Inferred type does not resolve completely, contains free variables

[pelotom]

TypeScript Version: 2.5.3

Code

function weird<T extends Record<string, { x: any }>>(o: T) {
  type Additional = T & { extra: { x: number } }
  type XValues = { [K in keyof Additional]: Additional[K]['x'] }
  const results: XValues = undefined as any // implementation not important
  return results
}

const results = weird({
  foo: { x: 'hello' },
  bar: { x: true },
})

Expected behavior:

The type of results should be

{
    foo: string;
    bar: boolean;
    extra: number;
}

and was, prior to 2.5.3.

Actual behavior:

The type of results is

{
    [K in keyof (T & {
        extra: {
            x: number;
        };
    })]: (T & {
        extra: {
            x: number;
        };
    })[K]["x"];
}

I'm not even sure what that type means, since it contains the free variable T that only existed in the scope of the function that produced the result. T should've resolved to { foo: { x: 'hello' }, bar: { x: true } }.

This is a minimal repro whittled down from a much larger body of code in our app, and the problem is preventing us from being able to migrate beyond TypeScript 2.5.2.

[pelotom]

Interestingly, this reformulation using a WithAdditional type constructor makes it resolve correctly:

type WithAdditional<T> = T & { extra: { x: number } }

function weird<T extends Record<string, { x: any }>>(o: T) {
  type XValues = { [K in keyof WithAdditional<T>]: WithAdditional<T>[K]['x'] }
  const results: XValues = undefined as any // implementation not important
  return results
}

const results = weird({
  foo: { x: 'hello' },
  bar: { x: true },
})
// results: {
//   foo: string;
//   bar: boolean;
//   extra: number;
// }

[ahejlsberg]

This is a duplicate of #19043. I will have a fix up soon.

[pelotom]

Duplicate of #19043

[pelotom]

GitHub has a first-class notion of issue duplication now, but apparently you have to be very specific in your wording :)