Announcing TypeScript 4.9 Beta

Daniel Rosenwasser

Today we’re announcing our beta release of TypeScript 4.9!

To get started using the beta, you can get it through NuGet, or- use npm with the following command:

npm install -D typescript@beta

You can also get editor support by

Here’s a quick list of what’s new in TypeScript 4.9!

The satisfies Operator

TypeScript developers are often faced with a dilemma: we want to ensure that some expression matches some type, but also want to keep the most specific type of that expression for inference purposes.

For example:

// Each property can be a string or an RGB tuple.
const palette = {
    red: [255, 0, 0],
    green: "#00ff00",
    bleu: [0, 0, 255]
//  ^^^^ sacré bleu - we've made a typo!
};

// We want to be able to use array methods on 'red'...
const redComponent = palette.red.at(0);

// or string methods on 'green'...
const greenNormalized = palette.green.toUpperCase();

Notice that we’ve written bleu, whereas we probably should have written blue. We could try to catch that bleu typo by using a type annotation on palette, but we’d lose the information about each property.

type Colors = "red" | "green" | "blue";

type RGB = [red: number, green: number, blue: number];

const palette: Record<Colors, string | RGB> = {
    red: [255, 0, 0],
    green: "#00ff00",
    bleu: [0, 0, 255]
//  ~~~~ The typo is now correctly detected
};

// But we now have an undesirable error here - 'palette.red' "could" be a string.
const redComponent = palette.red.at(0);

The new satisfies operator lets us validate that the type of an expression matches some type, without changing the resulting type of that expression. As an example, we could use satisfies to validate that all the properties of palette are compatible with string | number[]:

type Colors = "red" | "green" | "blue";

type RGB = [red: number, green: number, blue: number];

const palette = {
    red: [255, 0, 0],
    green: "#00ff00",
    bleu: [0, 0, 255]
//  ~~~~ The typo is now caught!
} satisfies Record<Colors, string | RGB>;

// Both of these methods are still accessible!
const redComponent = palette.red.at(0);
const greenNormalized = palette.green.toUpperCase();

satisfies can be used to catch lots of possible errors. For example, we could ensure that an object has all the keys of some type, but no more:

type Colors = "red" | "green" | "blue";

// Ensure that we have exactly the keys from 'Colors'.
const favoriteColors = {
    "red": "yes",
    "green": false,
    "blue": "kinda",
    "platypus": false
//  ~~~~~~~~~~ error - "platypus" was never listed in 'Colors'.
} satisfies Record<Colors, unknown>;

// All the information about the 'red', 'green', and 'blue' properties are retained.
const g: boolean = favoriteColors.green;

Maybe we don’t care about if the property names match up somehow, but we do care about the types of each property. In that case, we can also ensure that all of an object’s property values conform to some type.

type RGB = [red: number, green: number, blue: number];

const palette = {
    red: [255, 0, 0],
    green: "#00ff00",
    blue: [0, 0]
    //    ~~~~~~ error!
} satisfies Record<string, string | RGB>;

// Information about each property is still maintained.
const redComponent = palette.red.at(0);
const greenNormalized = palette.green.toUpperCase();

For more examples, you can see the issue proposing this and the implementing pull request. We’d like to thank Oleksandr Tarasiuk who implemented and iterated on this feature with us.

Unlisted Property Narrowing with the in Operator

As developers, we often need to deal with values that aren’t fully known at runtime. In fact, we often don’t know if properties exist, whether we’re getting a response from a server or reading a configuration file. JavaScript’s in operator can check whether a property exists on an object.

Previously, TypeScript allowed us to narrow away any types that don’t explicitly list a property.

interface RGB {
    red: number;
    green: number;
    blue: number;
}

interface HSV {
    hue: number;
    saturation: number;
    value: number;
}

function setColor(color: RGB | HSV) {
    if ("hue" in color) {
        // 'color' now has the type HSV
    }
    // ...
}

Here, the type RGB didn’t list the hue and got narrowed away, and leaving us with the type HSV.

But what about examples where no type listed a given property? In those cases, the language didn’t help us much. Let’s take the following example in JavaScript:

function tryGetPackageName(context) {
    const packageJSON = context.packageJSON;
    // Check to see if we have an object.
    if (packageJSON && typeof packageJSON === "object") {
        // Check to see if it has a string name property.
        if ("name" in packageJSON && typeof packageJSON.name === "string") {
            return packageJSON.name;
        }
    }

    return undefined;
}

Rewriting this to canonical TypeScript would just be a matter of defining and using a type for context; however, picking a safe type like unknown for the packageJSON property would cause issues in older versions of TypeScript.

interface Context {
    packageJSON: unknown;
}

function tryGetPackageName(context: Context) {
    const packageJSON = context.packageJSON;
    // Check to see if we have an object.
    if (packageJSON && typeof packageJSON === "object") {
        // Check to see if it has a string name property.
        if ("name" in packageJSON && typeof packageJSON.name === "string") {
        //                                              ~~~~
        // error! Property 'name' does not exist on type 'object.
            return packageJSON.name;
        //                     ~~~~
        // error! Property 'name' does not exist on type 'object.
        }
    }

    return undefined;
}

This is because while the type of packageJSON was narrowed from unknown to object, the in operator strictly narrowed to types that actually defined the property being checked. As a result, the type of packageJSON remained object.

