attrib is a performant data description and validation system for Python. It provides an intuitive API for describing and validating data structures using Python's descriptor protocol. attrib is designed to be deterministic, familiar, and flexible, allowing you to create complex data structures with no hidden runtime magic.
In most cases, especially heavy workflows, attrib can be 5 - 30% faster than attrs + cattrs, dataclasses + cattrs, or pydantic when deserializing + serializing data.
attrib dataclasses can be used in latency sensitive APIs, configuration definitions, and data pipelines.
- Declarative, intuitive and easily extensible API
- Fast and lightweight. Low OOP overhead and no hidden runtime magic.
- Recursive deserialization of complex or nested data structures.
- Composable, functional, and flexible validation.
- Enforces validation of field values on set.
- Context-aware deserialization and validation.
- Declarative serialization control with the
OptionsAPI - Growing set of custom fields, validators, and utilities.
- Adequately typed APIs for better IDE support.
- Detailed error handling with error history and context preservation.
- Type adapter API for adapted types (WIP)
- Supports Python 3.8+
This library is still in early development and may only be used for testing and experimentation.
To get started with attrib, clone the repository.
Please ensure you have uv installed. If not, visit uv's documentation for installation instructions.
Next, ensure you have python installed. If you do not, run these commands.
Check available python versions:
uv python listInstall version of choice:
uv python install <version>Proper documentation and tests will be included soon, but you can check the examples directory for usage patterns, or run:
uv sync --extra dev
uv add tzdata
uv run --python 3.10 examples/main.py 1000 -OO -Bto run benchmarks of attrib against attrs + cattrs and pydantic.
I would love to hear your feedback, suggestions, or contributions. Feel free to open an issue or a pull request.