Prettier 2.1.0 generates broken code with new ...infer T syntax in TypeScript 4.0

[papb]

Prettier 2.1.0
Playground link

--parser typescript

Input:

type Tail<T extends any[]> = T extends [infer U, ...infer R] ? R : never;

Output:

type Tail<T extends any[]> = T extends [infer U, ...(infer R)] ? R : never;

Expected behavior:

Output should be identical to input. The output given by Prettier does not compile in TypeScript 4.0.

Extra information:

This was mentioned by @bradennapier in #8646 but was forgotten.

I am working around this with // prettier-ignore for now.

---

Thank you for this awesome code formatter!

[papb]

By the way I am very surprised that TypeScript can't handle these parentheses... I feel the urge to put these parentheses myself, even if Prettier wouldn't put them...

Edit: I guess I just have to live with the fact that ...infer is actually another operator?

[bradennapier]

I am not sure the exact semantics atm. It is actually quite possible that there are cases where ...(infer U) would be used in a valid way, I would have to play with it some more. There isn't actually any direct information there in the release notes.

I think you'd be able to do ...(infer U)[] but have to confirm and that would give you a less precise type clearly if it works, since it'd be a union rather than a tuple.

[alexander-akait]

@bradennapier feel free to open a new issue

[bradennapier]

@bradennapier feel free to open a new issue

Just replying to the message previous.

I'll have to play with the syntax when I can, I'm pretty sure what I said would work but on mobile atm.

It seems likely that prettier shouldn't correct either use of it -- or if the () is literally only valid when used as an array then it could be smarter and remove () unless followed by [].

If anyone else has time to test to confirm as well that'd be good.

The ...infer style is the new syntax though and worst case could be converted to the other style if needed in meantime easy enough :).

[thorn0]

@bradennapier You're overthinking it. ...(infer U)[] looks similar to ...(infer U), but it doesn't have the same structure in the AST.

[brodybits]

I tried 2 more combos in tests/typescript/rest-type/infer-type.ts, don't know if they are relevant or not:

Input:

type Tail<T extends any[]> = T extends [infer U, ...infer R] ? R : never;
type Tail2<T extends any[]> = T extends [infer U, ...(infer R)] ? R : never;
type Tail3<T extends any[]> = T extends [infer U, ...(infer R)[]] ? R : never;

Output:

type Tail<T extends any[]> = T extends [infer U, ...infer R] ? R : never;
type Tail2<T extends any[]> = T extends [infer U, ...infer R] ? R : never;
type Tail3<T extends any[]> = T extends [infer U, ...(infer R)[]] ? R : never;

[ExE-Boss]

@brodybits The definition of Tail2 currently results in:

:2:51 - error TS2574: A rest element type must be an array type. 

2 | type Tail2<T extends any[]> = T extends [unknown, ...(infer R)] ? R : never;
                                                      ~~~~~~~~~~~~