Source code for rdflib.plugins.stores.sparqlstore

# -*- coding: utf-8 -*-
#
"""
This is an RDFLib store around Ivan Herman et al.'s SPARQL service wrapper.
This was first done in layer-cake, and then ported to RDFLib 3 and rdfextras

This version works with vanilla SPARQLWrapper installed by ``easy_install``,
``pip`` or similar. If you installed ``rdflib`` with a tool that understands
dependencies, it should have been installed automatically for you.

Changes:
    - Layercake adding support for namespace binding, I removed it again to
      work with vanilla SPARQLWrapper
    - JSON object mapping support suppressed
    - Replaced '4Suite-XML Domlette with Elementtree
    - Incorporated as an RDFLib store

"""
SPARQL_POST_UPDATE = "application/sparql-update"
SPARQL_POST_ENCODED = "application/x-www-form-urlencoded; charset=UTF-8"

# Defines some SPARQL keywords
LIMIT = 'LIMIT'
OFFSET = 'OFFSET'
ORDERBY = 'ORDER BY'

import re
# import warnings
try:
    from SPARQLWrapper import SPARQLWrapper, XML
except ImportError:
    raise Exception(
        "SPARQLWrapper not found! SPARQL Store will not work." +
        "Install with 'easy_install SPARQLWrapper'")

import sys
if getattr(sys, 'pypy_version_info', None) is not None \
    or sys.platform.startswith('java') \
        or sys.version_info[:2] < (2, 6):
    # import elementtree as etree
    from elementtree import ElementTree
    assert ElementTree
else:
    try:
        from xml.etree import ElementTree
        assert ElementTree
    except ImportError:
        from elementtree import ElementTree

from rdflib.plugins.stores.regexmatching import NATIVE_REGEX

from rdflib.store import Store
from rdflib.query import Result
from rdflib import Variable, Namespace, BNode, URIRef, Literal

