Daniel RosenwasserToday we're announcing the availability of TypeScript 4.1 Beta!
To get started using the beta, you can get it through NuGet, or use npm with the following command:
npm install typescript@beta
You can also get editor support by
- Downloading for Visual Studio 2019/2017
- Following directions for Visual Studio Code and Sublime Text.
For this release, we have some exciting new features, new checking flags, editor productivity updates, and speed improvements.
Let's get a look at what 4.1 has in store for us!
- Template Literal Types
- Key Remapping in Mapped Types
- Recursive Conditional Types
--noUncheckedIndexedAccesspathswithoutbaseUrlcheckJsImpliesallowJs- React 17 JSX Factories
- Editor Support for the JSDoc
@seeTag - Breaking Changes
Template Literal Types
String literal types in TypeScript allow us to model functions and APIs that expect a set of specific strings.
function setVerticalAlignment(color: "top" | "middle" | "bottom") {
// ...
}
setVerticalAlignment("middel");
// ~~~~~~~~
// error: Argument of type '"middel"' is not assignable to
// parameter of type '"top" | "middle" | "bottom"'.
This is pretty nice because string literal types can basically spell-check our string values.
We also like that string literals can be used as property names in mapped types.
In this sense, they're also usable as building blocks.
type Options = {
[K in "noImplicitAny" | "strictNullChecks" | "strictFunctionTypes"]?: boolean
};
// same as
// type Options = {
// noImplicitAny?: boolean,
// strictNullChecks?: boolean,
// strictFunctionTypes?: boolean
// };
But there's another place that that string literal types could be used as building blocks: building other string literal types.
That's why TypeScript 4.1 brings the template literal string type.
It has the same syntax as template literal strings in JavaScript, but is used in type positions.
When you use it with concrete literal types, it produces a new string literal type by concatenating the contents.
type World = "world";
type Greeting = `hello ${World}`;
// same as
// type Greeting = "hello world";
What happens when you have unions in substitution positions?
It produces the set of every possible string literal that could be represented by each union member.
type Color = "red" | "blue";
type Quantity = "one" | "two";
type SeussFish = `${Quantity | Color} fish`;
// same as
// type SeussFish = "one fish" | "two fish"
// | "red fish" | "blue fish";
This can be used beyond cute examples in release notes.
For example, several libraries for UI components have a way to specify both vertical and horizontal alignment in their APIs, often with both at once using a single string like "bottom-right".
Between vertically aligning with "top", "middle", and "bottom", and horizontally aligning with "left", "center", and "right", there are 9 possible strings where each of the former strings are connected with the latter strings using a dash.
type VerticalAlignment = "top" | "middle" | "bottom";
type HorizontalAlignment = "left" | "center" | "right";
// Takes
// | "top-left" | "top-center" | "top-right"
// | "middle-left" | "middle-center" | "middle-right"
// | "bottom-left" | "bottom-center" | "bottom-right"
declare function setAlignment(value: `${VerticalAlignment}-${HorizontalAlignment}`): void;
setAlignment("top-left"); // works!
setAlignment("top-middel"); // error!
setAlignment("top-pot"); // error! but good doughnuts if you're ever in Seattle
While there are lots of examples of this sort of API in the wild, this is still a bit of a toy example since we could write these out manually.
In fact, for 9 strings, this is likely fine; but when you need a ton of strings, you should consider automatically generating them ahead of time to save work on every type-check (or just use string, which will be much simpler to comprehend).
Some of the real value comes from dynamically creating new string literals.
For example, imagine a makeWatchedObject API that takes an object and produces a mostly identical object, but with a new on method to detect for changes to the properties.
let person = makeWatchedObject({
firstName: "Homer",
age: 42, // give-or-take
location: "Springfield",
});
person.on("firstNameChanged", () => {
console.log(`firstName was changed!`);
});
Notice that on listens on the event "firstNameChanged", not just "firstName".
How would we type this?
type PropEventSource<T> = {
on(eventName: `${string & keyof T}Changed`, callback: () => void): void;
};
/// Create a "watched object" with an 'on' method
/// so that you can watch for changes to properties.
declare function makeWatchedObject<T>(obj: T): T & PropEventSource<T>;
With this, we can build something that errors when we give the wrong property!
// error!
person.on("firstName", () => {
});
// error!
person.on("frstNameChanged", () => {
});
We can also do something special in template literal types: we can infer from substitution positions.
