Announcing: A Free Book, A New Course, A Huge Price Cut...
It's a massive ship day. We're launching a free TypeScript book, new course, giveaway, price cut, and sale.
TypeScript 5.2 has recently been released, and the team is already hard at work on TypeScript 5.3.
September 21st Update: I've been checking some of the PR's in the TS repo, and two really nice improvements have landed.
readonly
no longer required when checking if arrays satisfy as const
In TypeScript 5.2, this code would throw an error:
Type 'readonly []' does not satisfy the expected type 'string[]'.
const array = [] as const satisfies string[];
// Type 'readonly []' does not satisfy the expected type 'string[]'.
This would go away if you added a readonly
modifier to string[]
:
const array = [] as const satisfies readonly string[];
But this is a bit annoying. Since we're already specifying as const
, it feels redundant to also add readonly
.
Fortunately, in TypeScript 5.3 this error will no longer occur - it'll just work - thanks to this PR.
switch(true)
will be narrowed properlyswitch(true)
is a popular way to express complicated if/else in TypeScript. It lets you achieve a pattern-matching-like syntax:
function getNodeDescriptionSwitch(node: Node) {
switch (true) {
case isArrayLiteralExpression(node):
case isObjectLiteralExpression(node):
return "Array or object";
case isBigIntLiteral(node):
case isNumericLiteral(node):
return "Numberish";
case isNoSubstitutionTemplateLiteral(node):
case isRegularExpressionLiteral(node):
case isStringLiteral(node):
case isTemplateLiteral(node):
return "Stringlike";
default:
return "Some sort of node";
}
}
The unfortunate thing about this pattern is that TypeScript wouldn't do any narrowing in the case
statements.
function handleStringOrNumber(value: string | number) {
switch (true) {
case typeof value === "string":
// Error: value is still string | number
return value.toUpperCase();
case typeof value === "number":
// Error: value is still string | number
return value.toFixed(2);
}
}
In TypeScript 5.3, this will now just work out of the box, thanks to this PR from AndaristRake.
The TypeScript team has recently posted the TypeScript 5.3 Iteration Plan, the document they use to plan out what features might be in the next version of TypeScript.
The iteration plan is a great way to get a sneak peek at what's coming in TypeScript 5.3. It's not a guarantee that these features will land - but it's a good indication.
So - here are the most interesting features that might land in TS 5.3.
TypeScript 5.3 might implement Import Attributes, a TC39 proposal that recently reached Stage 3.
Import Attributes allow you to specify options for imports. For example, you can specify the type of a JSON import:
import json from './foo.json' with { type: 'json' };
It also lets you specify the type of a dynamic import:
import("foo.json", { with: { type: "json" } });
You can re-export a module with a validated type:
export { val } from './foo.js' with { type: "javascript" };
OR, instantiate a worker with a validated type:
new Worker("foo.wasm", {
type: "module",
with: { type: "webassembly" },
});
The motivation for this change is to provide JavaScript a way to validate what kinds of MIME types are being imported. The main reason is security: "to prevent a scenario where the responding server unexpectedly provides a different MIME type, causing code to be unexpectedly executed".
throw
ExpressionsOne of the syntaxes that has been proposed for JavaScript is the throw
expression. This is a way to throw an error without using a statement. For example, you could write:
const id = searchParams.id || throw new Error("id is required");
You might be surprised that this isn't available in JavaScript today - but it's not! It'll throw an error in TypeScript:
const id = searchParams .id || throw new Error ("id is required");Expression expected.1109Expression expected.
However, throw expressions are unlikely to land in TypeScript 5.3. The proposal is still in Stage 2, a little way off the Stage 3 benchmark needed to add them to TypeScript.
But the TypeScript iteration plan makes specific mention of 'championing' this proposal. This means they're actively working on it, so it could land in a future version of JavaScript/TypeScript.
I was lucky enough to have the chance to go to the Bloomberg offices in London a few weeks ago to chat with Titian, the author of this PR. I'm pretty excited about it.
In a monorepo with many packages, you might have packages that depend on each other. You could end up with an extremely deep, 'family tree'-like setup where package A depends on package B, which depends on package C, which depends on package D.
In these situations, TypeScript's checking can get pretty slow. Package D has to be checked first, then package C, then B, then A.
The reason for this is that printing declaration files (.d.ts files) for each package has to be done by TypeScript itself - which also means type checking them. This is a slow process.
