RFC 9535 | JSONPath | February 2024 |
Gössner, et al. | Standards Track | [Page] |
JSONPath defines a string syntax for selecting and extracting JSON (RFC 8259) values from within a given JSON value.¶
This is an Internet Standards Track document.¶
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc9535.¶
Copyright (c) 2024 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
JSON [RFC8259] is a popular representation format for structured data values. JSONPath defines a string syntax for selecting and extracting JSON values from within a given JSON value.¶
In relation to JSON Pointer [RFC6901], JSONPath is not intended as a replacement but as a more powerful companion. See Appendix C.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
The grammatical rules in this document are to be interpreted as ABNF,
as described in [RFC5234].
ABNF terminal values in this document define Unicode scalar values rather than
their UTF-8 encoding.
For example, the Unicode PLACE OF INTEREST SIGN (U+2318) would be defined
in ABNF as %x2318
.¶
Functions are referred to using the function name followed by a pair
of parentheses, as in fname()
.¶
The terminology of [RFC8259] applies except where clarified below. The terms "primitive" and "structured" are used to group different kinds of values as in Section 1 of [RFC8259]. JSON objects and arrays are structured; all other values are primitive. Definitions for "object", "array", "number", and "string" remain unchanged. Importantly, "object" and "array" in particular do not take on a generic meaning, such as they would in a general programming context.¶
The terminology of [RFC9485] applies.¶
Additional terms used in this document are defined below.¶
As per [RFC8259], a data item conforming to the generic data model of JSON, i.e., primitive data (numbers, text strings, and the special values null, true, and false), or structured data (JSON objects and arrays). [RFC8259] focuses on the textual representation of JSON values and does not fully define the value abstraction assumed here.¶
A name/value pair in an object. (A member is not itself a value.)¶
The name (a string) in a name/value pair constituting a member. This is also used in [RFC8259], but that specification does not formally define it. It is included here for completeness.¶
A value in a JSON array.¶
An integer that identifies a specific element in an array.¶
Short name for a JSONPath expression.¶
Short name for the value a JSONPath expression is applied to.¶
The position of a value within the query argument. This can be thought of as a sequence of names and indexes navigating to the value through the objects and arrays in the query argument, with the empty sequence indicating the query argument itself. A location can be represented as a Normalized Path (defined below).¶
The pair of a value along with its location within the query argument.¶
The unique node whose value is the entire query argument.¶
The expression $
, which refers to the root node of the query argument.¶
The expression @
, which refers to the current node in the context
of the evaluation of a filter expression (described later).¶
If the node is an array, the nodes of its elements; if the node is an object, the nodes of its member values. If the node is neither an array nor an object, it has no children.¶
The children of the node, together with the children of its children, and so forth recursively. More formally, the "descendants" relation between nodes is the transitive closure of the "children" relation.¶
The number of ancestors of the node within the value. The root node of the value has depth zero, the children of the root node have depth one, their children have depth two, and so forth.¶
A list of nodes. While a nodelist can be represented in JSON, e.g., as an array, this document does not require or assume any particular representation.¶
Formal parameter (of a function) that can take a function argument (an actual parameter) in a function expression.¶
A form of JSONPath expression that identifies a node in a value by providing a query that results in exactly that node. Each node in a query argument is identified by exactly one Normalized Path (we say that the Normalized Path is "unique" for that node), and to be a Normalized Path for a specific query argument, the Normalized Path needs to identify exactly one node. This is similar to, but syntactically different from, a JSON Pointer [RFC6901]. Note: This definition is based on the syntactical definition in Section 2.7; JSONPath expressions that identify a node in a value but do not conform to that syntax are not Normalized Paths.¶
Any Unicode [UNICODE] code point except high-surrogate and low-surrogate code points (in other words, integers in the inclusive base 16 ranges, either 0 to D7FF or E000 to 10FFFF). JSONPath queries are sequences of Unicode scalar values.¶
One of the constructs that selects children ([<selectors>]
)
or descendants (..[<selectors>]
) of an input value.¶
A single item within a segment that takes the input value and produces a nodelist consisting of child nodes of the input value.¶
A JSONPath expression built from segments that have been syntactically restricted in a certain way (Section 2.3.5.1) so that, regardless of the input value, the expression produces a nodelist containing at most one node. Note: JSONPath expressions that always produce a singular nodelist but do not conform to the syntax in Section 2.3.5.1 are not singular queries.¶
This document models the query argument as a tree of JSON values, each with its own node. A node is either the root node or one of its descendants.¶
This document models the result of applying a query to the query argument as a nodelist (a list of nodes).¶
Nodes are the selectable parts of the query argument. The only parts of an object that can be selected by a query are the member values. Member names and members (name/value pairs) cannot be selected. Thus, member values have nodes, but members and member names do not. Similarly, member values are children of an object, but members and member names are not.¶
This document is based on Stefan Gössner's popular JSONPath proposal (dated 2007-02-21) [JSONPath-orig], builds on the experience from the widespread deployment of its implementations, and provides a normative specification for it.¶
Appendix B describes how JSONPath was inspired by XML's XPath [XPath].¶
JSONPath was intended as a lightweight companion to JSON
implementations in programming languages such as PHP and JavaScript,
so instead of defining its own expression language, like XPath did,
JSONPath delegated parts of a query to the underlying
runtime, e.g., JavaScript's eval()
function.
As JSONPath was implemented in more environments, JSONPath
expressions became decreasingly portable.
For example, regular expression processing was often delegated to a
convenient regular expression engine.¶
This document aims to remove such implementation-specific dependencies and serve as a common JSONPath specification that can be used across programming languages and environments. This means that backwards compatibility is not always achieved; a design principle of this document is to go with a "consensus" between implementations even if it is rough, as long as that does not jeopardize the objective of obtaining a usable, stable JSON query language.¶
The term JSONPath was chosen because of the XPath inspiration and also because the outcome of a query consists of paths identifying nodes in the JSON query argument.¶
The JSON value a JSONPath query is applied to is, by definition, a valid JSON value. A JSON value is often constructed by parsing a JSON text.¶
The parsing of a JSON text into a JSON value and what happens if a JSON text does not represent valid JSON are not defined by this document. Sections 4 and 8 of [RFC8259] identify specific situations that may conform to the grammar for JSON texts but are not interoperable uses of JSON, as they may cause unpredictable behavior. This document does not attempt to define predictable behavior for JSONPath queries in these situations.¶
Specifically, the "Semantics" subsections of Sections 2.3.1, 2.3.2, 2.3.5, and 2.5.2 describe behavior that becomes unpredictable when the JSON value for one of the objects under consideration was constructed out of JSON text that exhibits multiple members for a single object that share the same member name ("duplicate names"; see Section 4 of [RFC8259]). Also, when selecting a child by name (Section 2.3.1) and comparing strings (Section 2.3.5.2.2), it is assumed these strings are sequences of Unicode scalar values; the behavior becomes unpredictable if they are not (Section 8.2 of [RFC8259]).¶
A JSONPath expression is applied to a JSON value, known as the query argument. The output is a nodelist.¶
A JSONPath expression consists of an identifier followed by a series of zero or more segments, each of which contains one or more selectors.¶
The root node identifier $
refers to the root node of the query argument,
i.e., to the argument as a whole.¶
The current node identifier @
refers to the current node in the context
of the evaluation of a filter expression (Section 2.3.5).¶
Segments select children ([<selectors>]
) or descendants (..[<selectors>]
) of an input value.¶
Segments can use bracket notation, for example:¶
$['store']['book'][0]['title']¶
or the more compact dot notation, for example:¶
$.store.book[0].title¶
Bracket notation contains one or more (comma-separated) selectors of any kind. Selectors are detailed in the next section.¶
A JSONPath expression may use a combination of bracket and dot notations.¶
This document treats the bracket notations as canonical and defines the shorthand dot notation in terms of bracket notation. Examples and descriptions use shorthand where convenient.¶
A name selector, e.g., 'name'
, selects a named child of an object.¶
An index selector, e.g., 3
, selects an indexed child of an array.¶
In the expression [*]
, a wildcard *
(Section 2.3.2) selects all children of a
node, and in the expression ..[*]
, it selects all descendants of a node.¶
An array slice start:end:step
(Section 2.3.4) selects a series of
elements from an array, giving a start position, an end position, and
