Constraint in generic type forgotten in assignability check

[ehoogeveen-medweb]

🔎 Search Terms

generic type constraint not applied inside type

🕗 Version & Regression Information

• This changed in commit or PR Divide-and-conquer strategy for intersections of unions #57871

⏯ Playground Link

https://www.typescriptlang.org/play/?ts=5.4.5#code/C4TwDgpgBAggdiA0hEUC8U4FcC2AjCAJygB8oBnYQgSzgHNSKR8B7AGwG4AoL0SKAEoRgWQnAAq4CMhAsAZgB4A8ngBWUCAA9gEOABNyUFmogBjYAD50UFeq079h46rPAuUD1AD8UANoBrFHkbNQBdDW1dAz84CAA3IlD3TxSfWITCZJSPAC4oAG8-GShaKEDZORD1AENDShp6UIBaLzyACgBKdCtigF9feto6JJS89KJuXikoGQBJOQBlajo4apFCCCU5W1cFLKqIh2jnVwAafeL7KMN4JBRzlKF1iSkZQwwn0RfIGXllNQs5ysH2EX0kPxQhiujignW6MxQ3lhbThaB6KC60OinzE4OkkKRxTG8SIXWJGUmWjALEIwCgfGgQj0WFMED0AEZ-nZIjCTuZTgjUFibggZAKAGrVNhYCAC2xvQ7XMpBSq2azlYK2YEXRHCwXzJYrNaiTbbEzmLkCmQWfapQWKmHygkAemdsDYAHdqiBDFQZQA6W3ZDw+Qq+RAlOCC0J5SXS6C9IPB8lEJNQFOESauqAAdRp-kM1EqphYcHq1VodOohg2TVMAAszIE9JHyNQ9NAGYGqTS6QzBGyWWyAExch3Hc3AK26nnRW5iqBxmVytQKvUa1VqdUqqralKXWciu4gNM+A9HQxzRbLVbrU07C22acgG3B4NnmcXqoK7MwT3e31CADNM31DIpI2jWMpRlKBEzfeCMxAzxEIQzASUzLggA

💻 Code

type AnyKey = number | string | symbol;

type ReturnTypeKeyof<Obj extends object> = Obj extends object
    ? [keyof Obj] extends [never]
        ? never
        : { [Key in keyof Obj as string]-?: () => Key }[string]
    : never;

type KeyIfSignatureOfObject<
    Obj extends object,
    Key extends AnyKey,
    ReturnTypeKeys = ReturnTypeKeyof<Obj>,
> = ReturnTypeKeys extends () => Key ? ((() => Key) extends ReturnTypeKeys ? Key : never) : never;

export type Reduced<Obj extends object, Key extends AnyKey, Value, ObjKeys extends keyof Obj = keyof Obj> =
    Key extends KeyIfSignatureOfObject<Obj, Key>
        ? Key extends ObjKeys // Always true.
            ? { [K in Key]: Value }
            : never
        : never;

🙁 Actual behavior

Gives error Type 'ObjKeys & KeyIfSignatureOfObject<Obj, Key, ReturnTypeKeyof<Obj>> & Key' is not assignable to type 'string | number | symbol'.

🙂 Expected behavior

No error; Key was constrained to AnyKey and should be usable as a key.

Additional information about the issue

I reduced this from a much larger example, so the code here doesn't make much sense (the names of the helper types might also not make sense anymore). It might be possible to reduce it further but I wasn't able to find anything obvious.

As the playground shows, simply wrapping the body of the type with a Key extends AnyKey check makes the error go away, which makes me think that this must be a bug (I guess it distributes the union, but the entire union should be valid as a key type). Checking Key extends ObjKeys before KeyIfSignatureOfObject also makes the error go away.

[ahejlsberg]

This one is quite subtle. We have logic in relationship checking that hoists union constraints of generic types in an intersection into a top-level union of intersections such that it can be properly related to another union. Because of the divide-and-conquer change, we break an intersection of four constituents into two intersections of two constituents. In the repro, one of those divided intersections so happens to qualify for constraint reduction (e.g. where T & string becomes just T because T extends string), and that in turn defeats the hoisting logic.

The fix is to not perform constraint reduction in intersections created for purposes of hoisting constraints. I'll put up a PR.

[ahejlsberg]

Meanwhile, here's a simplified repro:

type Test1<K1 extends keyof any, K2 extends keyof any> =
    MustBeKey<Extract<K1, keyof any> & K1 & K2>;  // No error

type Test2<K1 extends keyof any, K2 extends keyof any> =
    MustBeKey<K1 & K2 & Extract<K1, keyof any>>;  // Error, wat?

type MustBeKey<K extends keyof any> = K;