One way to speed this up would be to let a faster tool (like esbuild
or swc
) create the declaration files for each package. But this currently isn't possible. TypeScript is pretty loose about how few/many annotations you need to add to your code. Third-party tools aren't smart enough to generate declaration files based on inference.
Enter Isolated Declarations - a new, stricter mode of TypeScript that solves this problem.
It's an option you can add to your tsconfig.json
:
{
"compilerOptions": {
"isolatedDeclarations": true
}
}
And when enabled, it'll force you to be stricter about adding annotations. The exact level of strictness is still being worked out and might change over time. But as an example, return type annotations on exported functions will likely be mandatory to save TypeScript from having to infer them.
Before you kick up a fuss (I'm on record saying how little I like return type annotations), you only need to enable isolatedDeclarations
on shared packages - you won't need to enable it on your application code. The restrictions on shared packages will likely be desirable because in general you want to add more annotations to shared packages anyway.
A recent demo by Titian and team showed significant speedups for a large monorepo. I'm excited to see if this lands in TypeScript 5.3.
My one piece of advice when working with generic functions is "don't be afraid to use as
". TypeScript, as it exists now, doesn't do a great job of narrowing inside generic functions.
Check out this example below.
interface Example {
foo : string;
bar : number;
}
function exampleFunc <T extends keyof Example >(key : T ): Example [T ] {
if (key === "foo") {
return "abc";Type 'string' is not assignable to type 'Example[T]'.
Type 'string' is not assignable to type 'never'.2322Type 'string' is not assignable to type 'Example[T]'.
Type 'string' is not assignable to type 'never'. } else {
return 123;Type 'number' is not assignable to type 'Example[T]'.
Type 'number' is not assignable to type 'never'.2322Type 'number' is not assignable to type 'Example[T]'.
Type 'number' is not assignable to type 'never'. }
}
Here, we're trying to return a value from an object based on a key. If we pass in 'foo', we're returning a string. If we pass in 'bar', we're returning a number.
But TypeScript is giving an error, despite this code appearing to be valid.
The reason this doesn't happen is that TypeScript isn't narrowing Example[T]
to be the correct key. Any narrowing applied to Example[T]
will end up in it being typed as never
- hence the errors above.
The only current way to get this working is to type it as never
:
function exampleFunc <T extends keyof Example >(key : T ): Example [T ] {
if (key === "foo") {
return "abc" as never;
} else {
return 123 as never;
}
}
Which feels really, really bad.
TypeScript 5.3 might ship some changes here. There's a long-open issue that details exactly the motivations for this change.
This might be the thing I'm most excited about - the bad inference here is a huge barrier to people experimenting with generics. If TypeScript were smarter in these situations, it would make my job teaching generics a lot easier.
There's a famous hack with TypeScript where you can use a string & {}
to get 'loose autocomplete' on a string. For example:
type IconSize = "small" | "medium" | "large" | (string & {});
This annotation might look strange - but the reason it exists is so that you can pass anything to IconSize
while ALSO getting autocomplete for the three other values.
const icons : IconSize [] = [
"small",
"medium",
"large",
"extra-large",
"anything-goes",
];
TypeScript 5.3 might ship a new feature that makes this hack unnecessary. You'll be able to use string
as a type and get the same autocomplete:
type IconSize = "small" | "medium" | "large" | string;
This would be extremely welcome - especially because Webstorm users have had this for years.
fetch
in @types/node
On February 1st, 2022, the Node.js team merged a pull request to add the Fetch API to Node.js. This means that Node.js will have a fetch
function, just like the browser.
The trouble is, this hasn't yet been added to @types/node
. This papercut has resulted in a relatively heated back-and-forth on a DefinitelyTyped issue.
So, it's useful that the TypeScript team might be stepping in to take a look at it.
Share this article with your friends
It's a massive ship day. We're launching a free TypeScript book, new course, giveaway, price cut, and sale.
Learn why the order you specify object properties in TypeScript matters and how it can affect type inference in your functions.
Learn how to use corepack
to configure package managers in Node.js projects, ensuring you always use the correct one.
Learn how to strongly type process.env in TypeScript by either augmenting global type or validating it at runtime with t3-env.
Discover when it's appropriate to use TypeScript's any
type despite its risks. Learn about legitimate cases where any
is necessary.
Learn why TypeScript's types don't exist at runtime. Discover how TypeScript compiles down to JavaScript and how it differs from other strongly-typed languages.