3.1 library(semweb/rdf11): The RDF database

The library(semweb/rdf11) provides a new interface to the SWI-Prolog RDF database based on the RDF 1.1 specification.

3.1.1 Query the RDF database

[nondet]rdf(?S, ?P, ?O)

[nondet]rdf(?S, ?P, ?O, ?G)

True if an RDF triple <S,P,O> exists, optionally in the graph G. The object O is either a resource (atom) or one of the terms listed below. The described types apply for the case where O is unbound. If O is instantiated it is converted according to the rules described with rdf_assert/3.

Triples consist of the following three terms:

• Blank nodes are encoded by atoms that start with‘_:`.
• IRIs appear in two notations:
  • Full IRIs are encoded by atoms that do not start with‘_:`. Specifically, an IRI term is not required to follow the IRI standard grammar.
  • Abbreviated IRI notation that allows IRI prefix aliases that are registered by rdf_register_prefix/[2,3] to be used. Their notation is Alias:Local, where Alias and Local are atoms. Each abbreviated IRI is expanded by the system to a full IRI.

• Literals appear in two notations:

  String @ Lang

  A language-tagged string, where String is a Prolog string and Lang is an atom.

  Value ^^ Type

  A type qualified literal. For unknown types, Value is a Prolog string. If type is known, the Prolog representations from the table below are used.

  Datatype IRI	Prolog term
  xsd:float	float
  xsd:double	float
  xsd:decimal	float (1)
  xsd:integer	integer
  XSD integer sub-types	integer
  xsd:boolean	true or false
  xsd:date	date(Y,M,D)
  xsd:dateTime	date_time(Y,M,D,HH,MM,SS) (2,3)
  xsd:gDay	integer
  xsd:gMonth	integer
  xsd:gMonthDay	month_day(M,D)
  xsd:gYear	integer
  xsd:gYearMonth	year_month(Y,M)
  xsd:time	time(HH,MM,SS) (2)

Notes:

(1) The current implementation of xsd:decimal values as floats is formally incorrect. Future versions of SWI-Prolog may introduce decimal as a subtype of rational.

(2) SS fields denote the number of seconds. This can either be an integer or a float.

(3) The date_time structure can have a 7th field that denotes the timezone offset in seconds as an integer.

In addition, a ground object value is translated into a properly typed RDF literal using rdf_canonical_literal/2.

There is a fine distinction in how duplicate statements are handled in rdf/[3,4]: backtracking over rdf/3 will never return duplicate triples that appear in multiple graphs. rdf/4 will return such duplicate triples, because their graph term differs.

S	is the subject term. It is either a blank node or IRI.
P	is the predicate term. It is always an IRI.
O	is the object term. It is either a literal, a blank node or IRI (except for true and false that denote the values of datatype XSD boolean).
G	is the graph term. It is always an IRI.

See also

- Triple pattern querying
- xsd_number_string/2 and xsd_time_string/3 are used to convert between lexical representations and Prolog terms.

[nondet]rdf_has(?S, +P, ?O)

[nondet]rdf_has(?S, +P, ?O, -RealP)

Similar to rdf/3 and rdf/4, but P matches all predicates that are defined as an rdfs:subPropertyOf of P. This predicate also recognises the predicate properties inverse_of and symmetric. See rdf_set_predicate/2.

[nondet]rdf_reachable(?S, +P, ?O)

[nondet]rdf_reachable(?S, +P, ?O, +MaxD, -D)

True when O can be reached from S using the transitive closure of P. The predicate uses (the internals of) rdf_has/3 and thus matches both rdfs:subPropertyOf and the inverse_of and symmetric predicate properties. The version rdf_reachable/5 maximizes the steps considered and returns the number of steps taken.

If both S and O are given, these predicates are semidet. The number of steps D is minimal because the implementation uses breadth first search.

Constraints on literal values

[semidet]{}(+Where)

[semidet]rdf_where(+Where)

Formulate constraints on RDF terms, notably literals. These are intended to be used as illustrated below. RDF constraints are pure: they may be placed before, after or inside a graph pattern and, provided the code contains no commit operations (!, ->), the semantics of the goal remains the same. Preferably, constraints are placed before the graph pattern as they often help the RDF database to exploit its literal indexes. In the example below, the database can choose between using the subject and/or predicate hash or the ordered literal table.

    { Date >= "2000-01-01"^^xsd:date },
    rdf(S, P, Date)

The following constraints are currently defined:

> , >=,==,=<,<

The comparison operators are defined between numbers (of any recognised type), typed literals of the same type and langStrings of the same language.

prefix(String, Pattern)

substring(String, Pattern)

word(String, Pattern)

like(String, Pattern)

icase(String, Pattern)

Text matching operators that act on both typed literals and langStrings.

lang_matches(Term, Pattern)

Demands a full RDF term (Text@Lang) or a plain Lang term to match the language pattern Pattern.

