Home

Awesome

Readonly TypeScript Types

Build Status type-coverage codecov Mutation testing badge Known Vulnerabilities npm

A collection of readonly TypeScript types inspired by TypeScript's built-in readonly types (ReadonlyArray, ReadonlyMap, etc) and by is-immutable-type.

The types here are all fully Immutable following is-immutable-type#definitions.

This package assumes you have TypeScript's strict mode and noUncheckedIndexedAccess option turned on. eslint-plugin-total-functions provides an ESLint rule to ensure they're both on.

Installation

# yarn
yarn add readonly-types

# npm
npm install readonly-types

Usage

// Here's an example using ReadonlyURL.
import { ReadonlyURL } from "readonly-types";

// This is fine.
const hasFooSearchParam = (url: ReadonlyURL) => url.searchParams.has("foo");

// But this won't compile.
const setFooSearchParam = (url: ReadonlyURL) => url.searchParams.set("foo", "bar");

The Types

The second column contains the types provided by this library (which are all Immutable). The columns to the right of it show the types being replaced and what level of immutability they achieve by default.

The first column ("Even Better 🚀") contains types that are more than just immutable versions of the types in the later columns. These "even better" options require more effort to adopt than those in the second column (or may not even be generally available yet), but they're worth considering if you want something that is more closely aligned with a pure typeful functional programming approach.

Even Better 🚀ImmutableReadonlyDeepReadonlyShallowMutable
A dedicated Map type (good options below), see Objects vs. Maps for whyReadonlyRecordRecord
ReadonlyURLURL
ReadonlyURLSearchParamsURLSearchParams
Temporal (stage 3 proposal, aims to solve various problems in Date, including its mutability)ReadonlyDateDate
Chunk, PrincipledArray (does not return mutable arrays from methods like map), purpose-built immutable data structuresImmutableArrayReadonlyArrayArray
purpose-built immutable data structuresImmutableSetReadonlySetSet
purpose-built immutable data structuresImmutableMapReadonlyMapMap
ReadonlyWeakSetWeakSet
ReadonlyWeakMapWeakMap
Effect's Either, fp-ts's EitherReadonlyError (and friends)Error and friends
ReadonlyRegExpRegExp
Effect, fp-ts's TaskEitherReadonlyPromisePromise
DeepImmutableDeepReadonly from ts-essentials, which when used will produce a mix of Mutable and ReadonlyDeep types

Linting

You can ban the mutable counterparts to these readonly types using eslint-plugin-functional's prefer-immutable-types rule.

ImmutableArray and PrincipledArray

TypeScript's built-in ReadonlyArray isn't truly immutable. Observe:

const foo: ReadonlyArray<string> = [""] as const;

// This compiles
foo.every = () => false;
// So does this
foo.at = () => undefined;

is-immutable-type provides the answer in Making ReadonlyDeep types Immutable. We've reused that here to provide an ImmutableArray type.

import { ImmutableArray } from "readonly-types";

const foo: ImmutableArray<string> = [""] as const;

// These no longer compile
foo.every = () => false; // Cannot assign to 'every' because it is a read-only property. ts(2540)
foo.at = () => undefined; // Cannot assign to 'at' because it is a read-only property. ts(2540)

ReadonlyArray achieves the ReadonlyDeep level of immutability, ImmutableArray achieves the Immutable level.

It turns out that even ImmutableArray has cracks in its immutable armour. Here's a subtle one:

// This doesn't compile...
foo.at = () => undefined;

foo.map((value, index, array) => {
  // ... but this does!
  array.at = () => undefined;

  return value;
});

The array passed as the third argument to the map callback is typed as ReadonlyArray. Our ImmutableArray trick doesn't change that method's callback's argument's types. The same applies to filter, flatMap, find and so on.

To fix that issue we provide a type called PrincipledArray:

const foo: PrincipledArray<string> = [""] as const;

// This doesn't compile...
foo.at = () => undefined;

foo.map((value, index, array) => {
  // ... and neither does this!
  array.at = () => undefined;

  return value;
});

PrincipledArray makes a few other (type-incompatible) improvements while its at it, including:

import { principledArray } from "readonly-types";

// Given a principled array.
const foo = principledArray<string>([]);

// This does not compile.
// Property 'forEach' does not exist on type 'PrincipledArray<string>'. ts(2339)
foo.forEach(() => {});

// This would normally throw at runtime, but with PrincipledArray it does not compile
// Expected 2 arguments, but got 1. ts(2554)
// An argument for 'initialValue' was not provided.
const result = foo.reduce((p) => p);

The downside to PrincipledArray is that -- precisely because it changes the type in these ways -- you cannot assign it to a value of type ReadonlyArray. ImmutableArray doesn't have this downside. Choose whichever is most appropriate for you.

