Was this page helpful?

TypeScript 4.9

ts
// Each property can be a string or an RGB tuple.
constpalette = {
red: [255,0,0],
green:"#00ff00",
bleu: [0,0,255]
//  ^^^^ sacrebleu - we've made a typo!
};
// We want to be able to use string methods on 'green'...
constgreenNormalized =palette.green.toUpperCase();

ts
typeColors ="red" |"green" |"blue";
typeRGB = [red:number, green:number, blue:number];
constpalette: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.green' "could" be of type RGB and
// property 'toUpperCase' does not exist on type 'string | RGB'.
constgreenNormalized =palette.green.toUpperCase();

ts
typeColors ="red" |"green" |"blue";
typeRGB = [red:number, green:number, blue:number];
constpalette = {
red: [255,0,0],
green:"#00ff00",
bleu: [0,0,255]
//  ~~~~ The typo is now caught!
}satisfiesRecord<Colors,string | RGB>;
// toUpperCase() method is still accessible!
constgreenNormalized =palette.green.toUpperCase();

ts
typeColors ="red" |"green" |"blue";
// Ensure that we have exactly the keys from 'Colors'.
constfavoriteColors = {
"red":"yes",
"green":false,
"blue":"kinda",
"platypus":false
//  ~~~~~~~~~~ error - "platypus" was never listed in 'Colors'.
}satisfiesRecord<Colors,unknown>;
// All the information about the 'red', 'green', and 'blue' properties are retained.
constg:boolean =favoriteColors.green;

ts
typeRGB = [red:number, green:number, blue:number];
constpalette = {
red: [255,0,0],
green:"#00ff00",
blue: [0,0]
//    ~~~~~~ error!
}satisfiesRecord<string,string | RGB>;
// Information about each property is still maintained.
constredComponent =palette.red.at(0);
constgreenNormalized =palette.green.toUpperCase();

ts
interfaceRGB {
red:number;
green:number;
blue:number;
}
interfaceHSV {
hue:number;
saturation:number;
value:number;
}
functionsetColor(color:RGB |HSV) {
if ("hue"incolor) {
// 'color' now has the type HSV
    }
// ...
}

Rewriting this to canonical TypeScript would just be a matter of defining and using a type forcontext;however, picking a safe type likeunknown for thepackageJSON property would cause issues in older versions of TypeScript.

ts
interfaceContext {
packageJSON:unknown;
}
functiontryGetPackageName(context:Context) {
constpackageJSON =context.packageJSON;
// Check to see if we have an object.
if (packageJSON &&typeofpackageJSON ==="object") {
// Check to see if it has a string name property.
if ("name"inpackageJSON &&typeofpackageJSON.name ==="string") {
//                                              ~~~~
// error! Property 'name' does not exist on type 'object.
returnpackageJSON.name;
//                     ~~~~
// error! Property 'name' does not exist on type 'object.
        }
    }
returnundefined;
}

This is because while the type ofpackageJSON was narrowed fromunknown toobject, thein operator strictly narrowed to types that actually defined the property being checked.As a result, the type ofpackageJSON remainedobject.

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

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

ts
interfaceContext {
packageJSON:unknown;
}
functiontryGetPackageName(context:Context):string |undefined {
constpackageJSON =context.packageJSON;
// Check to see if we have an object.
if (packageJSON &&typeofpackageJSON ==="object") {
// Check to see if it has a string name property.
if ("name"inpackageJSON &&typeofpackageJSON.name ==="string") {
// Just works!
returnpackageJSON.name;
        }
    }
returnundefined;
}

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

For more information,read the implementing pull request

Auto-Accessors in Classes

TypeScript 4.9 supports an upcoming feature in ECMAScript called auto-accessors.Auto-accessors are declared just like properties on classes, except that they’re declared with theaccessor keyword.

ts
classPerson {
accessorname:string;
constructor(name:string) {
this.name =name;
    }
}

Under the covers, these auto-accessors “de-sugar” to aget andset accessor with an unreachable private property.

ts
classPerson {
#__name:string;
getname() {
returnthis.#__name;
    }
setname(value:string) {
this.#__name =value;
    }
constructor(name:string) {
this.name =name;
    }
}

You canread up more about the auto-accessors pull request on the original PR.

Checks For Equality onNaN

A major gotcha for JavaScript developers is checking against the valueNaN 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 toNaN - evenNaN!

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

But at least symmetricallyeverything is always not-equal toNaN.

js
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 inNaN.In turn, checking againstNaN ends up being fairly common, and the correct way to do so is to useNumber.isNaN -but as we mentioned, lots of people accidentally end up checking withsomeValue === NaN instead.

TypeScript now errors on direct comparisons againstNaN, and will suggest using some variation ofNumber.isNaN instead.

ts
functionvalidate(someValue:number) {
returnsomeValue !==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 toOleksandr Tarasiuk whocontributed this check.

File-Watching Now Uses File System Events

In earlier versions, TypeScript leaned heavily onpolling 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 innode_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 filesactually do change.Most modern platforms in use provide facilities and APIs likeCreateIoCompletionPort,kqueue,epoll, andinotify.Node.js mostly abstracts these away by providingfs.watch.File system events usually work great, but there arelots of caveats to using them, and in turn, to using thefs.watch API.A watcher needs to be careful to considerinode 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 tochoose 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 andwatchOptions - andsome editors like VS Code can supportwatchOptions 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 ofremote extensions to make this easier.

