memium: Linear Memory for TypeScript

Memium is a library for working with linear memory in TypeScript. It enables you to use binary structs and pointers^[1] easily, without having to worry about the underlying operations. Memium is designed with performance and a seamless user experience in mind.

^{[1] Work In Progress}

Installation

npm install memium

If you're using Memium, especially for big projects, please consider supporting the project.

Structs

The @struct decorator turns a class into a struct. Like the various typed array classes and DataView, Structs are ArrayBufferViews, meaning they have a buffer, byteOffset, and byteLength. They also have a similar constructor to typed arrays. You will need extend a class that implements these, or implement that functionality your self. Uint8array and Utilium's BufferView are good choices.

Once you have a struct, you can decorate members of the class as fields on the struct. The easiest way to do this is using the shortcuts for primitives, though you will need to use @field for nested structs and unions. With shortcuts, you write @types.type for a single value or @types.type(length) for an array. Importing types as t makes this even shorter and is more similar to C-style example_t types.

Structs can have attributes, just like in C. You pass these as arguments to the decorator. At the moment, you can use packed and align(n).

Putting all of it together, you could have something like this:

import { struct, types as t, packed } from 'memium';

@struct(packed)
class Person extends Uint8Array {
	@t.char(1) accessor magic = 'P';

	@t.uint16 accessor age = 0;

	@t.uint8 accessor name_length = 0;

	@t.char(32, { countedBy: 'name_length' }) accessor name: Uint8Array;
}

Structs in Memium have the same behavior as structs in C.

Inheritance

You can use normal class inheritance with structs, though you'll need to use @struct on any subclasses you want to be structs.

For example:

import { struct, types as t, packed, sizeof } from 'memium';

@struct()
class Animal extends Uint8Array {
	@t.char(64) accessor species: Uint8Array;
	@t.float32 accessor age: number;
	@t.float32 accessor weight: number;
}

@struct()
class Duck extends Animal {
	@t.uint8 accessor name_length: number;
	// Dynamically sized array
	@t.char(0, { countedBy: 'name_length' }) accessor name: Uint8Array;
}

const data = new ArrayBuffer(sizeof(Duck) + 32 /* starting memory for the name*/);

const duck = new Duck(data);

const encoder = new TextEncoder();
const encode = encoder.encode.bind(encoder);

Object.assign(duck, {
	species: encode('Mallard'),
	age: 1.5,
	name_length: 5,
	name: encode('Jerry'),
});

// References the same memory
const animal = new Animal(data);

console.log(animal.age); // 1.5

Nesting

As mentioned above, you'll need to use @field for non-primitive types (i.e. structs and unions). Below is a more complex example of a custom file format.

import { struct, types as t, packed, field } from 'memium';

const encoder = new TextEncoder();
const encode = encoder.encode.bind(encoder);

@struct(packed)
class Header extends Uint8Array {
	@t.char(8) accessor magic: Uint8Array = encode('EXAMPLE!');
	@t.uint16 accessor format: number;
	@t.uint16 accessor app_version: number;
	@t.uint8(16) accessor uuid: Uint8Array;
	@t.float64 accessor timestamp: number;
	@t.uint64(12) accessor padding: BigUint64Array; // pad to 128
}

@struct(packed)
class AppFile extends Uint8Array {
	@field(Header) accessor header: Header;
	@t.uint32 accessor n_sections: number;
	@field(Section, { countedBy: 'n_sections' }) accessor sections: Section[];
}

Pointers

Note

Pointers are a work in progress. What you see below is the planned API.

Pointers are centralized around Pointer, which is used along with array buffers.

A simple use case:

import { Animal, Duck } from 'the previous example';
import { sizeof, pointer, alloc } from 'memium';

const addr = alloc(sizeof(Duck));

const duck_ptr = new Pointer(Duck, addr);

const duck = duck_ptr.deref();