Type inference not working for this particular use of Generics?

[coreh]

Hey there. I was experimenting with generics, and noticed that type inference didn't work as I expected for this particular case. I'm not entirely sure if this is an actual blind spot of the type inference mechanism, or if my declaration doesn't mean what I think it means. it's a bit similar to the “Curiously recurring template pattern” described on #5949, but isn't actually the CRTP, since there's no actual recursion involved. Thanks in advance for looking into this issue.

TypeScript Version:

1.8.10

Code

interface IFoo {
    foo: 'foo'; // Could be anything
}

interface IWrapped<I> {
    getObject(): I;
}

function unwrap<I, T extends IWrapped<I>>(x: T): I {
    return x.getObject();
}

let x: IWrapped<IFoo>/* = ... */;
let y = unwrap<IFoo, IWrapped<IFoo>>(x); // Type is IFoo (correct)
let z = unwrap(x); // Type is {} (incorrect?)

Expected behavior:

I expected the compiler to infer the type of z as being IFoo, like when using explicit generic parameters on y.

Actual behavior:

The compiler inferred the type of z as being {}.

[Arnavion]

Note that in this particular case you can write it with just one generic parameter and not hit the issue:

function unwrap<I>(x: IWrapped<I>): I {

but yes, it does look like a bug.

[DanielRosenwasser]

I think this is a symptom of the issue @vsiao pointed out in #7234.

[ahejlsberg]

This is by design. Inference only infers from the arguments passed to the function to the declared parameters, it does not factor in constraints. Constrains are simply checks that are performed after inference completes. In your example there's really no reason to have two type parameters (and you should never have type parameters you don't need), but there are indeed situations where you want to infer multiple type parameters from a single argument. #7234 is about those scenarios.

[coreh]

Hey there, thanks for looking into this!

In hindsight I have no idea why I didn't think of

function unwrap<I>(x: IWrapped<I>): I {

and instead went with the way more complicated construct.

Really excited about how powerful the type system is becoming. The & trick to infer multiple types on #7234 is also really cool.