Regression in type inference for a union of array types inside an object type

[daitay4]

I have some schema like the following:

z.object({
    arrayProp: z.union([z.string().array(), z.number().array()]),
}),

this used to produce the following type for the arrayProp key: string[] | number[].

since version 3.21.2 this produces the following type:
string[] | (string[] & number[]) | (string[] & undefined) | (number[] & string[]) | number[] | (number[] & undefined)

this in turn causes many problems when trying to work with arrayProp, for example if I try to use an array predicate like map, I'm getting the following error:

TS2349: This expression is not callable.   Each member of the union type '(<U>(callbackfn: (value: string, index: number, array: string[]) => U, thisArg?: any) => U[]) | (<U>(callbackfn: (value: number, index: number, array: number[]) => U, thisArg?: any) => U[]) | ((<U>(callbackfn: (value: string, index: number, array: string[]) => U, thisArg?: any) => U[]) & (<U>(callbackfn: (value: num...' has signatures, but none of those signatures are compatible with each other.

However, if I remove the wrapper z.object and use the following schema: z.union([z.string().array(), z.number().array()]) then I'm getting the correct type inference of string[] | number[]

Looking at the code for the 3.21.2 PR (4d016b7#diff-a5a27c41f819170124f4fd4cf24b1d71b8ab2966a4bde235a7f8db53dc8ef486) suggests that this has to do with changes made to the internalDeepPartial type.

For now we're sticking with 3.21.1 version, but I would be happy to see that fixed as we use this pattern on multiple occasions across our project.

[jer-sen]

I face something similar:

const prop = z.string().or(z.object({}));
const obj = z.object({ prop });
type ObjectPropType = z.TypeOf<typeof obj>['prop'];
// type ObjectPropType = (string | {}) & (string | {} | undefined)
type PropType = z.TypeOf<typeof prop>;
// type PropType = string | {}

[Michal-Miky-Jankovsky]

Maybe the problem is in the typescript itself.

microsoft/TypeScript#44373

[eyalch]

Maybe the problem is in the typescript itself.

microsoft/TypeScript#44373

How could the problem be with TypeScript itself if it's a regression in Zod that started in version 3.21.2?

[dcfreire]

Interestingly calling .partial() on a z.object produces the correct type.

Edit: Probably because (string | string[] | undefined) & (string | string[] | undefined) is the same as (string | string[] | undefined)?

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[eyalch]

Bump.

As mentioned in the original description, this issue forces us to stay at version 3.21.1.

[xsjcTony]

Any updates? This is definitely something need to be fixed

[xsjcTony]

a simple one

const schema = z.union([
  z.string(),
  z.number(),
  z.union([z.string(), z.number()]).array()
]);

ends up

string | (string & (string | number)[]) | number | (number & (string | number)[]) | ((string | number)[] & string) | ((string | number)[] & number) | (string | number)[] | ((string | number)[] & undefined)

where

string | number | (string | number)[]

is expected

[jon-dez]

Simpler one:

foo is inferred as the expected type, while bar isn't.

[Joeri1324]

Is there any update on this issue? I'm running into the exact same problem.