The predicates rdf_where/1 and {}/1 are identical. The rdf_where/1 variant is provided to avoid ambiguity in applications where {}/1 is used for other purposes. Note that it is also possible to write rdf11:{...}.

3.1.2 Enumerating and testing objects

Enumerating objects by role

[nondet]rdf_subject(?S)

True when S is a currently known subject, i.e. it appears in the subject position of some visible triple. The predicate is semidet if S is ground.

[nondet]rdf_predicate(?P)

True when P is a currently known predicate, i.e. it appears in the predicate position of some visible triple. The predicate is semidet if P is ground.

[nondet]rdf_object(?O)

True when O is a currently known object, i.e. it appears in the object position of some visible triple. If Term is ground, it is pre-processed as the object argument of rdf_assert/3 and the predicate is semidet.

[nondet]rdf_node(?T)

True when T appears in the subject or object position of a known triple, i.e., is a node in the RDF graph.

[nondet]rdf_graph(?Graph)

True when Graph is an existing graph.

Enumerating objects by type

[nondet]rdf_literal(?Term)

True if Term is a known literal. If Term is ground, it is pre-processed as the object argument of rdf_assert/3 and the predicate is semidet.

[nondet]rdf_bnode(?BNode)

True if BNode is a currently known blank node. The predicate is semidet if BNode is ground.

[nondet]rdf_iri(?IRI)

True if IRI is a current IRI. The predicate is semidet if IRI is ground.

[nondet]rdf_name(?Name)

True if Name is a current IRI or literal. The predicate is semidet if Name is ground.

[nondet]rdf_term(?Term)

True if Term appears in the RDF database. Term is either an IRI, literal or blank node and may appear in any position of any triple. If Term is ground, it is pre-processed as the object argument of rdf_assert/3 and the predicate is semidet.

Testing objects types

[semidet]rdf_is_iri(@IRI)

True if IRI is an RDF IRI term.

For performance reasons, this does not check for compliance to the syntax defined in RFC 3987. This checks whether the term is (1) an atom and (2) not a blank node identifier.

Success of this goal does not imply that the IRI is present in the database (see rdf_iri/1 for that).

[semidet]rdf_is_bnode(@Term)

True if Term is an RDF blank node identifier.

A blank node is represented by an atom that starts with _:.

Success of this goal does not imply that the blank node is present in the database (see rdf_bnode/1 for that).

For backwards compatibility, atoms that are represented with an atom that starts with __ are also considered to be a blank node.

[semidet]rdf_is_literal(@Term)

True if Term is an RDF literal term.

An RDF literal term is of the form String@LanguageTag or Value^^Datatype.

Success of this goal does not imply that the literal is well-formed or that it is present in the database (see rdf_literal/1 for that).

[semidet]rdf_is_name(@Term)

True if Term is an RDF Name, i.e., an IRI or literal.

Success of this goal does not imply that the name is well-formed or that it is present in the database (see rdf_name/1 for that).

[semidet]rdf_is_object(@Term)

True if Term can appear in the object position of a triple.

Success of this goal does not imply that the object term in well-formed or that it is present in the database (see rdf_object/1 for that).

Since any RDF term can appear in the object position, this is equaivalent to rdf_is_term/1.

[semidet]rdf_is_predicate(@Term)

True if Term can appear in the predicate position of a triple.

Success of this goal does not imply that the predicate term is present in the database (see rdf_predicate/1 for that).

Since only IRIs can appear in the predicate position, this is equivalent to rdf_is_iri/1.

[semidet]rdf_is_subject(@Term)

True if Term can appear in the subject position of a triple.

Only blank nodes and IRIs can appear in the subject position.

Success of this goal does not imply that the subject term is present in the database (see rdf_subject/1 for that).

Since blank nodes are represented by atoms that start with‘_:` and an IRIs are atoms as well, this is equivalent to atom(Term).

[semidet]rdf_is_term(@Term)

True if Term can be used as an RDF term, i.e., if Term is either an IRI, a blank node or an RDF literal.

Success of this goal does not imply that the RDF term is present in the database (see rdf_term/1 for that).