an optional step value that moves the position from the start to the
end.¶
A filter expression ?<logical-expr>
selects certain children of an object or array, as in:¶
$.store.book[?@.price < 10].title¶
Table 1 provides a brief overview of JSONPath syntax.¶
Syntax Element | Description |
---|---|
$
|
root node identifier (Section 2.2) |
@
|
current node identifier (Section 2.3.5) (valid only within filter selectors) |
[<selectors>]
|
child segment (Section 2.5.1): selects zero or more children of a node |
.name
|
shorthand for ['name']
|
.*
|
shorthand for [*]
|
..[<selectors>]
|
descendant segment (Section 2.5.2): selects zero or more descendants of a node |
..name
|
shorthand for ..['name']
|
..*
|
shorthand for ..[*]
|
'name'
|
name selector (Section 2.3.1): selects a named child of an object |
*
|
wildcard selector (Section 2.3.2): selects all children of a node |
3
|
index selector (Section 2.3.3): selects an indexed child of an array (from 0) |
0:100:5
|
array slice selector (Section 2.3.4): start:end:step for arrays |
?<logical-expr>
|
filter selector (Section 2.3.5): selects particular children using a logical expression |
length(@.foo)
|
function extension (Section 2.4): invokes a function in a filter expression |
This section is informative. It provides examples of JSONPath expressions.¶
The examples are based on the simple JSON value shown in Figure 1, representing a bookstore (which also has a bicycle).¶
Table 2 shows some JSONPath queries that might be applied to this example and their intended results.¶
JSONPath | Intended Result |
---|---|
$.store.book[*].author
|
the authors of all books in the store |
$..author
|
all authors |
$.store.*
|
all things in the store, which are some books and a red bicycle |
$.store..price
|
the prices of everything in the store |
$..book[2]
|
the third book |
$..book[2].author
|
the third book's author |
$..book[2].publisher
|
empty result: the third book does not have a "publisher" member |
$..book[-1]
|
the last book in order |
$..book[0,1] $..book[:2]
|
the first two books |
$..book[?@.isbn]
|
all books with an ISBN number |
$..book[?@.price<10]
|
all books cheaper than 10 |
$..*
|
all member values and array elements contained in the input value |
A JSONPath expression is a string that, when applied to a JSON value (the query argument), selects zero or more nodes of the argument and outputs these nodes as a nodelist.¶
A query MUST be encoded using UTF-8. The grammar for queries given in this document assumes that its UTF-8 form is first decoded into Unicode scalar values as described in [RFC3629]; implementation approaches that lead to an equivalent result are possible.¶
A string to be used as a JSONPath query needs to be well-formed and valid. A string is a well-formed JSONPath query if it conforms to the ABNF syntax in this document. A well-formed JSONPath query is valid if it also fulfills both semantic requirements posed by this document, which are as follows:¶
A JSONPath implementation MUST raise an error for any query that is not well-formed and valid. The well-formedness and the validity of JSONPath queries are independent of the JSON value the query is applied to. No further errors relating to the well-formedness and the validity of a JSONPath query can be raised during application of the query to a value. This clearly separates well-formedness/validity errors in the query from mismatches that may actually stem from flaws in the data.¶
Mismatches between the structure expected by a valid query and the structure found in the data can lead to empty query results, which may be unexpected and indicate bugs in either. JSONPath implementations might therefore want to provide diagnostics to the application developer that aid in finding the cause of empty results.¶
Obviously, an implementation can still fail when executing a JSONPath query, e.g., because of resource depletion, but this is not modeled in this document. However, the implementation MUST NOT silently malfunction. Specifically, if a valid JSONPath query is evaluated against a structured value whose size is too large to process the query correctly (for instance, requiring the processing of numbers that fall outside the range of exact values), the implementation MUST provide an indication of overflow.¶
(Readers familiar with the HTTP error model may be reminded of 400 type errors when pondering well-formedness and validity, and they may recognize resource depletion and related errors as comparable to 500 type errors.)¶
Syntactically, a JSONPath query consists of a root identifier ($
), which
stands for a nodelist that contains the root node of the query argument,
followed by a possibly empty sequence of segments.¶
jsonpath-query = root-identifier segments segments = *(S segment) B = %x20 / ; Space %x09 / ; Horizontal tab %x0A / ; Line feed or New line %x0D ; Carriage return S = *B ; optional blank space¶
The syntax and semantics of segments are defined in Section 2.5.¶
In this document, the semantics of a JSONPath query define the required results and do not prescribe the internal workings of an implementation. This document may describe semantics in a procedural step-by-step fashion; however, such descriptions are normative only in the sense that any implementation MUST produce an identical result but not in the sense that implementers are required to use the same algorithms.¶
The semantics are that a valid query is executed against a value (the query argument) and produces a nodelist (i.e., a list of zero or more nodes of the value).¶
The query is a root identifier followed by a sequence of zero or more segments, each of which is applied to the result of the previous root identifier or segment and provides input to the next segment. These results and inputs take the form of nodelists.¶
The nodelist resulting from the root identifier contains a single node (the query argument). The nodelist resulting from the last segment is presented as the result of the query. Depending on the specific API, it might be presented as an array of the JSON values at the nodes, an array of Normalized Paths referencing the nodes, or both -- or some other representation as desired by the implementation. Note: An empty nodelist is a valid query result.¶
A segment operates on each of the nodes in its input nodelist in turn, and the resultant nodelists are concatenated in the order of the input nodelist they were derived from to produce the result of the segment. A node may be selected more than once and appears that number of times in the nodelist. Duplicate nodes are not removed.¶
A syntactically valid segment MUST NOT produce errors when executing the query. This means that some operations that might be considered erroneous, such as using an index lying outside the range of an array, simply result in fewer nodes being selected. (Additional discussion of this property can be found in the introduction of Section 2.1.)¶
As a consequence of this approach, if any of the segments produces an empty nodelist, then the whole query produces an empty nodelist.¶
If the semantics of a query give an implementation a choice of producing multiple possible orderings, a particular implementation may produce distinct orderings in successive runs of the query.¶
Consider this example. With the query argument {"a":[{"b":0},{"b":1},{"c":2}]}
, the
query $.a[*].b
selects the following list of nodes (denoted here by their values): 0
, 1
.¶
The query consists of $
followed by three segments: .a
, [*]
, and .b
.¶
First, $
produces a nodelist consisting of just the query argument.¶
Next, .a
selects from any object input node and selects the
node of any
member value of the input
node corresponding to the member name "a"
.
The result is again a list containing a single node: [{"b":0},{"b":1},{"c":2}]
.¶
Next, [*]
selects all the elements
from the input array node.
The result is a list of three nodes: {"b":0}
, {"b":1}
, and {"c":2}
.¶
Finally, .b
selects from any object input node with a member name
b
and selects the node of the member value of the input node corresponding to that name.
The result is a list containing 0
, 1
.
This is the concatenation of three lists: two of length one containing
0
, 1
, respectively, and one of length zero.¶
Every JSONPath query (except those inside filter expressions; see Section 2.3.5) MUST begin with the root identifier $
.¶
root-identifier = "$"¶
Selectors appear only inside child segments (Section 2.5.1) and descendant segments (Section 2.5.2).¶
A selector produces a nodelist consisting of zero or more children of the input value.¶
There are various kinds of selectors that produce children of objects, children of arrays, or children of either objects or arrays.¶
selector = name-selector / wildcard-selector / slice-selector / index-selector / filter-selector¶
The syntax and semantics of each kind of selector are defined below.¶
A name selector '<name>'
selects at most one object member value.¶
In contrast to JSON, the JSONPath syntax allows strings to be enclosed in single or double quotes.¶
name-selector = string-literal string-literal = %x22 *double-quoted %x22 / ; "string" %x27 *single-quoted %x27 ; 'string' double-quoted = unescaped / %x27 / ; ' ESC %x22 / ; \" ESC escapable single-quoted = unescaped / %x22 / ; " ESC %x27 / ; \' ESC escapable ESC = %x5C ; \ backslash unescaped = %x20-21 / ; see RFC 8259 ; omit 0x22 " %x23-26 / ; omit 0x27 ' %x28-5B / ; omit 0x5C \ %x5D-D7FF / ; skip surrogate code points %xE000-10FFFF escapable = %x62 / ; b BS backspace U+0008 %x66 / ; f FF form feed U+000C %x6E / ; n LF line feed U+000A %x72 / ; r CR carriage return U+000D %x74 / ; t HT horizontal tab U+0009 "/" / ; / slash (solidus) U+002F "\" / ; \ backslash (reverse solidus) U+005C (%x75 hexchar) ; uXXXX U+XXXX hexchar = non-surrogate / (high-surrogate "\" %x75 low-surrogate) non-surrogate = ((DIGIT / "A"/"B"/"C" / "E"/"F") 3HEXDIG) / ("D" %x30-37 2HEXDIG ) high-surrogate = "D" ("8"/"9"/"A"/"B") 2HEXDIG low-surrogate = "D" ("C"/"D"/"E"/"F") 2HEXDIG HEXDIG = DIGIT / "A" / "B" / "C" / "D" / "E" / "F"¶
Notes:¶
Double-quoted
strings follow the JSON string syntax (Section 7 of [RFC8259]);
single-quoted
strings follow an analogous pattern.
