private void EvalLeftJoinProduct(ISet x, BaseMultiset other, PartitionedMultiset partitionedSet, ISparqlExpression expr)
{
LeviathanLeftJoinBinder binder = new LeviathanLeftJoinBinder(partitionedSet);
SparqlEvaluationContext subcontext = new SparqlEvaluationContext(binder);
bool standalone = false, matched = false;
int id = partitionedSet.GetNextBaseID();
foreach (ISet y in other.Sets)
{
id++;
ISet z = x.Join(y);
z.ID = id;
try
{
partitionedSet.Add(z);
if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
{
partitionedSet.Remove(z.ID);
standalone = true;
}
else
{
matched = true;
}
}
catch
{
partitionedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone && !matched)
{
id++;
ISet z = x.Copy();
z.ID = id;
partitionedSet.Add(z);
}
}
/// <summary>
/// Does a Left Join of this Multiset to another Multiset where the Join is predicated on the given Expression
/// </summary>
/// <param name="other">Other Multiset</param>
/// <param name="expr">Expression</param>
/// <returns></returns>
public override BaseMultiset LeftJoin(BaseMultiset other, ISparqlExpression expr)
{
//If the Other is the Identity/Null Multiset the result is this Multiset
if (other is IdentityMultiset) return this;
if (other is NullMultiset) return this;
if (other.IsEmpty) return this;
Multiset joinedSet = new Multiset();
LeviathanLeftJoinBinder binder = new LeviathanLeftJoinBinder(joinedSet);
SparqlEvaluationContext subcontext = new SparqlEvaluationContext(binder);
//Find the First Variable from this Multiset which is in both Multisets
//If there is no Variable from this Multiset in the other Multiset then this
//should be a Join operation instead of a LeftJoin
List<String> joinVars = this._variables.Where(v => other.Variables.Contains(v)).ToList();
if (joinVars.Count == 0)
{
//Calculate a Product filtering as we go
foreach (ISet x in this.Sets)
{
bool standalone = false;
foreach (ISet y in other.Sets)
{
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone) joinedSet.Add(x.Copy());
}
}
else
{
//This is the old algorithm which is correct but has complexity O(n^2) so it scales terribly
//foreach (ISet x in this.Sets)
//{
// IEnumerable<ISet> ys = other.Sets.Where(s => joinVars.All(v => x[v] == null || s[v] == null || x[v].Equals(s[v])));
// //IEnumerable<ISet> ys = other.Sets.Where(s => s.IsCompatibleWith(x, joinVars));
// bool standalone = false;
// int i = 0;
// foreach (ISet y in ys)
// {
// i++;
// ISet z = x.Join(y);
// try
// {
// joinedSet.Add(z);
// if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
// {
// joinedSet.Remove(z.ID);
// standalone = true;
// }
// }
// catch
// {
// joinedSet.Remove(z.ID);
// standalone = true;
// }
// }
// if (standalone || i == 0) joinedSet.Add(x);
//}
//This is the new Join algorithm which is also correct but is O(2n) so much faster and scalable
//Downside is that it does require more memory than the old algorithm
List<HashTable<INode, int>> values = new List<HashTable<INode, int>>();
List<List<int>> nulls = new List<List<int>>();
foreach (String var in joinVars)
{
values.Add(new HashTable<INode, int>(HashTableBias.Enumeration));
nulls.Add(new List<int>());
}
//First do a pass over the LHS Result to find all possible values for joined variables
HashSet<int> matched = new HashSet<int>();
HashSet<int> standalone = new HashSet<int>();
foreach (ISet x in this.Sets)
{
int i = 0;
foreach (String var in joinVars)
{
INode value = x[var];
if (value != null)
{
values[i].Add(value, x.ID);
}
else
{
nulls[i].Add(x.ID);
}
i++;
}
}
//Then do a pass over the RHS and work out the intersections
foreach (ISet y in other.Sets)
{
IEnumerable<int> possMatches = null;
int i = 0;
foreach (String var in joinVars)
{
INode value = y[var];
if (value != null)
{
if (values[i].ContainsKey(value))
{
possMatches = (possMatches == null ? values[i].GetValues(value).Concat(nulls[i]) : possMatches.Intersect(values[i].GetValues(value).Concat(nulls[i])));
}
else
{
possMatches = Enumerable.Empty<int>();
break;
}
}
else
{
//Don't forget that a null will be potentially compatible with everything
possMatches = (possMatches == null ? this.SetIDs : possMatches.Intersect(this.SetIDs));
}
i++;
}
if (possMatches == null) continue;
//Now do the actual joins for the current set
//Note - We access the dictionary directly here because going through the this[int id] method
//incurs a Contains() call each time and we know the IDs must exist because they came from
//our dictionary originally!
