using Declarations and Explicit Resource Management

TypeScript 5.2 adds support for the upcomingExplicit Resource Management feature in ECMAScript.Let’s explore some of the motivations and understand what the feature brings us.

It’s common to need to do some sort of “clean-up” after creating an object.For example, you might need to close network connections, delete temporary files, or just free up some memory.

Let’s imagine a function that creates a temporary file, reads and writes to it for various operations, and then closes and deletes it.

ts
import*asfsfrom"fs";
exportfunctiondoSomeWork() {
constpath =".some_temp_file";
constfile =fs.openSync(path,"w+");
// use file...
// Close the file and delete it.
fs.closeSync(file);
fs.unlinkSync(path);
}

This is fine, but what happens if we need to perform an early exit?

ts
exportfunctiondoSomeWork() {
constpath =".some_temp_file";
constfile =fs.openSync(path,"w+");
// use file...
if (someCondition()) {
// do some more work...
// Close the file and delete it.
fs.closeSync(file);
fs.unlinkSync(path);
return;
    }
// Close the file and delete it.
fs.closeSync(file);
fs.unlinkSync(path);
}

We’re starting to see some duplication of clean-up which can be easy to forget.We’re also not guaranteed to close and delete the file if an error gets thrown.This could be solved by wrapping this all in atry/finally block.

ts
exportfunctiondoSomeWork() {
constpath =".some_temp_file";
constfile =fs.openSync(path,"w+");
try {
// use file...
if (someCondition()) {
// do some more work...
return;
        }
    }
finally {
// Close the file and delete it.
fs.closeSync(file);
fs.unlinkSync(path);
    }
}

While this is more robust, it’s added quite a bit of “noise” to our code.There are also other foot-guns we can run into if we start adding more clean-up logic to ourfinally block — for example, exceptions preventing other resources from being disposed.This is what theexplicit resource management proposal aims to solve.The key idea of the proposal is to support resource disposal — this clean-up work we’re trying to deal with — as a first class idea in JavaScript.

This starts by adding a new built-insymbol calledSymbol.dispose, and we can create objects with methods named bySymbol.dispose.For convenience, TypeScript defines a new global type calledDisposable which describes these.

ts
classTempFileimplementsDisposable {
#path:string;
#handle:number;
constructor(path:string) {
this.#path =path;
this.#handle =fs.openSync(path,"w+");
    }
// other methods
    [Symbol.dispose]() {
// Close the file and delete it.
fs.closeSync(this.#handle);
fs.unlinkSync(this.#path);
    }
}

Later on we can call those methods.

ts
exportfunctiondoSomeWork() {
constfile =newTempFile(".some_temp_file");
try {
// ...
    }
finally {
file[Symbol.dispose]();
    }
}

Moving the clean-up logic toTempFile itself doesn’t buy us much;we’ve basically just moved all the clean-up work from thefinally block into a method, and that’s always been possible.But having a well-known “name” for this method means that JavaScript can build other features on top of it.

That brings us to the first star of the feature:using declarations!using is a new keyword that lets us declare new fixed bindings, kind of likeconst.The key difference is that variables declared withusing get theirSymbol.dispose method called at the end of the scope!

So we could simply have written our code like this:

ts
exportfunctiondoSomeWork() {
usingfile =newTempFile(".some_temp_file");
// use file...
if (someCondition()) {
// do some more work...
return;
    }
}

Check it out — notry/finally blocks!At least, none that we see.Functionally, that’s exactly whatusing declarations will do for us, but we don’t have to deal with that.