No attempt was made to improve on this syntax, so if it is desired to
escape characters with
scalar values above 0xFFFF, such as U+1F3BC ("🎼", MUSICAL SCORE),
they need to be represented
by a pair of surrogate escapes ("\uD83C\uDFBC"
in this case).¶
\u
escapes (as specified in rules
referenced by hexchar
) can be either lowercase or uppercase,
while the u
in \u
needs to be lowercase (indicated as %x75
).¶
A name-selector
string MUST be converted to a
member name M
by removing the surrounding quotes and
replacing each escape sequence with its equivalent Unicode character, as
shown in Table 4:¶
Escape Sequence | Unicode Character | Description |
---|---|---|
\b
|
U+0008 | BS backspace |
\t
|
U+0009 | HT horizontal tab |
\n
|
U+000A | LF line feed |
\f
|
U+000C | FF form feed |
\r
|
U+000D | CR carriage return |
\"
|
U+0022 | quotation mark |
\'
|
U+0027 | apostrophe |
\/
|
U+002F | slash (solidus) |
\\
|
U+005C | backslash (reverse solidus) |
\uXXXX
|
see Section 2.3.1.1 | hexadecimal escape |
Applying the name-selector
to an object node
selects a member value whose name equals the member name M
or selects nothing if there is no such member value.
Nothing is selected from a value that is not an object.¶
Note: Processing the name selector requires comparing the member name string M
with member name strings in the JSON to which the selector is being applied.
Two strings MUST be considered equal if and only if they are identical
sequences of Unicode scalar values. In other words, normalization operations
MUST NOT be applied to either the member name string M
from the JSONPath or
the member name strings in the JSON prior to comparison.¶
JSON:¶
{ "o": {"j j": {"k.k": 3}}, "'": {"@": 2} }¶
Queries:¶
The examples in Table 5 show the name selector in use by child segments.¶
Query | Result | Result Paths | Comment |
---|---|---|---|
$.o['j j']
|
{"k.k": 3}
|
$['o']['j j']
|
Named value in a nested object |
$.o['j j']['k.k']
|
3
|
$['o']['j j']['k.k']
|
Nesting further down |
$.o["j j"]["k.k"]
|
3
|
$['o']['j j']['k.k']
|
Different delimiter in the query, unchanged Normalized Path |
$["'"]["@"]
|
2
|
$['\'']['@']
|
Unusual member names |
A wildcard selector selects the nodes of all children of an object or array. The order in which the children of an object appear in the resultant nodelist is not stipulated, since JSON objects are unordered. Children of an array appear in array order in the resultant nodelist.¶
Note that the children of an object are its member values, not its member names.¶
The wildcard selector selects nothing from a primitive JSON value (that is,
a number, a string, true
, false
, or null
).¶
JSON:¶
{ "o": {"j": 1, "k": 2}, "a": [5, 3] }¶
Queries:¶
The examples in Table 6 show the wildcard selector in use by a child segment.¶
Query | Result | Result Paths | Comment |
---|---|---|---|
$[*]
|
{"j": 1, "k": 2} [5, 3]
|
$['o'] $['a']
|
Object values |
$.o[*]
|
1 2
|
$['o']['j'] $['o']['k']
|
Object values |
$.o[*]
|
2 1
|
$['o']['k'] $['o']['j']
|
Alternative result |
$.o[*, *]
|
1 2 2 1
|
$['o']['j'] $['o']['k'] $['o']['k'] $['o']['j']
|
Non-deterministic ordering |
$.a[*]
|
5 3
|
$['a'][0] $['a'][1]
|
Array members |
The example above with the query $.o[*, *]
shows that the wildcard selector may produce nodelists in distinct
orders each time it appears in the child segment when it is applied to an object node with two or more
members (but not when it is applied to object nodes with fewer than two members or to array nodes).¶
An index selector <index>
matches at most one array element value.¶
index-selector = int ; decimal integer int = "0" / (["-"] DIGIT1 *DIGIT) ; - optional DIGIT1 = %x31-39 ; 1-9 non-zero digit¶
Applying the numerical index-selector
selects the corresponding
element. JSONPath allows it to be negative (see Section 2.3.3.2).¶
To be valid, the index selector value MUST be in the I-JSON range of exact values (see Section 2.1).¶
Notes:¶
A non-negative index-selector
applied to an array selects an array element using a zero-based index.
For example, the selector 0
selects the first, and the selector 4
selects the fifth element of a sufficiently long array.
Nothing is selected, and it is not an error, if the index lies outside the range of the array. Nothing is selected from a value that is not an array.¶
A negative index-selector
counts from the array end backwards,
obtaining an equivalent non-negative index-selector
by adding the
length of the array to the negative index.
For example, the selector -1
selects the last, and the selector -2
selects the penultimate element of an array with at least two elements.
As with non-negative indexes, it is not an error if such an element does
not exist; this simply means that no element is selected.¶
The array slice selector has the form <start>:<end>:<step>
.
It matches elements from arrays starting at index <start>
and ending at (but
not including) <end>
, while incrementing by step
with a default of 1
.¶
slice-selector = [start S] ":" S [end S] [":" [S step ]] start = int ; included in selection end = int ; not included in selection step = int ; default: 1¶
The slice selector consists of three optional decimal integers separated by colons. The second colon can be omitted when the third integer is omitted.¶
To be valid, the integers provided MUST be in the I-JSON range of exact values (see Section 2.1).¶
The slice selector was inspired by the slice operator that was proposed for ECMAScript 4 (ES4), which was never released, and that of Python.¶
This section is informative.¶
Array slicing is inspired by the behavior of the Array.prototype.slice
method
of the JavaScript language, as defined by the ECMA-262 standard [ECMA-262],
with the addition of the step
parameter, which is inspired by the Python slice expression.¶
The array slice expression start:end:step
selects elements at indices starting at start
,
incrementing by step
, and ending with end
(which is itself excluded).
So, for example, the expression 1:3
(where step
defaults to 1
)
selects elements with indices 1
and 2
(in that order), whereas
1:5:2
selects elements with indices 1
and 3
.¶
When step
is negative, elements are selected in reverse order. Thus,
for example, 5:1:-2
selects elements with indices 5
and 3
(in
that order), and ::-1
selects all the elements of an array in
reverse order.¶
When step
is 0
, no elements are selected.
(This is the one case that differs from the behavior of Python, which
raises an error in this case.)¶
The following section specifies the behavior fully, without depending on JavaScript or Python behavior.¶
A slice expression selects a subset of the elements of the input array in
the same order
as the array or the reverse order, depending on the sign of the step
parameter.
It selects no nodes from a node that is not an array.¶
A slice is defined by the two slice parameters, start
and end
, and
an iteration delta, step
.
Each of these parameters is
optional. In the rest of this section, len
denotes the length of the input array.¶
The default value for step
is 1
.
The default values for start
and end
depend on the sign of step
,
as shown in Table 8.¶
Condition | start | end |
---|---|---|
step >= 0 | 0 | len |
step < 0 | len - 1 | -len - 1 |
Slice expression parameters start
and end
are not directly usable
as slice bounds and must first be normalized.
Normalization for this purpose is defined as:¶
FUNCTION Normalize(i, len): IF i >= 0 THEN RETURN i ELSE RETURN len + i END IF¶
The result of the array index expression i
applied to an array
of length len
is the result of the array
slicing expression Normalize(i, len):Normalize(i, len)+1:1
.¶
Slice expression parameters start
and end
are used to derive slice bounds lower
and upper
.