import httplib
import urlparse

[docs]class NSSPARQLWrapper(SPARQLWrapper): nsBindings = {}
[docs] def setNamespaceBindings(self, bindings): """ A shortcut for setting namespace bindings that will be added to the prolog of the query @param bindings: A dictionary of prefixs to URIs """ self.nsBindings.update(bindings)
[docs] def setQuery(self, query): """ Set the SPARQL query text. Note: no check is done on the validity of the query (syntax or otherwise) by this module, except for testing the query type (SELECT, ASK, etc). Syntax and validity checking is done by the SPARQL service itself. @param query: query text @type query: string @bug: #2320024 """ self.queryType = self._parseQueryType(query) self.queryString = self.injectPrefixes(query)
[docs] def injectPrefixes(self, query): return '\n'.join( ['\n'.join(['PREFIX %s: <%s>' % (key, val) for key, val in self.nsBindings.items()]), query])
BNODE_IDENT_PATTERN = re.compile('(?P<label>_\:[^\s]+)') SPARQL_NS = Namespace('http://www.w3.org/2005/sparql-results#') sparqlNsBindings = {u'sparql': SPARQL_NS} ElementTree._namespace_map["sparql"] = SPARQL_NS
[docs]def TraverseSPARQLResultDOM(doc, asDictionary=False): """ Returns a generator over tuples of results """ # namespace handling in elementtree xpath sub-set is not pretty :( vars = [Variable(v.attrib["name"]) for v in doc.findall( './{http://www.w3.org/2005/sparql-results#}head/' + '{http://www.w3.org/2005/sparql-results#}variable')] for result in doc.findall( './{http://www.w3.org/2005/sparql-results#}results/' + '{http://www.w3.org/2005/sparql-results#}result'): currBind = {} values = [] for binding in result.findall( '{http://www.w3.org/2005/sparql-results#}binding'): varVal = binding.attrib["name"] var = Variable(varVal) term = CastToTerm(binding.findall('*')[0]) values.append(term) currBind[var] = term if asDictionary: yield currBind, vars else: def __locproc(values): if len(values) == 1: return values[0] else: return tuple(values) yield __locproc(values), vars
[docs]def localName(qname): # wtf - elementtree cant do this for me return qname[qname.index("}") + 1:]
[docs]def CastToTerm(node): """ Helper function that casts XML node in SPARQL results to appropriate rdflib term """ if node.tag == '{%s}bnode' % SPARQL_NS: return BNode(node.text) elif node.tag == '{%s}uri' % SPARQL_NS: return URIRef(node.text) elif node.tag == '{%s}literal' % SPARQL_NS: if 'datatype' in node.attrib: dT = URIRef(node.attrib['datatype']) if False: # not node.xpath('*'): return Literal('', datatype=dT) else: return Literal(node.text, datatype=dT) elif '{http://www.w3.org/XML/1998/namespace}lang' in node.attrib: return Literal(node.text, lang=node.attrib[ "{http://www.w3.org/XML/1998/namespace}lang"]) else: return Literal(node.text) else: raise Exception('Unknown answer type')
[docs]class SPARQLStore(NSSPARQLWrapper, Store): """ An RDFLib store around a SPARQL endpoint This is in theory context-aware, and should work OK when the context is specified. (I.e. for Graph objects) then all queries should work against the named graph with the identifier of the graph only. For ConjunctiveGraphs, reading is done from the "default graph" Exactly what this means depends on your endpoint. General SPARQL does not offer a simple way to query the union of all graphs. Fuseki/TDB has a flag for specifying that the default graph is the union of all graphs (tdb:unionDefaultGraph in the Fuseki config) If this is set this will work fine. .. warning:: The SPARQL Store does not support blank-nodes! As blank-nodes acts as variables in SPARQL queries there is no way to query for a particular blank node. See http://www.w3.org/TR/sparql11-query/#BGPsparqlBNodes """ formula_aware = False transaction_aware = False regex_matching = NATIVE_REGEX
[docs] def __init__(self, endpoint=None, bNodeAsURI=False, sparql11=True, context_aware=True): """ """ if endpoint: super(SPARQLStore, self).__init__(endpoint, returnFormat=XML) self.bNodeAsURI = bNodeAsURI self.nsBindings = {} self.sparql11 = sparql11 self.context_aware = context_aware
# Database Management Methods
[docs] def create(self, configuration): raise TypeError('The SPARQL store is read only')
[docs] def open(self, configuration, create=False): """ sets the endpoint URL for this SPARQLStore if create==True an exception is thrown. """ if create: raise Exception("Cannot create a SPARQL Endpoint") self.query_endpoint = configuration