You might be familiar withusing declarations in C#,with statements in Python, ortry-with-resource declarations in Java.These are all similar to JavaScript’s newusing keyword, and provide a similar explicit way to perform a “tear-down” of an object at the end of a scope.

using declarations do this clean-up at the very end of their containing scope or right before an “early return” like areturn or athrown error.They also dispose in a first-in-last-out order like a stack.

ts
functionloggy(id:string):Disposable {
console.log(`Creating${id}`);
return {
        [Symbol.dispose]() {
console.log(`Disposing${id}`);
        }
    }
}
functionfunc() {
usinga =loggy("a");
usingb =loggy("b");
    {
usingc =loggy("c");
usingd =loggy("d");
    }
usinge =loggy("e");
return;
// Unreachable.
// Never created, never disposed.
usingf =loggy("f");
}
func();
// Creating a
// Creating b
// Creating c
// Creating d
// Disposing d
// Disposing c
// Creating e
// Disposing e
// Disposing b
// Disposing a

using declarations are supposed to be resilient to exceptions;if an error is thrown, it’s rethrown after disposal.On the other hand, the body of your function might execute as expected, but theSymbol.dispose might throw.In that case, that exception is rethrown as well.

But what happens if both the logic before and during disposal throws an error?For those cases,SuppressedError has been introduced as a new subtype ofError.It features asuppressed property that holds the last-thrown error, and anerror property for the most-recently thrown error.

ts
classErrorAextendsError {
name ="ErrorA";
}
classErrorBextendsError {
name ="ErrorB";
}
functionthrowy(id:string) {
return {
        [Symbol.dispose]() {
thrownewErrorA(`Error from${id}`);
        }
    };
}
functionfunc() {
usinga =throwy("a");
thrownewErrorB("oops!")
}
try {
func();
}
catch (e:any) {
console.log(e.name);// SuppressedError
console.log(e.message);// An error was suppressed during disposal.
console.log(e.error.name);// ErrorA
console.log(e.error.message);// Error from a
console.log(e.suppressed.name);// ErrorB
console.log(e.suppressed.message);// oops!
}

You might have noticed that we’re using synchronous methods in these examples.However, lots of resource disposal involvesasynchronous operations, and we need to wait for those to complete before we continue running any other code.

That’s why there is also a newSymbol.asyncDispose, and it brings us to the next star of the show —await using declarations.These are similar tousing declarations, but the key is that they look up whose disposal must beawaited.They use a different method named bySymbol.asyncDispose, though they can operate on anything with aSymbol.dispose as well.For convenience, TypeScript also introduces a global type calledAsyncDisposable that describes any object with an asynchronous dispose method.

ts
asyncfunctiondoWork() {
// Do fake work for half a second.
awaitnewPromise(resolve=>setTimeout(resolve,500));
}
functionloggy(id:string):AsyncDisposable {
console.log(`Constructing${id}`);
return {
async [Symbol.asyncDispose]() {
console.log(`Disposing (async)${id}`);
awaitdoWork();
        },
    }
}
asyncfunctionfunc() {
awaitusinga =loggy("a");
awaitusingb =loggy("b");
    {
awaitusingc =loggy("c");
awaitusingd =loggy("d");
    }
awaitusinge =loggy("e");
return;
// Unreachable.
// Never created, never disposed.
awaitusingf =loggy("f");
}
func();
// Constructing a
// Constructing b
// Constructing c
// Constructing d
// Disposing (async) d
// Disposing (async) c
// Constructing e
// Disposing (async) e
// Disposing (async) b
// Disposing (async) a

Defining types in terms ofDisposable andAsyncDisposable can make your code much easier to work with if you expect others to do tear-down logic consistently.In fact, lots of existing types exist in the wild which have adispose() orclose() method.For example, the Visual Studio Code APIs even definetheir ownDisposable interface.APIs in the browser and in runtimes like Node.js, Deno, and Bun might also choose to useSymbol.dispose andSymbol.asyncDispose for objects which already have clean-up methods, like file handles, connections, and more.

Now maybe this all sounds great for libraries, but a little bit heavy-weight for your scenarios.If you’re doing a lot of ad-hoc clean-up, creating a new type might introduce a lot of over-abstraction and questions about best-practices.For example, take ourTempFile example again.

ts
classTempFileimplementsDisposable {
#path:string;
#handle:number;
constructor(path:string) {
this.#path =path;
this.#handle =fs.openSync(path,"w+");
    }
// other methods
    [Symbol.dispose]() {
// Close the file and delete it.
fs.closeSync(this.#handle);
fs.unlinkSync(this.#path);
    }
}
exportfunctiondoSomeWork() {
usingfile =newTempFile(".some_temp_file");
// use file...
if (someCondition()) {
// do some more work...
return;
    }
}

All we wanted was to remember to call two functions — but was this the best way to write it?Should we be callingopenSync in the constructor, create anopen() method, or pass in the handle ourselves?Should we expose a method for every possible operation we need to perform, or should we just make the properties public?