3.1.3 RDF literals

[det]rdf_canonical_literal(++In, -Literal)

Transform a relaxed literal specification as allowed for rdf_assert/3 into its canonical form. The following Prolog terms are translated:

Prolog Term	Datatype IRI
float	xsd:double
integer	xsd:integer
string	xsd:string
true or false	xsd:boolean
date(Y,M,D)	xsd:date
date_time(Y,M,D,HH,MM,SS)	xsd:dateTime
date_time(Y,M,D,HH,MM,SS,TZ)	xsd:dateTime
month_day(M,D)	xsd:gMonthDay
year_month(Y,M)	xsd:gYearMonth
time(HH,MM,SS)	xsd:time

For example:

?- rdf_canonical_literal(42, X).
X = 42^^'http://www.w3.org/2001/XMLSchema#integer'.

[det]rdf_lexical_form(++Literal, -Lexical:compound)

True when Lexical is the lexical form for the literal Literal. Lexical is of one of the forms below. The ntriples serialization is obtained by transforming String into a proper ntriples string using double quotes and escaping where needed and turning Type into a proper IRI reference.

• String^^Type
• String@Lang

[det]rdf_compare(-Diff, +Left, +Right)

True if the RDF terms Left and Right are ordered according to the comparison operator Diff. The ordering is defines as:

• Literal < BNode < IRI
• For literals
  • Numeric < non-numeric
  • Numeric literals are ordered by value. If both are equal, floats are ordered before integers.
  • Other data types are ordered lexicographically.

• BNodes and IRIs are ordered lexicographically.

Note that this ordering is a complete ordering of RDF terms that is consistent with the partial ordering defined by SPARQL.

Diff	is one of <, = or >

3.1.4 Accessing RDF graphs

[det]rdf_default_graph(-Graph)

[det]rdf_default_graph(-Old, +New)

Query/set the notion of the default graph. The notion of the default graph is local to a thread. Threads created inherit the default graph from their creator. See set_prolog_flag/2.

3.1.5 Modifying the RDF store

[det]rdf_assert(+S, +P, +O)

[det]rdf_assert(+S, +P, +O, +G)

Assert a new triple. If O is a literal, certain Prolog terms are translated to typed RDF literals. These conversions are described with rdf_canonical_literal/2.

If a type is provided using Value^^Type syntax, additional conversions are performed. All types accept either an atom or Prolog string holding a valid RDF lexical value for the type and xsd:float and xsd:double accept a Prolog integer.

[nondet]rdf_retractall(?S, ?P, ?O)

[nondet]rdf_retractall(?S, ?P, ?O, ?G)

Remove all matching triples from the database. Matching is performed using the same rules as rdf/3. The call does not instantiate any of its arguments.

rdf_create_bnode(--BNode)

Create a new BNode. A blank node is an atom starting with _:. Blank nodes generated by this predicate are of the form _:genid followed by a unique integer.

3.1.6 Accessing RDF collections

The following predicates are utilities to access RDF 1.1 collections. A collection is a linked list created from rdf:first and rdf:next triples, ending in rdf:nil.

[det]rdf_last(+RDFList, -Last)

True when Last is the last element of RDFList. Note that if the last cell has multiple rdf:first triples, this predicate becomes nondet.

[semidet]rdf_list(?RDFTerm)

True if RDFTerm is a proper RDF list. This implies that every node in the list has an rdf:first and rdf:rest property and the list ends in rdf:nil.

If RDFTerm is unbound, RDFTerm is bound to each maximal RDF list. An RDF list is maximal if there is no triple rdf(_, rdf:rest, RDFList).

[det]rdf_list(+RDFList, -PrologList)

True when PrologList represents the rdf:first objects for all cells in RDFList. Note that this can be non-deterministic if cells have multiple rdf:first or rdf:rest triples.

[nondet]rdf_length(+RDFList, -Length:nonneg)

True when Length is the number of cells in RDFList. Note that a list cell may have multiple rdf:rest triples, which makes this predicate non-deterministic. This predicate does not check whether the list cells have associated values (rdf:first). The list must end in rdf:nil.

[nondet]rdf_member(?Member, +RDFList)

True when Member is a member of RDFList

[nondet]rdf_nth0(?Index, +RDFList, ?X)

[nondet]rdf_nth1(?Index, +RDFList, ?X)

True when X is the Index-th element (0-based or 1-based) of RDFList. This predicate is deterministic if Index is given and the list has no multiple rdf:first or rdf:rest values.

[det]rdf_assert_list(+PrologList, ?RDFList)

[det]rdf_assert_list(+PrologList, ?RDFList, +Graph)

Create an RDF list from the given Prolog List. PrologList must be a proper Prolog list and all members of the list must be acceptable as object for rdf_assert/3. If RDFList is unbound and PrologList is not empty, rdf_create_bnode/1 is used to create RDFList.

[det]rdf_retract_list(+RDFList)

Retract the rdf:first, rdf:rest and rdf:type=rdf:'List' triples from all nodes reachable through rdf:rest. Note that other triples that exist on the nodes are left untouched.