ᯓ𝘁𝗿𝗮𝘃𝗲𝗿𝘀𝗮𝗯𝗹𝗲/𝘀𝗰𝗵𝗲𝗺𝗮

TypeScript schema rewriter

       

     

Overview

A schema is a syntax tree. ASTs lend themselves to (re)-interpretation. If you're not treating your TypeScript schemas like ASTs, you're missing out.

@traversable/schema makes it easy to do anything with a TypeScript schema.

What's a "schema rewriter"?

The idea of term rewriting comes from the programming language community. Languages like Racket and Lean invert control and give users a first-class API for rewriting and extending the language.

Unfortunately, we don't have that kind of power in TypeScript because we're limited by the target language (JavaScript). And frankly, given how flexible JavaScript already is, exposing that kind of API would be a recipe for disaster.

We do however have schemas, and schemas are basically ASTs.

Let's look at a concrete example of how @traversable/schema can be used as a rewriting tool.

Example

For this example, we'll be using @traversable/zod, since zod is the library most users are familiar with.

Let's write a function that takes an arbitrary zod schema as input and stringifies it.

import { zx } from '@traversable/schema'

const toString = zx.fold<string>((x) => {
  //                     𐙘____𐙘 this type parameter fills in the "holes" below
  switch (true) {
    case zx.tagged('null')(x): return 'z.null()'
    case zx.tagged('number')(x): return 'z.number()'
    case zx.tagged('string')(x): return 'z.string()'
    case zx.tagged('boolean')(x): return 'z.boolean()'
    case zx.tagged('undefined')(x): return 'z.undefined()'
    case zx.tagged('array')(x): return `${x._zod.def.element}.array()`
    //                                                 ^? method element: string
    case zx.tagged('optional')(x): return `${x._zod.def.innerType}.optional()`
    //                                                     ^? method innerType: string
    case zx.tagged('tuple')(x): return `z.tuple([${x._zod.def.items.join(', ')}])`
    //                                                         ^? method items: string[]
    case zx.tagged('record')(x): return `z.record(${x._zod.def.keyType}, ${x._zod.def.valueType})`
    //                                                            ^? method keyType: string
    case zx.tagged('object')(x): 
      return `z.object({ ${Object.entries(x._zod.def.shape).map(([k, v]) => `${k}: ${v}`).join(', ')} })`
    //                                                ^? method shape: { [x: string]: string }
    default: throw Error(`Unimplemented: ${x._zod.def.type}`)
    //              ^^ there's nothing stopping you from implementing the rest!
  }
})

// Let's test it out:

console.log(
  zx.toString(
    z.object({ A: z.array(z.string()), B: z.optional(z.tuple([z.number(), z.boolean()])) })
  )
)
// => z.object({ A: z.array(z.string()), B: z.optional(z.tuple([z.number(), z.boolean()])) })
Copy

Our "naive" implementation is actually more robust than it might seem -- in fact, that's how zx.toString is actually defined.

Off-the-shelf

@traversable/zod ships with a bunch of rewriters available off-the-shelf, including:

• zx.check
• zx.deepPartial
• zx.deepRequired
• zx.deepNullable
• zx.deepNonNullable
• zx.defaultValue
• zx.deepEqual
• zx.toPaths
• zx.toString
• zx.toType
• zx.generator (🧪)
• zx.makeLens (🧪)

Other packages

@traversable/schema supports other schema libraries too, but they are still being fuzz-tested and aren't ready for production yet.

Additionally, @traversable/schema publishes its own schema library that's been optimized for AST traversal, and which is documented below.

Schema library

@traversable/schema (the package) exploits a TypeScript feature called inferred type predicates to do what libaries like zod do, without the additional runtime overhead or abstraction.

Note:

These docs are a W.I.P.

We recommend jumping straight to the demo or playground.

Requirements

The only hard requirement is TypeScript 5.5. Since the core primitive that @traversable/schema is built on top of is inferred type predicates, we do not have plans to backport to previous versions.

Quick start

import { t } from '@traversable/schema'

declare let ex_01: unknown

if (t.bigint(ex_01)) {
    ex_01
    // ^? let ex_01: bigint
}

const schema_01 = t.object({
  abc: t.optional(t.string),
  def: t.tuple(
    t.eq(1),
    t.optional(t.eq(2)), // `t.eq` can be used to match any literal JSON value
    t.optional(t.eq(3)),
  )
})

if (schema_01(ex_01)) {
    ex_01
    // ^? let ex_01: { abc?: string, def: [ᵃ: 1, ᵇ?: 2, ᶜ?: 3] }
    //                                     ^ tuples are labeled to support optionality
}
Copy

Features

@traversable/schema is modular by schema (like valibot), but takes it a step further by making its feature set opt-in by default.

The ability to add features like this is a knock-on effect of traversable's extensible core.

First-class support for inferred type predicates

Note: This is the only feature on this list that is built into the core library.

The motivation for creating another schema library was to add native support for inferred type predicates, which no other schema library currently does (although please file an issue if that has changed!).

