Preprocess and parser Input type

[sbking]

When using z.preprocess, the parser's Input type is set to the wrapped parser's Input type. In most cases, this is probably incorrect, and this prevents using z.preprocess with frameworks like tRPC without type errors.

Example:

const parser = z.preprocess((data) => Number(data), z.number());

function handle(data: z.input<typeof parser>) {
  return parser.parse(data);
}

handle("42");
// ts-error: Argument of type 'string' is not assignable to parameter of type 'number'.

^ That's a very simplified example of how this works in tRPC. I'm not sure what the ideal API is here. Maybe if we could explicitly annotate the preprocess argument? e.g.:

const parser = z.preprocess((data: string | number) => Number(data), z.number());

type Input = z.input<typeof parser>;
assertEqual<Input, string | number>(true);

Or alternatively, should preprocess simply always have an Input type of unknown? This seems like it would make the most sense, because using preprocess assumes you can handle any unknown value.

const parser = z.preprocess((data) => Number(data), z.number());

type Input = z.input<typeof parser>;
assertEqual<Input, unknown>(true);

[sbking]

Thinking about this some more, I think a better solution would be to have some kind of chain API that allows you to pipe the result of one parser to another parser:

const numberParser = z.preprocess(Number, z.number().int().nonnegative());
const stringParser = z.string().chain(numberParser).transform((n) => [n]);

// Input: string
// Output: number[]

This would be essentially equivalent to:

const parser = z
  .string()
  .transform(Number)
  .refine((n) => z.number().int().nonnegative().safeParse(n).success)
  .transform((n) => [n]);

However using refine and safeParse is really tedious, and refine has incorrect types when used with type guards - see #1404

In fact I think a chain API would support all use cases for preprocess but in a more generic way:

const parser = z.unknown().transform(Number).chain(z.number().int().nonnegative());

[colinhacks]

Or alternatively, should preprocess simply always have an Input type of unknown?

This is definitely what it should be, updated in 3.19.1.

The chain API is on the roadmap (in my brain) and is discussed elsewhere so I'm gonna close this 👍