desert 2022.9.22

Deserialize to objects while staying DRY

Desert generates serialization schemas for dataclasses and attrs classes. Writing code that’s DRY (“don’t repeat yourself”) helps avoid bugs and improve readability. Desert helps you write code that’s DRY.

Installation

pip install desert

or with Poetry

poetry add desert

Usage

A simple example models two Person objects in a Car.

from dataclasses import dataclass

# Or using attrs
# from attr import dataclass

from typing import List

import desert

@dataclass
class Person:
    name: str
    age: int

@dataclass
class Car:
    passengers: List[Person]

# Load some simple data types.
data = {'passengers': [{'name': 'Alice', 'age': 21}, {'name': 'Bob', 'age': 22}]}


# Create a schema for the Car class.
schema = desert.schema(Car)

# Load the data.
car = schema.load(data)
assert car == Car(passengers=[Person(name='Alice', age=21), Person(name='Bob', age=22)])

Documentation

https://desert.readthedocs.io/

Limitations

String annotations and forward references inside of functions are not supported.

Acknowledgements

• This package began as an extension of marshmallow-dataclass to add support for attrs.

2022.09.22 (2022-09-22)

Backward-incompatible Changes

• Update all project dependencies and fix all deprecated warnings. Python 3.6 support was dropped to allow updating deprecated dependencies. #161

Changes

• It is now possible to use type-variant generics in your dataclasses, such as Sequence or MutableSequence instead of List, Mapping instead of Dict, etc.

  This allows you to hide implementation details from users of your dataclasses. If a field in your dataclass works just as fine with a tuple as a list, you no longer need to force your users to pass in a list just to satisfy type checkers.

  For example, by using Mapping or MutableMapping, users can pass OrderedDict to a Dict attribute without MyPy complaining.

  @dataclass
  class OldWay:
      str_list: List[str]
      num_map: Dict[str, float]
  
  
  # MyPy will reject this even though Marshmallow works just fine. If you use
  # type-variant generics, MyPy will accept this code.
  instance = OldClass([], collections.ChainMap(MY_DEFAULTS))
  
  
  @dataclass
  class NewWay:
      str_list: List[str]  # Type-invariants still work
      num_map: MutableMapping[str, float]  # Now generics do too

  #140

---

2020.11.18 (2020-11-18)

Changes

• Schemas no longer copy non-field dataclass attributes. Thanks to @sveinse for report and test. #79

---

2020.01.06 (2020-01-06)

Changes

• Additional metadata are supported in ib() and fields(). Thanks to @sveinse for reporting and testing. #28

---

2020.01.05 (2020-01-05)

Changes

• Add support for attrs factories that take self (attr.Factory(..., takes_self=True)). #27

---

2020.01.04 (2020-01-04)

Changes

• Add support for Tuple[int, …] per https://docs.python.org/3/library/typing.html#typing.Tuple #16 Thanks to @sveinse for reporting and testing.

• Use module imports internally. #18

• Desert version only needs to be updated in _version.py #19

• Fix doctests. #20

---

2020.01.03

Changes

• Optional fields allow None. #11. Thanks to @sveinse for reporting and testing.

2019.12.18

Changes

• Improve error message for unknown generics. #10

2019.12.10

Changes

• Add UnknownType exception with better error message for types that should be generic. #8

2019.12.09

Changes

• Marshmallow schema Meta arguments are accepted, allowing exclusion of unknown fields and other options. #3

2019.11.06 (2019-11-06)

Changes

• Add twine and wheel development dependencies. #2

---

2019.11.06 (2019-11-06)

Changes

• Switch to calver

Backward-incompatible Changes

• Non-optional fields without a default or factory now have required=True so raise marshmallow.exceptions.ValidationError when missing. #1

---

0.1.0 (2019-06-22)

Changes

• First release on PyPI.

—