/**
* Creates a relational expression which projects an array of expressions, and optionally
* optimizes.
*
* <p>The result may not be a {@link ProjectRel}. If the projection is trivial, <code>child</code>
* is returned directly; and future versions may return other formulations of expressions, such as
* {@link CalcRel}.
*
* @param child input relational expression
* @param exprs list of expressions for the input columns
* @param fieldNames aliases of the expressions, or null to generate
* @param optimize Whether to return <code>child</code> unchanged if the projections are trivial.
*/
public static RelNode createProject(
RelNode child, List<RexNode> exprs, List<String> fieldNames, boolean optimize) {
final RelOptCluster cluster = child.getCluster();
final RexProgram program =
RexProgram.create(child.getRowType(), exprs, null, fieldNames, cluster.getRexBuilder());
final List<RelCollation> collationList = program.getCollations(child.getCollationList());
if (DeprecateProjectAndFilter) {
return new CalcRel(
cluster, child.getTraitSet(), child, program.getOutputRowType(), program, collationList);
} else {
final RelDataType rowType =
RexUtil.createStructType(cluster.getTypeFactory(), exprs, fieldNames);
if (optimize && RemoveTrivialProjectRule.isIdentity(exprs, rowType, child.getRowType())) {
return child;
}
return new ProjectRel(
cluster,
cluster.traitSetOf(
collationList.isEmpty() ? RelCollationImpl.EMPTY : collationList.get(0)),
child,
exprs,
rowType,
ProjectRelBase.Flags.Boxed);
}
}
/**
* Creates a relational expression which projects an array of expressions, and optionally
* optimizes.
*
* <p>The result may not be a {@link ProjectRel}. If the projection is trivial, <code>child</code>
* is returned directly; and future versions may return other formulations of expressions, such as
* {@link CalcRel}.
*
* @param child input relational expression
* @param exprs list of expressions for the input columns
* @param fieldNames aliases of the expressions, or null to generate
* @param optimize Whether to return <code>child</code> unchanged if the projections are trivial.
*/
public static RelNode createProject(
RelNode child, List<RexNode> exprs, List<String> fieldNames, boolean optimize) {
final RelOptCluster cluster = child.getCluster();
final RexProgram program =
RexProgram.create(child.getRowType(), exprs, null, fieldNames, cluster.getRexBuilder());
final List<RelCollation> collationList = program.getCollations(child.getCollationList());
if (DEPRECATE_PROJECT_AND_FILTER) {
return new CalcRel(
cluster, child.getTraitSet(), child, program.getOutputRowType(), program, collationList);
} else {
final RelDataType rowType =
RexUtil.createStructType(
cluster.getTypeFactory(),
exprs,
fieldNames == null
? null
: SqlValidatorUtil.uniquify(fieldNames, SqlValidatorUtil.F_SUGGESTER));
if (optimize && RemoveTrivialProjectRule.isIdentity(exprs, rowType, child.getRowType())) {
return child;
}
return new ProjectRel(
cluster,
cluster.traitSetOf(
collationList.isEmpty() ? RelCollationImpl.EMPTY : collationList.get(0)),
child,
exprs,
rowType,
ProjectRelBase.Flags.BOXED);
}
}
public void onMatch(RelOptRuleCall call) {
assert matches(call);
final JoinRel join = (JoinRel) call.rels[0];
final List<Integer> leftKeys = new ArrayList<Integer>();
final List<Integer> rightKeys = new ArrayList<Integer>();
RelNode right = join.getRight();
final RelNode left = join.getLeft();
RexNode remainingCondition =
RelOptUtil.splitJoinCondition(left, right, join.getCondition(), leftKeys, rightKeys);
assert leftKeys.size() == rightKeys.size();
final List<CorrelatorRel.Correlation> correlationList =
new ArrayList<CorrelatorRel.Correlation>();
if (leftKeys.size() > 0) {
final RelOptCluster cluster = join.getCluster();
final RexBuilder rexBuilder = cluster.getRexBuilder();
int k = 0;
RexNode condition = null;
for (Integer leftKey : leftKeys) {
Integer rightKey = rightKeys.get(k++);
final String dyn_inIdStr = cluster.getQuery().createCorrel();
final int dyn_inId = RelOptQuery.getCorrelOrdinal(dyn_inIdStr);
// Create correlation to say 'each row, set variable #id
// to the value of column #leftKey'.
correlationList.add(new CorrelatorRel.Correlation(dyn_inId, leftKey));
condition =
RelOptUtil.andJoinFilters(
rexBuilder,
condition,
rexBuilder.makeCall(
SqlStdOperatorTable.equalsOperator,
rexBuilder.makeInputRef(
right.getRowType().getFieldList().get(rightKey).getType(), rightKey),
rexBuilder.makeCorrel(
left.getRowType().getFieldList().get(leftKey).getType(), dyn_inIdStr)));
}
right = CalcRel.createFilter(right, condition);
}
RelNode newRel =
new CorrelatorRel(
join.getCluster(),
left,
right,
remainingCondition,
correlationList,
join.getJoinType());
call.transformTo(newRel);
}
/**
* Creates a relational expression which filters according to a given condition, returning the
* same fields as its input.