def __set_query_endpoint(self, queryEndpoint): super(SPARQLStore, self).__init__(queryEndpoint, returnFormat=XML) self.endpoint = queryEndpoint def __get_query_endpoint(self): return self.endpoint query_endpoint = property(__get_query_endpoint, __set_query_endpoint)
[docs] def destroy(self, configuration): """ FIXME: Add documentation """ raise TypeError('The SPARQL store is read only')
# Transactional interfaces
[docs] def commit(self): """ """ raise TypeError('The SPARQL store is read only')
[docs] def rollback(self): """ """ raise TypeError('The SPARQL store is read only')
[docs] def add(self, (subject, predicate, obj), context=None, quoted=False): """ Add a triple to the store of triples. """ raise TypeError('The SPARQL store is read only')
[docs] def addN(self, quads): """ Adds each item in the list of statements to a specific context. The quoted argument is interpreted by formula-aware stores to indicate this statement is quoted/hypothetical. Note that the default implementation is a redirect to add. """ raise TypeError('The SPARQL store is read only')
[docs] def remove(self, (subject, predicate, obj), context): """ Remove a triple from the store """ raise TypeError('The SPARQL store is read only')
[docs] def query(self, query, initNs={}, initBindings={}, queryGraph=None, DEBUG=False): self.debug = DEBUG assert isinstance(query, basestring) self.setNamespaceBindings(initNs) if initBindings: if not self.sparql11: raise Exception( "initBindings not supported for SPARQL 1.0 Endpoints.") v = list(initBindings) # VALUES was added to SPARQL 1.1 on 2012/07/24 query += "\nVALUES ( %s )\n{ ( %s ) }\n"\ % (" ".join("?" + str(x) for x in v), " ".join(initBindings[x].n3() for x in v)) self.resetQuery() if self.context_aware and queryGraph and queryGraph != '__UNION__': self.addDefaultGraph(queryGraph) self.setQuery(query) return Result.parse(SPARQLWrapper.query(self).response)
[docs] def triples(self, (s, p, o), context=None): """ - tuple **(s, o, p)** the triple used as filter for the SPARQL select. (None, None, None) means anything. - context **context** the graph effectively calling this method. Returns a tuple of triples executing essentially a SPARQL like SELECT ?subj ?pred ?obj WHERE { ?subj ?pred ?obj } **context** may include three parameter to refine the underlying query: * LIMIT: an integer to limit the number of results * OFFSET: an integer to enable paging of results * ORDERBY: an instance of Variable('s'), Variable('o') or Variable('p') or, by default, the first 'None' from the given triple .. warning:: - Using LIMIT or OFFSET automatically include ORDERBY otherwise this is because the results are retrieved in a not deterministic way (depends on the walking path on the graph) - Using OFFSET without defining LIMIT will discard the first OFFSET - 1 results `` a_graph.LIMIT = limit a_graph.OFFSET = offset triple_generator = a_graph.triples(mytriple): #do something #Removes LIMIT and OFFSET if not required for the next triple() calls del a_graph.LIMIT del a_graph.OFFSET `` """ if ( isinstance(s, BNode) or isinstance(p, BNode) or isinstance(o, BNode) ): raise Exception("SPARQLStore does not support Bnodes! " "See http://www.w3.org/TR/sparql11-query/#BGPsparqlBNodes") vars = [] if not s: s = Variable('s') vars.append(s) if not p: p = Variable('p') vars.append(p) if not o: o = Variable('o') vars.append(o) if vars: v = ' '.join([term.n3() for term in vars]) else: v = '*' query = "SELECT %s WHERE { %s %s %s }" % \ (v, s.n3(), p.n3(), o.n3()) # The ORDER BY is necessary if hasattr(context, LIMIT) or hasattr(context, OFFSET) \ or hasattr(context, ORDERBY): var = None if isinstance(s, Variable): var = s elif isinstance(p, Variable): var = p elif isinstance(o, Variable): var = o elif hasattr(context, ORDERBY) \ and isinstance(getattr(context, ORDERBY), Variable): var = getattr(context, ORDERBY) query = query + ' %s %s' % (ORDERBY, var.n3()) try: query = query + ' LIMIT %s' % int(getattr(context, LIMIT)) except (ValueError, TypeError, AttributeError): pass try: query = query + ' OFFSET %s' % int(getattr(context, OFFSET)) except (ValueError, TypeError, AttributeError): pass self.resetQuery() if self.context_aware and context is not None: self.addDefaultGraph(context.identifier) self.setQuery(query) doc = ElementTree.parse(SPARQLWrapper.query(self).response) # ElementTree.dump(doc) for rt, vars in TraverseSPARQLResultDOM(doc, asDictionary=True): yield (rt.get(s, s), rt.get(p, p), rt.get(o, o)), None
