public void RetainAll(ResSet set)
{
foreach (Resource r in new ArrayList(this))
{
if (!set.Contains(r))
{
Remove(r);
}
}
}
private static bool MakeLeanMSG3(Store msg, ResSet predicates, StatementSink removed,
ResSet nodesremoved, SyncPath path) {
// The variable path has to be expanded by including the statements
// connected to the variables on the frontier. Statements
// mentioning a variable node have already been considered.
// The target of each such statement can be considered fixed
// or variable. If a variable is considered fixed, the edge
// must exist in the MSG substituting the variables for their
// values. If it's variable, it has to have at least one
// match in the MSG but not as any of the variable nodes.
// If all targets are considered fixed (and have matches),
// then the variables so far (and their edges) can all be
// removed and no more processing needs to be done.
// There are (2^N)-1 other considerations. For each of those,
// the targets considered variables all become the new
// frontier, and this is repeated.
// First, get a list of edges from the frontier that we
// haven't considered yet.
ArrayList alledges = new ArrayList();
foreach (BNode b in path.FrontierVariables) {
// Make sure all edges are kept because even the ones
// to literals have to be removed when duplication is found.
foreach (Statement s in msg.Select(new Statement(b, null, null)))
alledges.Add(new Edge(true, b, s.Predicate, s.Object));
foreach (Statement s in msg.Select(new Statement(null, null, b)))
alledges.Add(new Edge(false, b, s.Predicate, s.Subject));
}
ArrayList newedges = new ArrayList();
ResSet alltargets = new ResSet();
ResSet fixabletargetsset = new ResSet(); // can be fixed
ResSet variabletargetsset = new ResSet(); // must be variable
foreach (Edge e in alledges) {
if (path.Path.ContainsKey(e)) continue;
path.Path[e] = e;
// This checks if we can keep the target of this edge
// fixed, given the variable mappings we have so far.
bool isTargetFixable =
msg.Contains(e.AsStatement().Replace(path.Mapping));
// If the target of e is any of the following, we
// can check immediately if the edge is supported
// by the MSG under the variable mapping we have so far:
// a named node, literal, fixed node, or predicate
// a variable we've seen already
// If it's not supported, this path fails. If it is
// supported, we're done with this edge.
if (!(e.End is BNode)
|| path.FixedNodes.Contains(e.End)
|| predicates.Contains(e.End)
|| path.VariableNodes.Contains(e.End)) {
if (!isTargetFixable) return false;
continue; // this edge is supported, so we can continue
}
// The target of e is a new BNode.
// If this target is not fixable via this edge, it's
// not fixable at all.
if (!isTargetFixable) {
fixabletargetsset.Remove(e.End);
variabletargetsset.Add(e.End);
}
if (!alltargets.Contains(e.End)) {
alltargets.Add(e.End);
fixabletargetsset.Add(e.End);
}
newedges.Add(e);
}
// If all of the targets were fixable (trivially true also
// if there simple were no new edges/targets), then we've reached
// the end of this path. We can immediately remove
// the edges we've seen so far, under the variable mapping
// we've chosen.
if (variabletargetsset.Count == 0) {
foreach (Edge e in path.Path.Keys) {
Statement s = e.AsStatement();
msg.Remove(s);
if (removed != null) removed.Add(s);
}
foreach (Entity e in path.Mapping.Keys)
nodesremoved.Add(e);
return true;
}
// At this point, at least one target must be a variable
// and we'll have to expand the path in that direction.
// We might want to permute through the ways we can
// take fixable nodes as either fixed or variable, but
// we'll be greedy and assume everything fixable is
// fixed and everything else is a variable.
path.FixedNodes.AddRange(fixabletargetsset);
path.VariableNodes.AddRange(variabletargetsset);
// But we need to look at all the ways each variable target
// can be mapped to a new value, which means intersecting
// the possible matches for each relevant edge.
Entity[] variables = variabletargetsset.ToEntityArray();
ResSet[] values = new ResSet[variables.Length];
Entity[][] values_array = new Entity[variables.Length][];
int[] choices = new int[variables.Length];
for (int i = 0; i < variables.Length; i++) {
foreach (Edge e in newedges) {
if (e.End != variables[i]) continue;
// Get the possible values this edge allows
Resource[] vr;
if (e.Direction)
vr = msg.SelectObjects((Entity)path.Mapping[e.Start], e.Predicate);
else
vr = msg.SelectSubjects(e.Predicate, (Entity)path.Mapping[e.Start]);
// Filter out literals and any variables
// on the path! The two paths can't intersect
// except at fixed nodes.
