!Types

json-schema-library

> json-schema-library provides tools and utilities for working with JSON Schema - enabling creation, validation, and schema exploration. Unlike most validators and editors, which hide the inner workings, this library is designed for developers building custom tools around JSON Schema. It runs in both Node and browser environments, prioritizing flexibility and extensibility over minimal memory footprint or raw performance.

---

---

Quick start

npm install json-schema-library

_json-schema-library_ includes a compileSchema function that converts a JSON Schema into a SchemaNode, which gives you easy-to-use methods for working with the schema.

``ts
import { compileSchema, SchemaNode } from "json-schema-library";
import myJsonSchema from "./myJsonSchema.json";
import myData from "./myData.json";

const schema: SchemaNode = compileSchema(myJsonSchema);
// validate data and collect errors if invalid
const { valid, errors } = schema.validate(myData);
// create data which validates to the compiled JSON Schema
const defaultData = schema.getData();
// access a subschema at a specific JSON Pointer location
const { node, error } = schema.getNode("#/image/title");
node && console.log(node.schema);
`

Per default, compileSchema uses the draft-version referenced in $schema for _cross-draft_ support. In case $schema is omitted or invalid the latest schema (draft-2020-12) will be used. Customizing draft selection is documented in draft customization.

`ts
const schemaNode = compileSchema({ $schema: "draft-07" });
console.log(schemaNode.getDraftVersion()); // draft-07
`

Overview

$3

Use compileSchema once to turn a JSON Schema into a tree of SchemaNodes. After that, you'll work with individual nodes in the tree. You can also pass an options object to compileSchema to customize how the nodes are created.

`ts
type CompileOptions = {
// set of drafts to use
drafts: Draft[];
// a context to share
remote: SchemaNode;
// if format-validations should create errors. Defaults to true
formatAssertion: boolean | "meta-schema";
// default options for all calls to node.getData()
getDataDefaultOptions?: {
// Add all properties (required and optional) to the generated data
addOptionalProps?: boolean;
// Remove data that does not match input schema. Defaults to false
removeInvalidData?: boolean;
// Set to false to take default values as they are and not extend them. Defaults to true
extendDefaults?: boolean;
// Limits how often a $ref should be followed before aborting. Prevents infinite data-structure. Defaults to 1
recursionLimit?: number;
};
};
`

With this

`ts
import { compileSchema, draft04, draft06, draft07, draft2019, draft2020 } from "json-schema-library";

// only draft07 is used for all JSON schema
compileSchema(mySchema, { drafts: [draft07] });

// the created node will share a context with anotherSchemaNode enabling cross schema ref-resolution
// Note that anotherSchemaNode still uses its own drafts it was compiled with
compileSchema(mySchema, { remote: anotherSchemaNode });

// format validation is disabled
compileSchema(mySchema, { formatAssertion: false });

// for all calls to getData, addOptionalProps is true per default
compileSchema(mySchema, { getDataDefaultOptions: { addOptionalProps: true } });
`

Details on _drafts_ are documented in draft customization.
Details on getDataDefaultOptions are documented in getData.

$3

compileSchema builds a tree where each sub-schema becomes its own SchemaNode. Every node in the tree offers the same set of methods.
For example:

`ts
const root = compileSchema(mySchema);
const rootData = root.getData();
const { node: titleNode } = root.getNode("#/image/title");
const titleData = titleNode?.getData();
`

$3

_json-schema-library_ fully supports all core features of draft versions draft-04, draft-06, draft-07, draft-2019-09 and draft-2020-12. Additionally, most format-validations are supported per default besides the listed format below. You can always override or extend format validation as is documented in draft customization.

SchemaNode methods

addRemoteSchema ·
compileSchema ·
createSchema ·
getChildSelection ·
getData ·
getNode ·
getNodeChild ·
getNodeRef ·
getNodeRoot ·
reduceNode ·
toDataNodes ·
toSchemaNodes ·
validate

$3

addRemoteSchema lets you add additional schemas that can be referenced by an URL using $ref. Use this to combine multiple schemas without changing the actual schema.

