Repository navigation
Key in mapped variadic tuple should not be literal after the variadic partΒ #48856
Description
Bug Report
π Search Terms
Variadic tuple mapped type keyof
π Version & Regression Information
- This is the behavior in every version I tried
β― Playground Link
Playground link with relevant code
π» Code
type Keys<O extends unknown[]> = { [K in keyof O]: K };
type A = Keys<[string, ...string[]]>;
type B = Keys<[string, ...string[], number]>;π Actual behavior
A and B both have the variadic part typed as ..."1"[], which is wrong. B is further wrong because the number part becomes "2", while in practice it would rarely be "2".
π Expected behavior
A should be ["0", ...number[]]; B should be ["0", ...number[], number]
Activity
ahejlsberg commentedon Apr 27, 2022
I don't think this is a bug. Consider:
With the proposed change,
Pwould benumberinstead of"2"for the rest element, which means the resulting rest element would be...Box<string | number | boolean>[]instead of just...Box<boolean[]>. Not only is that less correct, it is a breaking change. Effectively,numberis a less precise way of expressing "element at index 2 or above", whereas"2"gets much closer. This speaks to the fact that in homomorphic mappings over array and tuple types, the key types are just exemplars and not an exact representation of some specific property.Josh-Cena commentedon Apr 28, 2022
I also thought about that. The problem is it's really hard to be correct for both the key type itself, and for homomorphic mapped types to continue be homomorphic. However, accessing
T[P]andPdirectly should be two orthogonal actions. In my actual case, I'm doingEntriesinstead ofKeys:In an ideal world, I'd like
Entries<[number, string, ...boolean[]]>to be[["0", number], ["1", string], ...[number, boolean][]]. Does that sound reasonable?If we have subtracted types, the index should rather by
Exclude<number, 0 | 1>, but since we are not there yet, I'd be more happy withnumberthan2as the index type.typescript-bot commentedon Apr 30, 2022
This issue has been marked 'Working as Intended' and has seen no recent activity. It has been automatically closed for house-keeping purposes.
ahejlsberg commentedon Jan 11, 2024
Joshua Chen (@Josh-Cena) FWIW, I now agree with your observations and suggested changes. See #57031. Better late than never! π