Juniper round trip

Juniper builds a JSON Schema and its type together. Once the schema is serialized, to a file or into an OpenAPI spec, only the JSON is left. genever reads that JSON back into the type Juniper inferred:

import { arraySchema, objectSchema, stringSchema } from 'juniper';

const user = objectSchema({
    properties: {
        id: stringSchema().startsWith('usr_'),
        tags: arraySchema(stringSchema()).minItems(1),
    },
    required: ['id'],
});

// Written to user.json, then `genever user.json --name User` writes:
// export type User = {
//     id: `usr_${string}`;
//     tags?: [string, ...string[]];
// };

genever's tests build schemas covering every Juniper schema type and keyword, serialize them to JSON text, genever them, and check the generated type is equivalent to Juniper's SchemaType, in both JSON Schema 2020-12 and OpenAPI 3.0. Juniper's own escape hatch, customSchema<T>(), takes a hand-written type, which is the type genever saves you from maintaining.

Differences

genever reads only the JSON. Where the JSON says more than Juniper's type does, or less, genever follows the JSON:

• An object schema with no properties is Record<string, unknown>, not Juniper's empty object type.
• Prefix items beyond minItems are optional, as a validator accepts a shorter array. Juniper's tupleSchema() always sets minItems, so its tuples are identical.
• In if/then/else, a property set to false in else is ?: never.
• A type that only exists in TypeScript, from .cast<T>() or customSchema<T>(), is read from the JSON the schema emits instead.
• minItems is typed up to 3 by default, as in Juniper. genever's --typed-min-items changes the cap.