Pysafe PickleSafe, Fast, Schema-Evolvable Object Graph Serialization

🛡️

Zero Arbitrary Code Execution

Unlike Python's standard pickle, pysafe-pickle never invokes __reduce__, __setstate__, exec, or eval during deserialization.

⚡

Rust-Native Performance

Core traversal, custom binary encoding, and decoding implemented in Rust with PyO3 bindings for maximum throughput.

🔄

Schema Evolution & Migrations

First-class support for evolving dataclass models with @psp.migrate decorators, automated chain resolution, and versioning.

📦

Drop-in Pickle API

Seamless dumps, loads, dump, load, and streaming Pickler / Unpickler classes matching standard pickle semantics.

🚀

Zero-Copy Tensors & Streaming

Native integration with NumPy and PyTorch tensors, plus PEP 574 PickleBuffer out-of-band streaming support.

🔐

Allowlist & HMAC Verification

Built-in strict type allowlisting and optional HMAC-SHA256 data tampering detection.

Quick Example

python

import pysafe_pickle as psp
from dataclasses import dataclass

@dataclass
class User:
    name: str
    age: int
    email: str = ""
    __pysafe_pickle_version__ = 1

# Serialize safely with Rust
data = psp.dumps(User(name="Alice", age=30, email="alice@example.com"))

# Deserialize without arbitrary code execution
user = psp.loads(data)
assert user.name == "Alice"

Community & Ecosystem

PyPI Package: pysafe-pickle
Repository: Abdullah-Masood-05/pygraph
License: AGPL-3.0