Each schema is referenced by their unique $id (since draft-06, previously id). Usually an $id is specified as an url, for example https://mydomain.com/schema/schema-name or with a file extension like https://mydomain.com/schema/schema-name.json.

On a compiled schema

`ts
const schemaNode = compileSchema({
$id: "https://sagold.com/local",
type: "object",
required: ["character"],
properties: {
character: {
$ref: "https://sagold.com/remote"
}
}
});
`

use the exposed method addRemoteSchema to add a remote schema for $ref-resolution:

`ts
schemaNode.addRemoteSchema("https://sagold.com/remote", {
$id: "https://sagold.com/remote",
title: "A character",
type: "string",
minLength: 1,
maxLength: 1
});
`

Note the given _url_ and $id on the root schema should match. If $id is omitted it will be added from the passed url.

To access the remote schema, add a $ref within your local schema and the remote schema will be resolved automatically:

`ts
schemaNode.validate({ character: "AB" }); // maxLength error
schemaNode.getData({}); // { character: "A" } - default value resolved
// returns remote schema (from compiled local schema):
const { node, error } = schemaNode.getNodeRef("https://sagold.com/remote");
`

Note JSON Schema $ref-keyword can become tricky when combined with $ids in sub-schemas. For more details, see json-schema.org: Structuring a complex schema.

$3

node.compileSchema creates a new schema node in the same context as node. With this, the created node will be able to resolve local $ref and remote $ref correctly. Note, the created schema will not be part of (linked) from any nodes in the schema-tree.

`ts
const someNode = node.compileSchema({ prefixItems: [{ type: "string" }, { $ref: "#/$defs/string" }] });
`

$3

createSchema returns a simple JSON Schema for the input data.

`ts
const schemaNode = compileSchema(mySchema);
const schema: JsonSchema = schemaNode.createSchema({ title: "initial value" });
console.log(schema); // { type: "string" }
`

$3

getChildSelection returns a list of available sub-schemas for the given property. In many cases, a single schema will be returned. For _oneOf_-schemas, a list of possible options is returned. This helper always returns a list of schemas.

`ts
const schemaNode = compileSchema(mySchema);
const schemas: SchemaNode[] = schemaNode.getChildSelection("content");
`

$3

getData creates input data from a JSON Schema that is valid to the schema. Where possible, the JSON Schema default property will be used to initially setup input data. Otherwise, the first values encountered (enum values, initial values, etc.) are used to build up the json-data.

`ts
const myData = compileSchema(myJsonSchema).getData();
`

Additionally, you can pass input data. getData will then complement any missing values from the schema, while keeping the initial values.

`ts
const myData = compileSchema(myJsonSchema).getData({ name: "input-data" });
`

Note If you are using references in your schema, getData will only resolve the first _$ref_ in each path, ensuring no infinite data structures are created. In case the limit of 1 _$ref_ resolution is too low, you can modify the value globally one by adjusting the json-schema-library settings:

`ts
const myData = compileSchema(myJsonSchema).getData(inputData, { recursionLimit: 2 });
`

$3

getNode returns the JSON Schema from data location specified by a JSON Pointer. In many cases the JSON Schema can be retrieved without passing any data, but in situations where the schema is dynamic (for example in _oneOf_, _dependencies_, etc.), input-data is required or getNode will return a _JsonError_ as is done when the JSON Pointer path is invalid.

