atomicx

atomicx is an easy-to-use atomics library for Python, providing atomic integer, boolean, and floats. It allows you to perform atomic operations on shared variables, ensuring thread-safety and preventing race conditions in concurrent programming. Everything is entirely lock-free and is backed by Rust's atomic types (or from portable-atomic). Builds are published for the standard GIL runtime and the free-threaded CPython variants.

Features

• Atomic integer operations: load, store, add, subtract, swap, compare and exchange, multiply, divide, increment, decrement.
• Atomic boolean operations: load, store, swap, compare and exchange, flip.
• Atomic floats: load, store, swap, compare and exchange.
• Strong typing provided as stubs for static type checkers.

Installation

Binary wheels are provided for Python 3.8 and above on Linux, macOS, and Windows:

pip install atomicx

Usage

See the documentation for more information. Here's a quick overview:

Atomic Integer

from atomicx import AtomicInt

# Create an atomic integer with an initial value of 0
atom = AtomicInt()

# Perform atomic operations
atom.store(10)
value = atom.load()
print(f"Value: {value}")

previous_value = atom.swap(20)
print(f"Previous Value: {previous_value}")

atom.add(5)
print(f"Result after addition: {atom}")

# Run a retry loop to clamp the value to at least 0
atom.update(lambda current: max(current, 0))

# Increment and decrement operations
atom.inc()
atom.dec()

Atomic Boolean

from atomicx import AtomicBool

# Create an atomic boolean with an initial value of False
atom = AtomicBool()

# Perform atomic operations
atom.store(True)
value = atom.load()
print(f"Value: {value}")

previous_value = atom.swap(False)
print(f"Previous Value: {previous_value}")

result = atom.compare_exchange(False, True)
print(f"Swap Result: {result}")

# Flip the value of the atomic boolean
atom.flip()

# Update using a callable
atom.update(lambda current: not current)

Atomic Float

from atomicx import AtomicFloat

# Create an atomic float with an initial value of 0.0
atom = AtomicFloat()
print(f"Initial Value: {atom.load()}")

# Perform atomic operations
atom.store(3.14)
value = atom.load()
print(f"Value: {value}")

# See docs for more operations

Thread Safety

Atomic variables are thread-safe and can be shared between threads. Each predefined operation on them are executed per thread as an indivisible unit. Here's an example of using an atomic integer in a multithreaded environment:

import threading
from atomicx import AtomicInt

x = AtomicInt(0)
def increment():
    for _ in range(1000):
        x.inc() # equivalent to x.add(1) or x += 1

threads = []
for _ in range(10):
    thread = threading.Thread(target=increment)
    thread.start()
    threads.append(thread)

for thread in threads:
    thread.join()

assert x.load() == 1000 * 10

The equivalent in vanilla Python without locks is not thread-safe in general.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

• The atomicx library is heavily dependent on and inspired by the Rust std::sync::atomic module.
• The floating point atomic operations are delegated to the portable-atomic crate.