The databind package provides a (de)serialization framework that understands most native Python types as well as dataclasses, as well as an implementation for serialize to/from JSON-like nested data structures.

Databind is intended mostly for flexible and easy to use configuration loading. It does not try achieve high-performance; you should look towards e.g. mashumaro for this usecase.

@dataclass
class Server:
    host: str
    port: int

@dataclass
class Config:
    server: Server

from databind.json import dump, load

dict_payload = {"server": {"host": "localhost", "port": 8080}}
loaded = Config(server=Server(host="localhost", port=8080))

assert load(dict_payload, Config) == loaded
assert dump(loaded, Config) == dict_payload

• Support for a plethora of builtin types, including Enum, Decimal, UUID, Path, datetime, date, time, timedelta
• Support for multiple union serialization modes (nested, flat, keyed, typing.Literal)
• Support for generic types, e.g. load([{"name": "Jane Doe"}], list[Person])
• Support for new-style type hints in older Python versions when using forward refererences (strings or __future__.annotations) thanks to typeapi
  • PEP 604 - Allow writing union types as X | Y
  • PEP585 - Type Hinting Generics in Standard Collections)
• Support for customized serialization and deserialization of types
• Support for flattening fields of a nested dataclass or collecting remaining fields in a dict
• Full runtime type checking during serialization
• Use "settings" to customize serialization behaviour
  • As global settings per load()/dump() call: load(..., settings=[ExtraKeys(True)])
  • As class-level settings using a decorator: @Union(style=Union.FLAT) or @ExtraKeys(True)
  • As type-hint level settings using typing.Annotated (or typing_extensions.Annotated): full_name: Annotated[str, Alias("fullName")] or FullNameField = Annotated[str, Alias("fullName")]

• Merged databind.core and databind.json packages into databind. The old PyPI packages will remain as proxies until the next minor version.
• Dropped support for Python 3.6 and 3.7.

Copyright © 2022 – Niklas Rosenstein