Explicitly annotated unique symbols are widened in return positions

[Andarist]

🔎 Search Terms

widening return unique symbol

🕗 Version & Regression Information

• This is the behavior in every version I tried

⏯ Playground Link

https://www.typescriptlang.org/play?ts=5.3.0-dev.20230814&ssl=6&ssc=2&pln=1&pc=1#code/KYDwDg9gTgLgBAYwgOwM7xsdcC8cDKAngLYBGEANgBQCUAUHaJLHAGYCuyCMAlim8lpwA3nTjjEKbAEMKPaagBccGITDAIrFVnh5M6ANxiJUYDHZRkcWfNR0AvkA

💻 Code

export const test = Symbol()

export function fn() {
    const alias: typeof test = test;
    return alias
}

🙁 Actual behavior

this declaration file is emitted:

export declare const test: unique symbol;
export declare function fn(): symbol;

🙂 Expected behavior

this declaration file to be emitted:

export declare const test: unique symbol;
export declare function fn(): typeof test;

Additional information

getWidenedLiteralLikeTypeForContextualReturnTypeIfNeeded calls getWidenedLiteralLikeTypeForContextualType and that calls getWidenedUniqueESSymbolType when there is no contextual type provided.

some issues that are (partially~) related:
#32242
#35896 (comment)
#36876

[fatcerberus]

I think you may have just found the underlying cause of #53276 😄

[Andarist]

This issue is actually different and I'm already fixing it here: #54778 😅 Different code paths handle widening from initializers and widening at return positions.

[RyanCavanaugh]

I'm not sure I'd call the existing behavior wrong, but it seems like a safe change to make

[ssalbdivad]

@RyanCavanaugh Are there other situations where casting the only return statement of a function doesn't dictate how it's inferred?

Could even be simplified down to just:

export const test = Symbol()

// inferred as () => symbol
const fn = () => test as typeof test

[Andarist]

I'm not sure I'd call the existing behavior wrong, but it seems like a safe change to make

Are there any other cases in the compiler where an explicit type annotation like this gets stomped by the compiler and the type gets widened? I'm really curious about this. I assume that there is no place like this and that's why I'm calling this a bug. I also understand that symbols are somewhat special but as long as the symbol is publicly available (so the declaration emit can actually reference it) I think the annotation should be preserved.