[docs] def triples_choices(self, (subject, predicate, object_), context=None): """ A variant of triples that can take a list of terms instead of a single term in any slot. Stores can implement this to optimize the response time from the import default 'fallback' implementation, which will iterate over each term in the list and dispatch to triples. """ raise NotImplementedError('Triples choices currently not supported')
[docs] def __len__(self, context=None): if not self.sparql11: raise NotImplementedError( "For performance reasons, this is not" + "supported for sparql1.0 endpoints") else: self.resetQuery() q = "SELECT (count(*) as ?c) WHERE {?s ?p ?o .}" if self.context_aware and context is not None: self.addDefaultGraph(context.identifier) self.setQuery(q) doc = ElementTree.parse(SPARQLWrapper.query(self).response) rt, vars = iter( TraverseSPARQLResultDOM(doc, asDictionary=True)).next() return int(rt.get(Variable("c")))
[docs] def contexts(self, triple=None): """ Iterates over results to SELECT ?NAME { GRAPH ?NAME { ?s ?p ?o } } returning instances of this store with the SPARQL wrapper object updated via addNamedGraph(?NAME) This causes a named-graph-uri key / value pair to be sent over the protocol """ if triple: s, p, o = triple else: s = p = o = None params = ((s if s else Variable('s')).n3(), (p if p else Variable('p')).n3(), (o if o else Variable('o')).n3()) self.setQuery( 'SELECT ?name WHERE { GRAPH ?name { %s %s %s }}' % params) doc = ElementTree.parse(SPARQLWrapper.query(self).response) return (rt.get(Variable("name")) for rt, vars in TraverseSPARQLResultDOM(doc, asDictionary=True))
# Namespace persistence interface implementation
[docs] def bind(self, prefix, namespace): self.nsBindings[prefix] = namespace
[docs] def prefix(self, namespace): """ """ return dict( [(v, k) for k, v in self.nsBindings.items()] ).get(namespace)
[docs] def namespace(self, prefix): return self.nsBindings.get(prefix)
[docs] def namespaces(self): for prefix, ns in self.nsBindings.items(): yield prefix, ns
[docs]class SPARQLUpdateStore(SPARQLStore): """ A store using SPARQL queries for read-access and SPARQL Update for changes This can be context-aware, if so, any changes will be to the given named graph only. For Graph objects, everything works as expected. .. warning:: The SPARQL Update Store does not support blank-nodes! As blank-nodes acts as variables in SPARQL queries there is no way to query for a particular blank node. See http://www.w3.org/TR/sparql11-query/#BGPsparqlBNodes """ where_pattern = re.compile(r"""(?P<where>WHERE\s*{)""", re.IGNORECASE)
[docs] def __init__(self, queryEndpoint=None, update_endpoint=None, bNodeAsURI=False, sparql11=True, context_aware=True, postAsEncoded=True): SPARQLStore.__init__(self, queryEndpoint, bNodeAsURI, sparql11, context_aware) self.connection = None self.update_endpoint = update_endpoint self.postAsEncoded = postAsEncoded self.headers = {'Content-type': SPARQL_POST_ENCODED, 'Connection': 'Keep-alive'} if not self.postAsEncoded: self.headers['Content-type'] = SPARQL_POST_UPDATE
def __set_update_endpoint(self, update_endpoint): self.__update_endpoint = update_endpoint if self.__update_endpoint: p = urlparse.urlparse(self.update_endpoint) assert not p.username, \ "SPARQL Update store does not support HTTP authentication" assert not p.password, \ "SPARQL Update store does not support HTTP authentication" assert p.scheme == "http", "SPARQL Update is an http protocol!" self.host = p.hostname self.port = p.port self.path = p.path self.connection = httplib.HTTPConnection( self.host, self.port) else: self.host = self.port = self.path = self.connection = None def __get_update_endpoint(self): return self.__update_endpoint update_endpoint = property( __get_update_endpoint, __set_update_endpoint, doc='the HTTP URL for the Update endpoint, typically' + 'something like http://server/dataset/update')
[docs] def open(self, configuration, create=False): """ sets the endpoint URLs for this SPARQLStore :param configuration: either a tuple of (queryEndpoint, update_endpoint), or a string with the query endpoint :param create: if True an exception is thrown. """ if create: raise Exception("Cannot create a SPARQL Endpoint") if isinstance(configuration, tuple): self.query_endpoint = configuration[0] if len(configuration) > 1: self.update_endpoint = configuration[1] else: self.endpoint = configuration if not self.update_endpoint: self.update_endpoint = self.endpoint
