Declaration emit errors (TS5088) on anonymous cyclic types that 6.0 elided to `any`

[ssalbdivad]

🔎 Search Terms

TS5088, cyclic structure, trivially serialized, declaration emit, elided any, TS 7

🕗 Version & Regression Information

Errors on 7.1.0-dev.20261006.1. 6.0.3 emits /*elided*/ any at the cycle instead. (7.0.2 also emitted /*elided*/ any for both repros below when I tried it, in case that helps narrow anything down.)

⏯ Playground Link

No response

💻 Code

// tsc --declaration --emitDeclarationOnly --strict repro.ts
type show<t> = { [k in keyof t]: t[k] } & unknown
type resolve<$, k extends keyof $> = show<{
	[p in keyof $[k]]: $[k][p] extends "ref" ? resolve<$, k> : $[k][p]
}>
declare const mod: <$>() => { [k in keyof $]: resolve<$, k> }
export const n = mod<{ Node: { value: number; next: "ref" } }>()

🙁 Actual behavior

error TS5088: The inferred type of 'n' references a type with a cyclic structure which cannot be trivially serialized. A type annotation is necessary.

🙂 Expected behavior

Same as 6.0.3, which emits:

export declare const n: {
    Node: {
        value: number;
        next: /*elided*/ any;
    };
};

Not sure whether the error is intended though! I can see the argument that an error is more honest than quietly writing any into a .d.ts, so if this is intentional, totally fair, but it might be worth calling out as a breaking change.

Additional information about the issue

Ran into this while moving ArkType's repo to TS 7. Here's what it looks like for an ArkType user exporting a scope with a recursive type:

import { scope } from "arktype"

const $ = scope({
	Node: { value: "number", "next?": "Node" },
	List: { head: "Node" }
})

// 6.0.3: emits, with `next?: /*elided*/ any` at the cycle
// 7.1.0-dev.20261006.1: error TS5088
export const types = $.export()

There's no reasonable annotation for users to write here, since the whole point is that the type is inferred from the definitions, so in practice this is a breaking change for anyone shipping declarations from code like this.

Thanks!

[ssalbdivad]

Mateusz Burzyński (@Andarist) mentioned it may be possible to emit the actual type using internal aliases like _Node here.

Would be fantastic if it is!