TypeScript: Mapped types for type maps

Stefan Baumgartner

Written by @ddprrt

More on TypeScript

Factory functions are a popular tool in JavaScript to create a diversity of objects with a single call. There’s a particular factory function that you might have used at some point:

document.createElement('video') // creates an HTMLVideoElement

document.createElement takes any string and creates HTMLElement instances. HTMLElement is one of the most derived objects in the DOM. Every available tag creates it’s on derivate. For example document.createElement('video') creates an instance of HTMLVideoElement.

But how do we type a factory function like that? One that has a couple of dozen different return types? Let’s try.

NOTE: TypeScript with the dom library activated in tsconfig.json knows of all HTMLElement derivates.

With conditional types #

The original typings for document.createElement take a string as parameter, and returns an object of type HTMLElement:

declare function createElement(tag: string, options?: any): HTMLElement

This is true, but not specific enough. We can be a lot more specific, since we know which tags implement corresponding HTMLElement derivates.

The first thing that came into my mind were conditional types. They were made for use cases like that! We know that tag is a subset of string, and more important: we know exactly which subset! The collection of tags available in your browser.

A possible generic type for the createElement factory function could look like this:

type CreatedElement<T extends string> =    /** 1 **/
T extends 'a' ? HTMLAnchorElement : /** 2 **/
T extends 'div' ? HTMLDivElement :
T extends 'video' ? HTMLVideoElement :
HTMLElement; /** 3 **/

Let’s go over this definition in detail:

  1. We start with a generic type CreatedElement<T>. The generic placeholder T has to be a subset of string
  2. We then test for a specific unit type from the string subset. For example, the string "a" is of type string, but also of type "a". You can think of string as the universe of all possible string unit types. If our generic placeholder T extends this particular subset of "a", we know that the result has to be of type HTMLAnchorElement. The else branch starts a cascade through all known HTML tags. In our example, we know of div and video.
  3. At the end of this cascade, when we went through all known HTML tags, we return the generic HTMLElement as a fallback. This is totally in tune with the way createElement works. When you create an element with a tag the browser doesn’t know, you get at least an HTMLElement.

So far, so good. This even looks like a map from string to HTMLElement derivate. Now the only thing we have to do is to extend this list with all available tags and return the respective element instance. We can even use things like union types to help with types that implement more than one tag:

type CreatedElement<T extends string> = 
T extends 'a' ? HTMLAnchorElement :
T extends 'div' ? HTMLDivElement :
T extends 'video' ? HTMLVideoElement :
T extends 'thead' | 'tfoot' | 'tbody' ? HTMLTableSectionElement : /** 1 **/
  1. All three of thead, tbody and tfoot implement HTMLTableSectionElement. We can use a union type of all three unit types to identify HTMLTableSectionElement

The solution is good and robust, but has one catch. A rather big one. The amount of comparisions is finite. Even though this looks like a map of types, in reality it’s a nested comparision chain:

type CreatedElement<T extends string> = 
T extends 'a' ? HTMLAnchorElement :
T extends 'div' ? HTMLDivElement :
T extends 'video' ? HTMLVideoElement :
T extends 'thead' | 'tfoot' | 'tbody' ? HTMLTableSectionElement :

So it’s nested. This also means that with every further comparision, there has to be reference to the original comparison. Internally, this can be best done via a recursion. And recursions take up memory.

This is why TypeScript gives you a hard limit of 50 nested comparisons to make sure memory and performance goals are met. If you extend your list beyond 50 comparisions, you get the error "Type instantiation is excessively deep and possibly infinite". Check out the issue #28663 on Github.

So that doesn’t solve our problem in the long run. So what shall we do?

Mapped types #

To find a proper solution for that problem, let’s think for one moment how you would’ve implemented the document.createElement factory function yourself in JavaScript. I would have used an object, where each key corresponds to the correct implementation. And I would’ve indexed the object dynamically with an index access. Something like that:

NOTE: This does not work. This is just for demonstration purposes

const elementMap = {
a: HTMLAnchorElement,
div: HTMLDivElement,
video: HTMLVideoElement

function createElement(tag) {
return tag in elementMap ? new elementMap[tag]()
: new HTMLElement()

This obviously doesn’t work, that’s what the document.createElement factory function is for. But it illustrates the way of accessing via the index access operator rather nice. Since every key in an object can be accessed using a string, we select the right constructor out of this list, and create a new instance. If we don’t have a constructor, let’s create a generic HTMLElement.

In TypeScript, we can create types that work in a similar manner. First, let’s create the AllElements type which is a map of all tags to their corresponding HTMLElement derivate:

type AllElements = {
'a': HTMLAnchorElement,
'div': HTMLDivElement,
'video': HTMLVideoElement,
... //extend as you need

This is what I like to call a type map. We could use this type to create an object of type AllElements, but in reality we most likely won’t need that. We only use this type as an helper type for CreatedElement:

type CreatedElement<T extends string> = 
T extends keyof AllElements ? AllElements[T] : /** 1 **/
HTMLElement; /** 2 **/
  1. The type signatur is the same as in the previous example. The generic placeholder T extends from string. But now we check if T is somewhere in the keys of AllElements. If so, we index the type that is stored with this particular key T. That’s how we get the correct derivate!
  2. In all other cases, we have “just” an HTMLElement.

Do you see how similar this type definition is to the JavaScript example above? Of course the way I wrote JavaScript earlier is just one way to express myself, and I used it deliberately to show the similarities with conditional types. But it shows how close TypeScript tries to be to JavaScript in terms of syntax and semantics.

The cool thing is: We are just moving in type space. No source created, just information to make your code a lot safer. Like that:

declare function createElement<T extends string>(tag: T, options?: any): CreatedElement<T>

We can use the function like that:

createElement('video') // It's an HTMLVideoElement
createElement('video').src = '' // `src` exists
createElement('a').href = '' // It's an HTMLAnchorElement with an `href`

We can even write our own factory functions, that can do a little more that just creating elements:

function elementFactory<T extends string>(tag: T, 
defaultProps: Partial<CreatedElement<T>>) : CreatedElement<T> {

const element = createElement(tag);
return Object.assign(element, defaultProps)

This factory takes a couple of default properties that need to be available in the generated output element. So things like:

elementFactory('video', { src: '' });

Can even be autocompleted. And TypeScript warns you if you want to specify a property that does not exist:

elementFactory('video', { source: '' }) // 💥 Error: this property does not exist

Pretty sweet, huh?

Bottom line #

Type maps are a good tool for factory functions which produce a ton of different results. And most likely for a lot, lot more. If you want to expand on the example shown above, take this gist. And look at this playground to play around.

More articles on TypeScript

this in JavaScript and TypeScript

TypeScript and ECMAScript Modules

TypeScript + React: Why I don't use React.FC

Stay up to date!

3-4 updates per month, spam-free, hand-crafted. Our newsletter gives you links, updates on fettblog.eu, conference talks, coding soundtracks, and much more.