[TS 5.4.0-beta] Tuple type members being preserved

[acutmore]

🔎 Search Terms

spread tuple type narrowing

🕗 Version & Regression Information

• This changed between versions 5.3.3 and 5.4.0-beta
• This changed in 5.4.0-dev.20240120

⏯ Playground Link

https://www.typescriptlang.org/play?ts=5.4.0-beta#code/CYUwxgNghgTiAEAzArgOzAFwJYHtXwHMQMAVAHhID4AKASgC54SBuAKFYAUYcBbLAZxAA6KBAjUA2kVJl+GGFlQEatADTwhm6eSioAnioC6tVvDNCMACxCpq1OP2QQMteAF5K8AN6mzfswD0AfAAegD8vv5geHLw1lDA7vAOThgSAAyGzP5+QfAArEIAzIy6euqFACyMcgpKkTnweeENZtGosRhQWBBJKc5C-BBYYCDUAIy02XmFRfBlFULV8GgA1qg4AO6oEoat-s0RjV098PxQ2PyIWCD8Z-KKBLvZB8FkALTv8Kggm-AgMG4MHgigKS0iAF8pqwgA

💻 Code

declare function getT<T>(): T;

Promise.all([getT<string>(), ...getT<any>()])
 .then((result) => {
 // ^?
 const head = result[0]; // 5.3: any, 5.4: string
 // ^?
 const tail = result.slice(1); // 5.3 any, 5.4: unknown[]
 // ^?
 tail satisfies string[]; // <-- new error in 5.4
 });

Additional information about the issue

More of a 'change report' than a 'bug report', I don't think this was part of the 5.4 beta announcement.

[Andarist]

This can be bisected to: #57031 and it's a bug.

When hovering over all we can see this:

(method) PromiseConstructor.all<[string, ...any[]]>(values: [string, ...any[]]): Promise<[string, ...unknown[]]> (+1 overload)

So how [string, ...any[]] was converted into [string, ...unknown[]] here? The reason is that now instantiateMappedTupleType has this ?? unknownType fallback when computing newElementTypes.

But it's not consistent with how mapped types continue to work today:

type Identity<T> = {
 [K in keyof T]: T[K]
}

type Result = Identity<[string, ...any[]]>
// ^? type Result: [string, ...any[]]

So the problem is that what we see in the original repro is actually not a true any but rather an errorType. And if we play around with it we can see the change in the behavior:

type Identity<T> = {
 [K in keyof T]: T[K]
}

type ErrorType = {}['prop']

type Result = Identity<[string, ...ErrorType[]]>
// ^? 5.3: any (displayed as `any` but it's an `errorType`) 5.4: [string, ...unknown[]] 

The PR to which this has been bisected improves the situation (the output should be a tuple and not an any/errorType) but accidentally maps the thing to [string, ...unknown[]].

Why the errorType has crawled into all of this though? Well, I already have a fix for this 😉 : #57116