This is possible because the traversable schemas are themselves just type predicates with a few additional properties that allow them to also be used for reflection.

• Instructions: To use this feature, define a predicate inline and @traversable/schema will figure out the rest.

Example

You can play with this example in the TypeScript Playground.

import { t } from '@traversable/schema'

export let Classes = t.object({
  promise: (v) => v instanceof Promise,
  set: (v) => v instanceof Set,
  map: (v) => v instanceof Map,
  weakMap: (v) => v instanceof WeakMap,
  date: (v) => v instanceof Date,
  regex: (v) => v instanceof RegExp,
  error: (v) => v instanceof Error,
  typeError: (v) => v instanceof TypeError,
  syntaxError: (v) => v instanceof SyntaxError,
  buffer: (v) => v instanceof ArrayBuffer,
  readableStream: (v) => v instanceof ReadableStream,
})

type Classes = t.typeof<typeof Classes>
//   ^? type Classes = {
//   promise: Promise<any>
//   set: Set<any>
//   map: Map<any, any>
//   weakMap: WeakMap<object, any>
//   date: Date
//   regex: RegExp
//   error: Error
//   typeError: TypeError
//   syntaxError: SyntaxError
//   buffer: ArrayBuffer
//   readableStream: ReadableStream<any>
// }

let Values = t.object({
  function: (v) => typeof v === 'function',
  successStatus: (v) => v === 200 || v === 201 || v === 202 || v === 204,
  clientErrorStatus: (v) => v === 400 || v === 401 || v === 403 || v === 404,
  serverErrorStatus: (v) => v === 500 || v === 502 || v === 503,
  teapot: (v) => v === 418,
  true: (v) => v === true,
  false: (v) => v === false,
  mixed: (v) => Array.isArray(v) || v === true,
  startsWith: (v): v is `bill${string}` => typeof v === 'string' && v.startsWith('bill'),
  endsWith: (v): v is `${string}murray` => typeof v === 'string' && v.endsWith('murral'),
})

type Values = t.typeof<typeof Values>
//   ^? type Values = {
//   function: Function
//   successStatus: 200 | 201 | 202 | 204
//   clientErrorStatus: 400 | 401 | 403 | 404
//   serverErrorStatus: 500 | 502 | 503
//   teapot: 418
//   true: true
//   false: false
//   mixed: true | any[]
//   startsWith: `bill${string}`
//   endsWith: `${string}murray`
// }

let Shorthand = t.object({
    nonnullable: Boolean,
    unknown: () => true,
    never: () => false,
})

type Shorthand = t.typeof<typeof Shorthand>
//   ^? type Shorthand = {
//   nonnullable: {}
//   unknown: unknown
//   never?: never
// }
Copy

.validate

.validate is similar to z.safeParse, except more than an order of magnitude faster*.

• Instructions: To install the .validate method to all schemas, simply import @traversable/schema-to-validator/install.
• [ ] TODO: add benchmarks + write-up

Example

Play with this example in the TypeScript playground.

import { t } from '@traversable/schema'
import '@traversable/schema-to-validator/install'
//      ↑↑ importing `@traversable/schema-to-validator/install` adds `.validate` to all schemas

let schema_01 = t.object({ 
  product: t.object({ 
    x: t.integer, 
    y: t.integer 
  }), 
  sum: t.union(
    t.tuple(t.eq(0), t.integer), 
    t.tuple(t.eq(1), t.integer),
  ),
})

let result = schema_01.validate({ product: { x: null }, sum: [2, 3.141592]})
//                     ↑↑ .validate is available

console.log(result)
// => 
// [
//   { "kind": "TYPE_MISMATCH", "path": [ "product", "x" ], "expected": "number", "got": null },
//   { "kind": "REQUIRED", "path": [ "product" ], "msg": "Missing key 'y'" },
//   { "kind": "TYPE_MISMATCH", "path": [ "sum", 0 ], "expected": 0, "got": 2 },
//   { "kind": "TYPE_MISMATCH", "path": [ "sum", 1 ], "expected": "number", "got": 3.141592 },
//   { "kind": "TYPE_MISMATCH", "path": [ "sum", 0 ], "expected": 1, "got": 2 },
//   { "kind": "TYPE_MISMATCH", "path": [ "sum", 1 ], "expected": "number", "got": 3.141592 },
// ]
Copy

.toString

One of @traversable/schema's primary goals is to remove as much friction from the code generation / metaprogramming workflow as possible.

To support that goal, all schemas shipped by the @traversable/schema package come with a .toString method that, when called, will return the schema as code.

This is also useful if you're ever in a situation where you're working with generated schemas, and you need to trouble shoot.

Example

import { t } from '@traversable/schema'

const CreateTodoAction = t.object({ type: t.eq('CREATE_TODO') })
const DeleteTodoAction = t.object({ type: t.eq('DELETE_TODO'), id: t.integer })
const TodoAction = t.union(
  CreateTodoAction,
  DeleteTodoAction,
)

console.log(TodoAction + '') 
// => t.union(t.object({ type: t.eq('CREATE_TODO') }), t.object({ type: t.eq('DELETE_TODO'), id: t.integer }))
Copy

.toType

The .toType method prints a stringified version of the type that the schema represents.

