#pragma once #include #include #include #include #include #include #include #include class NullMap; template struct StringRefWrapper { const ColumnType * column = nullptr; size_t row = 0; StringRef ref; StringRefWrapper(const ColumnType * column, size_t row) : column(column), row(row) {} StringRefWrapper(StringRef ref) : ref(ref) {} StringRefWrapper(const StringRefWrapper & other) = default; StringRefWrapper & operator =(int) { column = nullptr; return *this; } bool operator ==(int) const { return nullptr == column; } StringRefWrapper() {} operator StringRef() const { return column ? column->getDataAt(row) : ref; } bool operator==(const StringRefWrapper & other) const { return (column && column == other.column && row == other.row) || StringRef(*this) == other; } }; namespace ZeroTraits { template bool check(const StringRefWrapper x) { return nullptr == x.column; } template void set(StringRefWrapper & x) { x.column = nullptr; } }; namespace DB { template class ColumnUnique final : public COWPtrHelper> { friend class COWPtrHelper>; private: explicit ColumnUnique(MutableColumnPtr && holder); explicit ColumnUnique(const DataTypePtr & type); ColumnUnique(const ColumnUnique & other) : column_holder(other.column_holder), is_nullable(other.is_nullable) {} public: const ColumnPtr & getNestedColumn() const override; size_t uniqueInsert(const Field & x) override; size_t uniqueInsertFrom(const IColumn & src, size_t n) override; ColumnPtr uniqueInsertRangeFrom(const IColumn & src, size_t start, size_t length) override; size_t uniqueInsertData(const char * pos, size_t length) override; size_t uniqueInsertDataWithTerminatingZero(const char * pos, size_t length) override; size_t uniqueDeserializeAndInsertFromArena(const char * pos, const char *& new_pos) override; size_t getDefaultValueIndex() const override { return is_nullable ? 1 : 0; } size_t getNullValueIndex() const override; bool canContainNulls() const override { return is_nullable; } Field operator[](size_t n) const override { return (*column_holder)[n]; } void get(size_t n, Field & res) const override { column_holder->get(n, res); } StringRef getDataAt(size_t n) const override { return column_holder->getDataAt(n); } StringRef getDataAtWithTerminatingZero(size_t n) const override { return column_holder->getDataAtWithTerminatingZero(n); } UInt64 get64(size_t n) const override { return column_holder->get64(n); } UInt64 getUInt(size_t n) const override { return column_holder->getUInt(n); } Int64 getInt(size_t n) const override { return column_holder->getInt(n); } bool isNullAt(size_t n) const override { return column_holder->isNullAt(n); } ColumnPtr cut(size_t start, size_t length) const override { return column_holder->cut(start, length); } StringRef serializeValueIntoArena(size_t n, Arena & arena, char const *& begin) const override { return column_holder->serializeValueIntoArena(n, arena, begin); } void updateHashWithValue(size_t n, SipHash & hash) const override { return column_holder->updateHashWithValue(n, hash); } ColumnPtr filter(const IColumn::Filter & filt, ssize_t result_size_hint) const override { return column_holder->filter(filt, result_size_hint); } ColumnPtr permute(const IColumn::Permutation & perm, size_t limit) const override { return column_holder->permute(perm, limit); } int compareAt(size_t n, size_t m, const IColumn & rhs, int nan_direction_hint) const override { return column_holder->compareAt(n, m, rhs, nan_direction_hint); } void getPermutation(bool reverse, size_t limit, int nan_direction_hint, IColumn::Permutation & res) const override { column_holder->getPermutation(reverse, limit, nan_direction_hint, res); } ColumnPtr replicate(const IColumn::Offsets & offsets) const override { return column_holder->replicate(offsets); } std::vector scatter(IColumn::ColumnIndex num_columns, const IColumn::Selector & selector) const override { return column_holder->scatter(num_columns, selector); } void getExtremes(Field & min, Field & max) const override { column_holder->getExtremes(min, max); } bool valuesHaveFixedSize() const override { return column_holder->valuesHaveFixedSize(); } bool isFixedAndContiguous() const override { return column_holder->isFixedAndContiguous(); } size_t sizeOfValueIfFixed() const override { return column_holder->sizeOfValueIfFixed(); } bool isNumeric() const override { return column_holder->isNumeric(); } size_t byteSize() const override { return column_holder->byteSize(); } size_t allocatedBytes() const override { return column_holder->allocatedBytes() + (index ? index->getBufferSizeInBytes() : 0); } void forEachSubcolumn(IColumn::ColumnCallback callback) override { callback(column_holder); } private: using IndexMapType = HashMap, IndexType, StringRefHash>; ColumnPtr column_holder; /// For DataTypeNullable, nullptr otherwise. ColumnPtr nullable_column; NullMap * nullable_column_map = nullptr; /// Lazy initialized. std::unique_ptr index; bool is_nullable; size_t numSpecialValues() const { return is_nullable ? 