*
* @param child Child relational expression
* @param condition Condition
* @return Relational expression
*/
public static RelNode createFilter(RelNode child, RexNode condition) {
if (DeprecateProjectAndFilter) {
final RelOptCluster cluster = child.getCluster();
RexProgramBuilder builder =
new RexProgramBuilder(child.getRowType(), cluster.getRexBuilder());
builder.addIdentity();
builder.addCondition(condition);
final RexProgram program = builder.getProgram();
return new CalcRel(
cluster,
child.getTraitSet(),
child,
program.getOutputRowType(),
program,
Collections.<RelCollation>emptyList());
} else {
return new FilterRel(child.getCluster(), child, condition);
}
}
/**
* Creates a ProjectRel with no sort keys.
*
* @param cluster Cluster this relational expression belongs to
* @param child input relational expression
* @param exps set of expressions for the input columns
* @param fieldNames aliases of the expressions
* @param flags values as in {@link ProjectRelBase.Flags}
*/
public ProjectRel(
RelOptCluster cluster, RelNode child, RexNode[] exps, String[] fieldNames, int flags) {
this(
cluster,
child,
exps,
RexUtil.createStructType(cluster.getTypeFactory(), exps, fieldNames),
flags,
Collections.<RelCollation>emptyList());
}
private void onMatchRight(RelOptRuleCall call) {
final JoinRelBase topJoin = call.rel(0);
final JoinRelBase bottomJoin = call.rel(1);
final RelNode relC = call.rel(2);
final RelNode relA = bottomJoin.getLeft();
final RelNode relB = bottomJoin.getRight();
final RelOptCluster cluster = topJoin.getCluster();
// topJoin
// / \
// bottomJoin C
// / \
// A B
final int aCount = relA.getRowType().getFieldCount();
final int bCount = relB.getRowType().getFieldCount();
final int cCount = relC.getRowType().getFieldCount();
final BitSet bBitSet = BitSets.range(aCount, aCount + bCount);
// becomes
//
// newTopJoin
// / \
// newBottomJoin B
// / \
// A C
// If either join is not inner, we cannot proceed.
// (Is this too strict?)
if (topJoin.getJoinType() != JoinRelType.INNER
|| bottomJoin.getJoinType() != JoinRelType.INNER) {
return;
}
// Split the condition of topJoin into a conjunction. Each of the
// parts that does not use columns from B can be pushed down.
final List<RexNode> intersecting = new ArrayList<RexNode>();
final List<RexNode> nonIntersecting = new ArrayList<RexNode>();
split(topJoin.getCondition(), bBitSet, intersecting, nonIntersecting);
// If there's nothing to push down, it's not worth proceeding.
if (nonIntersecting.isEmpty()) {
return;
}
// Split the condition of bottomJoin into a conjunction. Each of the
// parts that use columns from B will need to be pulled up.
final List<RexNode> bottomIntersecting = new ArrayList<RexNode>();
final List<RexNode> bottomNonIntersecting = new ArrayList<RexNode>();
split(bottomJoin.getCondition(), bBitSet, bottomIntersecting, bottomNonIntersecting);
// target: | A | C |
// source: | A | B | C |
final Mappings.TargetMapping bottomMapping =
Mappings.createShiftMapping(
aCount + bCount + cCount, 0, 0, aCount, aCount, aCount + bCount, cCount);
List<RexNode> newBottomList = new ArrayList<RexNode>();
new RexPermuteInputsShuttle(bottomMapping, relA, relC)
.visitList(nonIntersecting, newBottomList);
final Mappings.TargetMapping bottomBottomMapping =
Mappings.createShiftMapping(aCount + bCount, 0, 0, aCount);
new RexPermuteInputsShuttle(bottomBottomMapping, relA, relC)
.visitList(bottomNonIntersecting, newBottomList);
final RexBuilder rexBuilder = cluster.getRexBuilder();
RexNode newBottomCondition = RexUtil.composeConjunction(rexBuilder, newBottomList, false);
final JoinRelBase newBottomJoin =
bottomJoin.copy(
bottomJoin.getTraitSet(), newBottomCondition, relA, relC, bottomJoin.getJoinType());
// target: | A | C | B |
// source: | A | B | C |
final Mappings.TargetMapping topMapping =
Mappings.createShiftMapping(
aCount + bCount + cCount,
0,
0,
aCount,
aCount + cCount,
aCount,
bCount,
aCount,
aCount + bCount,
cCount);
List<RexNode> newTopList = new ArrayList<RexNode>();
new RexPermuteInputsShuttle(topMapping, newBottomJoin, relB)
.visitList(intersecting, newTopList);
new RexPermuteInputsShuttle(topMapping, newBottomJoin, relB)
.visitList(bottomIntersecting, newTopList);
RexNode newTopCondition = RexUtil.composeConjunction(rexBuilder, newTopList, false);
@SuppressWarnings("SuspiciousNameCombination")
final JoinRelBase newTopJoin =
topJoin.copy(
topJoin.getTraitSet(), newTopCondition, newBottomJoin, relB, topJoin.getJoinType());
assert !Mappings.isIdentity(topMapping);
final RelNode newProject =
RelFactories.createProject(projectFactory, newTopJoin, Mappings.asList(topMapping));
call.transformTo(newProject);
}
/**
* Creates a new ValuesRel. Note that tuples passed in become owned by this rel (without a deep
* copy), so caller must not modify them after this call, otherwise bad things will happen.
*
* @param cluster .
* @param rowType row type for tuples produced by this rel
* @param tuples 2-dimensional array of tuple values to be produced; outer list contains tuples;
* each inner list is one tuple; all tuples must be of same length, conforming to rowType
*/
public ValuesRel(RelOptCluster cluster, RelDataType rowType, List<List<RexLiteral>> tuples) {
super(cluster, rowType, tuples, cluster.traitSetOf(Convention.NONE));
}