The direction of the iteration, defined
by the sign of step
, determines which of the parameters is the lower bound and which
is the upper bound:¶
FUNCTION Bounds(start, end, step, len): n_start = Normalize(start, len) n_end = Normalize(end, len) IF step >= 0 THEN lower = MIN(MAX(n_start, 0), len) upper = MIN(MAX(n_end, 0), len) ELSE upper = MIN(MAX(n_start, -1), len-1) lower = MIN(MAX(n_end, -1), len-1) END IF RETURN (lower, upper)¶
The slice expression selects elements with indices between the lower and
upper bounds.
In the following pseudocode, a(i)
is the i+1
th element of the array a
(i.e., a(0)
is the first element, a(1)
the second, and so forth).¶
IF step > 0 THEN i = lower WHILE i < upper: SELECT a(i) i = i + step END WHILE ELSE if step < 0 THEN i = upper WHILE lower < i: SELECT a(i) i = i + step END WHILE END IF¶
When step = 0
, no elements are selected, and the result array is empty.¶
JSON:¶
["a", "b", "c", "d", "e", "f", "g"]¶
Queries:¶
The examples in Table 9 show the array slice selector in use by a child segment.¶
Query | Result | Result Paths | Comment |
---|---|---|---|
$[1:3]
|
"b" "c"
|
$[1] $[2]
|
Slice with default step |
$[5:]
|
"f" "g"
|
$[5] $[6]
|
Slice with no end index |
$[1:5:2]
|
"b" "d"
|
$[1] $[3]
|
Slice with step 2 |
$[5:1:-2]
|
"f" "d"
|
$[5] $[3]
|
Slice with negative step |
$[::-1]
|
"g" "f" "e" "d" "c" "b" "a"
|
$[6] $[5] $[4] $[3] $[2] $[1] $[0]
|
Slice in reverse order |
Filter selectors are used to iterate over the elements or members of structured values, i.e., JSON arrays and objects. The structured values are identified in the nodelist offered by the child or descendant segment using the filter selector.¶
For each iteration (element/member), a logical expression (the filter expression) is evaluated, which decides whether the node of the element/member is selected. (While a logical expression evaluates to what mathematically is a Boolean value, this specification uses the term logical to maintain a distinction from the Boolean values that JSON can represent.)¶
During the iteration process, the filter expression receives the node of each array element or object member value of the structured value being filtered; this element or member value is then known as the current node.¶
The current node can be used as the start of one or more JSONPath
queries in subexpressions of the filter expression, notated
via the current-node-identifier @
.
Each JSONPath query can be used either for testing existence of a
result of the query, for obtaining a specific JSON value resulting
from that query that can then be used in a comparison, or as a
function argument.¶
Filter selectors may use function extensions, which are covered in Section 2.4. Within the logical expression for a filter selector, function expressions can be used to operate on nodelists and values. The set of available functions is extensible, with a number of functions predefined (see Section 2.4) and the ability to register further functions provided by the "Function Extensions" subregistry (Section 3.2). When a function is defined, it is given a unique name, and its return value and each of its parameters are given a declared type. The type system is limited in scope; its purpose is to express restrictions that, without functions, are implicit in the grammar of filter expressions. The type system also guides conversions (Section 2.4.2) that mimic the way different kinds of expressions are handled in the grammar when function expressions are not in use.¶
The filter selector has the form ?<logical-expr>
.¶
filter-selector = "?" S logical-expr¶
As the filter expression is composed of constituents free of side effects,
the order of evaluation does not need to be (and is not) defined.
Similarly, for conjunction (&&
) and disjunction (||
) (defined later),
both a short-circuiting and a fully evaluating
implementation will lead to the same result; both implementation
strategies are therefore valid.¶
The current node is accessible via the current node identifier @
.
This identifier addresses the current node of the filter-selector that
is directly enclosing the identifier. Note: Within nested
filter-selectors, there is no syntax to address the current node of
any other than the directly enclosing filter-selector (i.e., of
filter-selectors enclosing the filter-selector that is directly
enclosing the identifier).¶
Logical expressions offer the usual Boolean operators (||
for OR,
&&
for AND, and !
for NOT).
They have the normal semantics of Boolean algebra and obey its laws
(for example, see [BOOLEAN-LAWS]).
Parentheses MAY be used within logical-expr
for grouping.¶
It is not required that logical-expr
consist of
a parenthesized expression (which was required in [JSONPath-orig]),
although it can be, and the semantics are the same
as without the parentheses.¶
logical-expr = logical-or-expr logical-or-expr = logical-and-expr *(S "||" S logical-and-expr) ; disjunction ; binds less tightly than conjunction logical-and-expr = basic-expr *(S "&&" S basic-expr) ; conjunction ; binds more tightly than disjunction basic-expr = paren-expr / comparison-expr / test-expr paren-expr = [logical-not-op S] "(" S logical-expr S ")" ; parenthesized expression logical-not-op = "!" ; logical NOT operator¶
A test expression
either tests the existence of a node
designated by an embedded query (see Section 2.3.5.2.1) or tests the
result of a function expression (see Section 2.4).
In the latter case, if the function's declared result type is
LogicalType
(see Section 2.4.1), it tests whether the result
is LogicalTrue
; if the function's declared result type is
NodesType
, it tests whether the result is non-empty.
If the function's declared result type is ValueType
, its use in a
test expression is not well-typed (see Section 2.4.3).¶
test-expr = [logical-not-op S] (filter-query / ; existence/non-existence function-expr) ; LogicalType or NodesType filter-query = rel-query / jsonpath-query rel-query = current-node-identifier segments current-node-identifier = "@"¶
Comparison expressions are available for comparisons between primitive
values (that is, numbers, strings, true
, false
, and null
).
These can be obtained via literal values; singular queries, each of
which selects at most one node, the value of which is then used; or
function expressions (see Section 2.4) of type ValueType
.¶
comparison-expr = comparable S comparison-op S comparable literal = number / string-literal / true / false / null comparable = literal / singular-query / ; singular query value function-expr ; ValueType comparison-op = "==" / "!=" / "<=" / ">=" / "<" / ">" singular-query = rel-singular-query / abs-singular-query rel-singular-query = current-node-identifier singular-query-segments abs-singular-query = root-identifier singular-query-segments singular-query-segments = *(S (name-segment / index-segment)) name-segment = ("[" name-selector "]") / ("." member-name-shorthand) index-segment = "[" index-selector "]"¶
Literals can be notated in the way that is usual for JSON (with the extension that strings can use single-quote delimiters).¶
Note: Alphabetic characters in quoted strings are case-insensitive in ABNF, so within a floating point number, the ABNF expression "e" can be either the character 'e' or 'E'.¶
true
, false
, and null
are lowercase only (case-sensitive).¶
number = (int / "-0") [ frac ] [ exp ] ; decimal number frac = "." 1*DIGIT ; decimal fraction exp = "e" [ "-" / "+" ] 1*DIGIT ; decimal exponent true = %x74.72.75.65 ; true false = %x66.61.6c.73.65 ; false null = %x6e.75.6c.6c ; null¶
Table 10 lists filter expression operators in order of precedence from highest (binds most tightly) to lowest (binds least tightly).¶
Precedence | Operator type | Syntax |
---|---|---|
5 | Grouping Function Expressions |
(...) name (...)
|
4 | Logical NOT |
!
|
3 | Relations |
== != < <= > >=
|
2 | Logical AND |
&&
|
1 | Logical OR |
||
|
The filter selector works with arrays and objects exclusively. Its result is a list of (zero, one, multiple, or all) their array elements or member values, respectively. Applied to a primitive value, it selects nothing (and therefore does not contribute to the result of the filter selector).¶
In the resultant nodelist, children of an array are ordered by their position in the array. The order in which the children of an object (as opposed to an array) appear in the resultant nodelist is not stipulated, since JSON objects are unordered.¶
A query by itself in a logical context is an existence test that yields true if the query selects at least one node and yields false if the query does not select any nodes.¶
Existence tests differ from comparisons in that:¶
To examine the value of a node selected by a query, an explicit comparison is necessary.
For example, to test whether the node selected by the query @.foo
has the value null
, use @.foo == null
(see Section 2.6)
rather than the negated existence test !@.foo
(which yields false if @.foo
selects a node, regardless of the node's value).
