Skip to content

Commit 6274757

Browse files
committed
Additional cleanup for dynamic references
1 parent 5660c17 commit 6274757

File tree

1 file changed

+88
-96
lines changed

1 file changed

+88
-96
lines changed

specs/jsonschema-core.md

+88-96
Original file line numberDiff line numberDiff line change
@@ -240,7 +240,7 @@ A schema that itself describes a schema is called a meta-schema. Meta-schemas
240240
are used to validate JSON Schemas and specify the set of keywords those schemas
241241
are using.
242242

243-
#### Root Schema and Subschemas and Resources {#root}
243+
#### Root Schema, Subschemas, and Resources {#root}
244244

245245
A JSON Schema resource is a schema which is
246246
[canonically](https://www.rfc-editor.org/info/rfc6596) identified by an
@@ -334,9 +334,8 @@ NCNameChar = NCNameStartChar / "-" / "." / DIGIT
334334
All fragment identifiers that do not match the JSON Pointer syntax MUST be
335335
interpreted as plain name fragment identifiers.
336336

337-
Defining and referencing a plain name fragment identifier within an
338-
`application/schema+json` document are specified in the [`$anchor`
339-
keyword](#anchors) section.
337+
Defining a plain name fragment identifier within an `application/schema+json`
338+
document is specified in the [`$anchor` keyword](#anchors) section.
340339

341340
## General Considerations
342341

@@ -950,40 +949,32 @@ an [absolute IRI](https://www.rfc-editor.org/rfc/rfc3987.html#section-2.2)
950949

951950
#### Defining location-independent identifiers {#anchors}
952951

953-
Using JSON Pointer fragments requires knowledge of the structure of the schema.
954-
When writing schema documents with the intention to provide re-usable schemas,
955-
it may be preferable to use a plain name fragment that is not tied to any
956-
particular structural location. This allows a subschema to be relocated without
957-
requiring JSON Pointer references to be updated.
958-
959-
The `$anchor` and `$dynamicAnchor` keywords are used to specify such fragments.
960-
They are identifier keywords that can only be used to create plain name
961-
fragments, rather than absolute IRIs as seen with `$id`.
962-
963-
`$anchor` defines a reference target for `$ref`. The fragment defined by this
964-
keyword is appended to the IRI of the schema resource containing it. As
965-
discussed in {{id-keyword}}, this is either the nearest `$id` in the same or an
966-
ancestor schema object, or the base IRI for the document as determined according
967-
to [RFC 3987][rfc3987] and
968-
[RFC 3986][rfc3986].
969-
970-
In contrast, `$dynamicAnchor` operates independently of resource IRIs and is
971-
instead dependent on the dynamic scope of the evaluation. `$dynamicAnchor`
972-
defines a reference target for the `$dynamicRef` keyword. This advanced feature
973-
makes it easier to extend recursive schemas such as the meta-schemas, without
974-
imposing any particular semantics on that extension. See {{dynamic-ref}} for
975-
details.
976-
977-
In most cases, the normal fragment behavior both suffices and is more intuitive.
978-
Therefore it is RECOMMENDED that `$anchor` be used to create plain name
979-
fragments unless there is a clear need for `$dynamicAnchor`.
952+
Using JSON Pointers in IRI fragments to reference subschemas couples the IRI to
953+
the structure of the schema. Using plain name fragment identifiers in IRI
954+
fragments to identify subschemas is sometimes preferable because it is not tied
955+
to a particular structural location. This allows a subschema to be relocated
956+
without requiring references to be updated.
980957

981-
If present, the value of these keywords MUST be a string and MUST conform to the
982-
plain name fragment identifier syntax defined in {{fragments}}.[^4]
958+
The `$anchor` and `$dynamicAnchor` keywords are used to define
959+
location-independent identifiers for subschemas within a schema resource.
960+
961+
`$anchor` defines a plain name fragment identifier that can be used in IRI
962+
fragments as an alternative to JSON Pointers.[^4] See {{fragments}}.
983963

984964
[^4]: Note that the anchor string does not include the "#" character, as it is
985-
not a IRI reference. An `$anchor`: "foo" becomes the fragment `#foo` when used
986-
in a IRI. See below for full examples.
965+
just a fragment identifier not an IRI reference. To reference the "foo"
966+
`$anchor` from the same Schema Resource, you would use the fragment-only IRI
967+
`#foo`. See below for full examples.
968+
969+
`$dynamicAnchor` defines a different kind of identifier that only has meaning
970+
when used as a value for `$dynamicRef`. It's not a fragment identifier that can
971+
be used in an IRI. `$dynamicRef` uses a syntax similar to a fragment-only IRI,
972+
but the semantics are different. Unlike IRIs, the fragment identifies the
973+
primary resource (the schema resource) in addition to the secondary resource
974+
(the subschema within that schema resource). See {{dynamic-ref}} for details.
975+
976+
If present, the value of these keywords MUST be a string and MUST conform to the
977+
plain name fragment identifier syntax defined in {{fragments}}.
987978