`ts
const { node, error } = compileSchema(mySchema).getNode("/list/1/name", myData);
if (node) console.log(node.schema);
`

Note getNode will return a node=undefined for paths that lead to valid properties, but miss a schema definition. For example:

`ts
const { node, error } = compileSchema({ type: "object" }).getNode("/name");
console.log(node, error); // undefined, undefined
`

In case this is unwanted behaviour, use the withSchemaWarning option to return a json-error with code schema-warning instead:

`ts
const schemaNode = compileSchema({ type: "object" });
const { node, error } = schemaNode.getNode("/name", undefined, { withSchemaWarning: true });
console.log(node?.schema, error); // undefined, { type: "error", code: "schema-warning" }
`

Or set getNode to return a simple JSON Schema for the found data setting createSchema: true:

`ts
const schemaNode = compileSchema({ type: "object" });
const { node, error } = schemaNode.getNode("/name", { name: 123 }, { createSchema: true });
console.log(node?.schema, error); // { type: "number" }, undefined
`

$3

getNodeChild retrieves the SchemaNode of a child property or index. Using get it is possible to incrementally go through the data, retrieving the schema for each next item.

`ts
const mySchema = { type: "object", properties: { title: { type: "string" } } };
const root = compileSchema(mySchema);
const { node } = root.getNodeChild("title", { title: "value" });
if (node == null) return;
console.log(node.schema);
`

$3

getNodeRef retrieves the SchemaNode of a $ref string.

`ts
const root = compileSchema(mySchema);
root.addRemoteSchema("https://remote.com/schema", remoteSchema);

root.getNodeRef("#/$defs/title"); // title-schema of mySchema
root.getNodeRef("https://remote.com/schema"); // remoteSchema
`

$3

getNodeRoot returns the rootNode containing the initial JSON Schema

`ts
const root = compileSchema(mySchema);
const { node } = root.getNode("/image/title");

if (node) {
assert(node.getNodeRoot() === root); // success
}
`

$3

reduceNode compiles dynamic JSON schema keywords of a SchemaNode according to the given data.
This utility helps walking down the schema-tree with a set of data and it helps getting a mostly
complete json-schema for a specific data-value.

`ts
const { node: reducedNode } = compileSchema({
properties: {
trigger: { type: "boolean"}
}
dependentSchemas: {
trigger: {
required: ["title"],
properties: {
title: { type: "string" }
}
}
}
}).reduceNode({ trigger: true });

expect(reducedNode.schema).to.deep.eq({
required: ["title"],
properties: {
trigger: { type: "boolean"},
title: { type: "string" }
}
});
`

> ⚠️ Please be aware that certain schema-definitions are lost when resolving or merging sub-schemas.
> This mainly refers to validation-properties, but also some ambigiuous schema might get overriden.

$3

toDataNodes collects all data-items (_object_, _array_ and _value_) and their SchemaNode and return them as a list of DataNodes:

`ts
type DataNode = { pointer: string; value: unknown; node: SchemaNode };
const schemaNode = compileSchema(mySchema);
const nodes: DataNode[] = schemaNode.toDataNodes(myData);
`

$3

toSchemaNodes collects all sub-schema definitions, like in properties["property"], anyOf[1], contains, $defs["name"], etc. and returns them as a list of SchemaNodes:

`ts
const nodes: SchemaNode[] = compileSchema(mySchema).toSchemaNodes();
`

$3

validate is a complete _JSON Schema validator_ for your input data. Calling _validate_ will return a list of validation errors for the passed data.

`ts
const { valid, errors } = compileSchema(myJsonSchema).validate(myData);
// { valid: boolean, errors: JsonError[] }
`

You can also use async validators to validate data with json-schema. For this, another property asyncErrors is exposed on validate:

`ts
const { errorsAsync } = compileSchema(myJsonSchema).validate(myData);

if (errorsAsync.length > 0) {
const additionalErrors = (await Promise.all(errorsAsync)).filter((err) => err != null);
}
`

Per default _json-schema-library_ does not contain async validators, so errorsAsync is always empty. If you add async validators, a list of Promise is return and you need to resolve and filter non-errors (undefined) yourself.

> Note isValid only refers to errors. errorsAsync has to be evaluated separately

Draft Customization

Extending a Draft · Keyword

_json-schema-library_ uses the concept of drafts to support different versions of the JSON Schema specification — such as Draft 04, Draft 07, or 2020-12 — and to allow customization of schema behavior.