ImmutableNonEmptyArray and PrincipledNonEmptyArray

An array type that is verifiably non-empty (i.e. known to have at least one entry at compile time) is a useful type to have.

You can make such a type based on ReadonlyArray like this:

type ReadonlyNonEmptyArray<T> = readonly [T, ...(readonly T[])];

Like ReadonlyArray that type is only ReadonlyDeep, not truly Immutable.

We provide a truly immutable version in the form of ImmutableNonEmptyArray.

With PrincipledArray having removed the versions of reduce and reduceRight that do not require an initialValue, there becomes a need for another type that is verifiably non-empty (at compile time) which puts them back again.

We provide that type in the form of PrincipledNonEmptyArray, which you can think of as a mix between ImmutableNonEmptyArray and PrincipledArray:

// Given a principled non-empty array.
const foo = principledNonEmptyArray<string>(["a"]);

// This compiles, whereas it wouldn't have compiled for a regular principled array.
const result = foo.reduce((p) => p);

Array type compatibility

⬇️ can be assigned to ➡️ArrayReadonlyArrayImmutableArrayPrincipledArrayPrincipledNonEmptyArray
Array✅ ⚠️✅ ⚠️
ReadonlyArray✅ ⚠️
ImmutableArray✅ ⚠️
PrincipledArray
PrincipledNonEmptyArray

Assignments marked ⚠️ can lead to surprising mutation in whichever side of the assignment appears to have "more" immutability, via mutations made to the side that has "less". eslint-plugin-total-functions includes an ESLint rule to flag these unsafe assignments. https://github.com/eslint-functional/eslint-plugin-functional/issues/526 may play a part too. See https://github.com/Microsoft/TypeScript/issues/13347 for more.

Purpose-built immutable data structures

Types like ImmutableArray and PrincipledArray (and even the humble built-in ReadonlyArray) can help a lot with correctness but the underlying runtime type remains a mutable Array. The same goes for our immutable Set and Map types. In essence the data structures are the same, we're just constraining ourselves to an immutable subset of their mutable APIs.

One consequence of this is that if someone could get their hands on a mutable handle to one of our values, they could edit it as if it were mutable (e.g. via an as type assertion or via an Array.isArray check). This forces us to put a little asterisk next to any immutability guarantees we make. You might reach for Object.freeze in response to that risk, but that comes with its own issues (performance, compatibility, doesn't show up in the type system, ...).

Another consequence of this is that updating and copying values of these types is needlessly expensive (in terms of compute and memory). A copy of the entire structure must be taken to preserve correctness, even if all we want to do for example is update a single element.

There exist purpose-built immutable data structures that give us an immutable API without the associated performance cost of copying an underlying mutable structure (look for terms like 'structural sharing' and 'copy on write'). If performance is a factor for you, these can be a better choice than the immutable types provided by this package.

To get you started, check out the following:

A surprising irony of these types is that they typically aren't truly immutable, for the same reason that ReadonlyArray isn't truly immutable. Here's an example:

import { Map as ImmutableJsMap } from "immutable";
const foo = ImmutableJsMap([["key", "value"]]);
// This compiles
foo.delete = () => foo;

Because delete is implemented using method syntax it is necessarily mutable (TypeScript methods defined using method syntax cannot be readonly for "reasons"). This is so common that is-immutable-type#definitions defines a level of "readonly-ness" called ReadonlyDeep that sits below truly Immutable but above the mutable levels ReadonlyShallow and Mutable.

Depending on how strictly you wish to enforce immutability, ReadonlyDeep may or may not be acceptable to you. If it isn't, you can fix it like this:

import { Map as ImmutableJsMap } from "immutable";

type TrulyImmutableMap<K, V> = Readonly<ImmutableJsMap<K, V>>;

const foo: TrulyImmutableMap<string, string> = ImmutableJsMap([
  ["key", "value"],
]);

// No longer compiles
foo.delete = () => foo; // Cannot assign to 'delete' because it is a read-only property. ts(2540)

See Making ReadonlyDeep types Immutable for more on this.

See Also