ResSet v = new ResSet();
foreach (Resource r in vr) {
if (r is Literal) continue;
if (path.Mapping.ContainsKey(r)) continue;
v.Add(r);
}
// Intersect these with the values we have already.
if (values[i] == null)
values[i] = v;
else
values[i].RetainAll(v);
// If no values are available for this variable,
// we're totally done.
if (values[i].Count == 0) return false;
}
choices[i] = values[i].Count;
values_array[i] = values[i].ToEntityArray();
}
// Now we have to permute through the choice of values.
// Make an array of the number of choices for each variable.
Permutation p = new Permutation(choices);
int[] pstate;
while ((pstate = p.Next()) != null) {
SyncPath newpath = new SyncPath();
newpath.FixedNodes.AddRange(path.FixedNodes);
newpath.VariableNodes.AddRange(path.VariableNodes);
newpath.Mapping = (Hashtable)path.Mapping.Clone();
newpath.Path = (Hashtable)path.Path.Clone();
newpath.FrontierVariables = variabletargetsset;
for (int i = 0; i < variables.Length; i++) {
Entity value = values_array[i][pstate[i]];
newpath.Mapping[variables[i]] = value;
newpath.FixedNodes.Add(value);
}
if (MakeLeanMSG3(msg, predicates, removed,
nodesremoved, newpath)) return true;
}
return false;
}
public static void FindMSG(SelectableSource store, Entity node, StatementSink msg) {
if (node.Uri != null) throw new ArgumentException("node must be anonymous");
ResSet nodesSeen = new ResSet();
ResSet nodesToAdd = new ResSet();
nodesToAdd.Add(node);
while (nodesToAdd.Count > 0) {
ResSet nodes = nodesToAdd;
nodesToAdd = new ResSet();
Sink sink = new Sink(msg, nodesToAdd);
foreach (Entity n in nodes) {
if (nodesSeen.Contains(n)) continue;
nodesSeen.Add(n);
store.Select(new Statement(n, null, null, null), sink);
store.Select(new Statement(null, n, null, null), sink);
store.Select(new Statement(null, null, n, null), sink);
}
}
}
private static void MakeLeanMSG(Store msg, ICollection bnodecollection, StatementSink removed) {
// To make any graph lean, we try to eliminate duplicate
// paths through the graph, where duplicate means we
// take some subset of the bnodes and call them variables,
// and we relabel them as other bnodes from the remaining
// set (the fixed nodes). But there are 2^N subsets of bnodes
// we could choose as variables (N=number of bnodes), so we can't
// reasonably iterate through them.
// I'll make a simplifying assumption that bnode predicates
// in the graph will be considered always fixed.
// This lets us view the graph as actually a graph (with
// nodes and edges), and then we can make the observation that
// if variable node V is part of a subgraph that can be removed,
// if V directly connects to fixed node F via an edge labeled P,
// then F must connect to a fixed node G via an edge also
// labeled P. That is, we can start our search looking for
// nodes that project two edges with the same label.
// Also, we only want to consider contiguous 'paths' -- subsets
// of the bnodes connected only through those nodes --
// to see if there is another path in the MSG if we
// map bnodes in the first path to nodes in the MSG.
// So the strategy is to start at each node in the graph
// and consider it fixed. If it has two outgoing
// edges with the same property and one terminates on a
// bnode, this is the beginning of a possible pair
// of redundant paths (the one with the bnode being
// eliminable).
// However, the path with the bnode
// has to be incremented with all of that bnode's
// outgoing edges. The other path has to be
// incremented in parallel, following the same predicates
// to other nodes. If that can't be done, then these
// paths are not duplicates. If the parallel predicates
// terminate on the very same nodes, the bnode and its edges can
// be removed.
// From there, each of the nodes the bnode edges terminate on,
// besides the initial node, can be considered fixed or
// a variable. If it's a variable it might be able to have
// one of many possible values, but then the path has to
// be expanded to include all of the outgoing edges for this
// variable.
// Ok, here we go.
// If there is only one bnode in the MSG, then
// there are no subgraphs to check. That's nice.
if (bnodecollection.Count == 1) return;
// Remember which bnodes have been removed in
// due course.