Each draft describes how a schema should be parsed, validated, and interpreted. Drafts can also be extended or modified to change or enhance behavior, such as:

- Replacing or adding new keywords (oneOf, if/then, custom ones, etc.)
- Defining or overriding format validators (format: "email", etc.)
- Customizing or localizing error messages
- Tweaking how schema nodes behave during parsing or resolution

Out of the box, the library exports all compliant JSON Schema drafts:

`ts
import { draft04, draft06, draft07, draft2019, draft2020 } from "json-schema-library";
`

When you compile a schema, the library will automatically select the correct draft based on the $schema field — or fall back to the last draft in the list:

`ts
compileSchema(schema, { drafts: [draft04, draft07, draft2020] });
`

A Draft is an object that defines the core behavior and extensions for a schema. It includes:

`ts
type Draft = {
$schemaRegEx: string;
version: DraftVersion;
keywords: Keyword[];
errors: ErrorConfig;
formats: typeof formats;
methods: {};
};
`

Here’s a breakdown of what each piece does:

$schemaRegEx

A regex string that identifies whether a draft should be used for a given schema, based on the $schema property. For example:

`ts
draft.$schemaRegEx === "draft[-/]2020-12";
// matches "$schema": "https://json-schema.org/draft/2020-12/schema"
// matches "$schema": "draft-2020-12"
`

When compiling, drafts are matched from left to right — the first one that matches is used. If no match is found, the last draft in the list is used as a fallback. If you're only using one draft, the $schemaRegEx check is skipped.

version

Describes the draft version (e.g., "2020-12"). This is mostly used for debugging and logging.

keywords

A list of keyword handlers for that draft, such as properties, allOf, oneOf, $ref, and more. Each keyword defines how the library should parse and validate that keyword. You can override, extend, or remove any keyword.
Learn more in Keyword.

errors

An object mapping error types to either template strings or error functions. These can be used to customize error messages globally or define more intelligent error generation logic.

formats

An object mapping format names (like "email", "uuid", "date-time") to custom validation functions. You can override or add formats depending on your needs.

methods

Draft-specific implementations for certain core behaviors in SchemaNode, such as how child schemas are selected or how schemas are converted to data nodes. These can be overridden in custom drafts if needed.

$3

You may want to extend a draft when the default JSON Schema behavior does not fit your needs. Whether you want to add new keywords, modify error messages, or define custom formats for your validation, extendDraft helps you adjust the draft version to meet your specific requirements.

Examples:

`ts
import { extendDraft, draft2020, oneOfFuzzyKeyword, createCustomError, render, ErrorData } from "json-schema-library";

const myDraft = extendDraft(draft2020, {
// Match all $schema
$schemaRegEx: "",

// Register a custom "oneOf" keyword, replacing the existing one
keywords: [oneOfFuzzyKeyword],

formats: {
// Add a new "format": "test", which returns an error when the value is "test"
test: ({ data, node, pointer }) => {
if (data === "test") {
return node.createError("test-error", {
schema: node.schema,
pointer: pointer,
value: data,
customValue: "test"
});
}
}
},

errors: {
// Add a new custom error "test-error"
"test-error": "Test error for value {{value}} - {{customValue}}",

// Overwrite the existing MaxLengthError message
"max-length-error": "Too many characters",

// Add a dynamic MinLengthError with custom logic
"min-length-error": (data: ErrorData) => {
if (data.minLength === 1) {
return {
type: "error",
code: "min-length-one-error",
message: "Input is required",
data
};
}
return {
type: "error",
code: "min-length-error",
message: render("Value in {{pointer}} is {{length}}, but should be {{minimum}} at minimum", data),
data
};
}
}
});
`

$3

The built-in format validators may not always align with your specific requirements. For instance, you might need to validate the output of an , which produces values in formats like HH:MM or HH:MM:SS. In such cases, you can customize or overwrite the format validators to suit your needs using extendDraft