That brings us to the final stars of the feature:DisposableStack andAsyncDisposableStack.These objects are useful for doing both one-off clean-up, along with arbitrary amounts of cleanup.ADisposableStack is an object that has several methods for keeping track ofDisposable objects, and can be given functions for doing arbitrary clean-up work.We can also assign them tousing variables because — get this —they’re alsoDisposable!So here’s how we could’ve written the original example.

ts
functiondoSomeWork() {
constpath =".some_temp_file";
constfile =fs.openSync(path,"w+");
usingcleanup =newDisposableStack();
cleanup.defer(()=> {
fs.closeSync(file);
fs.unlinkSync(path);
    });
// use file...
if (someCondition()) {
// do some more work...
return;
    }
// ...
}

Here, thedefer() method just takes a callback, and that callback will be run oncecleanup is disposed of.Typically,defer (and otherDisposableStack methods likeuse andadopt)should be called immediately after creating a resource.As the name suggests,DisposableStack disposes of everything it keeps track of like a stack, in a first-in-last-out order, sodefering immediately after creating a value helps avoid odd dependency issues.AsyncDisposableStack works similarly, but can keep track ofasync functions andAsyncDisposables, and is itself anAsyncDisposable.

Thedefer method is similar in many ways to thedefer keyword inGo,Swift,Zig,Odin, and others, where the conventions should be similar.

Because this feature is so recent, most runtimes will not support it natively.To use it, you will need runtime polyfills for the following:

• Symbol.dispose
• Symbol.asyncDispose
• DisposableStack
• AsyncDisposableStack
• SuppressedError

However, if all you’re interested in isusing andawait using, you should be able to get away with only polyfilling the built-insymbols.Something as simple as the following should work for most cases:

ts
Symbol.dispose ??=Symbol("Symbol.dispose");
Symbol.asyncDispose ??=Symbol("Symbol.asyncDispose");

You will also need to set your compilationtarget toes2022 or below, and configure yourlib setting to either include"esnext" or"esnext.disposable".

json
{
"compilerOptions": {
"target":"es2022",
"lib": ["es2022","esnext.disposable","dom"]
    }
}

For more information on this feature,take a look at the work on GitHub!

Decorator Metadata

TypeScript 5.2 implementsan upcoming ECMAScript feature called decorator metadata.

The key idea of this feature is to make it easy for decorators to create and consume metadata on any class they’re used on or within.

Whenever decorator functions are used, they now have access to a newmetadata property on their context object.Themetadata property just holds a simple object.Since JavaScript lets us add properties arbitrarily, it can be used as a dictionary that is updated by each decorator.Alternatively, since everymetadata object will be identical for each decorated portion of a class, it can be used as a key into aMap.After all decorators on or in a class get run, that object can be accessed on the class viaSymbol.metadata.

ts
interfaceContext {
name:string;
metadata:Record<PropertyKey,unknown>;
}
functionsetMetadata(_target:any,context:Context) {
context.metadata[context.name] =true;
}
classSomeClass {
    @setMetadata
foo =123;
    @setMetadata
accessorbar ="hello!";
    @setMetadata
baz() { }
}
constourMetadata =SomeClass[Symbol.metadata];
console.log(JSON.stringify(ourMetadata));
// { "bar": true, "baz": true, "foo": true }

This can be useful in a number of different scenarios.Metadata could possibly be attached for lots of uses like debugging, serialization, or performing dependency injection with decorators.Since metadata objects are created per decorated class, frameworks can either privately use them as keys into aMap orWeakMap, or tack properties on as necessary.