988979
#### Duplicate schema identifiers {#duplicate-iris}
989980

@@ -1002,12 +993,13 @@ applicators, applying the referenced schema to the instance.
1002993
##### Direct References with `$ref` {#ref}
1003994

1004995
The `$ref` keyword is an applicator that is used to reference a statically
1005-
identified schema. Its results are the results of the referenced schema.[^5]
996+
identified schema. Its results are the results of evaluating the referenced
997+
schema.[^5]
1006998

1007999
[^5]: Note that this definition of how the results are determined means that
10081000
other keywords can appear alongside of `$ref` in the same schema object.
10091001

1010-
The value of the `$ref` keyword MUST be a string which is a IRI reference.
1002+
The value of the `$ref` keyword MUST be a string which is an IRI reference.
10111003
Resolved against the current IRI base, it produces the IRI of the schema to
10121004
apply. This resolution is safe to perform on schema load, as the process of
10131005
evaluating an instance cannot change how the reference resolves.
@@ -1019,28 +1011,25 @@ default to operating offline.
10191011

10201012
##### Dynamic References with `$dynamicRef` {#dynamic-ref}
10211013

1022-
The `$dynamicRef` keyword is an applicator that allows for deferring the full
1023-
resolution until runtime, at which point it is resolved each time it is
1024-
encountered while evaluating an instance.
1025-
1026-
Together with `$dynamicAnchor`, `$dynamicRef` implements a cooperative extension
1027-
mechanism that is primarily useful to to create open schemas, where
1028-
`$dynamicRef` defines the extension point and `$dynamicAnchor` defines the
1029-
target.
1014+
The `$dynamicRef` keyword is an applicator that is used when the referencing
1015+
schema might need to override where a reference in the referenced schema will
1016+
resolve. This is useful for authoring a recursive schema that can be extended or
1017+
a generic schema such as a list whose items are defined by the referencing
1018+
schema.
10301019

10311020
The value of the `$dynamicRef` property MUST be formatted as a valid
1032-
[IRI plain name fragment](#fragments).[^3]
1021+
[fragment-only IRI](#fragments).[^3]
10331022

1034-
[^3]: `$dynamicAnchor` defines the anchor with plain text, e.g. `foo`. Although
1035-
the value of `$dynamicRef` is not an IRI fragment, for historical reasons, the
1036-
value still uses an IRI fragment syntax, e.g. `#foo`.
1023+
[^3]: `$dynamicAnchor` defines the anchor with plain text, e.g. `foo`. Although,
1024+
for historical reasons, the value of `$dynamicRef` still uses a fragment-only
1025+
IRI syntax, e.g. `#foo`.
10371026

1038-
Resolution of `$dynamicRef` begins by identifying the outermost schema
1039-
resource in the [dynamic scope](#scopes) which defines a matching
1040-
`$dynamicAnchor`. The schema to apply is the subschema of this resource which
1041-
contains the matching `$dynamicAnchor`.
1027+
Resolution of `$dynamicRef` begins by identifying the outermost schema resource
1028+
in the [dynamic scope](#scopes) which defines a matching `$dynamicAnchor`. The
1029+
schema to apply is the subschema of this resource which contains the matching
1030+
`$dynamicAnchor`. If no matching `$dynamicAnchor` is found, see {{failed-refs}}.
10421031

1043-
For a full example using these keywords, see {{recursive-example}}.[^6]
1032+
For a full example using these keywords, see {{dynamic-example}}.[^6]
10441033

10451034
[^6]: The differences in the hyper-schema meta-schemas from draft-07 and draft
10461035
2019-09 dramatically demonstrates the utility of these keywords.
@@ -1205,19 +1194,19 @@ If an implementation has been configured to resolve that identifier to a schema
12051194
via pre-loading or other means, it can be used automatically; otherwise, the
12061195
behavior described in {{failed-refs}} MUST be used.
12071196

1208-
#### JSON Pointer fragments and embedded schema resources {#embedded}
1197+
#### JSON Pointer fragment identifiers and embedded schema resources {#embedded}
12091198

1210-
Since JSON Pointer IRI fragments are constructed based on the structure of the
1211-
schema document, an embedded schema resource and its subschemas can be
1212-
identified by JSON Pointer fragments relative to either its own canonical IRI,
1213-
or relative to any containing resource's IRI.
1199+
Since JSON Pointer fragment identifiers are based on the structure of the schema
1200+
document, an embedded schema resource and its subschemas can be identified using
1201+
JSON Pointer IRI fragments relative to either its own IRI, or relative to any
1202+
containing resource's IRI.
12141203

