docs/object-serialization/python/native.md

title: Python Native Serialization sidebar_position: 2 id: native license: | Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to You under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

 http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Python native serialization is the Python-only wire mode selected with xlang=False. Use it when every writer and reader is Python and the payload should follow Python's object model instead of the portable xlang type system.

Use Cross-Language Interoperability, the default Python mode, when bytes must be read by Java, C++, Go, Rust, JavaScript/TypeScript, C#, Swift, Dart, Scala, Kotlin, or another non-Python Fory implementation.

When To Use Native Serialization

Use native serialization when:

A payload is produced and consumed only by Python applications.
You are replacing pickle or cloudpickle for Python-only object graphs.
The data model includes functions, lambdas, local classes, methods, or Python reduction hooks.
The graph can contain shared objects or cycles that need Python reference tracking.
You need pickle protocol 5-style out-of-band buffers for large Python data objects.

Native mode can serialize Python-specific values such as global functions, local functions, lambdas, local classes, methods, and objects customized with __getstate__, __setstate__, __reduce__, or __reduce_ex__. Those values are not valid xlang payloads.

Create a Native-Mode Fory Instance

Create Fory with xlang=False:

import pyfory
fory = pyfory.Fory(xlang=False, ref=False, strict=True)

Keep strict=True for registered, trusted type surfaces. Use strict=False only when native-mode payloads need dynamic Python types such as functions, local classes, or objects reconstructed by reduction hooks.

Common Usage

import pyfory

fory = pyfory.Fory(xlang=False, ref=True, strict=False)

data = fory.dumps({"name": "Alice", "age": 30, "scores": [95, 87, 92]})
print(fory.loads(data))

from dataclasses import dataclass

@dataclass
class Person:
    name: str
    age: int

person = Person("Bob", 25)
data = fory.dumps(person)
print(fory.loads(data))  # Person(name='Bob', age=25)

Use dumps/loads for pickle-style APIs, or serialize/deserialize when matching the xlang API shape in code that switches modes explicitly.

Security And Dynamic Types

Native mode can reconstruct Python objects that execute import and construction logic during deserialization. Treat untrusted native-mode bytes the same way you would treat untrusted pickle bytes.

Keep strict=True when deserializing data that should contain only registered or built-in types.
Use strict=False only for trusted payloads that require dynamic Python classes or functions.
Provide a policy= deserialization policy when dynamic types are required but the accepted type surface should still be restricted.
Do not use xlang/native mode choice as a security control. Apply strict mode, policies, registration, and resource limits based on the payload source.

Python-specific values and hooks

See Functions, Classes, and Methods for callable and type values, then Serialization Hooks for reduction, state, construction, and pickle/cloudpickle migration.

References And Cycles

Enable ref=True when object identity, shared references, or cycles must round-trip:

import pyfory

fory = pyfory.Fory(xlang=False, ref=True, strict=True)

node = {}
node["self"] = node
data = fory.dumps(node)
decoded = fory.loads(data)
assert decoded["self"] is decoded

Disable reference tracking for value-shaped payloads that do not need identity preservation. It keeps the payload smaller and the hot path simpler.

Out-of-Band Buffers

Python native mode can use pickle protocol 5-style out-of-band buffers for large binary payloads and data structures backed by external memory:

import pickle
import pyfory

data = b"Large binary data"
pickle_buffer = pickle.PickleBuffer(data)

buffer_objects = []
fory = pyfory.Fory(xlang=False, ref=True, strict=False)
serialized = fory.dumps(pickle_buffer, buffer_callback=buffer_objects.append)
buffers = [obj.getbuffer() for obj in buffer_objects]
decoded = fory.loads(serialized, buffers=buffers)
assert bytes(decoded.raw()) == data

Use this when the payload stays in Python and large buffers should avoid extra copies. See Out-of-Band Serialization.

Native And Xlang Comparison

Requirement	Use native serialization	Use xlang serialization
Python-only payloads	Yes	Optional
Non-Python readers or writers	No	Yes
Functions, lambdas, local classes	Yes	No
`__reduce__` / `__getstate__` object hooks	Yes	No
Pickle/cloudpickle replacement	Yes	No
Portable type mapping across languages	No	Yes

Performance Comparison

import pyfory
import pickle
import timeit

fory = pyfory.Fory(xlang=False, ref=True, strict=False)

obj = {f"key{i}": f"value{i}" for i in range(10000)}
print(f"Fory: {timeit.timeit(lambda: fory.dumps(obj), number=1000):.3f}s")
print(f"Pickle: {timeit.timeit(lambda: pickle.dumps(obj), number=1000):.3f}s")

Troubleshooting

Another language cannot read the payload

The writer is using native serialization. Rebuild it with xlang=True, register portable schemas on every peer, and avoid Python-only values such as lambdas or local classes.

A dynamic class or function fails to deserialize

Use strict=False for trusted payloads and provide a deserialization policy= when only selected dynamic types should be accepted.

A cycle does not round-trip

Create the Fory instance with ref=True.

A value depends on pickle hooks

Keep the payload in native mode. Xlang mode does not execute Python __reduce__, __reduce_ex__, __getstate__, or __setstate__ object reconstruction hooks.