Similarly, @.foo == false
yields true only if @.foo
selects a node and
the value of that node is false
.¶
The comparison operators ==
and <
are defined first, and then these are used to define !=
, <=
, >
, and >=
.¶
When either side of a comparison results in an empty nodelist or the
special result Nothing
(see Section 2.4.1):¶
==
yields true if and only the
other side also results in an empty nodelist or the special result Nothing
.¶
<
yields false.¶
When any query or function expression on either side of a comparison results in a nodelist consisting of a single node, that side is replaced by the value of its node and then:¶
A comparison using the operator ==
yields true if and only if the comparison
is between:¶
equal objects with no duplicate names, that is, where:¶
A comparison using the operator <
yields true if and only if
the comparison is between values that are both numbers or both strings and that satisfy the comparison:¶
!=
, <=
, >
, and >=
are defined in terms of the other comparison operators. For any a
and b
:¶
a != b
yields true if and only if a == b
yields false.¶
a <= b
yields true if and only if a < b
yields true or a == b
yields true.¶
a > b
yields true if and only if b < a
yields true.¶
a >= b
yields true if and only if b < a
yields true or a == b
yields true.¶
The first set of examples shows some comparison expressions and their result with a given JSON value as input.¶
JSON:¶
{ "obj": {"x": "y"}, "arr": [2, 3] }¶
Comparisons:¶
Comparison | Result | Comment |
---|---|---|
$.absent1 == $.absent2
|
true | Empty nodelists |
$.absent1 <= $.absent2
|
true |
== implies <=
|
$.absent == 'g'
|
false | Empty nodelist |
$.absent1 != $.absent2
|
false | Empty nodelists |
$.absent != 'g'
|
true | Empty nodelist |
1 <= 2
|
true | Numeric comparison |
1 > 2
|
false | Numeric comparison |
13 == '13'
|
false | Type mismatch |
'a' <= 'b'
|
true | String comparison |
'a' > 'b'
|
false | String comparison |
$.obj == $.arr
|
false | Type mismatch |
$.obj != $.arr
|
true | Type mismatch |
$.obj == $.obj
|
true | Object comparison |
$.obj != $.obj
|
false | Object comparison |
$.arr == $.arr
|
true | Array comparison |
$.arr != $.arr
|
false | Array comparison |
$.obj == 17
|
false | Type mismatch |
$.obj != 17
|
true | Type mismatch |
$.obj <= $.arr
|
false | Objects and arrays do not offer < comparison |
$.obj < $.arr
|
false | Objects and arrays do not offer < comparison |
$.obj <= $.obj
|
true |
== implies <=
|
$.arr <= $.arr
|
true |
== implies <=
|
1 <= $.arr
|
false | Arrays do not offer < comparison |
1 >= $.arr
|
false | Arrays do not offer < comparison |
1 > $.arr
|
false | Arrays do not offer < comparison |
1 < $.arr
|
false | Arrays do not offer < comparison |
true <= true
|
true |
== implies <=
|
true > true
|
false | Booleans do not offer < comparison |
The second set of examples shows some complete JSONPath queries that make use of filter selectors and the results of evaluating these queries on a given JSON value as input. (Note: Two of the queries employ function extensions; please see Sections 2.4.6 and 2.4.7 for details about these.)¶
JSON:¶
{ "a": [3, 5, 1, 2, 4, 6, {"b": "j"}, {"b": "k"}, {"b": {}}, {"b": "kilo"} ], "o": {"p": 1, "q": 2, "r": 3, "s": 5, "t": {"u": 6}}, "e": "f" }¶
Queries:¶
The examples in Table 12 show the filter selector in use by a child segment.¶
Query | Result | Result Paths | Comment |
---|---|---|---|
$.a[?@.b == 'kilo']
|
{"b": "kilo"}
|
$['a'][9]
|
Member value comparison |
$.a[?(@.b == 'kilo')]
|
{"b": "kilo"}
|
$['a'][9]
|
Equivalent query with enclosing parentheses |
$.a[?@>3.5]
|
5 4 6
|
$['a'][1] $['a'][4] $['a'][5]
|
Array value comparison |
$.a[?@.b]
|
{"b": "j"} {"b": "k"} {"b": {}} {"b": "kilo"}
|
$['a'][6] $['a'][7] $['a'][8] $['a'][9]
|
Array value existence |
$[?@.*]
|
[3, 5, 1, 2, 4, 6, {"b": "j"}, {"b": "k"}, {"b": {}}, {"b": "kilo"}] {"p": 1, "q": 2, "r": 3, "s": 5, "t": {"u": 6}}
|
$['a'] $['o']
|
Existence of non-singular queries |
$[?@[?@.b]]
|
[3, 5, 1, 2, 4, 6, {"b": "j"}, {"b": "k"}, {"b": {}}, {"b": "kilo"}]
|
$['a']
|
Nested filters |
$.o[?@<3, ?@<3]
|
1 2 2 1
|
$['o']['p'] $['o']['q'] $['o']['q'] $['o']['p']
|
Non-deterministic ordering |
$.a[?@<2 || @.b == "k"]
|
1 {"b": "k"}
|
$['a'][2] $['a'][7]
|
Array value logical OR |
$.a[?match(@.b, "[jk]")]
|
{"b": "j"} {"b": "k"}
|
$['a'][6] $['a'][7]
|
Array value regular expression match |
$.a[?search(@.b, "[jk]")]
|
{"b": "j"} {"b": "k"} {"b": "kilo"}
|
$['a'][6] $['a'][7] $['a'][9]
|
Array value regular expression search |
$.o[?@>1 && @<4]
|
2 3
|
$['o']['q'] $['o']['r']
|
Object value logical AND |
$.o[?@>1 && @<4]
|
3 2
|
$['o']['r'] $['o']['q']
|
Alternative result |
$.o[?@.u || @.x]
|
{"u": 6}
|
$['o']['t']
|
Object value logical OR |
$.a[?@.b == $.x]
|
3 5 1 2 4 6
|
$['a'][0] $['a'][1] $['a'][2] $['a'][3] $['a'][4] $['a'][5]
|
Comparison of queries with no values |
$.a[?@ == @]
|
3 5 1 2 4 6 {"b": "j"} {"b": "k"} {"b": {}} {"b": "kilo"}
|
$['a'][0] $['a'][1] $['a'][2] $['a'][3] $['a'][4] $['a'][5] $['a'][6] $['a'][7] $['a'][8] $['a'][9]
|
Comparisons of primitive and of structured values |
The example above with the query $.o[?@<3, ?@<3]
shows that a filter selector may produce nodelists in distinct
orders each time it appears in the child segment.¶
Beyond the filter expression functionality defined in the preceding subsections, JSONPath defines an extension point that can be used to add filter expression functionality: "Function Extensions".¶
This section defines the extension point and some function extensions that use this extension point. While these mechanisms are designed to use the extension point, they are an integral part of the JSONPath specification and are expected to be implemented like any other integral part of this specification.¶
A function extension defines a registered name (see Section 3.2) that can be applied to a sequence of zero or more arguments, producing a result. Each registered function name is unique.¶
A function extension MUST be defined such that its evaluation is free of side effects, i.e., all possible orders of evaluation and choices of short-circuiting or full evaluation of an expression containing it MUST lead to the same result. (Note: Memoization or logging are not side effects in this sense as they are visible at the implementation level only -- they do not influence the result of the evaluation.)¶
function-name = function-name-first *function-name-char function-name-first = LCALPHA function-name-char = function-name-first / "_" / DIGIT LCALPHA = %x61-7A ; "a".."z" function-expr = function-name "(" S [function-argument *(S "," S function-argument)] S ")" function-argument = literal / filter-query / ; (includes singular-query) logical-expr / function-expr¶
Any function expressions in a query must be well-formed (by conforming to the above ABNF) and well-typed; otherwise, the JSONPath implementation MUST raise an error (see Section 2.1). To define which function expressions are well-typed, a type system is first introduced.¶
Each parameter and the result of a function extension must have a declared type.¶
Declared types enable checking a JSONPath query for well-typedness independent of any query argument the JSONPath query is applied to.¶
Table 13 defines the available types in terms of the instances they contain.¶
Type | Instances |
---|---|
ValueType
|
JSON values or Nothing
|
LogicalType
|
LogicalTrue or LogicalFalse
|
NodesType
|
Nodelists |
Notes:¶
ValueType
expressed as literals (which, in JSONPath, are limited to primitive values).¶
Nothing
represents the absence of a JSON value and is distinct from any JSON value, including null
.¶
LogicalTrue
and LogicalFalse
are unrelated to the JSON values expressed by the
literals true
and false
.¶
Just as queries can be used in logical expressions by testing for the
existence of at least one node (Section 2.3.5.2.1), a function expression of
declared type NodesType
can be used as a function argument for a
parameter of declared type LogicalType
, with the equivalent conversion rule:¶
LogicalTrue
.¶
LogicalFalse
.¶
Notes:¶
NodesType
to ValueType
may be surprising and has therefore not been defined.¶
NodesType
can
indirectly be used as an argument for a parameter of declared type
ValueType
by wrapping the expression in a call to a function extension,
such as value()
(see Section 2.4.8),
that takes a parameter of type NodesType
and returns a
result of type ValueType
.¶
The well-typedness of function expressions can now be defined in terms of this type system.¶
For a function expression to be well-typed:¶
Its declared type must be well-typed in the context in which it occurs.¶
As per the grammar, a function expression can occur in three different immediate contexts, which lead to the following conditions for well-typedness:¶
test-expr
in a logical expression:The function's declared result type is LogicalType
or
(giving rise to conversion as per Section 2.4.2) NodesType
.¶
comparable
in a comparison:The function's declared result type is ValueType
.¶
function-argument
in another function expression:The function's declared result type fulfills the following rules for the corresponding parameter of the enclosing function.¶
Its arguments must be well-typed for the declared type of the corresponding parameters.¶
The arguments of the function expression are well-typed when each argument of the function can be used for the declared type of the corresponding parameter, according to one of the following conditions:¶
When the declared type of the parameter is LogicalType
and the argument is one of the following:¶
NodesType
.
