use crate ::io ::PyFileLike ;
use crate ::model ::* ;
use crate ::sparql ::* ;
use crate ::store_utils ::* ;
use oxigraph ::io ::{ DatasetFormat , GraphFormat } ;
use oxigraph ::model ::GraphNameRef ;
use oxigraph ::store ::memory ::* ;
use pyo3 ::basic ::CompareOp ;
use pyo3 ::exceptions ::{ PyNotImplementedError , PyValueError } ;
use pyo3 ::prelude ::{
pyclass , pymethods , pyproto , Py , PyAny , PyCell , PyObject , PyRef , PyRefMut , PyResult , Python ,
} ;
use pyo3 ::{ PyIterProtocol , PyObjectProtocol , PySequenceProtocol } ;
use std ::convert ::TryFrom ;
use std ::io ::BufReader ;
/// In-memory store.
/// It encodes a `RDF dataset <https://www.w3.org/TR/rdf11-concepts/#dfn-rdf-dataset>`_ and allows to query it using SPARQL.
///
/// The :py:func:`str` function provides a serialization of the store in NQuads:
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g')))
/// >>> str(store)
/// '<http://example.com> <http://example.com/p> "1" <http://example.com/g> .\n'
#[ pyclass(name = " MemoryStore " , module = " oxigraph " ) ]
#[ derive(Eq, PartialEq, Clone) ]
#[ text_signature = " () " ]
pub struct PyMemoryStore {
inner : MemoryStore ,
}
#[ pymethods ]
impl PyMemoryStore {
#[ new ]
fn new ( ) -> Self {
Self {
inner : MemoryStore ::new ( ) ,
}
}
/// Adds a quad to the store
///
/// :param quad: the quad to add
/// :type quad: Quad
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g')))
/// >>> list(store)
/// [<Quad subject=<NamedNode value=http://example.com> predicate=<NamedNode value=http://example.com/p> object=<Literal value=1 datatype=<NamedNode value=http://www.w3.org/2001/XMLSchema#string>> graph_name=<NamedNode value=http://example.com/g>>]
#[ text_signature = " ($self, quad) " ]
fn add ( & self , quad : PyQuad ) {
self . inner . insert ( quad )
}
/// Removes a quad from the store
///
/// :param quad: the quad to remove
/// :type quad: Quad
///
/// >>> store = MemoryStore()
/// >>> quad = Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g'))
/// >>> store.add(quad)
/// >>> store.remove(quad)
/// >>> list(store)
/// []
#[ text_signature = " ($self, quad) " ]
fn remove ( & self , quad : & PyQuad ) {
self . inner . remove ( quad )
}
/// Looks for the quads matching a given pattern
///
/// :param subject: the quad subject or :py:const:`None` to match everything.
/// :type subject: NamedNode or BlankNode or None
/// :param predicate: the quad predicate or :py:const:`None` to match everything.
/// :type predicate: NamedNode or None
/// :param object: the quad object or :py:const:`None` to match everything.
/// :type object: NamedNode or BlankNode or Literal or None
/// :param graph: the quad graph name. To match only the default graph, use :py:class:`DefaultGraph`. To match everything use :py:const:`None`.
/// :type graph: NamedNode or BlankNode or DefaultGraph or None
/// :return: an iterator of the quads matching the pattern
/// :rtype: iter(Quad)
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g')))
/// >>> list(store.quads_for_pattern(NamedNode('http://example.com'), None, None, None))
/// [<Quad subject=<NamedNode value=http://example.com> predicate=<NamedNode value=http://example.com/p> object=<Literal value=1 datatype=<NamedNode value=http://www.w3.org/2001/XMLSchema#string>> graph_name=<NamedNode value=http://example.com/g>>]
#[ text_signature = " ($self, subject, predicate, object, graph_name = None) " ]
fn quads_for_pattern (
& self ,
subject : & PyAny ,
predicate : & PyAny ,
object : & PyAny ,
graph_name : Option < & PyAny > ,
) -> PyResult < QuadIter > {
let ( subject , predicate , object , graph_name ) =
extract_quads_pattern ( subject , predicate , object , graph_name ) ? ;
Ok ( QuadIter {
inner : self . inner . quads_for_pattern (
subject . as_ref ( ) . map ( | p | p . into ( ) ) ,
predicate . as_ref ( ) . map ( | p | p . into ( ) ) ,
object . as_ref ( ) . map ( | p | p . into ( ) ) ,
graph_name . as_ref ( ) . map ( | p | p . into ( ) ) ,
) ,
} )
}
/// Executes a `SPARQL 1.1 query <https://www.w3.org/TR/sparql11-query/>`_.
