TS 4.1 regression in type assertion of generic mapped type

[Igorbek]

TypeScript Version: 4.1.2

Search Terms: CFA, narrowing, generic, mapped type, conditional type, type assertion

Code

interface A { x: number }

declare function isA(a: unknown): a is A;

type FunctionsObj<T> = {
  [K in keyof T]: () => unknown
}

function g<
  T extends FunctionsObj<T>,
  M extends keyof T
>(a2: ReturnType<T[M]>) {
  if (isA(a2)) {
    // a2 is not narrowed
    a2.x // error, but should be ok
  }
}

Expected behavior:

Type of a2 should be narrowed to ReturnType<T[M]> & A
It works this way in TS 4.0

Actual behavior:

Type of a2 remains only ReturnType<T[M]>

Playground Link: https://www.typescriptlang.org/play?ts=4.1.0-beta#code/JYOwLgpgTgZghgYwgAgILIN7IB4C5kgCuAtgEbTIC+AUNQCYQIA2cUKMhICYwA9iMmABnVAAo4+TgGsQvAO4gAlPjiChaANy0wATwAOKAGKdufEEIDypAFYAeACoA+ZAF5M1ZMgDaAaUECpCB1eGGR7AF18UUVXZ2lZBWoaag4uHn5kAHNbDzDkCGxIEDp1YzSzSxsHRwAaXIBZfMKIYvVA4ND7akdxACZ8ACUIMEIoEHt9CAcvevDHGIxc4FDRYTE4XsUF3M8NgDpsZAB6I-yoKF4oXJoaIA

[Igorbek]

I tried to reduce the repro, but it reproduces only when all these present:

• conditional type that is being narrowed ReturnType<>
• generic mapped type that is constraints by itself T extends FunctionsObj<T>
• field accessed with generic type T[M]

[orta]

@typescript-bot run repros

[typescript-bot]

Heya @orta, I've started to run the code sample repros for you. Here's the link to my best guess at the log.

[typescript-bot]

👋 Hi, I'm the Repro bot. I can help narrow down and track compiler bugs across releases! This comment reflects the current state of the 2 repros in this issue running against the nightly TypeScript.

---

Issue body code block by @Igorbek

❌ Failed: -

• Property 'x' does not exist on type 'ReturnType<T[M]>'.

Comment by @weswigham

👍 Compiled

Historical Information

Issue body code block by @Igorbek

Version	Reproduction Outputs
4.1.2, Nightly	❌ Failed: - Property 'x' does not exist on type 'ReturnType<T[M]>'.
3.7.5, 3.8.2, 3.9.2, 4.0.2	👍 Compiled

Comment by @weswigham

Version	Reproduction Outputs
Nightly	👍 Compiled

[weswigham]

Here's the problem, but without control flow involved:

interface A { x: number }

declare function isA(a: unknown): a is A;

type FunctionsObj<T> = {
    [K in keyof T]: () => unknown
}

function g<
    T extends FunctionsObj<T>,
    M extends keyof T
>(a2: ReturnType<T[M]>, x: A) {
    x = a2; // shouldn't be allowed, but is
}

This is a bug in our subtyping rules, not control flow (which would undoubtedly be easier to fix).

@typescript-bot run repros

[typescript-bot]

Heya @weswigham, I've started to run the code sample repros for you. Here's the link to my best guess at the log.

[weswigham]

Looks like the root cause is our change in 4.1 to isDeeplyNestedType - it is now erroneously flagging a target side of the comparison, which doesn't change, as "deeply nested". Since the source side of the relation goes through about 6 transforms with the target side remaining unchanged, both sides get flagged as "deeply nesting", resulting in a Maybe result which eventually allows the assignment. I think this can be remedied by having separate source/target stack depths, so we don't increase the target stack depth when we're only transforming the source for a comparison.