duckdb 0.8.2-dev4711.0 → 0.8.2-dev4871.0

package/src/duckdb/src/common/row_operations/row_matcher.cpp

@@ -0,0 +1,408 @@
+#include "duckdb/common/row_operations/row_matcher.hpp"
+
+#include "duckdb/common/enum_util.hpp"
+#include "duckdb/common/exception.hpp"
+#include "duckdb/common/types/row/tuple_data_collection.hpp"
+
+namespace duckdb {
+
+using ValidityBytes = TupleDataLayout::ValidityBytes;
+
+template <class OP>
+struct RowMatchOperator {
+	static constexpr const bool COMPARE_NULL = false;
+
+	template <class T>
+	static inline bool Operation(const T &left, const T &right, bool left_null, bool right_null) {
+		if (right_null || left_null) {
+			return false;
+		}
+		return OP::template Operation<T>(left, right);
+	}
+};
+
+template <>
+struct RowMatchOperator<DistinctFrom> {
+	static constexpr const bool COMPARE_NULL = true;
+
+	template <class T>
+	static inline bool Operation(const T &left, const T &right, bool left_null, bool right_null) {
+		return DistinctFrom::template Operation<T>(left, right, left_null, right_null);
+	}
+};
+
+template <>
+struct RowMatchOperator<NotDistinctFrom> {
+	static constexpr const bool COMPARE_NULL = true;
+
+	template <class T>
+	static inline bool Operation(const T &left, const T &right, bool left_null, bool right_null) {
+		return NotDistinctFrom::template Operation<T>(left, right, left_null, right_null);
+	}
+};
+
+template <bool NO_MATCH_SEL, class T, class OP>
+static idx_t TemplatedMatch(Vector &, const TupleDataVectorFormat &lhs_format, SelectionVector &sel, const idx_t count,
+                            const TupleDataLayout &rhs_layout, Vector &rhs_row_locations, const idx_t col_idx,
+                            const vector<MatchFunction> &, SelectionVector *no_match_sel, idx_t &no_match_count) {
+	using MATCH_OP = RowMatchOperator<OP>;
+
+	// LHS
+	const auto &lhs_sel = *lhs_format.unified.sel;
+	const auto lhs_data = UnifiedVectorFormat::GetData<T>(lhs_format.unified);
+	const auto &lhs_validity = lhs_format.unified.validity;
+
+	// RHS
+	const auto rhs_locations = FlatVector::GetData<data_ptr_t>(rhs_row_locations);
+	const auto rhs_offset_in_row = rhs_layout.GetOffsets()[col_idx];
+	idx_t entry_idx;
+	idx_t idx_in_entry;
+	ValidityBytes::GetEntryIndex(col_idx, entry_idx, idx_in_entry);
+
+	idx_t match_count = 0;
+	for (idx_t i = 0; i < count; i++) {
+		const auto idx = sel.get_index(i);
+
+		const auto lhs_idx = lhs_sel.get_index(idx);
+		const auto lhs_null = lhs_validity.AllValid() ? false : !lhs_validity.RowIsValid(lhs_idx);
+
+		const auto &rhs_location = rhs_locations[idx];
+		const ValidityBytes rhs_mask(rhs_location);
+		const auto rhs_null = !rhs_mask.RowIsValid(rhs_mask.GetValidityEntryUnsafe(entry_idx), idx_in_entry);
+
+		if (MATCH_OP::template Operation<T>(lhs_data[lhs_idx], Load<T>(rhs_location + rhs_offset_in_row), lhs_null,
+		                                    rhs_null)) {
+			sel.set_index(match_count++, idx);
+		} else if (NO_MATCH_SEL) {
+			no_match_sel->set_index(no_match_count++, idx);
+		}
+	}
+	return match_count;
+}
+
+template <bool NO_MATCH_SEL, class OP>
+static idx_t StructMatchEquality(Vector &lhs_vector, const TupleDataVectorFormat &lhs_format, SelectionVector &sel,
+                                 const idx_t count, const TupleDataLayout &rhs_layout, Vector &rhs_row_locations,
+                                 const idx_t col_idx, const vector<MatchFunction> &child_functions,
+                                 SelectionVector *no_match_sel, idx_t &no_match_count) {
+	using MATCH_OP = RowMatchOperator<OP>;
+
+	// LHS
+	const auto &lhs_sel = *lhs_format.unified.sel;
+	const auto &lhs_validity = lhs_format.unified.validity;
+
+	// RHS
+	const auto rhs_locations = FlatVector::GetData<data_ptr_t>(rhs_row_locations);
+	idx_t entry_idx;
+	idx_t idx_in_entry;
+	ValidityBytes::GetEntryIndex(col_idx, entry_idx, idx_in_entry);
+
+	idx_t match_count = 0;
+	for (idx_t i = 0; i < count; i++) {
+		const auto idx = sel.get_index(i);
+
+		const auto lhs_idx = lhs_sel.get_index(idx);
+		const auto lhs_null = lhs_validity.AllValid() ? false : !lhs_validity.RowIsValid(lhs_idx);
+
+		const auto &rhs_location = rhs_locations[idx];
+		const ValidityBytes rhs_mask(rhs_location);
+		const auto rhs_null = !rhs_mask.RowIsValid(rhs_mask.GetValidityEntryUnsafe(entry_idx), idx_in_entry);
+
+		// For structs there is no value to compare, here we match NULLs and let recursion do the rest
+		// So we use the comparison only if rhs or LHS is NULL and COMPARE_NULL is true
+		if (!(lhs_null || rhs_null) ||
+		    (MATCH_OP::COMPARE_NULL && MATCH_OP::template Operation<uint32_t>(0, 0, lhs_null, rhs_null))) {
+			sel.set_index(match_count++, idx);
+		} else if (NO_MATCH_SEL) {
+			no_match_sel->set_index(no_match_count++, idx);
+		}
+	}
+
+	// Create a Vector of pointers to the start of the TupleDataLayout of the STRUCT
+	Vector rhs_struct_row_locations(LogicalType::POINTER);
+	const auto rhs_offset_in_row = rhs_layout.GetOffsets()[col_idx];
+	auto rhs_struct_locations = FlatVector::GetData<data_ptr_t>(rhs_struct_row_locations);
+	for (idx_t i = 0; i < match_count; i++) {
+		const auto idx = sel.get_index(i);
+		rhs_struct_locations[idx] = rhs_locations[idx] + rhs_offset_in_row;
+	}
+
+	// Get the struct layout and struct entries
+	const auto &rhs_struct_layout = rhs_layout.GetStructLayout(col_idx);
+	auto &lhs_struct_vectors = StructVector::GetEntries(lhs_vector);
+	D_ASSERT(rhs_struct_layout.ColumnCount() == lhs_struct_vectors.size());