In this case, the argument is converted to LogicalType as per Section 2.4.2.¶
logical-expr
that is not a function expression.¶
NodesType
and the argument is a query
(which includes singular query).¶
When the declared type of the parameter is ValueType
and the argument is one of the following:¶
length()
Function Extension
The length()
function extension provides a way to compute the length
of a value and make that available for further processing in the
filter expression:¶
$[?length(@.authors) >= 5]¶
Its only argument is an instance of ValueType
(possibly taken from a
singular query, as in the example above). The result is also an
instance of ValueType
: an unsigned integer or the special result Nothing
.¶
Nothing
.¶
count()
Function Extension
The count()
function extension provides a way to obtain the number of
nodes in a nodelist and make that available for further processing in
the filter expression:¶
$[?count(@.*.author) >= 5]¶
Its only argument is a nodelist. The result is a value (an unsigned integer) that gives the number of nodes in the nodelist.¶
Notes:¶
match()
Function Extension
The match()
function extension provides a way to check whether (the
entirety of; see Section 2.4.7) a given
string matches a given regular expression, which is in the form described in [RFC9485].¶
$[?match(@.date, "1974-05-..")]¶
Its arguments are instances of ValueType
(possibly taken from a
singular query, as for the first argument in the example above).
If the first argument is not a string or the second argument is not a
string conforming to [RFC9485], the result is LogicalFalse
.
Otherwise, the string that is the first argument is matched against
the I-Regexp contained in the string that is the second argument;
the result is LogicalTrue
if the string matches the I-Regexp and is
LogicalFalse
otherwise.¶
search()
Function Extension
The search()
function extension provides a way to check whether a
given string contains a substring that matches a given regular
expression, which is in the form described in [RFC9485].¶
$[?search(@.author, "[BR]ob")]¶
Its arguments are instances of ValueType
(possibly taken from a
singular query, as for the first argument in the example above).
If the first argument is not a string or the second argument is not a
string conforming to [RFC9485], the result is LogicalFalse
.
Otherwise, the string that is the first argument is searched for a
substring that matches the I-Regexp contained in the string
that is the second argument; the result is LogicalTrue
if at
least one such substring exists and is LogicalFalse
otherwise.¶
value()
Function Extension
The value()
function extension provides a way to convert an instance of NodesType
to a value and
make that available for further processing in the filter expression:¶
$[?value(@..color) == "red"]¶
Its only argument is an instance of NodesType
(possibly taken from a
filter-query
, as in the example above). The result is an
instance of ValueType
.¶
Nothing
.¶
Note: A singular query may be used anywhere where a ValueType is expected,
so there is no need to use the value()
function extension with a singular query.¶
Query | Comment |
---|---|
$[?length(@) < 3]
|
well-typed |
$[?length(@.*) < 3]
|
not well-typed since @.* is a non-singular query |
$[?count(@.*) == 1]
|
well-typed |
$[?count(1) == 1]
|
not well-typed since 1 is not a query or function expression |
$[?count(foo(@.*)) == 1]
|
well-typed, where foo() is a function extension with a parameter of type NodesType and result type NodesType
|
$[?match(@.timezone, 'Europe/.*')]
|
well-typed |
$[?match(@.timezone, 'Europe/.*') == true]
|
not well-typed as LogicalType may not be used in comparisons |
$[?value(@..color) == "red"]
|
well-typed |
$[?value(@..color)]
|
not well-typed as ValueType may not be used in a test expression |
$[?bar(@.a)]
|
well-typed for any function bar() with a parameter of any declared type and result type LogicalType
|
$[?bnl(@.*)]
|
well-typed for any function bnl() with a parameter of declared type NodesType or LogicalType and result type LogicalType
|
$[?blt(1==1)]
|
well-typed, where blt() is a function with a parameter of declared type LogicalType and result type LogicalType
|
$[?blt(1)]
|
not well-typed for the same function blt() , as 1 is not a query, logical-expr , or function expression |
$[?bal(1)]
|
well-typed, where bal() is a function with a parameter of declared type ValueType and result type LogicalType
|
For each node in an input nodelist, segments apply one or more selectors to the node and concatenate the results of each selector into per-input-node nodelists, which are then concatenated in the order of the input nodelist to form a single segment result nodelist.¶
It turns out that the more segments there are in a query, the greater the depth in the input value of the nodes of the resultant nodelist:¶
There are two kinds of segments: child segments and descendant segments.¶
segment = child-segment / descendant-segment¶
The syntax and semantics of each kind of segment are defined below.¶
The child segment consists of a non-empty, comma-separated sequence of selectors enclosed in square brackets.¶
Shorthand notations are also provided for when there is a single wildcard or name selector.¶
child-segment = bracketed-selection / ("." (wildcard-selector / member-name-shorthand)) bracketed-selection = "[" S selector *(S "," S selector) S "]" member-name-shorthand = name-first *name-char name-first = ALPHA / "_" / %x80-D7FF / ; skip surrogate code points %xE000-10FFFF name-char = name-first / DIGIT DIGIT = %x30-39 ; 0-9 ALPHA = %x41-5A / %x61-7A ; A-Z / a-z¶
.*
, a child-segment
directly built from a wildcard-selector
, is
shorthand for [*]
.¶
.<member-name>
, a child-segment
built from a
member-name-shorthand
, is shorthand for ['<member-name>']
.
Note: This can only be used with member names that are composed of certain
characters, as specified in the ABNF rule member-name-shorthand
.
Thus, for example, $.foo.bar
is shorthand for $['foo']['bar']
(but not for $['foo.bar']
).¶
A child segment contains a sequence of selectors, each of which selects zero or more children of the input value.¶
Selectors of different kinds may be combined within a single child segment.¶
For each node in the input nodelist, the resulting nodelist of a child segment is the concatenation of the nodelists from each of its selectors in the order that the selectors appear in the list. Note: Any node matched by more than one selector is kept as many times in the nodelist.¶
Where a selector can produce a nodelist in more than one possible order, each occurrence of the selector in the child segment may produce a nodelist in a distinct order.¶
In summary, a child segment drills down one more level into the structure of the input value.¶
The descendant segment consists of a double dot ..
followed by a child segment (using bracket notation).¶
Shorthand notations are also provided that correspond to the shorthand forms of the child segment.¶
descendant-segment = ".." (bracketed-selection / wildcard-selector / member-name-shorthand)¶
..*
, the descendant-segment
directly built from a
wildcard-selector
, is shorthand for ..[*]
.¶
..<member-name>
, a descendant-segment
built from a
member-name-shorthand
, is shorthand for ..['<member-name>']
.
Note: As with the similar shorthand of a child-segment
, this can
only be used with member names that are composed of certain
characters, as specified in the ABNF rule member-name-shorthand
.¶
Note: On its own, ..
is not a valid segment.¶
A descendant segment produces zero or more descendants of an input value.¶
For each node in the input nodelist, a descendant selector visits the input node and each of its descendants such that:¶
The order in which the children of an object are visited is not stipulated, since JSON objects are unordered.¶
Suppose the descendant segment is of the form ..[<selectors>]
(after converting any shorthand
form to bracket notation),
and the nodes, in the order visited, are D1
, ..., Dn
(where n >= 1
).
