Examples with UNION org.apache.calcite.sql.SqlKind.UNION used on opensource projects

Example 1 with UNION

use of org.apache.calcite.sql.SqlKind.UNION in project calcite by apache.

the class RelBuilder method rewriteAggregateWithDuplicateGroupSets.

/**
 * The {@code GROUP_ID()} function is used to distinguish duplicate groups.
 * However, as Aggregate normalizes group sets to canonical form (i.e.,
 * flatten, sorting, redundancy removal), this information is lost in RelNode.
 * Therefore, it is impossible to implement the function in runtime.
 *
 * <p>To fill this gap, an aggregation query that contains duplicate group
 * sets is rewritten into a Union of Aggregate operators whose group sets are
 * distinct. The number of inputs to the Union is equal to the maximum number
 * of duplicates. In the {@code N}th input to the Union, calls to the
 * {@code GROUP_ID} aggregate function are replaced by the integer literal
 * {@code N}.
 *
 * <p>This method also handles the case where group sets are distinct but
 * there is a call to {@code GROUP_ID}. That call is replaced by the integer
 * literal {@code 0}.
 *
 * <p>Also see the discussion in
 * <a href="https://issues.apache.org/jira/browse/CALCITE-1824">[CALCITE-1824]
 * GROUP_ID returns wrong result</a> and
 * <a href="https://issues.apache.org/jira/browse/CALCITE-4748">[CALCITE-4748]
 * If there are duplicate GROUPING SETS, Calcite should return duplicate
 * rows</a>.
 */
private RelBuilder rewriteAggregateWithDuplicateGroupSets(ImmutableBitSet groupSet, ImmutableSortedMultiset<ImmutableBitSet> groupSets, List<AggCall> aggregateCalls) {
    final List<String> fieldNamesIfNoRewrite = Aggregate.deriveRowType(getTypeFactory(), peek().getRowType(), false, groupSet, groupSets.asList(), aggregateCalls.stream().map(c -> ((AggCallPlus) c).aggregateCall()).collect(Util.toImmutableList())).getFieldNames();
    // If n duplicates exist for a particular grouping, the {@code GROUP_ID()}
    // function produces values in the range 0 to n-1. For each value,
    // we need to figure out the corresponding group sets.
    // 
    // For example, "... GROUPING SETS (a, a, b, c, c, c, c)"
    // (i) The max value of the GROUP_ID() function returns is 3
    // (ii) GROUPING SETS (a, b, c) produces value 0,
    // GROUPING SETS (a, c) produces value 1,
    // GROUPING SETS (c) produces value 2
    // GROUPING SETS (c) produces value 3
    final Map<Integer, Set<ImmutableBitSet>> groupIdToGroupSets = new HashMap<>();
    int maxGroupId = 0;
    for (Multiset.Entry<ImmutableBitSet> entry : groupSets.entrySet()) {
        int groupId = entry.getCount() - 1;
        if (groupId > maxGroupId) {
            maxGroupId = groupId;
        }
        for (int i = 0; i <= groupId; i++) {
            groupIdToGroupSets.computeIfAbsent(i, k -> Sets.newTreeSet(ImmutableBitSet.COMPARATOR)).add(entry.getElement());
        }
    }
    // AggregateCall list without GROUP_ID function
    final List<AggCall> aggregateCallsWithoutGroupId = new ArrayList<>(aggregateCalls);
    aggregateCallsWithoutGroupId.removeIf(RelBuilder::isGroupId);
    // For each group id value, we first construct an Aggregate without
    // GROUP_ID() function call, and then create a Project node on top of it.
    // The Project adds literal value for group id in right position.
    final Frame frame = stack.pop();
    for (int groupId = 0; groupId <= maxGroupId; groupId++) {
        // Create the Aggregate node without GROUP_ID() call
        stack.push(frame);
        aggregate(groupKey(groupSet, castNonNull(groupIdToGroupSets.get(groupId))), aggregateCallsWithoutGroupId);
        final List<RexNode> selectList = new ArrayList<>();
        final int groupExprLength = groupSet.cardinality();
        // Project fields in group by expressions
        for (int i = 0; i < groupExprLength; i++) {
            selectList.add(field(i));
        }
        // Project fields in aggregate calls
        int groupIdCount = 0;
        for (int i = 0; i < aggregateCalls.size(); i++) {
            if (isGroupId(aggregateCalls.get(i))) {
                selectList.add(getRexBuilder().makeExactLiteral(BigDecimal.valueOf(groupId), getTypeFactory().createSqlType(SqlTypeName.BIGINT)));
                groupIdCount++;
            } else {
                selectList.add(field(groupExprLength + i - groupIdCount));
            }
        }
        project(selectList, fieldNamesIfNoRewrite);
    }
    return union(true, maxGroupId + 1);
}

Example 2 with UNION

use of org.apache.calcite.sql.SqlKind.UNION in project calcite by apache.

the class RelBuilder method setOp.

private RelBuilder setOp(boolean all, SqlKind kind, int n) {
    List<RelNode> inputs = new ArrayList<>();
    for (int i = 0; i < n; i++) {
        inputs.add(0, build());
    }
    switch(kind) {
        case UNION:
        case INTERSECT:
        case EXCEPT:
            if (n < 1) {
                throw new IllegalArgumentException("bad INTERSECT/UNION/EXCEPT input count");
            }
            break;
        default:
            throw new AssertionError("bad setOp " + kind);
    }
    if (n == 1) {
        return push(inputs.get(0));
    }
    if (config.simplifyValues() && kind == UNION && inputs.stream().allMatch(r -> r instanceof Values)) {
        List<RelDataType> inputTypes = Util.transform(inputs, RelNode::getRowType);
        RelDataType rowType = getTypeFactory().leastRestrictive(inputTypes);
        requireNonNull(rowType, () -> "leastRestrictive(" + inputTypes + ")");
        final List<List<RexLiteral>> tuples = new ArrayList<>();
        for (RelNode input : inputs) {
            tuples.addAll(((Values) input).tuples);
        }
        final List<List<RexLiteral>> tuples2 = all ? tuples : Util.distinctList(tuples);
        return values(tuples2, rowType);
    }
    return push(struct.setOpFactory.createSetOp(kind, inputs, all));
}