Reference
Structuring a complex schema
When writing computer programs of even moderate complexity, it's commonly accepted that "structuring" the program into reusable functions is better than copying-and-pasting duplicate bits of code everywhere they are used. Likewise in JSON Schema, for anything but the most trivial schema, it's really useful to structure the schema into parts that can be reused in a number of places. This chapter will present the tools available for reusing and structuring schemas as well as some practical examples that use those tools.
Schema Identification
Like any other code, schemas are easier to maintain if they can be broken down into logical units that reference each other as necessary. In order to reference a schema, we need a way to identify a schema. Schema documents are identified by non-relative URIs.
Schema documents are not required to have an identifier, but you will need one if you want to reference one schema from another. In this documentation, we will refer to schemas with no identifier as "anonymous schemas".
In the following sections we will see how the "identifier" for a schema is determined.
URI terminology can sometimes be unintuitive. In this document, the following definitions are used.
- URI
[1] or
non-relative URI: A full URI containing a scheme (
https
). It may contain a URI fragment (#foo
). Sometimes this document will use "non-relative URI" to make it extra clear that relative URIs are not allowed. - relative reference
[2]:
A partial URI that does not contain a scheme (
https
). It may contain a fragment (#foo
). - URI-reference
[3]:
A relative reference or non-relative URI. It may contain a URI
fragment (
#foo
). - absolute URI
[4] A
full URI containing a scheme (
https
) but not a URI fragment (#foo
).
Even though schemas are identified by URIs, those identifiers are not
necessarily network-addressable. They are just identifiers. Generally,
implementations don't make HTTP requests (https://
) or read from the
file system (file://
) to fetch schemas. Instead, they provide a way to
load schemas into an internal schema database. When a schema is
referenced by it's URI identifier, the schema is retrieved from the
internal schema database.
Base URI
Using non-relative URIs can be cumbersome, so any URIs used in JSON Schema can be URI-references that resolve against the schema's base URI resulting in a non-relative URI. This section describes how a schema's base URI is determined.
Base URI determination and relative reference resolution is defined by RFC-3986. If you are familiar with how this works in HTML, this section should feel very familiar.
Retrieval URI
The URI used to fetch a schema is known as the "retrieval URI". It's often possible to pass an anonymous schema to an implementation in which case that schema would have no retrieval URI.
Let's assume a schema is referenced using the URI
https://example.com/schemas/address
and the following schema is
retrieved.
The base URI for this schema is the same as the retrieval URI, https://example.com/schemas/address
.
$id
You can set the base URI by using the $id
keyword at the root of the
schema. The value of $id
is a URI-reference without a fragment that
resolves against the retrieval-uri. The resulting URI is
the base URI for the schema.
$id
is just id
(without the dollar sign).This is analogous to the <base>
tag in HTML.
When the $id
keyword appears in a subschema, it means something slightly different. See the bundling section for more.
Let's assume the URIs https://example.com/schema/address
and
https://example.com/schema/billing-address
both identify the following
schema.
"type": "object", "properties": { "street_address": { "type": "string" }, "city": { "type": "string" }, "state": { "type": "string" } }, "required": ["street_address", "city", "state"]}
No matter which of the two URIs is used to retrieve this schema, the
base URI will be https://example.com/schemas/address
, which is the
result of the $id
URI-reference resolving against the
Retrieval URI.
However, using a relative reference when setting a base URI can be
problematic. For example, we couldn't use this schema as an anonymous
schema because there would be no Retrieval URI and you
can't resolve a relative reference against nothing. For this and other
reasons, it's recommended that you always use an absolute URI when
declaring a base URI with $id
.
The base URI of the following schema will always be
https://example.com/schemas/address
no matter what the
Retrieval URI was or if it's used as an anonymous schema.
"type": "object", "properties": { "street_address": { "type": "string" }, "city": { "type": "string" }, "state": { "type": "string" } }, "required": ["street_address", "city", "state"]}
JSON Pointer
In addition to identifying a schema document, you can also identify subschemas. The most common way to do that is to use a JSON Pointer in the URI fragment that points to the subschema.
A JSON Pointer describes a slash-separated path to traverse the keys in
the objects in the document. Therefore, /properties/street_address
means:
- 1) find the value of the key
properties
- 2) within that object, find the value of the key
street_address
The URI https://example.com/schemas/address#/properties/street_address
identifies the highlighted subschema in the following schema.
$anchor
A less common way to identify a subschema is to create a named anchor in
the schema using the $anchor
keyword and using that name in the URI
fragment. Anchors must start with a letter followed by any number of
letters, digits, -
, _
, :
, or .
.
In Draft 4, you declare an anchor the same way you do in Draft 6-7
except that $id
is just id
(without the dollar sign).
If a named anchor is defined that doesn't follow these naming rules, then behavior is undefined. Your anchors might work in some implementation, but not others.
The URI https://example.com/schemas/address#street_address
identifies
the subschema on the highlighted part of the following schema.
$ref
A schema can reference another schema using the $ref
keyword. The
value of $ref
is a URI-reference that is resolved against the
schema's Base URI. When evaluating a $ref
, an
implementation uses the resolved identifier to retrieve the referenced
schema and applies that schema to the instance.
$ref
behaves a little differently. When an
object contains a $ref
property, the object is considered a
reference, not a schema. Therefore, any other properties you put
in that object will not be treated as JSON Schema keywords and will
be ignored by the validator. $ref
can only be used where a
schema is expected.
For this example, let's say we want to define a customer record, where each customer may have both a shipping and a billing address. Addresses are always the same — they have a street address, city and state — so we don't want to duplicate that part of the schema everywhere we want to store an address. Not only would that make the schema more verbose, but it makes updating it in the future more difficult. If our imaginary company were to start doing international business in the future and we wanted to add a country field to all the addresses, it would be better to do this in a single place rather than everywhere that addresses are used.