ResSet nodesremoved = new ResSet();
// Remember which nodes are predicates and can't
// be considered variable.
ResSet predicates = new ResSet();
foreach (Statement s in msg.Select(Statement.All))
predicates.Add(s.Predicate);
// Start with each bnode to consider fixed.
foreach (BNode b in bnodecollection) {
if (nodesremoved.Contains(b)) continue;
MakeLeanMSG2(msg, predicates, removed, nodesremoved, b);
}
}
private static void MakeLeanMSG2(Store msg, ResSet predicates, StatementSink removed,
ResSet nodesremoved, BNode startingnode) {
// Find every pair of two distinct outgoing edges from startingnode
// with the same predicate, targeting entities only.
MultiMap edges = new MultiMap();
foreach (Statement s in msg.Select(new Statement(startingnode, null, null)))
if (s.Object is Entity)
edges.Put(new Edge(true, startingnode, s.Predicate, null), s.Object);
foreach (Statement s in msg.Select(new Statement(null, null, startingnode)))
edges.Put(new Edge(false, startingnode, s.Predicate, null), s.Subject);
foreach (Edge e in edges.Keys) {
// Make sure we have a distinct set of targets.
ResSet targets_set = new ResSet();
foreach (Entity r in edges.Get(e))
targets_set.Add(r);
if (targets_set.Count == 1) continue;
IList targets = targets_set.ToEntityArray();
// Take every pair of targets, provided
// one is a bnode that can be a variable.
for (int i = 0; i < targets.Count; i++) {
if (!(targets[i] is BNode) || predicates.Contains((BNode)targets[i])) continue;
if (nodesremoved.Contains((BNode)targets[i])) continue;
for (int j = 0; j < targets.Count; j++) {
if (i == j) continue;
// Create a new synchronous-path object.
SyncPath p = new SyncPath();
p.FixedNodes.Add((Resource)targets[j]);
p.FrontierVariables.Add((Resource)targets[i]);
p.Mapping[targets[i]] = targets[j];
p.Path[new Edge(e.Direction, e.Start, e.Predicate, (BNode)targets[i])] = p.Path;
if (MakeLeanMSG3(msg, predicates, removed, nodesremoved, p))
break; // the target was removed
}
}
}
}
static Entity[] GetClosure(Resource[] starts, Hashtable table) {
ResSet ret = new ResSet();
ResSet toadd = new ResSet(starts);
while (toadd.Count > 0) {
ResSet newadd = new ResSet();
foreach (Resource e in toadd) {
if (!(e is Entity)) continue;
if (ret.Contains(e)) continue;
ret.Add(e);
if (table.ContainsKey(e))
newadd.AddRange((ResSet)table[e]);
}
toadd.Clear();
toadd.AddRange(newadd);
}
return ret.ToEntityArray();
}
public void RetainAll(ResSet set) {
foreach (Resource r in new ArrayList(this))
if (!set.Contains(r))
Remove(r);
}
// The next few routines convert a set of axioms from a StatementSource
// into a data structure of use for the algorithm, with Sequents and things.
private static Hashtable RulesToCases(StatementSource rules)
{
Hashtable cases = new Hashtable();
MemoryStore rules_store = new MemoryStore(rules);
foreach (Statement p in rules_store) {
if (p.Meta == Statement.DefaultMeta) {
if (p.Predicate == entLOGIMPLIES && p.Object is Entity) {
MemoryStore body = new MemoryStore();
MemoryStore head = new MemoryStore();
rules_store.Select(new Statement(null, null, null, (Entity)p.Subject), new RemoveMeta(body));
rules_store.Select(new Statement(null, null, null, (Entity)p.Object), new RemoveMeta(head));
// Any variables in the head not bound in the body represent existentially closed bnodes.
// (Euler's OWL test case does this. Wish they had used bnodes instead of vars...)
ResSet bodyvars = new ResSet();
foreach (Statement b in body) {
if (b.Subject is Variable) bodyvars.Add(b.Subject);
if (b.Predicate is Variable) bodyvars.Add(b.Predicate);
if (b.Object is Variable) bodyvars.Add(b.Object);
}
foreach (Entity v in head.GetEntities()) {
if (v is Variable && !bodyvars.Contains(v))
head.Replace(v, new BNode(((Variable)v).LocalName));
}
// Replace (...) lists in the body that are tied to the subjects
// of user predicates with callArgs objects.