# Transactional interfaces
[docs] def commit(self): """ """ raise TypeError('The SPARQL Update store is not transaction aware!')
[docs] def rollback(self): """ """ raise TypeError('The SPARQL Update store is not transaction aware')
[docs] def add(self, spo, context=None, quoted=False): """ Add a triple to the store of triples. """ if not self.connection: raise Exception("UpdateEndpoint is not set - call 'open'") assert not quoted (subject, predicate, obj) = spo if ( isinstance(subject, BNode) or isinstance(predicate, BNode) or isinstance(obj, BNode) ): raise Exception("SPARQLStore does not support Bnodes! " "See http://www.w3.org/TR/sparql11-query/#BGPsparqlBNodes") triple = "%s %s %s ." % (subject.n3(), predicate.n3(), obj.n3()) if self.context_aware and context is not None: q = "INSERT DATA { GRAPH %s { %s } }" % ( context.identifier.n3(), triple) else: q = "INSERT DATA { %s }" % triple r = self._do_update(q) content = r.read() # we expect no content if r.status not in (200, 204): raise Exception("Could not update: %d %s\n%s" % ( r.status, r.reason, content))
[docs] def addN(self, quads): """ Add a list of quads to the store. """ if not self.connection: raise Exception("UpdateEndpoint is not set - call 'open'") data = "" for spoc in quads: (subject, predicate, obj, context) = spoc if ( isinstance(subject, BNode) or isinstance(predicate, BNode) or isinstance(obj, BNode) ): raise Exception("SPARQLStore does not support Bnodes! " "See http://www.w3.org/TR/sparql11-query/#BGPsparqlBNodes") triple = "%s %s %s ." % (subject.n3(), predicate.n3(), obj.n3()) data += "INSERT DATA { GRAPH <%s> { %s } }\n" % ( context.identifier, triple) r = self._do_update(data) content = r.read() # we expect no content if r.status not in (200, 204): raise Exception("Could not update: %d %s\n%s" % ( r.status, r.reason, content))
[docs] def remove(self, spo, context): """ Remove a triple from the store """ if not self.connection: raise Exception("UpdateEndpoint is not set - call 'open'") (subject, predicate, obj) = spo if not subject: subject = Variable("S") if not predicate: predicate = Variable("P") if not obj: obj = Variable("O") triple = "%s %s %s ." % (subject.n3(), predicate.n3(), obj.n3()) if self.context_aware and context is not None: q = "DELETE { GRAPH %s { %s } } WHERE { GRAPH %s { %s } }" % ( context.identifier.n3(), triple, context.identifier.n3(), triple) else: q = "DELETE { %s } WHERE { %s } " % (triple, triple) r = self._do_update(q) content = r.read() # we expect no content if r.status not in (200, 204): raise Exception("Could not update: %d %s\n%s" % ( r.status, r.reason, content))
def _do_update(self, update): import urllib if self.postAsEncoded: update = urllib.urlencode({'update': update.encode("utf-8")}) self.connection.request( 'POST', self.path, update.encode("utf-8"), self.headers) return self.connection.getresponse()
[docs] def update(self, query, initNs={}, initBindings={}, queryGraph=None, DEBUG=False): """ Perform a SPARQL Update Query against the endpoint, INSERT, LOAD, DELETE etc. Setting initNs adds PREFIX declarations to the beginning of the update. Setting initBindings adds inline VALUEs to the beginning of every WHERE clause. By the SPARQL grammar, all operations that support variables (namely INSERT and DELETE) require a WHERE clause. Important: initBindings fails if the update contains the substring 'WHERE {' which does not denote a WHERE clause, e.g. if it is part of a literal. """ self.debug = DEBUG assert isinstance(query, basestring) self.setNamespaceBindings(initNs) query = self.injectPrefixes(query) if initBindings: # For INSERT and DELETE the WHERE clause is obligatory # (http://www.w3.org/TR/2013/REC-sparql11-query-20130321/#rModify) # Other query types do not allow variables and don't # have a WHERE clause. This also works for updates with # more than one INSERT/DELETE. v = list(initBindings) values = "\nVALUES ( %s )\n{ ( %s ) }\n"\ % (" ".join("?" + str(x) for x in v), " ".join(initBindings[x].n3() for x in v)) query = self.where_pattern.sub("WHERE { " + values, query) r = self._do_update(query) content = r.read() # we expect no content if r.status not in (200, 204): raise Exception("Could not update: %d %s\n%s" % ( r.status, r.reason, content))