For example, let’s say we wanted to use decorators to keep track of which properties and accessors are serializable when usingJSON.stringify like so:

ts
import {serialize,jsonify }from"./serializer";
classPerson {
firstName:string;
lastName:string;
    @serialize
age:number
    @serialize
getfullName() {
return`${this.firstName}${this.lastName}`;
    }
toJSON() {
returnjsonify(this)
    }
constructor(firstName:string,lastName:string,age:number) {
// ...
    }
}

Here, the intent is that onlyage andfullName should be serialized because they are marked with the@serialize decorator.We define atoJSON method for this purpose, but it just calls out tojsonify which uses the metadata that@serialize created.

Here’s an example of how the module./serialize.ts might be defined:

ts
constserializables =Symbol();
typeContext =
    |ClassAccessorDecoratorContext
    |ClassGetterDecoratorContext
    |ClassFieldDecoratorContext
    ;
exportfunctionserialize(_target:any,context:Context):void {
if (context.static ||context.private) {
thrownewError("Can only serialize public instance members.")
    }
if (typeofcontext.name ==="symbol") {
thrownewError("Cannot serialize symbol-named properties.");
    }
constpropNames =
        (context.metadata[serializables]asstring[] |undefined) ??= [];
propNames.push(context.name);
}
exportfunctionjsonify(instance:object):string {
constmetadata =instance.constructor[Symbol.metadata];
constpropNames =metadata?.[serializables]asstring[] |undefined;
if (!propNames) {
thrownewError("No members marked with @serialize.");
    }
constpairStrings =propNames.map(key=> {
conststrKey =JSON.stringify(key);
conststrValue =JSON.stringify((instanceasany)[key]);
return`${strKey}:${strValue}`;
    });
return`{${pairStrings.join(", ")} }`;
}

This module has a localsymbol calledserializables to store and retrieve the names of properties marked@serializable.It stores a list of these property names on the metadata on each invocation of@serializable.Whenjsonify is called, the list of properties is fetched off of the metadata and used to retrieve the actual values from the instance, eventually serializing those names and values.

Using asymbol technically makes this data accessible to others.An alternative might be to use aWeakMap using the metadata object as a key.This keeps data private and happens to use fewer type assertions in this case, but is otherwise similar.

ts
constserializables =newWeakMap<object,string[]>();
typeContext =
    |ClassAccessorDecoratorContext
    |ClassGetterDecoratorContext
    |ClassFieldDecoratorContext
    ;
exportfunctionserialize(_target:any,context:Context):void {
if (context.static ||context.private) {
thrownewError("Can only serialize public instance members.")
    }
if (typeofcontext.name !=="string") {
thrownewError("Can only serialize string properties.");
    }
letpropNames =serializables.get(context.metadata);
if (propNames ===undefined) {
serializables.set(context.metadata,propNames = []);
    }
propNames.push(context.name);
}
exportfunctionjsonify(instance:object):string {
constmetadata =instance.constructor[Symbol.metadata];
constpropNames =metadata &&serializables.get(metadata);
if (!propNames) {
thrownewError("No members marked with @serialize.");
    }
constpairStrings =propNames.map(key=> {
conststrKey =JSON.stringify(key);
conststrValue =JSON.stringify((instanceasany)[key]);
return`${strKey}:${strValue}`;
    });
return`{${pairStrings.join(", ")} }`;
}

As a note, these implementations don’t handle subclassing and inheritance.That’s left as an exercise to you (and you might find that it is easier in one version of the file than the other!).

Because this feature is still fresh, most runtimes will not support it natively.To use it, you will need a polyfill forSymbol.metadata.Something as simple as the following should work for most cases:

ts
Symbol.metadata ??=Symbol("Symbol.metadata");

You will also need to set your compilationtarget toes2022 or below, and configure yourlib setting to either include"esnext" or"esnext.decorators".

json
{
"compilerOptions": {
"target":"es2022",
"lib": ["es2022","esnext.decorators","dom"]
    }
}

We’d like to thankOleksandr Tarasiuk for contributingthe implementation of decorator metadata for TypeScript 5.2!

