Boolean narrowed to true only when generic type is explicitly provided while it shouldn't

[Harpush]

🔎 Search Terms

specify type narrow union

🕗 Version & Regression Information

All versions I checked and nothing in FAQ that seems related

⏯ Playground Link

https://www.typescriptlang.org/play/?#code/LAKA9GCqDOCmAEAXAFggbgQwE4EsMDsBjBA-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-PqQILycKy7zPabZaBKRaQGs0Vysd1hVbCcTYkl4P6AErDgDyw4NFLpmqZTF1F0rZMNuPxsHNoXZ5F07CoSDXINgiK4ghOCA9Eq9+DdsgLnCG9sdsGd8Fd8DdvZ9IHZAbzocQ4aRtGOCxuwCZJrghCpumGbzlgJxnAQJjZNmGQ6LAgbBv+ABMhbYMWFKljyXSVtWJyzKKfgNjKBwtpubZ1OimISgRNbNNRZqgO23argSRIkkOYD6FgwzQL6YC-lhYYRlG1QgXGqDwIm+DJlBaZYJmcEIfBSG2K6Oi9q+XAcahxwYX+2gAMx4VgrGvhRnwLKR1a1sRYqXo2tEKvR3GdhK2K8SQxkeQQ-EDoJwmieyABCIX4NEvbxBgBjQAwkViSAQA

💻 Code

// Use the variance annotation to force this type to be invariant, even though it would otherwise be covariant
type Invariant<in out T> = () => T;

const foo = <T extends object>(obj: T): Invariant<{[P in keyof T]: Invariant<T[P]>}> => null!;
const bar = <T extends {[P in keyof T]: Invariant<any>}>(w: T): T => null!;

const inferred = bar({a: foo({b: true})}); // true is boolean as expected

// Explicitly specify the generic param
const explicit = bar<{
    a: Invariant<{
        b: Invariant<boolean>;
    }>
}>({a: foo({b: true})}); // narrowed to true instead of boolean and errors

🙁 Actual behavior

Narrows to true

🙂 Expected behavior

Should stay boolean

Additional information about the issue

Doing as boolean works...

[Andarist]

Sort of a duplicate of #48363 but with object types. It repros the same way when I wrap function arguments from that issue in objects: TS playground.

[Harpush]

Just adding that it also happened for unions

[Andarist]

Well, boolean is a union of true | false so it makes sense that unions are affected 😉 It's just that boolean is expected to behave more like a primitive than a union - hence the issue I linked.

The union case can be reproduced easily using the same kind of code (TS playground):

type Box<T> = {
  get: () => T;
  set: (value: T) => void;
};

declare function box<T>(value: T): Box<T>;

const bb1: Box<"foo" | "bar"> = box("foo"); // errors today
const bb2: Box<{ prop: "foo" | "bar" }> = box({ prop: "foo" }); // errors today

The algorithm can only keep the literal type based on the contextual type. There is no mechanism to widen it to the contextual union type when that position refers to an invariant type parameter. I think this issue here should stay focused on the invariant problem and I'll create a new one for this specific issue with boolean: #59754

[RyanCavanaugh]

What's a repro that doesn't depend on an incorrect variance annotation?

[Andarist]

@RyanCavanaugh I think it's the one I posted above, the one with "foo" | "bar".

[BalaM314]

What's a repro that doesn't depend on an incorrect variance annotation?

Replace

//Use the variance annotation to force this type to be invariant, even though it would otherwise be covariant
type Invariant<in out T> = () => T;

with

type Invariant = (_:T) => T;

and the error is the same.

[RyanCavanaugh]

I don't see an error after removing the variance annotation

[BalaM314]

https://www.typescriptlang.org/play/?#code/LAKALgngDgpgBASQHYDcCGAnAlmpYA8AKgHxwC8cAFAPoBchAlOaYQNyigDGA9kgM5g4AM27dycInBgAPMDCQATPnG4AjAFYxOYYpTXracRoeTpsuAgG8A2gAU4WJHADWMCNyFGAuidSYceER2XsQAvqRkpEgArgA2sQCE7CA8-IKqmOKSMnKKyjb2ji5uHt6+ZgEEuBBhugDuhsZGzHAx8UkcKbwCDkhCMBgYMAriGRiUlmiGItwTqoZgGNEwoQyryQD0G3A7u3v7AHoA-KBbcKk9Y4b4lqfbe1OIfuaBtyBn+7vzTxUW+KqiWIwXDETb3XbhMFwcJwAB08Lu526ghm1zeHy+hgB3CBuCh4URi2WDmUjjkGCgQzkIzQygARNjcUg6Z0tgBRaRQWJYThYMCxCBwPiwXlCQVgAAW8AA5vIBjy4FBMGgALZcZFSTnc3mCChjG6gB7lfx-N6fHbfUwmwKM4FIUGGiHEUDhCaPGZzBZLFZrBisODsgBKgYA8oHEXtjoiLijRGiI5i4ESYPjne8NkhuHUkxK0IJk60YMNlGAxKp4AzAXa6QAaByCLCkvADSkwalwWlwOnJlkgO4crk8vkCoUirBinMyuXYTiK5Uq4TcDBwMYdxSL7jqtKawc6gBMo0wBpARp+1qsjs+luelX+VZByT2BJArsm01Ex7231teJduks3zJqs6wBhsdRLs4fCshsA7asOgrClo47ilKcCykg8qzkqGCqouy6rrgIycLS+beh2yg-kgW49DIu58gAzIeGCfrsjxWi8F4nua16-Da972o+Tp-m676zABXrEp2lHAX6oHgRgkF3AAQvxlDJkwaCxHwYjyYpIBAA

[Harpush]

@RyanCavanaugh anything else needed here? It is still marked with needs more info