vld-ts

vld-ts reads the JSON Schema of a vld schema (or a vld::schema! type) and prints a Zod 3 or Valibot 1 expression that accepts the same values. It has no runtime dependency besides vld (with openapi) and serde_json; run it from a build script, a test or a small binary and write the output to your frontend.

Install

[dependencies]
vld = { version = "0.5", features = ["openapi"] }
vld-ts = "0.5"

No optional features. MSRV: 1.85 (see MSRV).

Example

use vld_ts::{impl_to_valibot, impl_to_zod, to_zod, ToValibot, ToZod};

vld::schema! {
    pub struct Address {
        pub city: String => vld::string().min(1),
    }
}

vld::schema! {
    pub struct User {
        pub name: String => vld::string().min(2).max(50),
        pub address: Address => vld::nested!(Address),
    }
}

impl_to_zod!(Address);
impl_to_zod!(User, "UserBody"); // custom export name
impl_to_valibot!(User);

// A single schema value
assert_eq!(to_zod(&vld::string().email()), "z.string().email()");

// A module with named exports; the import line is yours.
let module = format!(
    "import {{ z }} from \"zod\";\n\n{}\n{}",
    Address::to_zod_export(),
    User::to_zod_export(),
);
std::fs::write("frontend/src/schemas.ts", module).unwrap();

let valibot = User::to_valibot(); // "v.object({...})"

Generated schemas.ts:

import { z } from "zod";

export const AddressSchema = z.object({
  city: z.string().min(1)
});
export type Address = z.infer<typeof AddressSchema>;

export const UserBodySchema = z.object({
  name: z.string().min(2).max(50),
  address: z.lazy(() => AddressSchema)
});
export type UserBody = z.infer<typeof UserBodySchema>;

Valibot exports use v.InferOutput<typeof …> and pipes (v.pipe(v.string(), v.minLength(2), v.maxLength(50))).

API

What is generated

Dates (date, date-time, time) use vld's exact ISO grammar as regexes rather than .datetime(), so both sides accept the same strings.

Gotchas

• 0.5 change: with_default now emits x.nullish().transform((v) => v ?? d) / v.nullish(x, d), because vld accepts null for a defaulted field. Regenerate client schemas after upgrading.
• Field order: required fields come out in declaration order (taken from required). Optional fields keep their place in the JSON map, which serde_json sorts unless its preserve_order feature is on, so turn it on to get the full declared order.
• Zod output targets the Zod 3 API (z.string().email(), .ip({ version })).
• $refs point at <TypeName>Schema, so export every nested type too, under its type name. Rename (impl_to_zod!(T, "Name")) only top-level types.
• email uses .email() / v.email(), which require a dot in the domain (vld accepts a@b). url_strict's public-host check and phone_e164_strict's numbering-plan check are shape-only on the client.
• Rust regex features JS cannot express (\p{…}, scoped flags, class set operations, …) are not checked: the output contains a /* pattern not portable to JS, not checked */ note instead of a wrong regex.
• Refinements are attached last (a Zod refinement returns ZodEffects, which has no .min()).
• The crate's tests can run the generated code under node: set VLD_TS_NODE_DIR to a directory whose node_modules has zod@3 and valibot@1 (see Contributing).