[Regression 3.9 -> 4.X] Variadic tuple type is not correctly inferred

Bug Report

🔎 Search Terms

variadic tuple, regression, inference

🕗 Version & Regression Information

• This changed between versions 3.9.X and 4.X.X

⏯ Playground Link

Typescript 3.9 (works as expected)

Typescript 4.0 (does not work any more)

💻 Code

type Sink<T> = (t: number, payload: T) => void;

type Producer<T> = (t: number, sink: Sink<T>) => void;

type ExtractContent<T extends [Producer<unknown>, ...Producer<unknown>[]]> = {
    [k in keyof T]: T[k] extends Producer<infer U> ? U : never;
}

function combineWith<T extends [Producer<unknown>, ...Producer<unknown>[]], U>(
    fn: (...args: ExtractContent<T>) => U,
    ...sources: T
): Producer<U> {
    return undefined as any;
}

function of<T>(n: T): Producer<T> {
    return undefined as any;
}

const x = combineWith((x, y) => x + y, of(1), of(2));

With typescript 3.9, the arguments of the lambda passed to combineWith are correctly read as (number, number). With 4.0 everything is just unknown, even the Producer arguments later

🙁 Actual behavior

x and y (of the lambda function) are types unknown

🙂 Expected behavior

x and y (of the lambda function) are types number

Start with the linked TypeScript 3.9 and 4.0 Playground reproductions and the combineWith, ExtractContent, Producer, and of definitions in the issue. Compare inference behavior between versions and confirm completion when the callback parameters and later Producer arguments infer as number rather than unknown.