foreach (int poss in possMatches)
{
if (this._sets[poss].IsCompatibleWith(y, joinVars))
{
ISet z = this._sets[poss].Join(y);
joinedSet.Add(z);
try
{
if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
{
joinedSet.Remove(z.ID);
standalone.Add(poss);
}
else
{
matched.Add(poss);
}
}
catch
{
joinedSet.Remove(z.ID);
standalone.Add(poss);
}
}
}
}
//Finally add in unmatched sets from LHS
foreach (int id in this.SetIDs)
{
if (!matched.Contains(id) || standalone.Contains(id)) joinedSet.Add(this._sets[id].Copy());
}
}
return joinedSet;
}
/// <summary>
/// Does a Left Join of this Multiset to another Multiset where the Join is predicated on the given Expression.
/// </summary>
/// <param name="other">Other Multiset.</param>
/// <param name="expr">Expression.</param>
/// <returns></returns>
public virtual BaseMultiset LeftJoin(BaseMultiset other, ISparqlExpression expr)
{
// If the Other is the Identity/Null Multiset the result is this Multiset
if (other is IdentityMultiset)
{
return(this);
}
if (other is NullMultiset)
{
return(this);
}
if (other.IsEmpty)
{
return(this);
}
Multiset joinedSet = new Multiset();
LeviathanLeftJoinBinder binder = new LeviathanLeftJoinBinder(joinedSet);
SparqlEvaluationContext subcontext = new SparqlEvaluationContext(binder);
// Find the First Variable from this Multiset which is in both Multisets
// If there is no Variable from this Multiset in the other Multiset then this
// should be a Join operation instead of a LeftJoin
List <String> joinVars = Variables.Where(v => other.Variables.Contains(v)).ToList();
if (joinVars.Count == 0)
{
#if NET40
if (Options.UsePLinqEvaluation && expr.CanParallelise)
{
PartitionedMultiset partitionedSet = new PartitionedMultiset(this.Count, other.Count + 1);
this.Sets.AsParallel().ForAll(x => EvalLeftJoinProduct(x, other, partitionedSet, expr));
return(partitionedSet);
}
#endif
// Do a serial Left Join Product
// Calculate a Product filtering as we go
foreach (ISet x in Sets)
{
bool standalone = false;
bool matched = false;
foreach (ISet y in other.Sets)
{
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
{
joinedSet.Remove(z.ID);
standalone = true;
}
else
{
matched = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone && !matched)
{
joinedSet.Add(x.Copy());
}
}
#if NET40
#endif
}
else
{
// This is the new Join algorithm which is also correct but is O(2n) so much faster and scalable
// Downside is that it does require more memory than the old algorithm
List <MultiDictionary <INode, List <int> > > values = new List <MultiDictionary <INode, List <int> > >();
List <List <int> > nulls = new List <List <int> >();
foreach (String var in joinVars)
{
joinedSet.AddVariable(var);
values.Add(new MultiDictionary <INode, List <int> >(new FastVirtualNodeComparer()));
nulls.Add(new List <int>());
}
// First do a pass over the RHS Result to find all possible values for joined variables
foreach (ISet y in other.Sets)
{
int i = 0;
foreach (String var in joinVars)
{
INode value = y[var];
if (value != null)
{
if (values[i].TryGetValue(value, out List <int> ids))
{
ids.Add(y.ID);
}
else
{
values[i].Add(value, new List <int> {
y.ID
});
}
}
else
{
nulls[i].Add(y.ID);
}
i++;
}
}
// Then do a pass over the LHS and work out the intersections
#if NET40
if (Options.UsePLinqEvaluation && expr.CanParallelise)
{
this.Sets.AsParallel().ForAll(x => EvalLeftJoin(x, other, joinVars, values, nulls, joinedSet, subcontext, expr));
}
else
{
// Use a Serial Left Join
foreach (ISet x in this.Sets)
{
this.EvalLeftJoin(x, other, joinVars, values, nulls, joinedSet, subcontext, expr);
}
}
#else
// Use a Serial Left Join
foreach (var x in Sets)
{
EvalLeftJoin(x, other, joinVars, values, nulls, joinedSet, subcontext, expr);
}
#endif
}
return(joinedSet);
}
/// <summary>
/// Does a Left Join of this Multiset to another Multiset where the Join is predicated on the given Expression
/// </summary>
/// <param name="other">Other Multiset</param>