We can make our last example generic to infer from parts of the eventName string to figure out the associated property.
type PropEventSource<T> = {
on<K extends string & keyof T>
(eventName: `${K}Changed`, callback: (newValue: T[K]) => void ): void;
};
declare function makeWatchedObject<T>(obj: T): T & PropEventSource<T>;
let person = makeWatchedObject({
firstName: "Homer",
age: 42,
location: "Springfield",
});
// works! 'newName' is typed as 'string'
person.on("firstNameChanged", newName => {
// 'newName' has the type of 'firstName'
console.log(`new name is ${newName.toUpperCase()}`);
});
// works! 'newAge' is typed as 'number'
person.on("ageChanged", newAge => {
if (newAge < 0) {
console.log("warning! negative age");
}
})
Here we made on into a generic method.
When a user calls with the string "firstNameChanged', TypeScript will try to infer the right type for K.
To do that, it will match K against the content prior to "Changed" and infer the string "firstName".
Once TypeScript figures that out, the on method can fetch the type of firstName on the original object, which is string in this case.
Similarly, when we call with "ageChanged", it finds the type for the property age which is number).
Inference can be combined in different ways, often to deconstruct strings, and reconstruct them in different ways.
In fact, to help with modifying these string literal types, we've added a few new utilities for modifying casing in letters (i.e. converting to lowercase and uppercase characters).
type EnthusiasticGreeting<T extends string> = `${uppercase T}`
type HELLO = EnthusiasticGreeting<"hello">;
// same as
// type HELLO = "HELLO";
Currently, these utilities are type operators named uppercase, lowercase, capitalize, and uncapitalize.
The first two transform every character in a string, and the latter two transform only the first character in a string.
They can only be used in template substitution positions, and in 4.1's final release, they will likely be switched over to type aliases named Uppercase, Lowercase, Capitalize and Uncapitalize.
For more details, see the original pull request and the in-progress pull request to switch to type alias helpers.
Key Remapping in Mapped Types
Just as a refresher, a mapped type can create new object types based on arbitrary keys
type Options = {
[K in "noImplicitAny" | "strictNullChecks" | "strictFunctionTypes"]?: boolean
};
// same as
// type Options = {
// noImplicitAny?: boolean,
// strictNullChecks?: boolean,
// strictFunctionTypes?: boolean
// };
or new object types based on other object types.
/// 'Partial<T>' is the same as 'T', but with each property marked optional.
type Partial<T> = {
[K in keyof T]?: T[K]
};
Until now, mapped types could only produce new object types with keys that you provided them; however, lots of the time you want to be able to create new keys, or filter out keys, based on the inputs.
That's why TypeScript 4.1 allows you to re-map keys in mapped types with a new as clause.
type MappedTypeWithNewKeys<T> = {
[K in keyof T as NewKeyType]: T[K]
// ^^^^^^^^^^^^^
// This is the new syntax!
}
With this new as clause, you can leverage features like template literal types to easily create property names based off of old ones.
type Getters<T> = {
[K in keyof T as `get${capitalize K}`]: () => T[K]
};
interface Person {
name: string;
age: number;
location: string;
}
type LazyPerson = Getters<Person>;
and you can even filter out keys by producing never.
That means you don't have to use an extra Omit helper type in some cases.
// Remove the 'kind' property
type RemoveKindField<T> = {
[K in keyof T as Exclude<K, "kind">]: T[K]
};
interface Circle {
kind: "circle";
radius: number;
}
type KindlessCircle = RemoveKindField<Circle>;
// same as
// type KindlessCircle = {
// radius: number;
// };
For more information, take a look at the original pull request over on GitHub.
Recursive Conditional Types
In JavaScript it's fairly common to see functions that can flatten and build up container types at arbitrary levels.
For example, consider the .then() method on instances of Promise.
.then(...) unwraps each promise until it finds a value that's not "promise-like", and passes that value to a callback.
There's also a relatively new flat method on Arrays that can take a depth of how deep to flatten.
Expressing this in TypeScript's type system was, for all practical intents and purposes, not possible.
While there were hacks to achieve this, the types ended up looking very unreasonable.
That's why TypeScript 4.1 eases some restrictions on conditional types - so that they can model these patterns.
In TypeScript 4.1, conditional types can now immediately reference themselves within their branches, making it easier to write recursive type aliases.