"type": "object", "properties": { "first_name": { "type": "string" }, "last_name": { "type": "string" }, "shipping_address": { "$ref": "/schemas/address" }, "billing_address": { "$ref": "/schemas/address" } }, "required": ["first_name", "last_name", "shipping_address", "billing_address"]}
The URI-references in $ref
resolve against the schema's
Base URI (https://example.com/schemas/customer
) which
results in https://example.com/schemas/address
. The implementation
retrieves that schema and uses it to evaluate the "shipping_address"
and "billing_address" properties.
When using $ref
in an anonymous schema, relative references may not be
resolvable. Let's assume this example is used as an anonymous schema
The $ref
at /properties/shipping_address
can resolve just fine
without a non-relative base URI to resolve against, but the $ref
at
/properties/billing_address
can't resolve to a non-relative URI and
therefore can't be used to retrieve the address schema.
$defs
Sometimes we have small subschemas that are only intended for use in the
current schema and it doesn't make sense to define them as separate
schemas. Although we can identify any subschema using JSON Pointers or
named anchors, the $defs
keyword gives us a standardized place to keep
subschemas intended for reuse in the current schema document.
Let's extend the previous customer schema example to use a common
schema for the name properties. It doesn't make sense to define a new
schema for this and it will only be used in this schema, so it's a good
candidate for using $defs
.
"type": "object", "properties": { "first_name": { "$ref": "#/$defs/name" }, "last_name": { "$ref": "#/$defs/name" }, "shipping_address": { "$ref": "/schemas/address" }, "billing_address": { "$ref": "/schemas/address" } }, "required": ["first_name", "last_name", "shipping_address", "billing_address"],
"$defs": { "name": { "type": "string" } }}
$defs
isn't just good for avoiding duplication. It can also be useful
for writing schemas that are easier to read and maintain. Complex parts
of the schema can be defined in $defs
with descriptive names and
referenced where it's needed. This allows readers of the schema to more
quickly and easily understand the schema at a high level before diving
into the more complex parts.
It's possible to reference an external subschema, but generally you
want to limit a $ref
to referencing either an external schema or an
internal subschema defined in $defs
.
Recursion
The $ref
keyword may be used to create recursive schemas that refer to
themselves. For example, you might have a person
schema that has an
array of children
, each of which are also person
instances.
A snippet of the British royal family tree
Above, we created a schema that refers to itself, effectively creating a
"loop" in the validator, which is both allowed and useful. Note,
however, that a $ref
referring to another $ref
could cause an
infinite loop in the resolver, and is explicitly disallowed.
Extending Recursive Schemas
Documentation Coming Soon
Bundling
Working with multiple schema documents is convenient for development,
but it's often more convenient for distribution to bundle all of your
schemas into a single schema document. This can be done using the $id
keyword in a subschema. When $id
is used in a subschema, it indicates
an embedded schema. The identifier for the embedded schema is the value
of $id
resolved against the Base URI of the schema it
appears in. A schema document that includes embedded schemas is called a
Compound Schema Document. Each schema with an $id
in a Compound Schema
Document is called a Schema Resource.
$id
is just id
(without the dollar sign).This is analogous to the <iframe>
tag in HTML.
It is unusual to use embedded schemas when developing schemas. It's generally best not to use this feature explicitly and use schema bundling tools to construct bundled schemas if such a thing is needed. :::
This example shows the customer schema example and the address schema example bundled into a Compound Schema Document.
"type": "object", "properties": { "first_name": { "type": "string" }, "last_name": { "type": "string" }, "shipping_address": { "$ref": "/schemas/address" }, "billing_address": { "$ref": "/schemas/address" } }, "required": ["first_name", "last_name", "shipping_address", "billing_address"],
"$defs": { "address": { "$id": "https://example.com/schemas/address", "$schema": "http://json-schema.org/draft-07/schema#",
"type": "object", "properties": { "street_address": { "type": "string" }, "city": { "type": "string" }, "state": { "$ref": "#/definitions/state" } }, "required": ["street_address", "city", "state"],
"definitions": { "state": { "enum": ["CA", "NY", "... etc ..."] } } } }}
All references in a Compound Schema Document need to be the same whether
the Schema Resources are bundled or not. Notice that the $ref
keywords
from the customer schema have not changed. The only difference is that
the address schema is now defined at /$defs/address
instead of a
separate schema document. You couldn't use #/$defs/address
to
reference the address schema because if you unbundled the schema, that
reference would no longer point to the address schema.
$id
only represented a base URI change, not an embedded schema.
However, even though it's allowed, it's still highly recommended
that JSON Pointers don't cross a schema with a base URI change.
You should also see that "$ref": "#/definitions/state"
resolves to the
definitions
keyword in the address schema rather than the one at the
top level schema like it would if the embedded schema wasn't used.
Each Schema Resource is evaluated independently and may use different
JSON Schema dialects. The example above has the address Schema Resource
using Draft 7 while the customer Schema Resource uses Draft 2020-12. If
no $schema
is declared in an embedded schema, it defaults to using the
dialect of the parent schema.
$id
is just a base URI change and not
considered an independent Schema Resource. Because $schema
is
only allowed at the root of a Schema Resource, all schemas bundled
using subschema $id
must use the same dialect.
$schema
with a different value than the parent schema) was added.Need Help?
Did you find these docs helpful?
Help us make our docs great!
At JSON Schema, we value docs contributions as much as every other type of contribution!
Still Need Help?
Learning JSON Schema is often confusing, but don't worry, we are here to help!.