2 : 1; } void buildIndex(); ColumnType * getRawColumnPtr() { return static_cast(column_holder->assumeMutable().get()); } const ColumnType * getRawColumnPtr() const { return static_cast(column_holder.get()); } IndexType insert(const StringRefWrapper & ref, IndexType value); }; template ColumnUnique::ColumnUnique(const DataTypePtr & type) : is_nullable(type->isNullable()) { if (is_nullable) { nullable_column = type->createColumn()->cloneResized(numSpecialValues()); auto & column_nullable = static_cast(nullable_column->assumeMutableRef()); column_holder = column_nullable.getNestedColumnPtr(); nullable_column_map = &column_nullable.getNullMapData(); (*nullable_column_map)[1] = 0; } else column_holder = type->createColumn()->cloneResized(numSpecialValues()); } template ColumnUnique::ColumnUnique(MutableColumnPtr && holder) : column_holder(std::move(holder)) { if (column_holder->isColumnNullable()) { nullable_column = std::move(column_holder); auto & column_nullable = static_cast(nullable_column->assumeMutableRef()); column_holder = column_nullable.getNestedColumnPtr(); nullable_column_map = &column_nullable.getNullMapData(); is_nullable = true; } } template const ColumnPtr& ColumnUnique::getNestedColumn() const { if (is_nullable) { nullable_column_map->resize_fill(column_holder->size()); return nullable_column; } return column_holder; } template size_t ColumnUnique::getNullValueIndex() const { if (!is_nullable) throw Exception("ColumnUnique can't contain null values."); return 0; } template void ColumnUnique::buildIndex() { if (index) return; auto column = getRawColumnPtr(); index = std::make_unique(); (*index)[StringRefWrapper(column, getDefaultValueIndex())] = getDefaultValueIndex(); for (auto row : ext::range(numSpecialValues(), column->size())) { (*index)[StringRefWrapper(column, row)] = row; } } template IndexType ColumnUnique::insert(const StringRefWrapper & ref, IndexType value) { if (!index) buildIndex(); using IteratorType = typename IndexMapType::iterator; IteratorType it; bool inserted; index->emplace(ref, it, inserted); if (inserted) it->second = value; return it->second; } template size_t ColumnUnique::uniqueInsert(const Field & x) { if (x.getType() == Field::Types::Null) return getNullValueIndex(); auto column = getRawColumnPtr(); auto prev_size = static_cast(column->size()); if ((*column)[getDefaultValueIndex()] == x) return getDefaultValueIndex(); column->insert(x); auto pos = insert(StringRefWrapper(column, prev_size), prev_size); if (pos != prev_size) column->popBack(1); return static_cast(pos); } template size_t ColumnUnique::uniqueInsertFrom(const IColumn & src, size_t n) { if (is_nullable && src.isNullAt(n)) return getNullValueIndex(); auto ref = src.getDataAt(n); return uniqueInsertData(ref.data, ref.size); } template size_t ColumnUnique::uniqueInsertData(const char * pos, size_t length) { auto column = getRawColumnPtr(); if (column->getDataAt(getDefaultValueIndex()) == StringRef(pos, length)) return getDefaultValueIndex(); auto size = static_cast(column->size()); if (!index->has(StringRefWrapper(StringRef(pos, length)))) { column->insertData(pos, length); return static_cast(insert(StringRefWrapper(StringRef(pos, length)), size)); } return size; } template size_t ColumnUnique::uniqueInsertDataWithTerminatingZero(const char * pos, size_t length) { if (std::is_same::value) return uniqueInsertData(pos, length - 1); if (column_holder->valuesHaveFixedSize()) return uniqueInsertData(pos, length); /// Don't know if data actually has terminating zero. So, insert it firstly. auto column = getRawColumnPtr(); size_t prev_size = column->size(); column->insertDataWithTerminatingZero(pos, length); if (column->compareAt(getDefaultValueIndex(), prev_size, *column, 1) == 0) { column->popBack(1); return getDefaultValueIndex(); } auto position = insert(StringRefWrapper(column, prev_size), prev_size); if (position != prev_size) column->popBack(1); return static_cast(position); } template size_t ColumnUnique::uniqueDeserializeAndInsertFromArena(const char * pos, const char *& new_pos) { auto column = getRawColumnPtr(); size_t prev_size = column->size(); new_pos = column->deserializeAndInsertFromArena(pos); if (column->compareAt(getDefaultValueIndex(), prev_size, *column, 1) == 0) { column->popBack(1); return getDefaultValueIndex(); } auto index_pos = insert(StringRefWrapper(column, prev_size), prev_size); if (index_pos != prev_size) column->popBack(1); return static_cast(index_pos); } template ColumnPtr ColumnUnique::uniqueInsertRangeFrom(const IColumn & src, size_t start, size_t length) { if (!index) buildIndex(); const ColumnType * src_column; const NullMap * null_map = nullptr; if (src.isColumnNullable()) { auto nullable_column = static_cast(&src); src_column = static_cast(&nullable_column->getNestedColumn()); null_map = &nullable_column->getNullMapData(); } else src_column = static_cast(&src); auto column = getRawColumnPtr(); auto positions_column = ColumnVector::create(length); auto & positions = positions_column->getData(); size_t next_position = column->size(); for (auto i : ext::range(0, length)) { auto row = start + i; if (null_map && (*null_map)[row]) positions[i] = getNullValueIndex(); else if (column->compareAt(getDefaultValueIndex(), row, *src_column, 1) == 0) positions[i] = getDefaultValueIndex(); else { auto it = index->find(StringRefWrapper(src_column, row)); if (it == index->end()) { positions[i] = next_position; auto ref = src_column->getDataAt(row); column->insertData(ref.data, ref.size); (*index)[StringRefWrapper(column, next_position)] = next_position; ++next_position; } else positions[i] = it->second; } } return positions_column; } }