@wry/tuple

Immutable finite list objects with constant-time equality testing (===) and no hidden memory leaks.

Installation & Usage

First install the package from npm:

1npm install @wry/tuple

This package has a default export that can be imported using any name, but is typically named tuple:

1import assert from "assert";
2import tuple from "@wry/tuple";
3
4// Tuples are array-like:
5assert(tuple(1, 2, 3).length === 3);
6assert(tuple("a", "b")[1] === "b");
7
8// Deeply equal tuples are also === equal!
9assert.strictEqual(
10  tuple(1, tuple(2, 3), 4),
11  tuple(1, tuple(2, 3), 4),
12);

In addition to the default export, @wry/tuple exports the Tuple class, whose Tuple.from function provides the default export; and the WeakTrie class, which you can learn more about by reading its code. You probably will not need to use these exports directly:

import tuple, { Tuple, WeakTrie } from "@wry/tuple";

assert(tuple === Tuple.from);

Constructing tuples

The tuple function takes any number of arguments and returns a unique, immutable object that inherits from Tuple.prototype and is guaranteed to be === any other Tuple object created from the same sequence of arguments:

1const obj = { asdf: 1234 };
2const t1 = tuple(1, "asdf", obj);
3const t2 = tuple(1, "asdf", obj);
4
5assert.strictEqual(t1 === t2, true);
6assert.strictEqual(t1, t2);

Own properties

A tuple has a fixed numeric length property, and its elements may be accessed using array index notation:

1assert.strictEqual(t1.length, 3);
2
3t1.forEach((x, i) => {
4  assert.strictEqual(x, t2[i]);
5});

Nested tuples

Since Tuple objects are just another kind of JavaScript object, naturally tuples can contain other tuples:

1assert.strictEqual(
2  tuple(t1, t2),
3  tuple(t2, t1)
4);
5
6assert.strictEqual(
7  tuple(1, t2, 3)[1][2],
8  obj
9);

However, because tuples are immutable and always distinct from any of their arguments, it is not possible for a tuple to contain itself, nor to contain another tuple that contains the original tuple, and so forth.

Constant time === equality

Since Tuple objects are identical when (and only when) their elements are identical, any two tuples can be compared for equality in constant time, regardless of how many elements they contain.

This behavior also makes Tuple objects useful as keys in a Map, or elements in a Set, without any extra hashing or equality logic:

1const map = new Map;
2
3map.set(tuple(1, 12, 3), {
4  author: tuple("Ben", "Newman"),
5  releaseDate: Date.now()
6});
7
8const version = "1.12.3";
9const info = map.get(tuple(...version.split(".").map(Number)));
10if (info) {
11  console.log(info.author[1]); // "Newman"
12}

Shallow immutability

While the identity, number, and order of elements in a tuple is fixed, please note that the contents of the individual elements are not frozen in any way:

1const obj = { asdf: 1234 };
2tuple(1, "asdf", obj)[2].asdf = "oyez";
3assert.strictEqual(obj.asdf, "oyez");

Iterability

Every Tuple object is array-like and iterable, so ... spreading and destructuring work as they should:

1func(...tuple(a, b));
2func.apply(this, tuple(c, d, e));
3
4assert.deepEqual(
5  [1, ...tuple(2, 3), 4],
6  [1, 2, 3, 4]
7);
8
9assert.strictEqual(
10  tuple(1, ...tuple(2, 3), 4),
11  tuple(1, 2, 3, 4)
12);
13
14const [a, [_, b]] = tuple(1, tuple(2, 3), 4);
15assert.strictEqual(a, 1);
16assert.strictEqual(b, 3);

Instance pooling (internalization)

Any data structure that guarantees === equality based on structural equality must maintain some sort of internal pool of previously encountered instances.

Implementing such a pool for tuples is fairly straightforward (though feel free to give it some thought before reading this code, if you like figuring things out for yourself):

1const pool = new Map;
2
3function tuple(...items) {
4  let node = pool;
5
6  items.forEach(item => {
7    let child = node.get(item);
8    if (!child) node.set(item, child = new Map);
9    node = child;
10  });
11
12  // If we've created a tuple instance for this sequence of elements before,
13  // return that instance again. Otherwise create a new immutable tuple instance
14  // with the same (frozen) elements as the items array.
15  return node.tuple || (node.tuple = Object.create(
16    tuple.prototype,
17    Object.getOwnPropertyDescriptors(Object.freeze(items))
18  ));
19}

This implementation is pretty good, because it requires only linear time (O(items.length)) to determine if a tuple has been created previously for the given items, and you can't do better than linear time (asymptotically speaking) because you have to look at all the items.

This code is also useful as an illustration of exactly how the tuple constructor behaves, in case you weren't satisfied by my examples in the previous section.

Garbage collection

The simple implementation above has a serious problem: in a garbage-collected language like JavaScript, the pool itself will retain references to all Tuple objects ever created, which prevents Tuple objects and their elements (which can be arbitrarily large) from ever being reclaimed by the garbage collector, even after they become unreachable by any other means. In other words, storing objects in this kind of Tuple would inevitably cause memory leaks.

To solve this problem, it's tempting to try changing Map to WeakMap here:

1const pool = new WeakMap;

and here:

1if (!child) node.set(item, child = new WeakMap);

This approach is appealing because a WeakMap should allow its keys to be reclaimed by the garbage collector. That's the whole point of a WeakMap, after all. Once a tuple becomes unreachable because the program has stopped using it anywhere else, its elements are free to disappear from the pool of WeakMaps whenever they too become unreachable. In other words, something like a WeakMap is exactly what we need here.

Unfortunately, this strategy stumbles because a tuple can contain primitive values as well as object references, whereas a WeakMap only allows keys that are object references.

In other words, node.set(item, ...) would fail whenever item is not an object, if node is a WeakMap. To see how the @wry/tuple library cleverly gets around this WeakMap limitation, have a look at this module.