Named and Anonymous Tuple Elements

Tuple types have supported optional labels or names for each element.

ts
typePair<T> = [first:T, second:T];

These labels don’t change what you’re allowed to do with them — they’re solely to help with readability and tooling.

However, TypeScript previously had a rule that tuples could not mix and match between labeled and unlabeled elements.In other words, either no element could have a label in a tuple, or all elements needed one.

ts
// ✅ fine - no labels
typePair1<T> = [T,T];
// ✅ fine - all fully labeled
typePair2<T> = [first:T, second:T];
// ❌ previously an error
typePair3<T> = [first:T,T];
//                         ~
// Tuple members must all have names
// or all not have names.

This could be annoying for rest elements where we’d be forced to just add a label likerest ortail.

ts
// ❌ previously an error
typeTwoOrMore_A<T> = [first:T, second:T, ...T[]];
//                                          ~~~~~~
// Tuple members must all have names
// or all not have names.
// ✅
typeTwoOrMore_B<T> = [first:T, second:T, rest: ...T[]];

It also meant that this restriction had to be enforced internally in the type system, meaning TypeScript would lose labels.

ts
typeHasLabels = [a:string, b:string];
typeHasNoLabels = [number,number];
typeMerged = [...HasNoLabels, ...HasLabels];
//   ^ [number, number, string, string]
//
//     'a' and 'b' were lost in 'Merged'

In TypeScript 5.2, the all-or-nothing restriction on tuple labels has been lifted.The language can now also preserve labels when spreading into an unlabeled tuple.

We’d like to extend our thanks toJosh Goldberg andMateusz Burzyński whocollaborated to lift this restriction.

Easier Method Usage for Unions of Arrays

In previous versions of TypeScript, calling a method on a union of arrays could end in pain.

ts
declareletarray:string[] |number[];
array.filter(x=> !!x);
//    ~~~~~~ error!
// This expression is not callable.
//   Each member of the union type '...' has signatures,
//   but none of those signatures are compatible
//   with each other.

In this example, TypeScript would try to see if each version offilter is compatible acrossstring[] andnumber[].Without a coherent strategy, TypeScript threw its hands in the air and said “I can’t make it work”.

In TypeScript 5.2, before giving up in these cases, unions of arrays are treated as a special case.A new array type is constructed out of each member’s element type, and then the method is invoked on that.

Taking the above example,string[] | number[] is transformed into(string | number)[] (orArray<string | number>), andfilter is invoked on that type.There is a slight caveat which is thatfilter will produce anArray<string | number> instead of astring[] | number[];but for a freshly produced value there is less risk of something “going wrong”.

This means lots of methods likefilter,find,some,every, andreduce should all be invokable on unions of arrays in cases where they were not previously.

You canread up more details on the implementing pull request.

Type-Only Import Paths with TypeScript Implementation File Extensions

TypeScript now allows both declarationand implementation file extensions to be included in type-only import paths, regardless of whetherallowImportingTsExtensions is enabled.

This means that you can now writeimport type statements that use.ts,.mts,.cts, and.tsx file extensions.

ts
importtype {JustAType }from"./justTypes.ts";
exportfunctionf(param:JustAType) {
// ...
}

It also means thatimport() types, which can be used in both TypeScript and JavaScript with JSDoc, can use those file extensions.

js
/**
 *@param{import("./justTypes.ts").JustAType}param
 */
exportfunctionf(param) {
// ...
}

For more information,see the change here.

Comma Completions for Object Members

It can be easy to forget to add a comma when adding a new property to an object.Previously, if you forgot a comma and requested auto-completion, TypeScript would confusingly give poor unrelated completion results.

TypeScript 5.2 now gracefully provides object member completions when you’re missing a comma.But to just skip past hitting you with a syntax error, it willalso auto-insert the missing comma.

For more information,see the implementation here.

Inline Variable Refactoring

TypeScript 5.2 now has a refactoring to inline the contents of a variable to all usage sites.

.

Using the “inline variable” refactoring will eliminate the variable and replace all the variable’s usages with its initializer.Note that this may cause that initializer’s side-effects to run at a different time, and as many times as the variable has been used.

For more details,see the implementing pull request.

