Reference: 4.1 RDF Term Syntax
This section discusses IRIs, variables, blank nodes, and literals in Flint and SPARQL. Many of the conventions here, including for prefixed IRIs, variables, and blank nodes, were borrowed from the Datomic query and update grammar.
NOTE: As of v0.2.0 Flint supports Unicode. Certain Unicode characters are forbidden by the SPARQL grammar in symbols and keywords (e.g. the Greek question mark and most mathematical symbols). For details on allowed characters, see the SPARQL grammar section on literals.
Flint does not support non-Basic Multilingual Plane code points (i.e. those with a value above 0xFFFF
) in symbols and keywords, whether directly or as surrogate code points (the latter being disallowed in the SPARQL grammar).
Internationalized Resource Identifiers (IRIs) and their subset Universal Resource Identifiers (URIs) are represented in two ways: as full IRIs or as prefixed IRIs.
Examples: <http://absolute-iri-example.com/>
, <relative-iri>
, (java.net.URL. <http://foo.org>)
Full IRIs in Flint are written as one of the following:
<my-iri-string>
. The string inside the angle bracket pair can include any characters except for whitespace, ^
, <
, >
, "
, \
, |
, or `
. Translating to SPARQL does not affect full IRIs.java.net.URI
instances (Clojure). The inner string must follow the above restrictions, i.e. no whitespace, <
, >
, etc.js/URL
objects (ClojureScript). The inner string must follow the above restrictions.NOTE: This can mean that any string can become a IRI in Flint, though in practice they should conform to the specification for IRIs after expansion.
Examples: :my-prefix/foo
, :bar
Prefixed IRIs in Flint are written as keywords of the form :prefix/name
, where the prefix is optional. When translating to SPARQL, prefixed IRIs are transformed into the form prefix:name
.
NOTE: Digits not allowed as the first character of a prefix keyword namespace. The .
character is not allowed as the first or last character of the keyword namespace or name.
NOTE: Prefixed IRIs must have the prefix be an entry in the :prefixes
map; otherwise, validation will fail (unless :validate?
is set to false
).
a
Examples: a
and :a
.
There is a third IRI representation allowed in Flint for predicates: the keyword :a
/symbol a
. This is provided as syntactic sugar to compactly represent the IRI:
<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
Translating to SPARQL does not affect :a
/a
other than stringifying it.
Therefore, in the the following query:
{:prefixes {:rdf "<http://www.w3.org/1999/02/22-rdf-syntax-ns#>"}
:ask []
:where [[?x "<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>" ?y]
[?x :rdf/type ?y]
[?x :a ?y]
[?x a ?y]]}
the :where
triples are all equivalent.
Examples: ?var
Variables are written as symbols prefixed with a question mark ?
. Translating to SPARQL does not change the variable other than stringifying it.
Examples: _
, _b0
Blank nodes are written as symbols prefixed with an underscore _
. When translating a blank node to SPARQL, a colon is added after the underscore, e.g. _:b0
. The exception if the symbol is a single _
character; then it is rewritten as []
instead.
NOTE: The .
character is not permitted as the first or last character after the _
.
NOTE: Blank nodes have certain restrictions: they cannot be used in any delete-related clauses, nor can the same blank node be repeated across different basic graph patterns or SPARQL updates.
Examples: *
and :*
The wildcard is used in certain query clauses and expressions in order to "return everything". It can be written as either a symbol or keyword.
Flint supports the following literals by default: simple strings, language-tagged strings, numbers, booleans, and timestamps.
During SPARQL translation, an IRI denoting the datatype is added as a suffix
if :force-iris?
is true
or if it is not a primitive type (e.g. timestamps). For example, the following is a stringified dateTime
timestamp with an appended datatype IRI:
"2022-01-01T10:10:10Z"^^<http://www.w3.org/2001/XMLSchema#dateTime>
The user can shorten the IRI string by including an entry for <http://www.w3.org/2001/XMLSchema#>
in their prefixes map. For example, if the :prefixes
map has the following:
{:xsd "<http://www.w3.org/2001/XMLSchema#>"}
then that IRI prefix is associated with :xsd
and the formatted literal becomes:
"2022-01-01T10:10:10Z"^^xsd:dateTime
Examples: 0
, -2
, 3.14
Numbers cover both integers and doubles, which are represented as integer and double literals in SPARQL, respectively. In addition to primitive values, java.math.BigInteger
, java.math.BigDecimal
, and clojure.lang.BigInt
classes are also supported in Clojure.
Numbers are not transformed during SPARQL translation beyond stringification. Note that the datatype IRI will vary on the numeric value's underlying type (e.g. Clojure integers, as Java Long
values, will be associated with the IRI xsd:long
by default).
Examples: true
and false
Booleans are not transformed during SPARQL translation beyond stringification.