Works on both the term- and type-level.

• Instructions: To install the .toType method on all schemas, simply import @traversable/schema-to-string/install.

• Caveat: type-level functionality is provided as a heuristic only; since object keys are unordered in the TS type system, the order that the keys are printed at runtime might differ from the order they appear on the type-level.

Example

Play with this example in the TypeScript playground

import { t } from '@traversable/schema'
import '@traversable/schema-to-string/install'
//      ↑↑ importing `@traversable/schema-to-string/install` adds the upgraded `.toType` method on all schemas

const schema_02 = t.intersect(
  t.object({
    bool: t.optional(t.boolean),
    nested: t.object({
      int: t.integer,
      union: t.union(t.tuple(t.string), t.null),
    }),
    key: t.union(t.string, t.symbol, t.number),
  }),
  t.object({
    record: t.record(t.string),
    maybeArray: t.optional(t.array(t.string)),
    enum: t.enum('x', 'y', 1, 2, null),
  }),
)

let ex_02 = schema_02.toType()
//  ^? let ex_02: "({ 
//       'bool'?: (boolean | undefined), 
//       'nested': { 'int': number, 'union': ([string] | null) }, 
//       'key': (string | symbol | number) } 
//     & { 
//        'record': Record<string, string>, 
//        'maybeArray'?: ((string)[] | undefined), 
//        'enum': 'x' | 'y' | 1 | 2 | null 
//     })"
Copy

.toJsonSchema

• Instructions: To install the .toJsonSchema method on all schemas, simply import @traversable/schema-to-json-schema/install.

Example

Play with this example in the TypeScript playground.

import * as vi from 'vitest'

import { t } from '@traversable/schema'
import '@traversable/schema-to-json-schema/install'
//      ↑↑ importing `@traversable/schema-to-json-schema/install` adds `.toJsonSchema` on all schemas

const schema_02 = t.intersect(
  t.object({
    stringWithMaxExample: t.optional(t.string.max(255)),
    nestedObjectExample: t.object({
      integerExample: t.integer,
      tupleExample: t.tuple(
        t.eq(1),
        t.optional(t.eq(2)),
        t.optional(t.eq(3)),
      ),
    }),
    stringOrNumberExample: t.union(t.string, t.number),
  }),
  t.object({
    recordExample: t.record(t.string),
    arrayExample: t.optional(t.array(t.string)),
    enumExample: t.enum('x', 'y', 1, 2, null),
  }),
)

vi.assertType<{
  allOf: [
    {
      type: "object"
      required: ("nestedObjectExample" | "stringOrNumberExample")[]
      properties: {
        stringWithMaxExample: { type: "string", minLength: 3 }
        stringOrNumberExample: { anyOf: [{ type: "string" }, { type: "number" }] }
        nestedObjectExample: {
          type: "object"
          required: ("integerExample" | "tupleExample")[]
          properties: {
            integerExample: { type: "integer" }
            tupleExample: {
              type: "array"
              minItems: 1
              maxItems: 3
              items: [{ const: 1 }, { const: 2 }, { const: 3 }]
              additionalItems: false
            }
          }
        }
      }
    },
    {
      type: "object"
      required: ("recordExample" | "enumExample")[]
      properties: {
        recordExample: { type: "object", additionalProperties: { type: "string" } }
        arrayExample: { type: "array", items: { type: "string" } }
        enumExample: { enum: ["x", "y", 1, 2, null] }
      }
    }
  ]
}>(schema_02.toJsonSchema())
//           ↑↑ importing `@traversable/schema-to-json-schema` installs `.toJsonSchema`
Copy

Codec (.pipe, .extend, .parse, .decode & .encode)

• Instructions: to install the .codec method on all schemas, all you need to do is import @traversable/schema-codec.
  • To create a covariant codec (similar to zod's .transform), use .codec.pipe
  • To create a contravariant codec (similar to zod's .preprocess), use .codec.extend (WIP)

Example

Play with this example in the TypeScript playground.

import { t } from '@traversable/schema'
import '@traversable/schema-codec/install'
//      ↑↑ importing `@traversable/schema-codec/install` adds `.codec` on all schemas

let User = t
  .object({ name: t.optional(t.string), createdAt: t.string })
  .codec // <-- notice we're pulling off the `.codec` property
  .pipe((user) => ({ ...user, createdAt: new Date(user.createdAt) }))
  .unpipe((user) => ({ ...user, createdAt: user.createdAt.toISOString() }))

let fromAPI = User.parse({ name: 'Bill Murray', createdAt: new Date().toISOString() })
//   ^?  let fromAPI: Error | { name?: string, createdAt: Date}

if (fromAPI instanceof Error) throw fromAPI
fromAPI
// ^? { name?: string, createdAt: Date }

let toAPI = User.encode(fromAPI)
//  ^? let toAPI: { name?: string, createdAt: string }
Copy