/// <param name="expr">Expression</param>
/// <returns></returns>
public override BaseMultiset LeftJoin(BaseMultiset other, ISparqlExpression expr)
{
//If the Other is the Identity/Null Multiset the result is this Multiset
if (other is IdentityMultiset) return this;
if (other is NullMultiset) return this;
if (other.IsEmpty) return this;
Multiset joinedSet = new Multiset();
LeviathanLeftJoinBinder binder = new LeviathanLeftJoinBinder(joinedSet);
SparqlEvaluationContext subcontext = new SparqlEvaluationContext(binder);
//Find the First Variable from this Multiset which is in both Multisets
//If there is no Variable from this Multiset in the other Multiset then this
//should be a Join operation instead of a LeftJoin
List<String> joinVars = this._variables.Where(v => other.Variables.Contains(v)).ToList();
if (joinVars.Count == 0)
{
//Calculate a Product filtering as we go
foreach (ISet x in this.Sets)
{
bool standalone = false;
foreach (ISet y in other.Sets)
{
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.EffectiveBooleanValue(subcontext, z.ID))
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone) joinedSet.Add(x);
}
}
else
{
foreach (ISet x in this.Sets)
{
IEnumerable<ISet> ys = other.Sets.Where(s => joinVars.All(v => x[v] == null || s[v] == null || x[v].Equals(s[v])));
//IEnumerable<ISet> ys = other.Sets.Where(s => s.IsCompatibleWith(x, joinVars));
bool standalone = false;
int i = 0;
foreach (ISet y in ys)
{
i++;
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.EffectiveBooleanValue(subcontext, z.ID))
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone || i == 0) joinedSet.Add(x);
}
}
return joinedSet;
}
/// <summary>
/// Does a Left Join of this Multiset to another Multiset where the Join is predicated on the given Expression
/// </summary>
/// <param name="other">Other Multiset</param>
/// <param name="expr">Expression</param>
/// <returns></returns>
public override BaseMultiset LeftJoin(BaseMultiset other, ISparqlExpression expr)
{
//If the Other is the Identity/Null Multiset the result is this Multiset
if (other is IdentityMultiset)
{
return(this);
}
if (other is NullMultiset)
{
return(this);
}
if (other.IsEmpty)
{
return(this);
}
Multiset joinedSet = new Multiset();
LeviathanLeftJoinBinder binder = new LeviathanLeftJoinBinder(joinedSet);
SparqlEvaluationContext subcontext = new SparqlEvaluationContext(binder);
//Find the First Variable from this Multiset which is in both Multisets
//If there is no Variable from this Multiset in the other Multiset then this
//should be a Join operation instead of a LeftJoin
List <String> joinVars = this._variables.Where(v => other.Variables.Contains(v)).ToList();
if (joinVars.Count == 0)
{
//Calculate a Product filtering as we go
foreach (ISet x in this.Sets)
{
bool standalone = false;
foreach (ISet y in other.Sets)
{
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone)
{
joinedSet.Add(x.Copy());
}
}
}
else
{
//This is the old algorithm which is correct but has complexity O(n^2) so it scales terribly
//foreach (ISet x in this.Sets)
//{
// IEnumerable<ISet> ys = other.Sets.Where(s => joinVars.All(v => x[v] == null || s[v] == null || x[v].Equals(s[v])));
// //IEnumerable<ISet> ys = other.Sets.Where(s => s.IsCompatibleWith(x, joinVars));
// bool standalone = false;
// int i = 0;
// foreach (ISet y in ys)
// {
// i++;
// ISet z = x.Join(y);
// try
// {
// joinedSet.Add(z);
// if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
// {
// joinedSet.Remove(z.ID);
// standalone = true;
// }
// }
// catch
// {
// joinedSet.Remove(z.ID);
// standalone = true;
// }
// }
// if (standalone || i == 0) joinedSet.Add(x);
//}
//This is the new Join algorithm which is also correct but is O(2n) so much faster and scalable
//Downside is that it does require more memory than the old algorithm
List <HashTable <INode, int> > values = new List <HashTable <INode, int> >();
List <List <int> > nulls = new List <List <int> >();