12151204
Conceptually, a set of linked schema resources should behave identically whether
12161205
each resource is a separate document connected with [schema
12171206
references](#referenced), or is structured as a single document with one or more
12181207
schema resources embedded as subschemas.
12191208

1220-
Since IRIs involving JSON Pointer fragments relative to the parent schema
1209+
Since IRIs with JSON Pointer fragments relative to the parent schema
12211210
resource's IRI cease to be valid when the embedded schema is moved to a separate
12221211
document and referenced, applications and schemas SHOULD NOT use such IRIs to
12231212
identify embedded schema resources or locations within them.
@@ -1244,7 +1233,7 @@ For the `additionalProperties` schema within that embedded resource, the IRI
12441233
object, but that object's IRI relative to its resource's canonical IRI is
12451234
`https://example.com/bar#/additionalProperties`.
12461235

1247-
Now consider the following two schema resources linked by reference using a IRI
1236+
Now consider the following two schema resources linked by reference using an IRI
12481237
value for `$ref`:
12491238

12501239
```jsonschema
@@ -1264,10 +1253,11 @@ value for `$ref`:
12641253
```
12651254

12661255
Here we see that `https://example.com/bar#/additionalProperties`, using a JSON
1267-
Pointer fragment appended to the canonical IRI of the "bar" schema resource, is
1268-
still valid, while `https://example.com/foo#/items/additionalProperties`, which
1269-
relied on a JSON Pointer fragment appended to the canonical IRI of the "foo"
1270-
schema resource, no longer resolves to anything.
1256+
Pointer fragment identifier appended to the canonical IRI of the "bar" schema
1257+
resource, is still valid, while
1258+
`https://example.com/foo#/items/additionalProperties`, which relied on a JSON
1259+
Pointer fragment identifier appended to the canonical IRI of the "foo" schema
1260+
resource, no longer resolves to anything.
12711261

12721262
Note also that `https://example.com/foo#/items` is valid in both arrangements,
12731263
but resolves to a different value. This IRI ends up functioning similarly to a
@@ -1282,14 +1272,15 @@ undefined. Schema authors SHOULD NOT rely on such IRIs, as using them may
12821272
reduce interoperability.[^8]
12831273

12841274
[^8]: This is to avoid requiring implementations to keep track of a whole stack
1285-
of possible base IRIs and JSON Pointer fragments for each, given that all but
1286-
one will be fragile if the schema resources are reorganized. Some have argued
1287-
that this is easy so there is no point in forbidding it, while others have
1288-
argued that it complicates schema identification and should be forbidden.
1289-
Feedback on this topic is encouraged. After some discussion, we feel that we
1290-
need to remove the use of "canonical" in favour of talking about JSON Pointers
1291-
which reference across schema resource boundaries as undefined or even forbidden
1292-
behavior (<https://github.com/json-schema-org/json-schema-spec/issues/937>,
1275+
of possible base IRIs and JSON Pointer fragment identifiers for each, given
1276+
that all but one will be fragile if the schema resources are reorganized. Some
1277+
have argued that this is easy so there is no point in forbidding it, while
1278+
others have argued that it complicates schema identification and should be
1279+
forbidden. Feedback on this topic is encouraged. After some discussion, we feel
1280+
that we need to remove the use of "canonical" in favour of talking about JSON
1281+
Pointers which reference across schema resource boundaries as undefined or even
1282+
forbidden behavior
1283+
(<https://github.com/json-schema-org/json-schema-spec/issues/937>,
12931284
<https://github.com/json-schema-org/json-schema-spec/issues/1183>)
12941285

12951286
Further examples of such non-canonical IRI construction, as well as the
@@ -1578,9 +1569,9 @@ subschema, then validation succeeds against this keyword if the instance also
15781569
successfully validates against this keyword's subschema.
15791570

15801571
This keyword has no effect when `if` is absent, or when the instance fails to
1581-
validate against the `if` subschema. Implementations MUST NOT evaluate the instance
1582-
against this keyword, for either validation or annotation collection purposes,
1583-
in such cases.
1572+
validate against the `if` subschema. Implementations MUST NOT evaluate the
1573+
instance against this keyword, for either validation or annotation collection
1574+
purposes, in such cases.
15841575

15851576
##### `else`
15861577

@@ -1591,8 +1582,8 @@ then validation succeeds against this keyword if the instance successfully
15911582
validates against this keyword's subschema.
15921583

15931584
This keyword has no effect when `if` is absent, or when the instance
1594-
successfully validates against the `if` subschema. Implementations MUST NOT evaluate
1595-
the instance against this keyword, for either validation or annotation
1585+
successfully validates against the `if` subschema. Implementations MUST NOT
1586+
evaluate the instance against this keyword, for either validation or annotation
15961587
collection purposes, in such cases.
15971588