///
/// :param query: the query to execute
/// :type query: str
/// :param use_default_graph_as_union: if the SPARQL query should look for triples in all the dataset graphs by default (i.e. without `GRAPH` operations). Disabled by default.
/// :type use_default_graph_as_union: bool, optional
/// :param default_graph: list of the graphs that should be used as the query default graph. By default, the store default graph is used.
/// :type default_graph: NamedNode or BlankNode or DefaultGraph or list(NamedNode or BlankNode or DefaultGraph) or None, optional
/// :param named_graphs: list of the named graphs that could be used in SPARQL `GRAPH` clause. By default, all the store named graphs are available.
/// :type named_graphs: list(NamedNode or BlankNode) or None, optional
/// :return: a :py:class:`bool` for ``ASK`` queries, an iterator of :py:class:`Triple` for ``CONSTRUCT`` and ``DESCRIBE`` queries and an iterator of :py:class:`QuerySolution` for ``SELECT`` queries.
/// :rtype: QuerySolutions or QueryTriples or bool
/// :raises SyntaxError: if the provided query is invalid
///
/// ``SELECT`` query:
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1')))
/// >>> list(solution['s'] for solution in store.query('SELECT ?s WHERE { ?s ?p ?o }'))
/// [<NamedNode value=http://example.com>]
///
/// ``CONSTRUCT`` query:
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1')))
/// >>> list(store.query('CONSTRUCT WHERE { ?s ?p ?o }'))
/// [<Triple subject=<NamedNode value=http://example.com> predicate=<NamedNode value=http://example.com/p> object=<Literal value=1 datatype=<NamedNode value=http://www.w3.org/2001/XMLSchema#string>>>]
///
/// ``ASK`` query:
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1')))
/// >>> store.query('ASK { ?s ?p ?o }')
/// True
#[ text_signature = " ($self, query, *, use_default_graph_as_union, default_graph, named_graphs) " ]
#[ args(
query ,
"*" ,
use_default_graph_as_union = "false" ,
default_graph = "None" ,
named_graphs = "None"
) ]
fn query (
& self ,
query : & str ,
use_default_graph_as_union : bool ,
default_graph : Option < & PyAny > ,
named_graphs : Option < & PyAny > ,
py : Python < ' _ > ,
) -> PyResult < PyObject > {
let query = parse_query (
query ,
use_default_graph_as_union ,
default_graph ,
named_graphs ,
) ? ;
let results = self . inner . query ( query ) . map_err ( map_evaluation_error ) ? ;
query_results_to_python ( py , results )
}
/// Executes a `SPARQL 1.1 update <https://www.w3.org/TR/sparql11-update/>`_.
///
/// :param update: the update to execute
/// :type update: str
/// :raises SyntaxError: if the provided update is invalid
///
/// The store does not track the existence of empty named graphs.
/// This method has no ACID guarantees.