You canread up more on this change on GitHub.

“Remove Unused Imports” and “Sort Imports” Commands for Editors

Previously, TypeScript only supported two editor commands to manage imports.For our examples, take the following code:

ts
import {Zebra,Moose,HoneyBadger }from"./zoo";
import {foo,bar }from"./helper";
letx:Moose |HoneyBadger =foo();

The first was called “Organize Imports” which would remove unused imports, and then sort the remaining ones.It would rewrite that file to look like this one:

ts
import {foo }from"./helper";
import {HoneyBadger,Moose }from"./zoo";
letx:Moose |HoneyBadger =foo();

In TypeScript 4.3, we introduced a command called “Sort Imports” which wouldonly sort imports in the file, but not remove them - and would rewrite the file like this.

ts
import {bar,foo }from"./helper";
import {HoneyBadger,Moose,Zebra }from"./zoo";
letx:Moose |HoneyBadger =foo();

The caveat with “Sort Imports” was that in Visual Studio Code, this feature was only available as an on-save command - not as a manually triggerable command.

TypeScript 4.9 adds the other half, and now provides “Remove Unused Imports”.TypeScript will now remove unused import names and statements, but will otherwise leave the relative ordering alone.

ts
import {Moose,HoneyBadger }from"./zoo";
import {foo }from"./helper";
letx:Moose |HoneyBadger =foo();

This feature is available to all editors that wish to use either command;but notably, Visual Studio Code (1.73 and later) will have support built inand will surface these commands via its Command Palette.Users who prefer to use the more granular “Remove Unused Imports” or “Sort Imports” commands should be able to reassign the “Organize Imports” key combination to them if desired.

You canview specifics of the feature here.

Go-to-Definition onreturn Keywords

In the editor, when running a go-to-definition on thereturn keyword, TypeScript will now jump you to the top of the corresponding function.This can be helpful to get a quick sense of which function areturn belongs to.

We expect TypeScript will expand this functionality to more keywordssuch asawait andyield orswitch,case, anddefault.

This feature was implemented thanks toOleksandr Tarasiuk.

Performance Improvements

TypeScript has a few small, but notable, performance improvements.

First, TypeScript’sforEachChild function has been rewritten to use a function table lookup instead of aswitch statement across all syntax nodes.forEachChild is a workhorse for traversing syntax nodes in the compiler, and is used heavily in the binding stage of our compiler, along with parts of the language service.The refactoring offorEachChild yielded up to a 20% reduction of time spent in our binding phase and across language service operations.

Once we discovered this performance win forforEachChild, we tried it out onvisitEachChild, a function we use for transforming nodes in the compiler and language service.The same refactoring yielded up to a 3% reduction in time spent in generating project output.

The initial exploration inforEachChild wasinspired by a blog post byArtemis Everfree.While we have some reason to believe the root cause of our speed-up might have more to do with function size/complexity than the issues described in the blog post, we’re grateful that we were able to learn from the experience and try out a relatively quick refactoring that made TypeScript faster.

Finally, the way TypeScript preserves the information about a type in the true branch of a conditional type has been optimized.In a type like

ts
interfaceZoo<TextendsAnimal> {
// ...
}
typeMakeZoo<A> =AextendsAnimal ?Zoo<A> :never;

TypeScript has to “remember” thatA must also be anAnimal when checking ifZoo<A> is valid.This is basically done by creating a special type that used to hold the intersection ofA withAnimal;however, TypeScript previously did this eagerly which isn’t always necessary.Furthermore, some faulty code in our type-checker prevented these special types from being simplified.TypeScript now defers intersecting these types until it’s necessary.For codebases with heavy use of conditional types, you might witness significant speed-ups with TypeScript, but in our performance testing suite, we saw a more modest 3% reduction in type-checking time.

You can read up more on these optimizations on their respective pull requests:

• forEachChild as a jump-table
• visitEachChild as a jump-table
• Optimize substitition types

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 andlib.d.ts updates, but there may be some small ones.

Better Types forPromise.resolve

Promise.resolve now uses theAwaited type to unwrap Promise-like types passed to it.This means that it more often returns the rightPromise type, but that improved type can break existing code if it was expectingany orunknown instead of aPromise.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.

js
// Input:
import {someValue,SomeClass }from"some-module";
/**@type{SomeClass} */
letval =someValue;
// Previous Output:
import {someValue }from"some-module";
/**@type{SomeClass} */
letval =someValue;
// Current Output:
import {someValue,SomeClass }from"some-module";
/**@type{SomeClass} */
letval =someValue;

More information is available atthe implementing change.

exports is Prioritized OvertypesVersions

Previously, TypeScript incorrectly prioritized thetypesVersions field over theexports field when resolving through apackage.json under--moduleResolution node16.If this change impacts your library, you may need to addtypes@ version selectors in yourpackage.json’sexports field.

diff
  {
      "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 Withconstraint onSubstitutionTypes

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

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