For example, if we wanted to write a type to get the element types of nested arrays, we could write the following deepFlatten type.
type ElementType<T> =
T extends ReadonlyArray<infer U> ? ElementType<U> : T;
function deepFlatten<T extends readonly unknown[]>(x: T): ElementType<T>[] {
throw "not implemented";
}
// All of these return the type 'number[]':
deepFlatten([1, 2, 3]);
deepFlatten([[1], [2, 3]]);
deepFlatten([[1], [[2]], [[[3]]]]);
Similarly, in TypeScript 4.1 we can write an Awaited type to deeply unwrap Promises.
type Awaited<T> = T extends PromiseLike<infer U> ? Awaited<U> : T;
/// Like `promise.then(...)`, but more accurate in types.
declare function customThen<T, U>(
p: Promise<T>,
onFulfilled: (value: Awaited<T>) => U
): Promise<Awaited<U>>;
Keep in mind that while these recursive types are powerful, but they should be used responsibly and sparingly.
First off, these types can do a lot of work which means that they can increase type-checking time.
Trying to model numbers in the Collatz conjecture or Fibonacci sequence might be fun, but don't ship that in .d.ts files on npm.
But apart from being computationally intensive, these types can hit an internal recursion depth limit on sufficiently-complex inputs.
When that recursion limit is hit, that results in a compile-time error.
In general, it's better not to use these types at all than to write something that fails on more realistic examples.
See more at the implementation.
Pedantic Index Signature Checks (--noUncheckedIndexedAccess)
TypeScript has a feature called index signatures.
These signatures are a way to signal to the type system that users can access arbitrarily-named properties.
interface Options {
path: string;
permissions: number;
// Extra properties are caught by this index signature.
[propName: string]: string | number;
}
function checkOptions(opts: Options) {
opts.path // string
opts.permissions // number
// These are all allowed too!
// They have the type 'string | number'.
opts.yadda.toString();
opts["foo bar baz"].toString();
opts[Math.random()].toString();
}
In the above example, Options has an index signature that says any accessed property that's not already listed should have the type string | number.
This is often convenient for optimistic code that assumes you know what you're doing, but the truth is that most values in JavaScript do not support every potential property name.
Most types will not, for example, have a value for a property key created by Math.random() like in the previous example.
For many users, this behavior was undesirable, and felt like it wasn't leveraging the full strict-checking of --strictNullChecks.
That's why TypeScript 4.1 ships with a new flag called --noUncheckedIndexedAccess.
Under this new mode, every property access (like foo.bar) or indexed access (like foo["bar"]) is considered potentially undefined.
That means that in our last example, opts.yadda will have the type string | number | undefined as opposed to just string | number.
If you need to access that property, you'll either have to check for its existence first or use a non-null assertion operator (the postfix ! character).
// Checking if it's really there first.
if (opts.yadda) {
console.log(opts.yadda.toString());
}
// Basically saying "trust me I know what I'm doing"
// with the '!' non-null assertion operator.
opts.yadda!.toString();
One consequence of using --noUncheckedIndexedAccess is that indexing into an array is also more strictly checked, even in a bounds-checked loop.
function screamLines(strs: string[]) {
// this will have issues
for (let i = 0; i < strs.length; i++) {
console.log(strs[i].toUpperCase());
// ~~~~~~~
// error! Object is possibly 'undefined'.
}
}
If you don't need the indexes, you can iterate over individual elements by using a for-of loop or a forEach call.
function screamLines(strs: string[]) {
// this works fine
for (const str of strs) {
console.log(str.toUpperCase());
}
// this works fine
strs.forEach(str => {
console.log(str.toUpperCase());
});
}
This flag can be handy for catching out-of-bounds errors, but might be noisy for a lot of code, so it is not automatically enabled by the --strict flag.
You can learn more at the implementing pull request.
paths without baseUrl
Using path-mapping is fairly common - often it's to have nicer imports, often it's to simulate monorepo linking behavior.
Unfortunately, specifying paths to enable path-mapping required also specifying an option called baseUrl, which allows bare specifier paths to be reached relative to the baseUrl too.
This also often caused poor paths to be used by auto-imports.
In TypeScript 4.1, the paths option can be used without baseUrl.
This helps avoid some of these issues.
checkJs Implies allowJs
Previously if you were starting a checked JavaScript project, you had to set both allowJs and checkJs.
This was a slightly annoying bit of friction in the experience, so checkJs now implies allowJs by default.
See more details at the pull request.
React 17 JSX Factories
TypeScript 4.1 supports React 17's upcoming jsx and jsxs factory functions through two new options for the jsx compiler option:
react-jsxreact-jsxdev
These options are intended for production and development compiles respectively.