15981589
##### `dependentSchemas` {#dependent-schemas}
@@ -1856,8 +1847,8 @@ Omitting this keyword has the same assertion behavior as an empty schema.
18561847

18571848
The value of `unevaluatedProperties` MUST be a valid JSON Schema.
18581849

1859-
This keyword applies to object instances by applying its subschema to the object's
1860-
property values.
1850+
This keyword applies to object instances by applying its subschema to the
1851+
object's property values.
18611852

18621853
The behavior of this keyword depends on all adjacent keywords as well as
18631854
keywords in successfully validated subschemas that apply to the same instance
@@ -1872,9 +1863,9 @@ subschema validates against all applicable property values.
18721863
The annotation result of this keyword is the set of instance property names
18731864
validated by this keyword's subschema.
18741865

1875-
The presence of this keyword affects the behavior of other `unevaluatedProperties`
1876-
keywords found earlier in the dynamic scope that apply to the same instance
1877-
location.
1866+
The presence of this keyword affects the behavior of other
1867+
`unevaluatedProperties` keywords found earlier in the dynamic scope that apply
1868+
to the same instance location.
18781869

18791870
Omitting this keyword has the same assertion behavior as an empty schema.
18801871

@@ -2115,7 +2106,8 @@ determines the canonical nature of the resulting full IRI.[^18]
21152106
and direct you to read the CREF located in {{embedded}} for further comments.
21162107

21172108
While the following IRIs do correctly indicate specific schemas, per the reasons
2118-
outlined in {{embedded}}, they are to be avoided as they may not work in all implementations:
2109+
outlined in {{embedded}}, they are to be avoided as they may not work in all
2110+
implementations:
21192111

21202112
Document location `/$defs/B`:
21212113
- canonical (and base) `IRI: https://example.com/other.json`
@@ -2183,7 +2175,7 @@ scope of this specification to determine or provide a set of safe `$ref` removal
21832175
transformations, as they depend not only on the schema structure but also on the
21842176
intended usage.
21852177

2186-
## %appendix% Example of recursive schema extension {#recursive-example}
2178+
## %appendix% Example of recursive schema extension {#dynamic-example}
21872179

21882180
Consider the following two schemas describing a simple recursive tree structure,
21892181
where each node in the tree can have a "data" field of any type. The first
@@ -2235,7 +2227,7 @@ the following full schema IRIs:
22352227
- `https://example.com/strict-tree#node`
22362228

22372229
In addition, JSON Schema implementations keep track of the fact that these
2238-
fragments were created with `$dynamicAnchor`.
2230+
fragment identifiers were created with `$dynamicAnchor`.
22392231

22402232
If we apply the "strict-tree" schema to the instance, we will follow the `$ref`
22412233
to the "tree" schema, examine its "children" subschema, and find the
@@ -2253,25 +2245,25 @@ At this point, the evaluation path is
22532245
1. `https://example.com/tree#/properties/children`
22542246
1. `https://example.com/tree#/properties/children/items`
22552247

2256-
Since we are looking for a plain name fragment, which can be defined anywhere
2257-
within a schema resource, the JSON Pointer fragments are irrelevant to this
2258-
check. That means that we can remove those fragments and eliminate consecutive
2259-
duplicates, producing:
2248+
Since we are looking for a plain name fragment identifier, which can be defined
2249+
anywhere within a schema resource, the JSON Pointer IRI fragments are irrelevant
2250+
to this check. That means that we can remove the fragments and eliminate
2251+
consecutive duplicates, producing:
22602252

22612253
1. `https://example.com/strict-tree`
22622254
1. `https://example.com/tree`
22632255

2264-
In this case, the outermost resource also has a "node" fragment defined by
2265-
`$dynamicAnchor`. Therefore instead of resolving the `$dynamicRef` to
2256+
In this case, the outermost resource also has a "node" fragment identifier
2257+
defined by `$dynamicAnchor`. Therefore instead of resolving the `$dynamicRef` to
22662258
`https://example.com/tree#node`, we resolve it to
22672259
`https://example.com/strict-tree#node`.
22682260

2269-
This way, the recursion in the "tree" schema recurses to the root of
2270-
"strict-tree", instead of only applying "strict-tree" to the instance root, but
2271-
applying "tree" to instance children.
2261+
The reference in the "tree" schema resolves to the root of "strict-tree", so
2262+
"strict-tree" is applied not only to the tree instance's root, but also its
2263+
children.
22722264

22732265
This example shows both `$dynamicAnchor`s in the same place in each schema,
2274-
specifically the resource root schema. Since plain-name fragments are
2266+
specifically the resource root schema. Since plain-name fragment identifiers are
22752267
independent of the JSON structure, this would work just as well if one or both
22762268
of the node schema objects were moved under `$defs`. It is the matching
22772269
`$dynamicAnchor` values which tell us how to resolve the dynamic reference, not

0 commit comments

Comments
 (0)