foreach (String var in joinVars)
{
values.Add(new HashTable <INode, int>(HashTableBias.Enumeration));
nulls.Add(new List <int>());
}
//First do a pass over the LHS Result to find all possible values for joined variables
HashSet <int> matched = new HashSet <int>();
HashSet <int> standalone = new HashSet <int>();
foreach (ISet x in this.Sets)
{
int i = 0;
foreach (String var in joinVars)
{
INode value = x[var];
if (value != null)
{
values[i].Add(value, x.ID);
}
else
{
nulls[i].Add(x.ID);
}
i++;
}
}
//Then do a pass over the RHS and work out the intersections
foreach (ISet y in other.Sets)
{
IEnumerable <int> possMatches = null;
int i = 0;
foreach (String var in joinVars)
{
INode value = y[var];
if (value != null)
{
if (values[i].ContainsKey(value))
{
possMatches = (possMatches == null ? values[i].GetValues(value).Concat(nulls[i]) : possMatches.Intersect(values[i].GetValues(value).Concat(nulls[i])));
}
else
{
possMatches = Enumerable.Empty <int>();
break;
}
}
else
{
//Don't forget that a null will be potentially compatible with everything
possMatches = (possMatches == null ? this.SetIDs : possMatches.Intersect(this.SetIDs));
}
i++;
}
if (possMatches == null)
{
continue;
}
//Now do the actual joins for the current set
//Note - We access the dictionary directly here because going through the this[int id] method
//incurs a Contains() call each time and we know the IDs must exist because they came from
//our dictionary originally!
foreach (int poss in possMatches)
{
if (this._sets[poss].IsCompatibleWith(y, joinVars))
{
ISet z = this._sets[poss].Join(y);
joinedSet.Add(z);
try
{
if (!expr.Evaluate(subcontext, z.ID).AsSafeBoolean())
{
joinedSet.Remove(z.ID);
standalone.Add(poss);
}
else
{
matched.Add(poss);
}
}
catch
{
joinedSet.Remove(z.ID);
standalone.Add(poss);
}
}
}
}
//Finally add in unmatched sets from LHS
foreach (int id in this.SetIDs)
{
if (!matched.Contains(id) || standalone.Contains(id))
{
joinedSet.Add(this._sets[id].Copy());
}
}
}
return(joinedSet);
}
/// <summary>
/// Does a Left Join of this Multiset to another Multiset where the Join is predicated on the given Expression
/// </summary>
/// <param name="other">Other Multiset</param>
/// <param name="expr">Expression</param>
/// <returns></returns>
public override BaseMultiset LeftJoin(BaseMultiset other, ISparqlExpression expr)
{
//If the Other is the Identity/Null Multiset the result is this Multiset
if (other is IdentityMultiset)
{
return(this);
}
if (other is NullMultiset)
{
return(this);
}
if (other.IsEmpty)
{
return(this);
}
Multiset joinedSet = new Multiset();
LeviathanLeftJoinBinder binder = new LeviathanLeftJoinBinder(joinedSet);
SparqlEvaluationContext subcontext = new SparqlEvaluationContext(binder);
//Find the First Variable from this Multiset which is in both Multisets
//If there is no Variable from this Multiset in the other Multiset then this
//should be a Join operation instead of a LeftJoin
List <String> joinVars = this._variables.Where(v => other.Variables.Contains(v)).ToList();
if (joinVars.Count == 0)
{
//Calculate a Product filtering as we go
foreach (ISet x in this.Sets)
{
bool standalone = false;
foreach (ISet y in other.Sets)
{
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.EffectiveBooleanValue(subcontext, z.ID))
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone)
{
joinedSet.Add(x);
}
}
}
else
{
foreach (ISet x in this.Sets)
{
IEnumerable <ISet> ys = other.Sets.Where(s => joinVars.All(v => x[v] == null || s[v] == null || x[v].Equals(s[v])));
//IEnumerable<ISet> ys = other.Sets.Where(s => s.IsCompatibleWith(x, joinVars));
bool standalone = false;
int i = 0;
foreach (ISet y in ys)
{
i++;
ISet z = x.Join(y);
try
{
joinedSet.Add(z);
if (!expr.EffectiveBooleanValue(subcontext, z.ID))
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
catch
{
joinedSet.Remove(z.ID);
standalone = true;
}
}
if (standalone || i == 0)
{
joinedSet.Add(x);
}
}
}
return(joinedSet);
}