///
/// ``INSERT DATA`` update:
///
/// >>> store = MemoryStore()
/// >>> store.update('INSERT DATA { <http://example.com> <http://example.com/p> "1" }')
/// >>> list(store)
/// [<Quad subject=<NamedNode value=http://example.com> predicate=<NamedNode value=http://example.com/p> object=<Literal value=1 datatype=<NamedNode value=http://www.w3.org/2001/XMLSchema#string>> graph_name=<DefaultGraph>>]
///
/// ``DELETE DATA`` update:
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1')))
/// >>> store.update('DELETE DATA { <http://example.com> <http://example.com/p> "1" }')
/// >>> list(store)
/// []
///
/// ``DELETE`` update:
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1')))
/// >>> store.update('DELETE WHERE { <http://example.com> ?p ?o }')
/// >>> list(store)
/// []
#[ text_signature = " ($self, update) " ]
fn update ( & self , update : & str ) -> PyResult < ( ) > {
self . inner . update ( update ) . map_err ( map_evaluation_error )
}
/// Loads an RDF serialization into the store
///
/// It currently supports the following formats:
///
/// * `N-Triples <https://www.w3.org/TR/n-triples/>`_ (``application/n-triples``)
/// * `N-Quads <https://www.w3.org/TR/n-quads/>`_ (``application/n-quads``)
/// * `Turtle <https://www.w3.org/TR/turtle/>`_ (``text/turtle``)
/// * `TriG <https://www.w3.org/TR/trig/>`_ (``application/trig``)
/// * `RDF/XML <https://www.w3.org/TR/rdf-syntax-grammar/>`_ (``application/rdf+xml``)
///
/// It supports also some MIME type aliases.
/// For example ``application/turtle`` could also be used for `Turtle <https://www.w3.org/TR/turtle/>`_
/// and ``application/xml`` for `RDF/XML <https://www.w3.org/TR/rdf-syntax-grammar/>`_.
///
/// :param input: The binary I/O object to read from. For example, it could be a file opened in binary mode with ``open('my_file.ttl', 'rb')``.
/// :type input: io.RawIOBase or io.BufferedIOBase
/// :param mime_type: the MIME type of the RDF serialization
/// :type mime_type: str
/// :param base_iri: the base IRI used to resolve the relative IRIs in the file or :py:const:`None` if relative IRI resolution should not be done
/// :type base_iri: str or None, optional
/// :param to_graph: if it is a file composed of triples, the graph in which store the triples. By default, the default graph is used.
/// :type to_graph: NamedNode or BlankNode or DefaultGraph or None, optional
/// :raises ValueError: if the MIME type is not supported or the `to_graph` parameter is given with a quad file.
/// :raises SyntaxError: if the provided data is invalid
///
/// >>> store = MemoryStore()
/// >>> store.load(io.BytesIO(b'<foo> <p> "1" .'), "text/turtle", base_iri="http://example.com/", to_graph=NamedNode("http://example.com/g"))
/// >>> list(store)
/// [<Quad subject=<NamedNode value=http://example.com/foo> predicate=<NamedNode value=http://example.com/p> object=<Literal value=1 datatype=<NamedNode value=http://www.w3.org/2001/XMLSchema#string>> graph_name=<NamedNode value=http://example.com/g>>]
#[ text_signature = " ($self, input, /, mime_type, *, base_iri = None, to_graph = None) " ]
#[ args(input, mime_type, " * " , base_iri = " None " , to_graph = " None " ) ]
fn load (
& self ,
input : PyObject ,
mime_type : & str ,
base_iri : Option < & str > ,
to_graph : Option < & PyAny > ,
) -> PyResult < ( ) > {
let to_graph_name = if let Some ( graph_name ) = to_graph {
Some ( PyGraphNameRef ::try_from ( graph_name ) ? )
} else {
None
} ;
let input = BufReader ::new ( PyFileLike ::new ( input ) ) ;
if let Some ( graph_format ) = GraphFormat ::from_media_type ( mime_type ) {
self . inner
. load_graph (
input ,
graph_format ,
& to_graph_name . unwrap_or ( PyGraphNameRef ::DefaultGraph ) ,
base_iri ,
)
. map_err ( map_io_err )
} else if let Some ( dataset_format ) = DatasetFormat ::from_media_type ( mime_type ) {
if to_graph_name . is_some ( ) {
return Err ( PyValueError ::new_err (
"The target graph name parameter is not available for dataset formats" ,
) ) ;
}
self . inner
. load_dataset ( input , dataset_format , base_iri )
. map_err ( map_io_err )
} else {
Err ( PyValueError ::new_err ( format! (
"Not supported MIME type: {}" ,
mime_type
) ) )
}
}
/// Dumps the store quads or triples into a file
///
/// It currently supports the following formats:
///
/// * `N-Triples <https://www.w3.org/TR/n-triples/>`_ (``application/n-triples``)
/// * `N-Quads <https://www.w3.org/TR/n-quads/>`_ (``application/n-quads``)
/// * `Turtle <https://www.w3.org/TR/turtle/>`_ (``text/turtle``)
/// * `TriG <https://www.w3.org/TR/trig/>`_ (``application/trig``)
/// * `RDF/XML <https://www.w3.org/TR/rdf-syntax-grammar/>`_ (``application/rdf+xml``)
///
/// It supports also some MIME type aliases.
