Contextual inference not working in "nested" scenarios, inferred type is never

[kucharskimaciej]

Bug Report

🔎 Search Terms

Contextual inference
Does not exist on type 'never'

🕗 Version & Regression Information

Tested in 4+, and this is the behavior in every version I tried.

⏯ Playground Link

Playground link with relevant code

💻 Code

type ModuleWithState<TState> = {
    state: TState;
}

type State = {
    a: number;
}

type MoreState = {
    z: string;
}

function createModule<TState, TActions>(state: TState, actions: TActions): ModuleWithState<TState> & TActions {
    return {
        state,
        ...actions
    }
}

function convert<TState, TActions>(m: ModuleWithState<TState> & TActions): ModuleWithState<TState & MoreState> & TActions {
    return {
        ...m,
        state: {
            ...m.state,
            z: "boo"
        },

    }
}


const breaks = convert(
    createModule({
        a: 12
    }, {
        foo() {
            return true
        }
    })
);

breaks.a; // Property 'a' does not exist on type 'never'.
breaks.z; // Property 'z' does not exist on type 'never'.
breaks.foo() // Property 'foo' does not exist on type 'never'.

const mod = createModule({
    a: 12
}, {
    foo() {
        return true
    }
})

const works = convert(mod);
works.a; // number
works.z // string
works.foo();

🙁 Actual behavior

The first scenario (breaks) fails the infer the correct type of the passed argument. Second scenario works works as expected. The only difference between the two is that in the latter the argument is first assigned to a variable.

🙂 Expected behavior

Both scenarios work exactly the same.

[devanshj]

You probably meant to write breaks.state.a, breaks.state.z, works.state.a, works.state.z instead (because that is what is expected to work according to the runtime code and types if I'm not reading it incorrectly). Also your playground link is inconsistent with the code.

This seems to be related to #40439. Because if you make foo a non-function say true then it starts working. TypeScript seems to topple with inference when there's a functional property in complex cases.

Here's a workaround

[RyanCavanaugh]

@ahejlsberg the behavior in the playground link seems bad. Thoughts? Possibly related to #44476 (not sure) ?

[ahejlsberg]

@RyanCavanaugh It's a simple issue of not properly propagating the special silentNeverType in intersections. Trivial fix, will have a PR up soon.