Note: D1
is the input value.¶
For each i
such that 1 <= i <= n
, the nodelist Ri
is defined to be a result of applying
the child segment [<selectors>]
to the node Di
.¶
For each node in the input nodelist,
the result of the descendant segment is the concatenation of R1
,
..., Rn
(in that order).
These results are then concatenated in input nodelist order to form
the result of the segment.¶
In summary, a descendant segment drills down one or more levels into the structure of each input value.¶
JSON:¶
{ "o": {"j": 1, "k": 2}, "a": [5, 3, [{"j": 4}, {"k": 6}]] }¶
Queries:¶
(Note that the fourth example can be expressed in two equivalent queries, shown in Table 16 in one table row instead of two almost-identical rows.)¶
Query | Result | Result Paths | Comment |
---|---|---|---|
$..j
|
1 4
|
$['o']['j'] $['a'][2][0]['j']
|
Object values |
$..j
|
4 1
|
$['a'][2][0]['j'] $['o']['j']
|
Alternative result |
$..[0]
|
5 {"j": 4}
|
$['a'][0] $['a'][2][0]
|
Array values |
$..[*] or $..*
|
{"j": 1, "k": 2} [5, 3, [{"j": 4}, {"k": 6}]] 1 2 5 3 [{"j": 4}, {"k": 6}] {"j": 4} {"k": 6} 4 6
|
$['o'] $['a'] $['o']['j'] $['o']['k'] $['a'][0] $['a'][1] $['a'][2] $['a'][2][0] $['a'][2][1] $['a'][2][0]['j'] $['a'][2][1]['k']
|
All values |
$..o
|
{"j": 1, "k": 2}
|
$['o']
|
Input value is visited |
$.o..[*, *]
|
1 2 2 1
|
$['o']['j'] $['o']['k'] $['o']['k'] $['o']['j']
|
Non-deterministic ordering |
$.a..[0, 1]
|
5 3 {"j": 4} {"k": 6}
|
$['a'][0] $['a'][1] $['a'][2][0] $['a'][2][1]
|
Multiple segments |
Note: The ordering of the results for the $..[*]
and $..*
examples above is not guaranteed, except that:¶
{"j": 1, "k": 2}
must appear before 1
and 2
,¶
[5, 3, [{"j": 4}, {"k": 6}]]
must appear before 5
, 3
, and [{"j": 4}, {"k": 6}]
,¶
5
must appear before 3
, which must appear before [{"j": 4}, {"k": 6}]
,¶
5
and 3
must appear before {"j": 4}
, 4
, {"k": 6}
, and 6
,¶
[{"j": 4}, {"k": 6}]
must appear before {"j": 4}
and {"k": 6}
,¶
{"j": 4}
must appear before {"k": 6}
,¶
{"k": 6}
must appear before 4
, and¶
4
must appear before 6
.¶
The example above with the query $.o..[*, *]
shows that a selector may produce nodelists in distinct orders
each time it appears in the descendant segment.¶
The example above with the query $.a..[0, 1]
shows that the child segment [0, 1]
is applied to each node
in turn (rather than the nodes being visited once per selector, which is the case for some JSONPath implementations
that do not conform to this specification).¶
null
Note: JSON null
is treated the same as any other JSON value, i.e., it is not taken to mean "undefined" or "missing".¶
JSON:¶
{"a": null, "b": [null], "c": [{}], "null": 1}¶
Queries:¶
Query | Result | Result Paths | Comment |
---|---|---|---|
$.a
|
null
|
$['a']
|
Object value |
$.a[0]
|
null used as array |
||
$.a.d
|
null used as object |
||
$.b[0]
|
null
|
$['b'][0]
|
Array value |
$.b[*]
|
null
|
$['b'][0]
|
Array value |
$.b[?@]
|
null
|
$['b'][0]
|
Existence |
$.b[?@==null]
|
null
|
$['b'][0]
|
Comparison |
$.c[?@.d==null]
|
Comparison with "missing" value | ||
$.null
|
1
|
$['null']
|
Not JSON null at all, just a member name string |
A Normalized Path is a unique representation of the location of a node in a value that
uniquely identifies the node in the value.
Specifically, a Normalized Path is a JSONPath query with restricted syntax (defined below),
e.g., $['book'][3]
, which when applied to the value, results in a nodelist consisting
of just the node identified by the Normalized Path.
Note: A Normalized Path represents the identity of a node in a specific value.
There is precisely one Normalized Path identifying any particular node in a value.¶
A nodelist may be represented compactly in JSON as an array of strings, where the strings are Normalized Paths.¶
Normalized Paths provide a predictable format that simplifies testing and post-processing of nodelists, e.g., to remove duplicate nodes. Normalized Paths are used in this document as result paths in examples.¶
Normalized Paths use the canonical bracket notation, rather than dot notation.¶
Single quotes are used in Normalized Paths to delimit string member names. This reduces the number of characters that need escaping when Normalized Paths appear in strings delimited by double quotes, e.g., in JSON texts.¶
Certain characters are escaped in Normalized Paths in one and only one way; all other characters are unescaped.¶
normalized-path = root-identifier *(normal-index-segment) normal-index-segment = "[" normal-selector "]" normal-selector = normal-name-selector / normal-index-selector normal-name-selector = %x27 *normal-single-quoted %x27 ; 'string' normal-single-quoted = normal-unescaped / ESC normal-escapable normal-unescaped = ; omit %x0-1F control codes %x20-26 / ; omit 0x27 ' %x28-5B / ; omit 0x5C \ %x5D-D7FF / ; skip surrogate code points %xE000-10FFFF normal-escapable = %x62 / ; b BS backspace U+0008 %x66 / ; f FF form feed U+000C %x6E / ; n LF line feed U+000A %x72 / ; r CR carriage return U+000D %x74 / ; t HT horizontal tab U+0009 "'" / ; ' apostrophe U+0027 "\" / ; \ backslash (reverse solidus) U+005C (%x75 normal-hexchar) ; certain values u00xx U+00XX normal-hexchar = "0" "0" ( ("0" %x30-37) / ; "00"-"07" ; omit U+0008-U+000A BS HT LF ("0" %x62) / ; "0b" ; omit U+000C-U+000D FF CR ("0" %x65-66) / ; "0e"-"0f" ("1" normal-HEXDIG) ) normal-HEXDIG = DIGIT / %x61-66 ; "0"-"9", "a"-"f" normal-index-selector = "0" / (DIGIT1 *DIGIT) ; non-negative decimal integer¶
Since there can only be one Normalized Path identifying a given node, the syntax
stipulates which characters are escaped and which are not.
So the definition of normal-hexchar
is designed for hex escaping of characters
that are not straightforwardly printable, for example, U+000B LINE TABULATION, but
for which no standard JSON escape, such as \n
, is available.¶
Path | Normalized Path | Comment |
---|---|---|
$.a
|
$['a']
|
Object value |
$[1]
|
$[1]
|
Array index |
$[-3]
|
$[2]
|
Negative array index for an array of length 5 |
$.a.b[1:2]
|
$['a']['b'][1]
|
Nested structure |
$["\u000B"]
|
$['\u000b']
|
Unicode escape |
$["\u0061"]
|
$['a']
|
Unicode character |
IANA has registered the following media type [RFC6838]:¶
application¶
jsonpath¶
N/A¶
N/A¶
binary (UTF-8)¶
See the Security Considerations section of RFC 9535.¶
N/A¶
RFC 9535¶
Applications that need to convey queries in JSON data¶
N/A¶
COMMON¶
N/A¶
JSONPath WG¶
IETF¶
Per this specification, IANA has created a new "Function Extensions" subregistry in a new "JSONPath" registry. The "Function Extensions" subregistry has the policy "Expert Review" (Section 4.5 of [RFC8126]).¶
The experts are instructed to be frugal in the allocation of function extension names that are suggestive of generally applicable semantics, keeping them in reserve for functions that are likely to enjoy wide use and can make good use of their conciseness. The expert is also instructed to direct the registrant to provide a specification (Section 4.6 of [RFC8126]) but can make exceptions, for instance, when a specification is not available at the time of registration but is likely forthcoming. If the expert becomes aware of function extensions that are deployed and in use, they may also initiate a registration on their own if they deem such a registration can avert potential future collisions.¶
Each entry in the subregistry must include the following:¶
A lowercase ASCII [RFC0020] string that starts with a letter and can
contain letters, digits, and underscore characters afterwards
([a-z][_a-z0-9]*
). No other entry in the subregistry can have the
same function name.¶
A brief description¶
A comma-separated list of zero or more declared types, one for each of the arguments expected for this function extension¶
The declared type of the result for this function extension¶
See Section 2.3 of [RFC8126].¶
A reference document that provides a description of the function extension¶
The initial entries in this subregistry are listed in Table 19; the entries in the "Change Controller" column all have the value "IETF", and the entries in the "Reference" column all have the value "Section 2.4 of RFC 9535":¶
Function Name | Brief Description | Parameters | Result |
---|---|---|---|
length | length of string, array, or object |
ValueType
|
ValueType
|
count | size of nodelist |
NodesType
|
ValueType
|
match | regular expression full match |
ValueType , ValueType
|
LogicalType
|
search | regular expression substring match |
ValueType , ValueType
|
LogicalType
|
value | value of the single node in nodelist |
NodesType
|
ValueType
|
Security considerations for JSONPath can stem from:¶
Historically, JSONPath has often been implemented by feeding parts of
the query to an underlying programming language engine, e.g.,
JavaScript's eval()
function.
