Generics and type inference

[Roam-Cooper]

TypeScript Version: 2.2.2 / nightly (2.2.0-dev.201xxxxx)

Code

export class Thing<T>{
    constructor(input: Fruit<T>, mapping: () => T) { }
}

new Thing(Fruit<apple>null, ()=>{
return {};
});

Expected behavior:
Compiler complains that {} is not assignable to Fruit as the compiler sees T as a Fruit<apple> FIRST.
Actual behavior:
Compiler finds best common type between Fruit<apple> and {} for T which ends up being {}. However, this means that the mapping function is not forced to return Fruit<apple>.

Additionally, if I specify mapping as:

mapping: () => T & {}

Then it works, and the mapping function must return a structure meeting the shape of Fruit<apple>.

If the appropriate generic types are explicitly provided, however, it also works. It just seems to try to best case guess T from all of its uses, and that best case almost always seems to be {} which means that this will always be broken...

Is this a problem related to the compiler picking the best common type/type inference for the generic, or am I wrong? (Also, sorry for opening so many issues recently, I'm building something new for work...)

[Roam-Cooper]

Found that this is related to #5254

[Roam-Cooper]

I feel like the only way to solve this is something like

let a: {} = { b: 1 };

Causing the error: { b: 1 } is not assignable to {} because b: number does not exist on {}
But I have a funny feeling that this is already "working as intended"...

[mhegazy]

But I have a funny feeling that this is already "working as intended"...

yup. unfortunately , a type with no members ( i.e. {}), is a super type of all types. more in https://github.com/Microsoft/TypeScript/wiki/FAQ#why-are-all-types-assignable-to-empty-interfaces