I saw this proposal #367 and the relevant context. I understand the need of recurseIntoArrays: false, but still get a little bit confused.

This #356 (comment) talks about the situation about what problem recurseIntoArrays: true would cause.

Array types like (string|undefined)[] are thorny because JSON.parse(JSON.stringify(['a', undefined, 'b'])) becomes ['a', null, 'b'] on the other side and null is not assignable to undefined. This is where PartialDeep was correctly causing a type failure.

I agree with

In some sense I think PartialDeep is going a little too far. {foo: string[]} becomes {foo?: (string|undefined)[]} but the user probably just intended {foo?: string[]} which would be fine for serialization.

But I think we didn't consider all the type possibilities.

For serialization,

• { foo: string[] } should become { foo?: string[] } instead of { foo?: (string | undefined)[] }
• { foo: { bar: string }[] } should become { foo?: { bar?: string }[] } instead of { foo?: ({ bar?: string } | undefined)[] } or { foo?: { bar: string }[] }

Type-generically, there are 3 results that { foo: T[] } can become:

1. { foo?: T[] }
2. { foo?: PartialDeep<T, Options>[] }
3. { foo?: PartialDeep<T | undefined, Options>[] }

I think No.2 is what most users want when serializing, not No.1.