This approach is well known to lead to injection attacks and would
require perfect input validation to prevent these attacks (see
Section 12 of [RFC8259] for similar considerations for JSON itself).
Instead, JSONPath implementations need to implement the entire syntax
of the query without relying on the parsers of programming language
engines.¶
Attacks on availability may attempt to trigger unusually expensive runtime performance exhibited by certain implementations in certain cases. (See Section 10 of [RFC8949] for issues in hash-table implementations and Section 8 of [RFC9485] for performance issues in regular expression implementations.) Implementers need to be aware that good average performance is not sufficient as long as an attacker can choose to submit specially crafted JSONPath queries or query arguments that trigger surprisingly high, possibly exponential, CPU usage or, for example, via a naive recursive implementation of the descendant segment, stack overflow. Implementations need to have appropriate resource management to mitigate these attacks.¶
JSONPath queries are often not static but formed from variables that provide index values, member names, or values to compare with in a filter expression. These variables need to be validated (e.g., only allowing specific constructs such as .name to be formed when the given values allow that) and translated (e.g., by escaping string delimiters). Not performing these validations and translations correctly can lead to unexpected failures, which can lead to availability, confidentiality, and integrity breaches, in particular, if an adversary has control over the values (e.g., by entering them into a web form). The resulting class of attacks, injections (e.g., SQL injections), is consistently found among the top causes of application security vulnerabilities and requires particular attention.¶
Where JSONPath is used as a part of a security mechanism, attackers can attempt to provoke unexpected or unpredictable behavior or take advantage of differences in behavior between JSONPath implementations.¶
Unexpected or unpredictable behavior can arise from a query argument with certain constructs described as unpredictable by [RFC8259]. Predictable behavior can be expected, except in relation to the ordering of objects, for any query argument conforming with [RFC7493].¶
Other attacks can target the behavior of underlying technologies, such as UTF-8 (see Section 10 of [RFC3629]) and the Unicode character set.¶
This appendix collects the ABNF grammar from the ABNF passages used throughout the document.¶
Figure 2 contains the collected ABNF grammar that defines the syntax of a JSONPath query.¶
Figure 3 contains the collected ABNF grammar that
defines the syntax of a JSONPath Normalized Path while also using the rules
root-identifier
, ESC
, DIGIT
, and DIGIT1
from Figure 2.¶
This appendix is informative.¶
At the time JSONPath was invented, XML was noted for the availability of powerful tools to analyze, transform, and selectively extract data from XML documents. [XPath] is one of these tools.¶
In 2007, the need for something solving the same class of problems for the emerging JSON community became apparent, specifically for:¶
(Note: XPath has evolved since 2007, and recent versions even nominally support operating inside JSON values. This appendix only discusses the more widely used version of XPath that was available in 2007.)¶
JSONPath picks up the overall feeling of XPath but maps the concepts to syntax (and partially semantics) that would be familiar to someone using JSON in a dynamic language.¶
For example, in popular dynamic programming languages such as JavaScript, Python, and PHP, the semantics of the XPath expression:¶
/store/book[1]/title¶
can be realized in the expression:¶
x.store.book[0].title¶
or in bracket notation:¶
x['store']['book'][0]['title']¶
with the variable x holding the query argument.¶
The JSONPath language was designed to:¶
JSONPath expressions apply to JSON values in the same way
as XPath expressions are used in combination with an XML document.
JSONPath uses $
to refer to the root node of the query argument, similar
to XPath's /
at the front.¶
JSONPath expressions move further down the hierarchy using dot notation
($.store.book[0].title
)
or the bracket notation
($['store']['book'][0]['title']
); both replace XPath's /
within query expressions, where dot notation serves as a lightweight but limited syntax while bracket notation is a
heavyweight but more general syntax.¶
Both JSONPath and XPath use *
for a wildcard.
JSONPath's descendant segment notation, starting with ..
, borrowed from [E4X], is similar to XPath's //
.
The array slicing construct [start:end:step]
is unique to JSONPath,
inspired by [SLICE] from ECMASCRIPT 4.¶
Filter expressions are supported via the syntax ?<logical-expr>
as in:¶
$.store.book[?@.price < 10].title¶
Table 20 extends Table 1 by providing a comparison with similar XPath concepts.¶
XPath | JSONPath | Description |
---|---|---|
/
|
$
|
the root XML element |
.
|
@
|
the current XML element |
/
|
. or []
|
child operator |
..
|
n/a | parent operator |
//
|
..name , ..[index] , ..* , or ..[*]
|
descendants (JSONPath borrows this syntax from E4X) |
*
|
*
|
wildcard: All XML elements regardless of their names |
@
|
n/a | attribute access: JSON values do not have attributes |
[]
|
[]
|
subscript operator used to iterate over XML element collections and for predicates |
|
|
[,]
|
Union operator (results in a combination of node sets); called list operator in JSONPath, allows combining member names, array indices, and slices |
n/a |
[start:end:step]
|
array slice operator borrowed from ES4 |
[]
|
?
|
applies a filter (script) expression |
seamless | n/a | expression engine |
()
|
n/a | grouping |
For further illustration, Table 21 shows some XPath expressions and their JSONPath equivalents.¶
XPath | JSONPath | Result |
---|---|---|
/store/book/author
|
$.store.book[*].author
|
the authors of all books in the store |
//author
|
$..author
|
all authors |
/store/*
|
$.store.*
|
all things in store, which are some books and a red bicycle |
/store//price
|
$.store..price
|
the prices of everything in the store |
//book[3]
|
$..book[2]
|
the third book |
//book[last()]
|
$..book[-1]
|
the last book in order |
//book[position()<3]
|
$..book[0,1] $..book[:2]
|
the first two books |
//book[isbn]
|
$..book[?@.isbn]
|
filter all books with an ISBN number |
//book[price<10]
|
$..book[?@.price<10]
|
filter all books cheaper than 10 |
//*
|
$..*
|
all elements in an XML document; all member values and array elements contained in input value |
XPath has a lot more functionality (location paths in unabbreviated syntax, operators, and functions) than listed in this comparison. Moreover, there are significant differences in how the subscript operator works in XPath and JSONPath:¶
This appendix is informative.¶
In relation to JSON Pointer [RFC6901], JSONPath is not intended as a replacement but as a more powerful companion. The purposes of the two standards are different.¶
JSON Pointer is for identifying a single value within a JSON value whose structure is known.¶
JSONPath can identify a single value within a JSON value, for example, by using a Normalized Path. But JSONPath is also a query syntax that can be used to search for and extract multiple values from JSON values whose structure is known only in a general way.¶
A Normalized JSONPath can be converted into a JSON Pointer by converting the syntax, without knowledge of any JSON value. The inverse is not generally true, i.e., a numeric reference token (path component) in a JSON Pointer may identify a member value of an object or an element of an array. For conversion to a JSONPath query, knowledge of the structure of the JSON value is needed to distinguish these cases.¶
This document is based on Stefan Gössner's original online article defining JSONPath [JSONPath-orig].¶
The books example was taken from course material that Bielefeld University, Germany used in 2002.¶
This work is indebted to Christoph Burgmer for the superb JSONPath comparison project [COMPARISON] that details the behavior of over forty JSONPath implementations applied to numerous queries.¶