| 660 | } |
| 661 | |
| 662 | MergeTreeSetIndex::MergeTreeSetIndex(const Columns & set_elements, std::vector<KeyTuplePositionMapping> && indexes_mapping_) |
| 663 | : has_all_keys(set_elements.size() == indexes_mapping_.size()), indexes_mapping(std::move(indexes_mapping_)) |
| 664 | { |
| 665 | ::sort(indexes_mapping.begin(), indexes_mapping.end(), |
| 666 | [](const KeyTuplePositionMapping & l, const KeyTuplePositionMapping & r) |
| 667 | { |
| 668 | return std::tie(l.key_index, l.tuple_index) < std::tie(r.key_index, r.tuple_index); |
| 669 | }); |
| 670 | |
| 671 | /// Deduplicate key columns. This can make the condition weaker, e.g.: |
| 672 | /// (x + 1, x + 2) IN (10, 10) |
| 673 | /// effectively turns into: |
| 674 | /// (x + 1) IN (10) |
| 675 | indexes_mapping.erase(std::unique( |
| 676 | indexes_mapping.begin(), indexes_mapping.end(), |
| 677 | [](const KeyTuplePositionMapping & l, const KeyTuplePositionMapping & r) |
| 678 | { |
| 679 | return l.key_index == r.key_index; |
| 680 | }), indexes_mapping.end()); |
| 681 | |
| 682 | size_t tuple_size = indexes_mapping.size(); |
| 683 | ordered_set.resize(tuple_size); |
| 684 | |
| 685 | for (size_t i = 0; i < tuple_size; ++i) |
| 686 | ordered_set[i] = set_elements[indexes_mapping[i].tuple_index]; |
| 687 | |
| 688 | Block block_to_sort; |
| 689 | SortDescription sort_description; |
| 690 | for (size_t i = 0; i < tuple_size; ++i) |
| 691 | { |
| 692 | String column_name = "_" + toString(i); |
| 693 | block_to_sort.insert({ordered_set[i], nullptr, column_name}); |
| 694 | sort_description.emplace_back(column_name, 1, 1); |
| 695 | } |
| 696 | |
| 697 | sortBlock(block_to_sort, sort_description); |
| 698 | |
| 699 | for (size_t i = 0; i < tuple_size; ++i) |
| 700 | ordered_set[i] = block_to_sort.getByPosition(i).column; |
| 701 | } |
| 702 | |
| 703 | /** Return the BoolMask where: |
| 704 | * 1: the intersection of the set and the range is non-empty |