rememory 0.3.3

A set of lightweight shared memory variable types for safe multiprocessing access

Rememory

Rememory provides shared-memory data structures (RememoryDict and RememoryList) as well as basic types (str, int, float, and bool) that work safely across multiple processes – even across completely independent Python interpreters – using OS-level named locks.

This is designed for scenarios where you need a simple, Python-native way to share structured state between processes without relying on an external database or message broker. Data is serialized in shared memory using Python's pickle format.

Features

• Shared-memory backend using multiprocessing.shared_memory
• Cross-process synchronization with OS-level locks (Windows mutex / POSIX semaphore)
• Works across separate scripts and interpreters, not just multiprocessing.Process
• Drop-in replacements for str, int, float, bool, Dict and List
• Type-safe generics for editors and type checkers (e.g. RememoryDict[str, int])

Installation

pip install rememory

Dependencies

On Linux/macOS posix_ipc, on Windows pywin32:

Basic Usage

RememoryDict

from rememory import Dict # RMDict, or RememoryDict also work

# Create or attach to a shared dict
shared: Dict[str, dict[str, int]] = Dict("game_state")

# Write to it (any process using the same name sees these changes)
shared["player1"] = {"score": 10, "level": 2}

# Read from it
print(shared["player1"])

# Iterate
for key, value in shared.items():
    print(key, value)

Any process that does Dict("game_state") will connect to the same shared memory block.

RememoryList

from rememory import List # RMList, or RememoryList also work

shared_list: List[str] = List("chat")

# Append items
shared_list.append("hello")
shared_list.append("world")

# Read
print(shared_list[0])  # "hello"
for item in shared_list:
    print(item)

RememoryInt

from rememory import int, IntTypes # RMInt, or RememoryInt also work

counter = int("shared_counter", IntTypes.INT32)

counter.value = 42

print(counter.value)  # 42

RememoryFloat

from rememory import float, FloatTypes # RMFloat, or RememoryFloat also work

pi = float("pi", FloatTypes.FLOAT32)

pi.value = 3.14

print(pi.value)

RememoryBlock

from noexcept import NoBaseException
from rememory import Block, BlockSize

# Store arbitrary pickleable objects in shared memory
class CustomError:
    pass

err_block: Block[CustomError] = Block("err_block", BlockSize.s256)

err_block.value = CustomError("boom")

print(err_block.value)

RememoryString

from rememory import str, BlockSize # RMString, or RememoryString also work

msg = str("message", BlockSize.S128)

msg.value = "Hello"

print(msg.value)

RememoryBool

from rememory import bool

flag = bool("ready")

flag.value = True

print(flag.value)

Multiprocessing Example

from multiprocessing import Process
from rememory import Dict, List

SHARED_DICT = "game"
SHARED_LIST = "events"

def worker():
    d = Dict[str, dict[str, int]](SHARED_DICT)
    pid = str(os.getpid())
    d[pid] = {"score": 1, "level": 1}
    d[pid]["score"] += 1

    l = List[str](SHARED_LIST)
    l.append(f"{pid}-joined")

if __name__ == "__main__":
    d = Dict[str, dict[str, int]](SHARED_DICT)
    l = List[str](SHARED_LIST)

    # Clear previous state
    d._write_data({})
    l._write_data([])

    procs = [Process(target=worker) for _ in range(4)]
    for p in procs: p.start()
    for p in procs: p.join()

    print("Shared dict:", dict(d.items()))
    print("Shared list:", list(l))

Locking

Rememory uses OS-named synchronization primitives:

• Windows: Named mutex via pywin32
• Linux/macOS: POSIX named semaphore via posix_ipc

This ensures that even separate Python scripts respect locking when using the same shared memory name.

When to Use

• Game engines or simulations needing shared state between processes
• Multi-process pipelines without a heavy DB or broker
• Debugging tools that need live shared state

When Not to Use

• When your data doesn’t fit in memory
• When you need persistence beyond process lifetime (shared memory is volatile)

License

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

Author

Nichola Walch [e-mail]