Hashtable callArgs = new Hashtable();
CollectCallArgs(body, callArgs);
// Rules can't have more than one statement in their
// consequent. The best we can do is break up
// the consequent into multiple rules. (Since all head
// variables are bound in body, it's equivalent...?)
foreach (Statement h in head)
AddSequent(cases, new Sequent(h, body.ToArray(), callArgs));
} else {
AddSequent(cases, new Sequent(p, new Statement[0], null));
}
}
}
return cases;
}
static Entity[] GetClosure(Resource[] starts, Hashtable table, bool includeStarts) {
ResSet ret = new ResSet();
ResSet toadd = new ResSet(starts);
bool firstRound = true;
while (toadd.Count > 0) {
ResSet newadd = new ResSet();
foreach (Resource e in toadd) {
if (!(e is Entity)) continue;
if (ret.Contains(e)) continue;
if (!(firstRound && !includeStarts)) ret.Add(e);
if (table.ContainsKey(e))
newadd.AddRange((ResSet)table[e]);
}
toadd.Clear();
toadd.AddRange(newadd);
firstRound = false;
}
return ret.ToEntityArray();
}
public void Query(Statement[] graph, QueryOptions options, QueryResultSink sink) {
if (options.DistinguishedVariables != null && options.DistinguishedVariables.Count == 0)
throw new ArgumentException("options.DistinguishedVariables cannot be an empty list.");
StringBuilder query = new StringBuilder();
query.Append("SELECT ");
// Get a list of variables and map them to fresh names
#if !DOTNET2
Hashtable variableNames = new Hashtable();
#else
Dictionary<Variable,string> variableNames = new Dictionary<Variable,string>();
#endif
Hashtable variableNames2 = new Hashtable();
foreach (Statement s in graph) {
for (int j = 0; j < 3; j++) {
Variable v = s.GetComponent(j) as Variable;
if (v == null) continue;
if (variableNames.ContainsKey(v)) continue;
variableNames2["v" + variableNames.Count] = v;
variableNames[v] = "?v" + variableNames.Count;
}
}
// What variables will we select on?
ArrayList selectedVars = new ArrayList();
foreach (Variable v in
options.DistinguishedVariables != null
? options.DistinguishedVariables
: variableNames.Keys) {
if (!variableNames.ContainsKey(v)) continue; // in case distinguished variables list
// has more than what actually appears in query
if (selectedVars.Contains(v)) continue; // don't select more than once
query.Append(variableNames[v]);
query.Append(' ');
selectedVars.Add(v);
}
if (selectedVars.Count == 0) {
if (options.DistinguishedVariables == null)
throw new ArgumentException("There were no variables in the query.");
else
throw new ArgumentException("None of the variables in the query were distinguished.");
}
// Bnodes are not allowed here -- we can't query on them.
foreach (Statement s in graph) {
for (int j = 0; j < 3; j++) {
if (s.GetComponent(j) is BNode && !(s.GetComponent(j) is Variable)) {
Variable[] varArray = (Variable[])selectedVars.ToArray(typeof(Variable));
sink.Init(varArray);
sink.Finished();
return;
}
}
}
// Build the graph pattern.
query.Append("WHERE {\n");
ResSet firstVarUse = new ResSet();
foreach (Statement s in graph) {
for (int j = 0; j < 3; j++) {
Resource r = s.GetComponent(j);
query.Append(S(r, r is Variable && variableNames.ContainsKey((Variable)r) ? (string)variableNames[(Variable)r] : null));
query.Append(" ");
}
query.Append(" . \n");
if (options.VariableKnownValues != null) {
for (int j = 0; j < 3; j++) {
Resource r = s.GetComponent(j);
if (firstVarUse.Contains(r)) continue;
firstVarUse.Add(r);
if (r is Variable && variableNames.ContainsKey((Variable)r) &&
#if !DOTNET2
options.VariableKnownValues.Contains(r)
#else
options.VariableKnownValues.ContainsKey((Variable)r)
#endif
)
query.Append(SL(options.VariableKnownValues[(Variable)r], (string)variableNames[(Variable)r], true));
}
}
// And what about meta...?
}
query.Append("}");
if (options.Limit > 0) {
query.Append(" LIMIT ");
query.Append(options.Limit);
}
Load(query.ToString(), new QueryResultsWrapper(sink, variableNames2));
}
