"'T' is declared but never used" on type parameter

[jakebailey]

Steps to reproduce

export function wowee<T>() {
  throw new Error("TODO");
}

Behavior with typescript@5.8

No error.

Behavior with tsgo

'T' is declared but never used.ts(6196)

[RyanCavanaugh]

Are these minimal repros? 👀

[jakebailey]

No, sorry.

export function wowee<T>() {
  throw new Error("TODO");
}

[jakebailey]

Test case:

// @strict: true
// @noUnusedLocals: true
// @noUnusedParameters: true
// @target: esnext

export function wowee<T>() {
  throw new Error("TODO");
}

This will produce an error baseline with 'T' is declared but never used., which is incorrect.

[jakebailey]

I'm apparently dumb. Strada does get mad:

However! The original test I had does error, albeit inconsistently: Playground Link

// @strict: true
// @noUnusedLocals: true
// @noUnusedParameters: true
// @target: esnext

export function tryParseJson<T>(text: string): unknown;
export function tryParseJson<T>(text: string, predicate: (parsed: unknown) => parsed is T): T | undefined;
export function tryParseJson<T>() {
    throw new Error("noop")
}

export function tryParseJson2<T>(text: string): unknown;
export function tryParseJson2<T>(text: string, predicate: (parsed: unknown) => parsed is T): T | undefined;
export function tryParseJson2() {
    throw new Error("noop")
}

export function tryParseJson3<T>(_text: string): unknown {
    throw new Error("noop")
}

Corsa errors on all of these, but Strada only errors on the first and third.

Did Strada have a carve-out for overloads, and Corsa is missing that?

[ahejlsberg]

Corsa's behavior is correct here. We have multiple issues related to duplicate identifier checking in Strada. They were fixed in Corsa in #693.

[jakebailey]

I guess; this is just real code in the wild from dt-tools. Would be good to check on Strada how breaky this might be.

[ahejlsberg]

Regarding #1618, I don't see how having a function body should change whether we check for unused type parameters. I mean, if type parameters aren't used anywhere in a signature, what's their purpose? And why would it be meaningful if there isn't a body? If anything, I feel like we should tighten the check even further and error on type parameters that aren't used in the signature, regardless of whether they're used in the body. Use in the body without any occurrence in the signature is highly suspicious and serves no purpose I can think of.

[jakebailey]

Yeah, I mistakenly forgot that I could just delete the type variable in my example. I think you're right about erroring elsewhere too.