ZodFormattedError does not propagate the custom error format to recursive fields

[carlgieringer]

ZodFormattedError accepts two type params.

export type ZodFormattedError<T, U = string>

T is the type of the input schema and U is the type of each error entry. U is important when providing a custom issue mapper. ZodFormattedError uses T to define the shape of the formatted error object, but it does not propagate U to the fields of this object. E.g.:

... ZodFormattedError<NonNullable<T>[K]>

This means that typing fails when trying to access custom error types of a `ZodFormattedError.

Example: the following code should pass typechecking sandbox:

import { z } from "zod";

const Thing = z.object({
  name: z.string().min(2)
});

const result = Thing.safeParse({ name: "a" });
if (!result.success) {
  const formatted = result.error.format((issue) => ({
    code: issue.code,
    message: issue.message
  }));
  console.log(formatted.name?._errors?.code);
}

But instead it fails with `Property 'code' does not exist on type 'string[]'.ts(2339)'

[carlgieringer]

Hi, this issue has a simple fix, as presented in #1617.

[carlgieringer]

@colinhacks Hi, should I do anything in particular to have the PR for this issue reviewed? Thanks!

[JacobWeisenburger]

sometimes you just need to be a squeaky wheel. :D

I approved and merged it

[carlgieringer]

I wasn't sure how squeaky I should be. Thanks, @JacobWeisenburger!

[colinhacks]

@carlgieringer Unfortunately this has been reverted in 3.21 for causing infinite recursion & compiler slowdowns. It's possible this can be re-implemented in a way that avoids these issues (maybe) but I had to put out a fix asap and didn't have time to look into it.

[carlgieringer]

@colinhacks too bad! I see the PR that removed the feature and the references to slowness. If I re-attempt, I'll check those.

Also could someone re-open this issue? Or should I just create a new one?

[romellem]

I know this is an old issue / thread, but if anyone has time, could they explain what the actual "recursive type" issue was? As mentioned, since it is just a "6 character change," it is hard to understand how that was able to have such a large performance impact