Optimized Checks for Ongoing Type Compatibility

Because TypeScript is a structural type system, types occasionally need to be compared in a member-wise fashion;however, recursive types add some issues here.For example:

ts
interfaceA {
value:A;
other:string;
}
interfaceB {
value:B;
other:number;
}

When checking whether the typeA is compatible with the typeB, TypeScript will end up checking whether the types ofvalue inA andB are respectively compatible.At this point, the type system needs to stop checking any further and proceed to check other members.To do this, the type system has to track when any two types are already being related.

Previously TypeScript already kept a stack of type pairs, and iterated through that to determine whether those types are being related.When this stack is shallow that’s not a problem; but when the stack isn’t shallow, that, uh,is a problem.

In TypeScript 5.3, a simpleSet helps track this information.This reduced the time spent on a reported test case that used thedrizzle library by over 33%!

Benchmark 1: old
  Time (mean ± σ):      3.115 s ±  0.067 s    [User: 4.403 s, System: 0.124 s]
  Range (min … max):    3.018 s …  3.196 s    10 runs
Benchmark 2: new
  Time (mean ± σ):      2.072 s ±  0.050 s    [User: 3.355 s, System: 0.135 s]
  Range (min … max):    1.985 s …  2.150 s    10 runs
Summary
  'new' ran
    1.50 ± 0.05 times faster than 'old'

Read more on the change here.

Breaking Changes and Correctness Fixes

TypeScript strives not to unnecessarily introduce breaks;however, occasionally we must make corrections and improvements so that code can be better-analyzed.

lib.d.ts Changes

Types generated for the DOM may have an impact on your codebase.For more information,see the DOM updates for TypeScript 5.2.

labeledElementDeclarations May Holdundefined Elements

In orderto support a mixture of labeled and unlabeled elements, TypeScript’s API has changed slightly.ThelabeledElementDeclarations property ofTupleType may holdundefined for at each position where an element is unlabeled.

diff
  interface TupleType {
-     labeledElementDeclarations?: readonly (NamedTupleMember | ParameterDeclaration)[];
+     labeledElementDeclarations?: readonly (NamedTupleMember | ParameterDeclaration | undefined)[];
  }

module andmoduleResolution Must Match Under Recent Node.js settings

The--module and--moduleResolution options each support anode16 andnodenext setting.These are effectively “modern Node.js” settings that should be used on any recent Node.js project.What we’ve found is that when these two options don’t agree on whether they are using Node.js-related settings, projects are effectively misconfigured.

In TypeScript 5.2, when usingnode16 ornodenext for either of the--module and--moduleResolution options, TypeScript now requires the other to have a similar Node.js-related setting.In cases where the settings diverge, you’ll likely get an error message like either

Option 'moduleResolution' must be set to 'NodeNext' (or left unspecified) when option 'module' is set to 'NodeNext'.

or

Option 'module' must be set to 'Node16' when option 'moduleResolution' is set to 'Node16'.

So for example--module esnext --moduleResolution node16 will be rejected — but you may be better off just using--module nodenext alone, or--module esnext --moduleResolution bundler.

For more information,see the change here.

Consistent Export Checking for Merged Symbols

When two declarations merge, they must agree on whether they are both exported.Due to a bug, TypeScript missed specific cases in ambient contexts, like in declaration files ordeclare module blocks.For example, it would not issue an error on a case like the following, wherereplaceInFile is declared once as an exported function, and one as an un-exported namespace.

ts
declaremodule'replace-in-file' {
exportfunctionreplaceInFile(config:unknown):Promise<unknown[]>;
export {};
namespacereplaceInFile {
exportfunctionsync(config:unknown):unknown[];
  }
}

In an ambient module, adding anexport { ... } or a similar construct likeexport default ... implicitly changes whether all declarations are automatically exported.TypeScript now recognizes these unfortunately confusing semantics more consistently, and issues an error on the fact that all declarations ofreplaceInFile need to agree in their modifiers, and will issue the following error:

Individual declarations in merged declaration 'replaceInFile' must be all exported or all local.

For more information,see the change here.