TypeScript 4.9 makes the in operator a little bit more powerful when narrowing types that don’t list the property at all. Instead of leaving them as-is, the language will intersect their types with Record<"property-key-being-checked", unknown>.

So in our example, packageJSON will have its type narrowed from unknown to object to object & Record<"name", unknown> That allows us to access packageJSON.name directly and narrow that independently.

interface Context {
    packageJSON: unknown;
}

function tryGetPackageName(context: Context): string | undefined {
    const packageJSON = context.packageJSON;
    // Check to see if we have an object.
    if (packageJSON && typeof packageJSON === "object") {
        // Check to see if it has a string name property.
        if ("name" in packageJSON && typeof packageJSON.name === "string") {
            // Just works!
            return packageJSON.name;
        }
    }

    return undefined;
}

TypeScript 4.9 also tightens up a few checks around how in is used, ensuring that the left side is assignable to the type string | number | symbol, and the right side is assignable to object. This helps check that we’re using valid property keys, and not accidentally checking primitives.

For more information, read the implementing pull request

Checks For Equality on NaN

A major gotcha for JavaScript developers is checking against the value NaN using the built-in equality operators.

For some background, NaN is a special numeric value that stands for "Not a Number". Nothing is ever equal to NaN – even NaN!

console.log(NaN == 0)  // false
console.log(NaN === 0) // false

console.log(NaN == NaN)  // false
console.log(NaN === NaN) // false

But at least symmetrically everything is always not-equal to NaN.

console.log(NaN != 0)  // true
console.log(NaN !== 0) // true

console.log(NaN != NaN)  // true
console.log(NaN !== NaN) // true

This technically isn’t a JavaScript-specific problem, since any language that contains IEEE-754 floats has the same behavior; but JavaScript’s primary numeric type is a floating point number, and number parsing in JavaScript can often result in NaN. In turn, checking against NaN ends up being fairly common, and the correct way to do so is to use Number.isNaNbut as we mentioned, lots of people accidentally end up checking with someValue === NaN instead.

TypeScript now errors on direct comparisons against NaN, and will suggest using some variation of Number.isNaN instead.

function validate(someValue: number) {
    return someValue !== NaN;
    //     ~~~~~~~~~~~~~~~~~
    // error: This condition will always return 'true'.
    //        Did you mean '!Number.isNaN(someValue)'?
}

We believe that this change should strictly help catch beginner errors, similar to how TypeScript currently issues errors on comparisons against object and array literals.

We’d like to extend our thanks to Oleksandr Tarasiuk who contributed this check.

File-Watching Now Uses File System Events

In earlier versions, TypeScript leaned heavily on polling for watching individual files. Using a polling strategy meant checking the state of a file periodically for updates. On Node.js, fs.watchFile is the built-in way to get a polling file-watcher. While polling tends to be more predictable across platforms and file systems, it means that your CPU has to periodically get interrupted and check for updates to the file, even when nothing’s changed. For a few dozen files, this might not be noticeable; but on a bigger project with lots of files – or lots of files in node_modules – this can become a resource hog.

Generally speaking, a better approach is to use file system events. Instead of polling, we can announce that we’re interested in updates of specific files and provide a callback for when those files actually do change. Most modern platforms in use provide facilities and APIs like CreateIoCompletionPort, kqueue, epoll, and inotify. Node.js mostly abstracts these away by providing fs.watch. File system events usually work great, but there are lots of caveats to using them, and in turn, to using the fs.watch API. A watcher needs to be careful to consider inode watching, unavailability on certain file systems (e.g.networked file systems), whether recursive file watching is available, whether directory renames trigger events, and even file watcher exhaustion! In other words, it’s not quite a free lunch, especially if you’re looking for something cross-platform.

As a result, our default was to pick the lowest common denominator: polling. Not always, but most of the time.

Over time, we’ve provided the means to choose other file-watching strategies. This allowed us to get feedback and harden our file-watching implementation against most of these platform-specific gotchas. As TypeScript has needed to scale to larger codebases, and has improved in this area, we felt swapping to file system events as the default would be a worthwhile investment.

