizard):
    class _(DataclassWizard.Meta):
        skip_if = IS_NOT(True)  # Skip fields if the value is not `True`

    my_bool: bool
    my_str: 'str | None'


print(Example(my_bool=True, my_str=None).to_dict())
# Output: {'my_bool': True}

Skip Defaults Based on a Condition

Skip fields with default values matching a specific condition using Meta.skip_defaults_if:

from __future__ import annotations  # Can remove in PY 3.10+

from dataclass_wizard import DataclassWizard
from dataclass_wizard.conditions import IS


class Example(DataclassWizard):
    class _(DataclassWizard.Meta):
        skip_defaults_if = IS(None)  # Skip default `None` values.

    str_with_no_default: str | None
    my_str: str | None = None
    my_bool: bool = False


print(Example(str_with_no_default=None, my_str=None).to_dict())
# > {'str_with_no_default': None, 'my_bool': False}

Note

Setting skip_defaults_if also enables skip_defaults=True automatically.

Per-Field Conditional Skipping

Apply skip rules to specific fields with annotations or skip_if_field:

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

from dataclasses import dataclass
from typing import Annotated

from dataclass_wizard import JSONWizard, skip_if_field
from dataclass_wizard.conditions import SkipIfNone, EQ


@dataclass
class Example(JSONWizard):
    my_str: Annotated[str | None, SkipIfNone]  # Skip if `None`.
    other_str: str | None = skip_if_field(EQ(''), default=None)  # Skip if empty.

print(Example(my_str=None, other_str='').to_dict())
# Output: {}

Skip Fields Based on Truthy or Falsy Values

Use the IS_TRUTHY and IS_FALSY helpers to conditionally skip fields based on their truthiness:

from dataclasses import dataclass, field
from dataclass_wizard import JSONWizard
from dataclass_wizard.conditions import IS_FALSY


@dataclass
class ExampleWithFalsy(JSONWizard):
    class _(JSONWizard.Meta):
        skip_if = IS_FALSY()  # Skip fields if they evaluate as "falsy".

    my_bool: bool
    my_list: list = field(default_factory=list)
    my_none: None = None

print(ExampleWithFalsy(my_bool=False, my_list=[], my_none=None).to_dict())
# > {}

Note

Special Cases

SkipIfNone: Alias for SkipIf(IS(None)), skips fields with a value of None.
Condition Helpers:
- IS, IS_NOT: Identity checks.
- EQ, NE, LT, LE, GT, GE: Comparison operators.
- IS_TRUTHY, IS_FALSY: Skip fields based on truthy or falsy values.

Combine these helpers for flexible serialization rules!

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 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' = 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 == '{"my_str": "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 skip=True can be passed in to the Alias helper function. 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 typing import Annotated

from dataclass_wizard import DataclassWizard, Alias


class MyClass(DataclassWizard):
    my_str: str
    my_int: int
    other_str: Annotated[str, Alias('AnotherStr', skip=True)]
    my_bool: bool = Alias('TestBool', skip=True)


data = {'my_str': 'my string',
        'my_int': 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 == {'my_str': 'my string'}

`Environ` Magic

Easily map environment variables to Python dataclasses with EnvWizard:

import os
from dataclass_wizard import EnvWizard

# Set up environment variables
os.environ.update({
    'APP_NAME': 'Env Wizard',
    'MAX_CONNECTIONS': '10',
    'DEBUG_MODE': 'true'
})

# Define dataclass using EnvWizard
class AppConfig(EnvWizard):
    app_name: str
    max_connections: int
    debug_mode: bool

# Load config from environment variables
config = AppConfig()
print(config.app_name)    #> Env Wizard
print(config.debug_mode)  #> True
assert config.max_connections == 10

# Override with keyword arguments
config = AppConfig(app_name='Dataclass Wizard Rocks!', debug_mode='false')
print(config.app_name)    #> Dataclass Wizard Rocks!
assert config.debug_mode is False

Note

EnvWizard simplifies environment variable mapping with type validation, .env file support, and secret file handling (file names become keys).

Key Features:

Auto Parsing: Supports complex types and nested structures.
Configurable: Customize variable names, prefixes, and dotenv files.
Validation: Errors for missing or malformed variables.

📖 Full Documentation

Advanced Example: Dynamic Prefix Handling

EnvWizard supports dynamic prefix application, ideal for customizable environments:

import os
from dataclass_wizard import EnvWizard, Env


# Define dataclass with custom prefix support
class AppConfig(EnvWizard):
    class _(EnvWizard.Meta):
        env_prefix = 'APP_'  # Default prefix for env vars

    name: str = Env('A_NAME')  # Looks for `A_NAME` (explicit override)
    debug: bool


# Set environment variables
os.environ['A_NAME'] = 'Test!'
os.environ['CUSTOM_DEBUG'] = 'yes'

# Apply a dynamic prefix at runtime
config = AppConfig(__env__={'prefix': 'CUSTOM_'})  # Looks for `A_NAME` and `CUSTOM_DEBUG`

print(config)
# > AppConfig(name='Test!', debug=True)

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 JSONWizard 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.

Name		Name	Last commit message	Last commit date
Latest commit History 231 Commits
.github		.github
benchmarks		benchmarks
dataclass_wizard		dataclass_wizard
docs		docs
images		images
recipe		recipe
tests		tests
.bumpversion.toml		.bumpversion.toml
.editorconfig		.editorconfig
.env		.env
.gitignore		.gitignore
.pre-commit-config.yaml		.pre-commit-config.yaml
.pyup.yml		.pyup.yml
.readthedocs.yaml		.readthedocs.yaml
CONTRIBUTING.rst		CONTRIBUTING.rst
HISTORY.rst		HISTORY.rst
LICENSE		LICENSE
MANIFEST.in		MANIFEST.in
Makefile		Makefile
README.rst		README.rst
pyproject.toml		pyproject.toml
run_bench.py		run_bench.py

Uh oh!

Folders and files

Latest commit

History

Repository files navigation

Why Dataclass Wizard?

Quick Examples

Installation

Wizard Mixins ✨

Supported Types 🧑‍💻

Custom Type Hooks 🔌

JSON Marshalling

No Inheritance Needed

Custom Key Mappings

Mapping Nested JSON Keys

Custom Paths for Complex JSON

Extending from Meta

Other Uses for Meta

Debug Mode

Handle Unknown JSON Keys

Save or "Catch-All" Unknown JSON Keys

Date and Time with Custom Patterns

Recursive Types and Dataclasses with Cyclic References

Example Usage

Recursive Union

Recursive Union with Nested dataclasses

Official References

Dataclasses in Union Types

Supercharged Union Parsing

Conditional Field Skipping

Quick Examples

Serialization Options

Skip Defaults

Exclude Fields

Environ Magic

Advanced Example: Dynamic Prefix Handling

Field Properties

Contributing

TODOs

Credits

About

Topics

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Sponsor this project

Uh oh!

Packages 0

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

Extending from `Meta`

Other Uses for `Meta`

Recursive `Union`

Recursive `Union` with Nested `dataclasses`

Dataclasses in `Union` Types

Supercharged `Union` Parsing

`Environ` Magic

Packages