Instantiation expressions don't filter signatures with non-matching constraints

[Andarist]

Bug Report

🔎 Search Terms

instantiation expression, overload

🕗 Version & Regression Information

• This is the behavior in every version I tried

⏯ Playground Link

Playground link with relevant code

💻 Code

declare function foo<T extends keyof HTMLElementTagNameMap>(arg: T): HTMLElementTagNameMap[T];
declare function foo<T extends HTMLElement>(arg: T): T;

type FirstFoo = typeof foo<"div">; // Type 'string' does not satisfy the constraint 'HTMLElement'.(2344)

🙁 Actual behavior

This instantiation expression errors despite the fact that a matching signature exists.

🙂 Expected behavior

It sounds likely that most people would assume that this should work similarly to how overloads are chosen - but instead an intersection is being created based on all signatures with the same type arguments arity.

originally reported by @herjiict here

[fatcerberus]

an intersection is being created based on all signatures with the same type arguments arity

Is it that, or is it just picking the last overload, similar to what happens with Parameters?

[Andarist]

IIRC what I’ve seen when debugging this - it’s an intersection, it’s not picking a single matching signature.

I think that it makes sense per the second part of this comment: #51695 (comment) . It’s just that when mapping those signatures reportErrors was always set to true and non-matching signatures were not rejected

[Andarist]

I went back to recheck the implementation, "intersection" definitely caught my eye somewhere around this code but it seems that it was in some different parts.

So this isn't actually becoming an intersection, but the effect is kinda similar - it's an object type with multiple signatures (and more):
https://github.dev/microsoft/TypeScript/blob/70d5cb28275d35e45a0f84aa981b45e6fb059160/src/compiler/checker.ts#L33667-L33679

[RyanCavanaugh]

@ahejlsberg any thoughts?

[apendua]

What would need to happen from the TypeScript team perspective for this change to be accepted? It seems to be a very reasonable improvement and it also addresses a couple other problems (linked above).