/// For example ``application/turtle`` could also be used for `Turtle <https://www.w3.org/TR/turtle/>`_
/// and ``application/xml`` for `RDF/XML <https://www.w3.org/TR/rdf-syntax-grammar/>`_.
///
/// :param output: The binary I/O object to write to. For example, it could be a file opened in binary mode with ``open('my_file.ttl', 'wb')``.
/// :type input: io.RawIOBase or io.BufferedIOBase
/// :param mime_type: the MIME type of the RDF serialization
/// :type mime_type: str
/// :param from_graph: if a triple based format is requested, the store graph from which dump the triples. By default, the default graph is used.
/// :type from_graph: NamedNode or BlankNode or DefaultGraph or None, optional
/// :raises ValueError: if the MIME type is not supported or the `from_graph` parameter is given with a quad syntax.
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g')))
/// >>> output = io.BytesIO()
/// >>> store.dump(output, "text/turtle", from_graph=NamedNode("http://example.com/g"))
/// >>> output.getvalue()
/// b'<http://example.com> <http://example.com/p> "1" .\n'
#[ text_signature = " ($self, output, /, mime_type, *, from_graph = None) " ]
#[ args(output, mime_type, " * " , from_graph = " None " ) ]
fn dump ( & self , output : PyObject , mime_type : & str , from_graph : Option < & PyAny > ) -> PyResult < ( ) > {
let from_graph_name = if let Some ( graph_name ) = from_graph {
Some ( PyGraphNameRef ::try_from ( graph_name ) ? )
} else {
None
} ;
let output = PyFileLike ::new ( output ) ;
if let Some ( graph_format ) = GraphFormat ::from_media_type ( mime_type ) {
self . inner
. dump_graph (
output ,
graph_format ,
& from_graph_name . unwrap_or ( PyGraphNameRef ::DefaultGraph ) ,
)
. map_err ( map_io_err )
} else if let Some ( dataset_format ) = DatasetFormat ::from_media_type ( mime_type ) {
if from_graph_name . is_some ( ) {
return Err ( PyValueError ::new_err (
"The target graph name parameter is not available for dataset formats" ,
) ) ;
}
self . inner
. dump_dataset ( output , dataset_format )
. map_err ( map_io_err )
} else {
Err ( PyValueError ::new_err ( format! (
"Not supported MIME type: {}" ,
mime_type
) ) )
}
}
/// Returns an iterator over all the store named graphs
///
/// :return: an iterator of the store graph names
/// :rtype: iter(NamedNode or BlankNode)
///
/// >>> store = MemoryStore()
/// >>> store.add(Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g')))
/// >>> list(store.named_graphs())
/// [<NamedNode value=http://example.com/g>]
#[ text_signature = " ($self) " ]
fn named_graphs ( & self ) -> GraphNameIter {
GraphNameIter {
inner : self . inner . named_graphs ( ) ,
}
}
/// Adds a named graph to the store
///
/// :param graph_name: the name of the name graph to add
/// :type graph_name: NamedNode or BlankNode
///
/// >>> store = MemoryStore()
/// >>> store.add_graph(NamedNode('http://example.com/g'))
/// >>> list(store.named_graphs())
/// [<NamedNode value=http://example.com/g>]
#[ text_signature = " ($self, graph_name) " ]
fn add_graph ( & self , graph_name : PyGraphName ) {
match graph_name {
PyGraphName ::DefaultGraph ( _ ) = > ( ) ,
PyGraphName ::NamedNode ( graph_name ) = > self . inner . insert_named_graph ( graph_name ) ,
PyGraphName ::BlankNode ( graph_name ) = > self . inner . insert_named_graph ( graph_name ) ,
}
}
/// Removes a graph from the store
///
/// The default graph will not be remove but just cleared.