In TypeScript 4.9, file watching is powered by file system events by default, only falling back to polling if we fail to set up event-based watchers. For most developers, this should provide a much less resource-intensive experience when running in --watch mode, or running with a TypeScript-powered editor like Visual Studio or VS Code.

The way file-watching works can still be configured through environment variables and watchOptions – and some editors like VS Code can support watchOptions independently. Developers using more exotic set-ups where source code resides on a networked file systems (like NFS and SMB) may need to opt back into the older behavior; though if a server has reasonable processing power, it might just be better to enable SSH and run TypeScript remotely so that it has direct local file access. VS Code has plenty of remote extensions to make this easier.

You can read up more on this change on GitHub.

Correctness Fixes and Breaking Changes

lib.d.ts Updates

While TypeScript strives to avoid major breaks, even small changes in the built-in libraries can cause issues. We don’t expect major breaks as a result of DOM and lib.d.ts updates, but there may be some small ones.

Better Types for Promise.resolve

Promise.resolve now uses the Awaited type to unwrap Promise-like types passed to it. This means that it more often returns the right Promise type, but that improved type can break existing code if it was expecting any or unknown instead of a Promise. For more information, see the original change.

JavaScript Emit No Longer Elides Imports

When TypeScript first supported type-checking and compilation for JavaScript, it accidentally supported a feature called import elision. In short, if an import is not used as a value, or the compiler can detect that the import doesn’t refer to a value at runtime, the compiler will drop the import during emit.

This behavior was questionable, especially the detection of whether the import doesn’t refer to a value, since it means that TypeScript has to trust sometimes-inaccurate declaration files. In turn, TypeScript now preserves imports in JavaScript files.

// Input:
import { someValue, SomeClass } from "some-module";

/** @type {SomeClass} */
let val = someValue;

// Previous Output:
import { someValue } from "some-module";

/** @type {SomeClass} */
let val = someValue;

// Current Output:
import { someValue, SomeClass } from "some-module";

/** @type {SomeClass} */
let val = someValue;

More information is available at the implementing change.

exports is Prioritized Over typesVersions

Previously, TypeScript incorrectly prioritized the typesVersions field over the exports field when resolving through a package.json under --moduleResolution node16. If this change impacts your library, you may need to add types@ version selectors in your package.json‘s exports field.

  {
      "type": "module",
      "main": "./dist/main.js"
      "typesVersions": {
          "<4.8": { ".": ["4.8-types/main.d.ts"] },
          "*": { ".": ["modern-types/main.d.ts"] }
      },
      "exports": {
          ".": {
+             "types@<4.8": "4.8-types/main.d.ts",
+             "types": "modern-types/main.d.ts",
              "import": "./dist/main.js"
          }
      }
  }

For more information, see this pull request.

substitute Replaced With constraint on SubstitutionTypes

As part of an optimization on substitution types, SubstitutionType objects no longer contain the substitute property representing the effective substitution (usually an intersection of the base type and the implicit constraint) – instead, they just contain the constraint property.

For more details, read more on the original pull request.

What’s Next

Over the next few weeks, we’ll be continuing development on TypeScript 4.9, eventually shipping a release candidate and then a final release. That work will primarily include bug fixes, polish, and low-risk editor-oriented features. So while the TypeScript beta represents a mostly-feature-complete point of the release cycle, the best way to get an idea of the state of TypeScript 4.9 is to try a nightly build!

And if you’re interested in planning around the release, be sure to keep an eye on our iteration plan. In the meantime, please try out the beta today and let us know what you think!

Happy Hacking!

– Daniel Rosenwasser and the TypeScript Team

8 comments

Discussion is closed. Login to edit/delete existing comments.

  • MinskLeo 0

    I really like this updates, especially highlighting move from pooling to fs-watch. It will be probably a huge resource saving. Thanks, great job!

    • Szymon Dolny 0

      it will get in the way while working with containers

    • Christopher 0

      Yeah that’s great !

  • chukwuemeka Igbokwe 0

    I like the file watch feature, it will provide much less resource-intensive experience when running in watch mode. 👍🏾

  • David Podolsky 1

    @Daniel
    I believe you have a slight issue with your satisfies example. This line:

    // But we now have an undesirable error here - 'palette.red' "could" be a string.
    const redComponent = palette.red.at(0);
    

    This wouldn’t be a typechecking error as the .at function exists on both the array and string prototype so it satisfies both types. For your example I would pick a function that is only found on arrays. I believe pop() would work better here in this example and get your point across.

    • Stevan Skrbic 0

      👍🏾

  • Jon Lepage 0

    satisfies Operator look awesome!

    • Christopher 0

      I don’t like the satisfies that much.
      I prefer to have object and check typeof it’s better.

Feedback usabilla icon