Examples: "Hello World!"
, "你好世界"
, "cat: \\\"meow\\\""
, "foo\\nbar"
String literals can contain any characters except unescaped line breaks, carriage returns, backslashes, or double quotes; this is in order to prevent SPARQL injection attacks. (Therefore strings like "cat: \"meow\"
and "foo\nbar"
are not allowed.) Strings are not transformed during SPARQL translation.
Examples: {:en "Hello World!"}
, {:zh "你好世界"}
Strings with language tags are represented by a map between one language tag keyword and the string literal (which follows the same restrictions as simple strings). Note that this is the only literal that cannot have a datatype IRI appended, even if :force-iris?
is true
.
Examples: #inst "2022-01-01T10:10:10Z"
Timestamp literal values include the following classes:
java.time.Temporal
classes (Clojure). These include (in the java.time
package) Instant
, ZonedDateTime
, OffsetDateTime
, and LocalDateTime
. In addition, the OffsetTime
, LocalTime
, and LocalDate
classes are supported if the user only wants to represented time or date information.java.util.Date
classes (Clojure). These include, in addition to that class itself, its java.sql
package subclasses Timestamp
, Date
, and Time
.js/Date
(ClojureScript).As mentioned above, timestamps will be stringified as ISO 8601 timestamps and will have (depending on the class) the xsd:dateTime
, xsd:date
, or xsd:time
IRI appended, regardless of the value of :force-iris?
.
A user can implement a custom literal by extending com.yetanalytics.flint.axiom.protocol/Literal
and defining the following:
-valid-literal?
to validate the value.-format-literal
to format the entire string.-format-literal-strval
, -format-literal-lang-tag
, and -format-literal-url
to format individual aspects of the literal value.Note that -format-literal
and -format-literal-url
also accept an opts
map for optional arguments that affect formatting. Currently implement literals accept the args :force-iri?
(to force-append datatype IRIs) and :iri-prefix-m
(a map from base IRI strings to prefixes to shorten datatype IRIs).
Here is an example of an implementation of a Rational
literal (inspired by an example given by the Apache Jena documentation):
(require '[com.yetanalytics.flint.axiom.protocol :as p])
(defrecord Rational [numerator denominator]
p/Literal
(p/-valid-literal? [_rational]
(and (int? numerator)
(int? denominator)
(not (zero? denominator))))
(p/-format-literal [rational]
(p/-format-literal rational {}))
(p/-format-literal [rational opts]
(str "\"" (p/-format-literal-strval rational)
"\"^^" (p/-format-literal-url rational opts)))
(p/-format-literal-strval [_rational]
(str numerator "/" denominator))
(p/-format-literal-lang-tag [_rational]
nil)
(p/-format-literal-url [rational]
(p/-format-literal-url rational {}))
(p/-format-literal-url [_rational {:keys [iri-prefix-m]}]
(if-some [prefix (get iri-prefix-m "http://foo.org/literals#")]
(str (name prefix) ":rational")
"<http://foo.org/literals#rational>")))
which can then be used in a SPARQL query as so:
(def rational-value (map->Rational {:numerator 5 :denominator 6}))
{:prefixes {:foo "<http://foo.org/literals#>"}
:select ['?x]
:where [['?x '?y rational-value]]}}
and thus becomes:
PREFIX foo: <http://foo.org/literals#>
SELECT ?x
WHERE {
?x ?y \"5/6\"^^foo:rational .
}
Alternately you can extend a pre-existing Clojure(Script) type using extend-protocol
or extend-type
, e.g. with Clojure's Ratio
type:
(extend-protocol p/Literal
clojure.lang.Ratio
(p/-valid-literal?
[ratio]
(not= java.math.BigInteger/ZERO (.denominator ratio)))
(p/-format-literal
([ratio]
(p/-format-literal ratio {}))
([ratio opts]
(str "\"" (p/-format-literal-strval ratio)
"\"^^" (p/-format-literal-url ratio opts))))
(p/-format-literal-strval
[ratio]
(str (.numerator ratio) "/" (.denominator ratio)))
(p/-format-literal-lang-tag
[_ratio]
nil)
(p/-format-literal-url
([ratio]
(p/-format-literal-url ratio {}))
([_ratio {:keys [iri-prefix-m]}]
(if-some [prefix (get iri-prefix-m "http://foo.org/literals#")]
(str (name prefix) ":ratio")
"<http://foo.org/literals#ratio>"))))
NOTE: A user can also extend other protocols in the flint.axiom.protocol
namespace, e.g. IRI
or PrefixedIRI
to create custom implementations of other SPARQL values.
Can you improve this documentation?Edit on GitHub
cljdoc is a website building & hosting documentation for Clojure/Script libraries
× close