$3

Keywords hold the main logic for JSON Schema functionality. Each Keyword corresponds to a JSON Schema keyword like properties, prefixItems, oneOf, etc and offers implementations to parse, validate, resolve and reduce. Note that support for each implementation is optional, dependending on the feature requirements. The main properties of a Keyword:

- a Keyword is only processed if the specified keyword is available as property on the JSON Schema
- an optional order property may be added as order of keyword execution is sometimes important (additionalItems last, $ref evaluation first)
- the list of keywords is unique by property-value keyword
- for a given function addX, a function X must be present

`ts
type Keyword = {
id: string;
keyword: string;
order?: number;
parse?: (node: SchemaNode) => void;
addResolve?: (node: SchemaNode) => boolean;
resolve?: JsonSchemaResolver;
addValidate?: (node: SchemaNode) => boolean;
validate?: JsonSchemaValidator;
addReduce?: (node: SchemaNode) => boolean;
reduce?: JsonSchemaReducer;
};
`

For examples on keyword implementations refer to ./src/keywords.

parse

parse will be executed on compile time, usually to add a compiled sub-schema on the parent-node.

resolve

A resolver returns a child-schema for a property-key, item-index or undefined if the key does not apply.

reduce

A reducer replaces the JSON Schema keyword to a simple, static JSON Schema based on the current data

Currently keywords are not exposed per default. You can still access any keyword implementation by retrieving them from a draft:

`ts
import { draft07 } from "json-schema-library";
const dependentSchemasKeyword = draft2020.keywords.find((f) => f.keyword === "dependentSchemas");
`

Keyword extensions

$3

For oneOf resolution, JSON Schema states that data is valid if it validates against exactly one of those sub-schemas. In some scenarios this is unwanted behaviour, as the actual oneOf schema is known and only validation errors of this exact sub-schema should be returned.

For an explicit oneOf resolution, the JSON Schema may be extended by a property oneOfProperty. This will always associate an entry with a matching value (instead of schema validation) and return only this schema or validation errors, depending on the current task. For example:

`ts
const schema = {
oneOfProperty: "id",
oneOf: [
{
type: "object",
properties: { id: { const: "1" }, title: { type: "number" } }
},
{
type: "object",
properties: { id: { const: "2" }, title: { type: "number" } }
},
{
type: "object",
properties: { id: { const: "3" }, title: { type: "number" } }
}
]
};

const resolvedNode = compileSchema(schema).reduce({ id: "2", title: "not a number" });

// will always return (even if invalid)
expect(resolvedNode?.schema).to.deep.eq({
type: "object",
properties: { id: { const: "2" }, title: { type: "number" } }
});
`

$3

If you're working with complex schemas that use the oneOf keyword to validate multiple options, oneOfFuzzyKeyword offers an alternative approach. It scores the schemas to return the best match, even if none of the schemas fully validate the input data. This makes error messages more readable and helps identify the most appropriate schema when multiple options exist.

oneOfFuzzyKeyword helps when no schema fully validates the data but you want to prioritize schemas based on how well they fit the input. This makes it easier to interpret validation results for complex conditions.

oneOfFuzzyKeyword is exposed by _json-schema-library_ and can be used to extend any draft.

`ts
import { extendDraft, oneOfFuzzyKeyword, draft2020 } from "json-schema-library";

const myDraft = extendDraft(draft2020, {
keywords: [oneOfFuzzyKeyword]
});
`

$3

You can set custom errors messages locally by using the errors-keyword:

`ts
const { errors } = compileSchema({
type: "array",
minItems: 2,
errorMessages: {
"min-items-error": "Custom error {{minItems}}"
}
}).validate([1]);

assert.deepEqual(errors[0].message, "Custom error 2");
`

$3

In order to allow customization of regular expressions a schema property regexFlags is supported for pattern, patternProperties and format regex:

`ts
const { errors } = compileSchema({
type: "string",
pattern: "^[a-zA-Z0-9+_.-]+",
regexFlags: "v"
}).validate("-");
`

Per default, a regexFlags "u" is used. To change this setting globally, change REGEX_FLAGS in settings:

