Set empty object when zodDiscriminatedUnion is specified.

[taku-hatano]

Hello team,

In parseWithZod, value coerce to empty object, if schema is ZodObject.
ref. https://github.com/edmundhung/conform/blob/v1.3.0/packages/conform-zod/coercion.ts#L240-L243

I think value should also coerce to empty object, if schema is ZodDiscriminatedUnion.
because ZodDiscriminatedUnionOption is extended type of ZodObject.
like this taku-hatano@c8514a4

What do you think about it?

[chimame]

Thank you for your suggestion.

I don't understand everything, so can you tell us why undefined is not good and should be changed to an empty object? From what I saw in your code, does it mean you want to change the error message?

I used Zod's playground to define a nested schema using discriminatedUnion and checked the behavior, and I confirmed that the Required error is displayed just like the current behavior.

Please explain in more detail if I don't understand your problem well.

[chimame]

Thank you for explaining again. I understand what you are trying to say.

I think this problem is not a problem with discriminatedUnion, but a problem with object. First, I created a code to verify the two patterns in this case only with Zod.

1. Pattern with nested object
2. Pattern with nested discriminatedUnion

In both cases, a Required error occurs for nest. However, in the case of Conform, the pattern with nested discriminatedUnion behaves the same as Zod, but the pattern with nested object behaves differently.

Therefore, I think we should conform to Zod's behavior. The problematic code is here.

conform/packages/conform-zod/coercion.ts

Lines 239 to 246 in 077215b

	.transform((value) => {
	if (typeof value === 'undefined') {
	// Defaults it to an empty object
	return {};
	}
	return value;
	})

@edmundhung
Should you follow Zod's behavior or should discriminatedUnion behave like object? I think it is better to follow Zod's behavior and reduce implicit conversions. This decision may also require modifications to the Valibot side, as Valibot does not perform implicit conversions for object.

[edmundhung]

Sorry for the late response!

The object coercion logic was intentional and implemented in #733. This is usually not necessary but it helps with cases like checkbox which doesn't contribute to the form data and people ends up with the "required" error (e.g. #391) which is quite similar to the example here.

Personally, I would prefer adding object coercion for discriminatedUnion as well as the whole idea of the coerceFormValue helper is to remove the need from the users to add special logic to handle form data in their schema.

[chimame]

@edmundhung

Thank you for explaining how this behavior was created. I think it would be good to add the same behavior to discriminatedUnion based on the concept of coerceFormValue.

What bothers me a little is that discriminatedUnion and union behave similarly, so it's a bit worrying to see a difference there. The following schema definition should behave almost the same as discriminatedUnion.

const schema = z.object({
  nest: z.union([
    z.object({
      type: z.literal('a')
    }),
    z.object({
      type: z.literal('b')
    })
  ]),
});

I think it's fine to use discriminatedUnion in rare cases.

@taku-hatano
My response has caused your change proposal to get lost.
Your proposal seems to be acceptable, so would you mind creating a pull request?

[edmundhung]

What bothers me a little is that discriminatedUnion and union behave similarly, so it's a bit worrying to see a difference there.

I understand and I feel the same at first when looking at this issue. But I think discriminatedUnion does accept object only (at least in v3), so it's an optimization we can do. It maybe worth looking if v4 expects the same or not 🤔

[taku-hatano]

I'm sorry for late response.

I created PR.
#909

[chimame]

I found something very interesting.
It seems that in Zod v4, discriminatedUnion becomes the same as union. So, if Zod v4 is supported, this problem will be naturally solved.

https://github.com/colinhacks/zod/blob/v4/v4.md

[edmundhung]

Is it exactly the same? I wonder the fact that discriminatedUnion might means that it works slightly different from union. It might just be the API signature looks the same now?

[chimame]

From what I saw in the code, I think it is defined the same as union internally. I haven't seen everything yet, so it may be wrong.

https://github.com/colinhacks/zod/blob/c87e0d045596c2737416807c51e7a31e7ff03de9/packages/core/src/api.ts#L1156-L1166