///
/// :param graph_name: the name of the name graph to remove
/// :type graph_name: NamedNode or BlankNode or DefaultGraph
///
/// >>> store = MemoryStore()
/// >>> quad = Quad(NamedNode('http://example.com'), NamedNode('http://example.com/p'), Literal('1'), NamedNode('http://example.com/g'))
/// >>> store.remove_graph(NamedNode('http://example.com/g'))
/// >>> list(store)
/// []
#[ text_signature = " ($self, graph_name) " ]
fn remove_graph ( & self , graph_name : & PyAny ) -> PyResult < ( ) > {
match PyGraphNameRef ::try_from ( graph_name ) ? {
PyGraphNameRef ::DefaultGraph = > self . inner . clear_graph ( GraphNameRef ::DefaultGraph ) ,
PyGraphNameRef ::NamedNode ( graph_name ) = > self
. inner
. remove_named_graph ( & PyNamedOrBlankNodeRef ::NamedNode ( graph_name ) ) ,
PyGraphNameRef ::BlankNode ( graph_name ) = > self
. inner
. remove_named_graph ( & PyNamedOrBlankNodeRef ::BlankNode ( graph_name ) ) ,
}
Ok ( ( ) )
}
}
#[ pyproto ]
impl PyObjectProtocol for PyMemoryStore {
fn __str__ ( & self ) -> String {
self . inner . to_string ( )
}
fn __richcmp__ ( & self , other : & PyCell < Self > , op : CompareOp ) -> PyResult < bool > {
let other : & PyMemoryStore = & other . borrow ( ) ;
match op {
CompareOp ::Eq = > Ok ( self = = other ) ,
CompareOp ::Ne = > Ok ( self ! = other ) ,
_ = > Err ( PyNotImplementedError ::new_err (
"Ordering is not implemented" ,
) ) ,
}
}
fn __bool__ ( & self ) -> bool {
! self . inner . is_empty ( )
}
}
#[ pyproto ]
impl < ' p > PySequenceProtocol < ' p > for PyMemoryStore {
fn __len__ ( & self ) -> usize {
self . inner . len ( )
}
fn __contains__ ( & self , quad : PyQuad ) -> bool {
self . inner . contains ( & quad )
}
}
#[ pyproto ]
impl PyIterProtocol for PyMemoryStore {
fn __iter__ ( slf : PyRef < Self > ) -> QuadIter {
QuadIter {
inner : slf . inner . iter ( ) ,
}
}
}
#[ pyclass(unsendable, module = " oxigraph " ) ]
pub struct QuadIter {
inner : MemoryQuadIter ,
}
#[ pyproto ]
impl PyIterProtocol for QuadIter {
fn __iter__ ( slf : PyRefMut < Self > ) -> Py < Self > {
slf . into ( )
}
fn __next__ ( mut slf : PyRefMut < Self > ) -> Option < PyQuad > {
slf . inner . next ( ) . map ( | q | q . into ( ) )
}
}
#[ pyclass(unsendable, module = " oxigraph " ) ]
pub struct GraphNameIter {
inner : MemoryGraphNameIter ,
}
#[ pyproto ]
impl PyIterProtocol for GraphNameIter {
fn __iter__ ( slf : PyRefMut < Self > ) -> Py < Self > {
slf . into ( )
}
fn __next__ ( mut slf : PyRefMut < Self > ) -> Option < PyNamedOrBlankNode > {
slf . inner . next ( ) . map ( | q | q . into ( ) )
}
}