feat(union-to-intersection): allow indexing by the resulting type

[niieani]

Previously, using UnionToIntersection in the key position would result in the error:

Type 'unknown' cannot be used as an index type.

type Objects = { a: string } | { b: string } | { c: string }
type X = Objects[UnionToIntersection<keyof Objects>]

After the change the error goes away, and it doesn't affect how the type works otherwise.
The use case is when you use generics and want to ensure the type is never for unions of objects.

[sindresorhus]

Can you add a test?

[sindresorhus]

@niieani Bump :)

[niieani]

Hi @sindresorhus, sorry I'm an OSS maintainer too, didn't yet have time to follow up. It's on my list.

[niieani]

Hi @sindresorhus! Added a test.

[JGJP]

@niieani @sindresorhus

Sorry for bringing this up but why was this done? Why should you be able to pass a string to UnionToIntersection?

The example uses

type X = Objects[UnionToIntersection<keyof Objects>]

(keyof Objects actually resolves to never due to there being no keys in common)

but I believe that this would be closer to a real world usage?

type X = Objects[keyof UnionToIntersection<Objects>]

It seems that the issue of there being no keys in common was actually recognized, as the eventually added test includes the key z which is common between all objects:

type-fest/test-d/union-to-intersection.ts

Lines 11 to 14 in eb96609

	// It's possible to index by the resulting type.
	type ObjectsUnion = {a: string; z: string} | {b: string; z: string} | {c: string; z: string};
	declare const value: ObjectsUnion[UnionToIntersection<keyof ObjectsUnion>];
	expectType<string>(value);

In this test though UnionToIntersection<keyof ObjectsUnion> just evaluates to 'z', so UnionToIntersection is totally unnecessary.

[sindresorhus]

@JGJP It's generally better to open a new issue than to comment on an old pull request.