Types forgetting key information for infinite keys

[GuiltyDolphin]

Bug Report

🔎 Search Terms

• key number string
• type key loses
• etc. (I found this hard to search for, please let me know if there's another issues regarding this)

🕗 Version & Regression Information

• This is the behavior in every version I tried, and I reviewed the FAQ index but couldn't see any relevant entries, let me know if there are!

⏯ Playground Link

Playground link with relevant code

💻 Code

type KeyIsFinite = {
    [k in 0 | 1 | 2]: k
}

type KeyIsNum = {
    [k in number]: k
}

// if KeyIsFinite[1]: 1 then this will be true
const keyIsFiniteTest: KeyIsFinite[1] extends 1 ? true : false = true;
// if KeyIsNum[1]: 1 then this will be true, but this is actually false, because in KeyIsNum, we are losing information about what `k` is
const keyIsNumTest: KeyIsNum[1] extends 1 ? true : false = true;

// similarly, you can do this with strings
type KeyIsFiniteStr = {
    [k in '0' | '1' | '2']: k
}

type KeyIsStr = {
    [k in string]: k
}

const keyIsFiniteStrTest: KeyIsFiniteStr['1'] extends '1' ? true : false = true;
const keyIsStrTest: KeyIsStr['1'] extends '1' ? true : false = true;

🙁 Actual behavior

We see a type error informing us that keyIsNumTest has type false but is the value true, this is because KeyIsNum is not preserving information about k (I assume), and thus is saying that KeyIsNum[1] = number rather than KeyIsNum[1] = 1. We see that KeyIsFinite[1] = 1, so this appears to be an inconsistency.

🙂 Expected behavior

I would expect KeyIsNum to behave like KeyIsFinite and thus for KeyIsNum[1] = 1 rather than KeyIsNum[1] = number

[jcalz]

Duplicate of (or strongly related to) #22509

[RyanCavanaugh]

Strong duplicate energy but really just working as intended: mapped types over finite types produce object types with properties, but mapped types over infinite types produce index signatures. Different input different output.

[GuiltyDolphin]

@jcalz @RyanCavanaugh Yep, I think with "Here I expect string to behave as if it were the union of all possible string literal types." in #22509 this is likely a duplicate. I'd like to just promote that it'd be good for this to work for things like numbers too, but I'll close this in favour of that issue. Thanks!