Often, the options from one can extend from the other.
For example, a tsconfig.json for production builds might look like the following:
// ./src/tsconfig.json
{
"compilerOptions": {
"module": "esnext",
"target": "es2015",
"jsx": "react-jsx",
"strict": true
},
"include": [
"./**/*"
]
}
and one for development builds might look like the following:
// ./src/tsconfig.dev.json
{
"extends": "./tsconfig.json",
"compilerOptions": {
"jsx": "react-jsxdev"
}
}
For more information, check out the corresponding PR.
Editor Support for the JSDoc @see Tag
The JSDoc tag @see tag now has better support in editors for TypeScript and JavaScript.
This allows you to use functionality like go-to-definition in a dotted name following the tag.
For example, going to definition on first or C in the JSDoc comment just works in the following example:
// @filename: first.ts
export class C { }
// @filename: main.ts
import * as first from './first';
/**
* @see first.C
*/
function related() { }
Thanks to frequent contributor Wenlu Wang for implementing this!
Breaking Changes
abstract Members Can't Be Marked async
Members marked as abstract can no longer be marked as async.
The fix here is to remove the async keyword, since callers are only concerned with the return type.
any/unknown Are Propagated in Falsy Positions
Previously, for an expression like foo && somethingElse, the type of foo was any or unknown, the type of the whole that expression would be the type of somethingElse.
For example, previously the type for x here was { someProp: string }.
declare let foo: unknown;
declare let somethingElse: { someProp: string };
let x = foo && somethingElse;
However, in TypeScript 4.1, we are more careful about how we determine this type.
Since nothing is known about the type on the left side of the &&, we propagate any and unknown outward instead of the type on the right side.
The most common pattern we saw of this tended to be when checking compatibility with booleans, especially in predicate functions.
function isThing(x: any): boolean {
return x && typeof x === 'object' && x.blah === 'foo';
}
Often the appropriate fix is to switch from foo && someExpression to !!foo && someExpression.
--declaration and --outFile Requires a Package Name Root
When you have a project which uses both outFile plus declaration to emit a single .js file for your project, alongside a corresponding .d.ts file, that declaration file would usually require some sort of post-processing of the module identifiers to make sense to external consumers.
For example, a project like
// @filename: projectRoot/index.ts
export * from "./nested/base";
// @filename: projectRoot/nested/base.ts
export const a = "123"
Would produce a .d.ts file that looks like the following:
declare module "nested/base" {
export const a = "123";
}
declare module "index" {
export * from "nested/base";
}
Which is technically accurate, but not that useful.
TypeScript 4.1 requires bundledPackageName to be specified when a single .d.ts file is requested to be produced.
declare module "hello/nested/base" {
export const a = "123";
}
declare module "hello" {
export * from "hello/nested/base";
}
Without this option, you may receive an error like
The `bundledPackageName` option must be provided when using outFile and node module resolution with declaration emit.
resolve's Parameters Are No Longer Optional in Promises
When writing code like the following
new Promise(resolve => {
doSomethingAsync(() => {
doSomething();
resolve();
})
})
You may get an error like the following:
resolve()
^^^^^^^^^
error TS2554: Expected 1 arguments, but got 0.
An argument for 'value' was not provided.
This is because resolve no longer has an optional parameter, so by default, it must now be passed a value.
Often this catches legitimate bugs with using Promises.
The typical fix is to pass it the correct argument, and sometimes to add an explicit type argument.
new Promise<number>(resolve => {
// ^^^^^^^^
doSomethingAsync(value => {
doSomething();
resolve(value);
// ^^^^^
})
})
However, sometimes resolve() realy does need to be called without an argument.
In these cases, we can give Promise an explicit void generic type argument (i.e. write it out as Promise<void>).
This leverages new functionality in TypeScript 4.1 where a potentially-void trailing parameter can become optional.
new Promise<void>(resolve => {
// ^^^^^^
doSomethingAsync(() => {
doSomething();
resolve();
})
})
TypeScript 4.1 ships with a quick fix to help fix this break.
What's Next?
As always, we'll be working on smoothing out the experience until 4.1 comes out.
We'd love to hear what your thoughts are we finalize our Release Candidate next month, so please give TypeScript 4.1 and let us know if you run into anything!
Happy Hacking!
- Daniel Rosenwasser and the TypeScript Team
The post Announcing TypeScript 4.1 Beta appeared first on TypeScript.