`ts
import settings from "json-schema-library";

settings.REGEX_FLAGS = "v";
`

Breaking Changes

$3

- added support for ref resolution in getSchemaType

$3

- introduced esm module export
- fixed typo in argument ~~disableRecusionLimit~~ disableRecursionLimit
- added settings to exports for global changes of settings

$3

- introduce setting REGEX_FLAGS and schema-property regexFlags to customize regex flags to use evaluating regex
- fixed an issue resolving non-URI compatible $ref-targets containing #/definitions

$3

- introduce getData setting useTypeDefaults
- introduce support to merge meta-properties using $ref-resolution

$3

- replaced node.additionalItems by node.items for drafts below 2020-12
- fixed additionalItems behaviour to be ignored when schema.items is not an array

$3

> This update involves some significant changes in how you work with the library, so please carefully review the migration guide and adjust your implementation accordingly.

In version v10.0.0, we've made significant changes to the library’s API, particularly in how we handle drafts and schemas. These changes are required to support features like dynamicAnchor, unevaluatedItems, and oneOfIndex and to integrate with the headless-json-editor. The previous approach of directly working with JSON schema objects lacked the flexibility needed for more advanced features and extensibility.

The new implementation revolves around compiling schemas into a SchemaNode tree. This change offers a more fitting, simpler, and extensible approach to working with JSON schemas.

#### Key Changes:

- Compile Schema: The compileSchema function now replaces the previous Draft-Class approach.
- SchemaNode Representation: All schemas are now represented as SchemaNode, which holds the schema and provides an easier way to work with them.

#### Breaking Changes:

compileSchema is now a standalone function and replaces the Draft class. All return values for JSON Schema are now SchemaNode objects that contain a schema property.

`ts
// PREVIOUSLY
const draft = new Draft(schema);

// NOW
const node = compileSchema(schema);
`

Changed Methods:

- draft.createSchemaOf(schema) → node.createSchema(schema)
- draft.each(data, callback) → const nodes = node.toDataNodes(data)
- draft.eachSchema(callback) → const nodes = node.toSchemaNodes()
- draft.getChildSchemaSelection(property) → node.getChildSelection(property)
- draft.getNode(options) → node.getNode(pointer, data, options)
- draft.getTemplate(inputData) → node.getData(inputData)
- draft.isValid(data) → node.validate(data).valid
- draft.step(property, data) → node.getNodeChild(property, data)

Renamed Properties: templateDefaultOptions → getDataDefaultOptions

Draft Customization: Customizing drafts has changed completely. The previous methods of extending drafts are no longer valid, and draft handling is now centered around SchemaNode.

Removed Error Property name: Error property name has been removed from JsonError in favor of code.

Removed Configuration Option: The templateDefaultOptions property has been removed from the global settings object. You should now configure it using the compileSchema options:

`ts
compileSchema(schema, {
getDataDefaultOptions: {
addOptionalProps: false,
removeInvalidData: false,
extendDefaults: true
}
});
`

Changed remote $id support in addRemoteSchema. An $id has to be a valid url (previously any value was accepted)

$3

breaking changes:

- _getSchema_ signature changed in favour of an options object. Instead of draft.getNode(pointer, data) arguments have to be passed as an object draft.getNode({ pointer, data }). This removes setting unwanted optional arguments and keeps the api more stable in the future (e.g. withSchemaWarning option)
- _JsonError_ now must expose pointer, schema and value consistently on data property

updates

- _getSchema_ consistently returns errors and can return errors for empty schema using withSchemaWarning option

$3

With version v8.0.0, _getData_ was improved to better support optional properties and utilize existing core logic, making it more reliable. Breaking changes:

- Renamed JSONError to JsonError and JSONSchema to JsonSchema
- getData only adds required properties. Behaviour can be changed by getData default options
- Internal schema property oneOfSchema has been replaced by schema.getOneOfOrigin()
- Changed unique-items-error to point to error for duplicated item and changed data-properties
- Removed SchemaService as it was no longer used nor tested