oneOf

oneOf is essentially useless and a pitfall for bad practices of schema building. I can't think of a single case where you wouldn't want to use a discriminated union instead. Not saying there aren't any, but... yaknow. If anyone comes across this in the future, consider the following:

Zod

const unionSchema = z.union([z.string(), z.string()])

With the above Zod schema, any string will pass validation.

Current output

{
  "anyOf": [
    {
      "type": "string"
    },
    {
      "type": "string"
    }
  ]
}

With the above JSON schema, any string will pass validation, just like the source Zod schema.

Output with proposed change in issue

{
  "oneOf": [
    {
      "type": "string"
    },
    {
      "type": "string"
    }
  ]
}

With the above JSON schema, no value will ever pass validation. The schema is "broken". 😿

from zod-to-json-schema.

Why? Please provide a problem case

from zod-to-json-schema.

Sorry about the noise. I misunderstood the problem I was having.

I was using https://github.com/drwpow/openapi-typescript and it converts anyOf as Partial<number> & Partial<string>. So the actual issue is caused by the openapi-typescript library.

from zod-to-json-schema.