Convert dictionaries to dataclasses and back

pip install howard==1.12.1



Python datatype marshalling

This library marshalls dictionaries [read json] into instances of defined dataclasses and back.


from dataclasses import dataclass

import howard

class Person:
    name: str
    age: int

my_dict = {'name': 'Bob', 'age': 24}
person = howard.from_dict(my_dict, Person)
assert == 'Bob'
assert person.age == 24

The main purpose is to use python dataclasses as a schema definition rather than having to hand-write a schema. Howard does not currently include logic itself for generating json-schema (swagger) object documentation, but that is a long term goal of it.

Howard differs from a standard dataclass because it can recursively marshall and unmarshall. It also supports more types out of the box than dataclasses.asdict does. Some supported types:

  • Enums
  • TypedDict
  • Collections (lists/dictionaries)
  • Datetime
  • all primitives (int/string/boolean/float)

All of the logic for howard is in your dataclass definition, not in howard. Howard just has a to_dict and from_dict method, and it bases all decisions off of your dataclass. There is no inheritance on custom types, everything is standard, built-in python. (3.7+)


pip install howard

More examples

For more examples, you can go look at the tests at tests/

Here is a basic example of recursive types and how it can work with howard:

from enum import Enum
from dataclasses import dataclass, field
from typing import List

import howard

class Suit(Enum):
    heart = 'h'
    spade = 's'
    diamond = 'd'
    club = 'c'

def validate_rank(i: int) -> int:
    lb, ub = 1, 13
    if lb <= i <= ub:
        return i
    raise ValueError(f'{i} is not between {lb} and {ub}')

class Card:
    rank: int = field(metadata={'howard': {'decoder': validate_rank}})
    suit: Suit

class Hand:
    hand_id: int = 0
    cards: List[Card] = field(default_factory=list)

d = {'hand_id': 2, 'cards': [{'rank': 2, 'suit': 'c'}, {'rank': 10, 'suit': 'h'}]}

# d is a dictionary, now we turn it into the dataclass
obj = howard.from_dict(d, Hand)

assert isinstance(obj, Hand)
assert obj.hand_id == 2
assert len( == 2
assert isinstance([0], Card)
assert[0].suit ==

# and back to a dictionary
json_dict = howard.to_dict(obj)

In the above example, you can see a couple things.

  1. A Hand contains a list of Card.
  2. The sub-object Card also gets unmarshalled correctly.
  3. The Suit object is an enum, which is like a string in json form, but only has 4 possible values.
  4. The Card has a field called rank which has its own custom decoder. In this case, the decoder acts as a validator, but can also be used for custom decode logic.


  • Why not just use dataclasses.asdict and MyDataclass(**my_dict)? dataclasses.asdict doesn't work on all types, for example, Enums or datetimes. MyDataclass(**my_dict) will not recursively turn the subobjects into their respective datatype.

  • What about custom types? You can specify custom decoders and encoders in a dataclass field.metadata section. See example above.