+
+	for (idx_t struct_col_idx = 0; struct_col_idx < rhs_struct_layout.ColumnCount(); struct_col_idx++) {
+		auto &lhs_struct_vector = *lhs_struct_vectors[struct_col_idx];
+		auto &lhs_struct_format = lhs_format.children[struct_col_idx];
+		const auto &child_function = child_functions[struct_col_idx];
+		match_count = child_function.function(lhs_struct_vector, lhs_struct_format, sel, match_count, rhs_struct_layout,
+		                                      rhs_struct_row_locations, struct_col_idx, child_function.child_functions,
+		                                      no_match_sel, no_match_count);
+	}
+
+	return match_count;
+}
+
+template <typename OP>
+static idx_t SelectComparison(Vector &, Vector &, const SelectionVector &, idx_t, SelectionVector *,
+                              SelectionVector *) {
+	throw NotImplementedException("Unsupported list comparison operand for RowMatcher::GetMatchFunction");
+}
+
+template <>
+idx_t SelectComparison<Equals>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                               SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::NestedEquals(left, right, sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<NotEquals>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                  SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::NestedNotEquals(left, right, sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<DistinctFrom>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                     SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::DistinctFrom(left, right, &sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<NotDistinctFrom>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                        SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::NotDistinctFrom(left, right, &sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<GreaterThan>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                    SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::DistinctGreaterThan(left, right, &sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<GreaterThanEquals>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                          SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::DistinctGreaterThanEquals(left, right, &sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<LessThan>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                 SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::DistinctLessThan(left, right, &sel, count, true_sel, false_sel);
+}
+
+template <>
+idx_t SelectComparison<LessThanEquals>(Vector &left, Vector &right, const SelectionVector &sel, idx_t count,
+                                       SelectionVector *true_sel, SelectionVector *false_sel) {
+	return VectorOperations::DistinctLessThanEquals(left, right, &sel, count, true_sel, false_sel);
+}
+
+template <bool NO_MATCH_SEL, class OP>
+static idx_t GenericNestedMatch(Vector &lhs_vector, const TupleDataVectorFormat &, SelectionVector &sel,
+                                const idx_t count, const TupleDataLayout &rhs_layout, Vector &rhs_row_locations,
+                                const idx_t col_idx, const vector<MatchFunction> &, SelectionVector *no_match_sel,
+                                idx_t &no_match_count) {
+	const auto &type = rhs_layout.GetTypes()[col_idx];
+
+	// Gather a dense Vector containing the column values being matched
+	Vector key(type);
+	const auto gather_function = TupleDataCollection::GetGatherFunction(type);
+	gather_function.function(rhs_layout, rhs_row_locations, col_idx, sel, count, key,
+	                         *FlatVector::IncrementalSelectionVector(), key, gather_function.child_functions);
+
+	// Densify the input column
+	Vector sliced(lhs_vector, sel, count);
+
+	if (NO_MATCH_SEL) {
+		SelectionVector no_match_sel_offset(no_match_sel->data() + no_match_count);
+		auto match_count = SelectComparison<OP>(sliced, key, sel, count, &sel, &no_match_sel_offset);
+		no_match_count += count - match_count;
+		return match_count;
+	}
+	return SelectComparison<OP>(sliced, key, sel, count, &sel, nullptr);
+}
+
+void RowMatcher::Initialize(const bool no_match_sel, const TupleDataLayout &layout, const Predicates &predicates) {
+	match_functions.reserve(predicates.size());
+	for (idx_t col_idx = 0; col_idx < predicates.size(); col_idx++) {
+		match_functions.push_back(GetMatchFunction(no_match_sel, layout.GetTypes()[col_idx], predicates[col_idx]));
+	}
+}
+
+idx_t RowMatcher::Match(DataChunk &lhs, const vector<TupleDataVectorFormat> &lhs_formats, SelectionVector &sel,
+                        idx_t count, const TupleDataLayout &rhs_layout, Vector &rhs_row_locations,
+                        SelectionVector *no_match_sel, idx_t &no_match_count) {
+	D_ASSERT(!match_functions.empty());
+	for (idx_t col_idx = 0; col_idx < match_functions.size(); col_idx++) {
+		const auto &match_function = match_functions[col_idx];
+		count =
+		    match_function.function(lhs.data[col_idx], lhs_formats[col_idx], sel, count, rhs_layout, rhs_row_locations,
+		                            col_idx, match_function.child_functions, no_match_sel, no_match_count);
+	}
+	return count;
+}
+
+MatchFunction RowMatcher::GetMatchFunction(const bool no_match_sel, const LogicalType &type,
+                                           const ExpressionType predicate) {
+	return no_match_sel ? GetMatchFunction<true>(type, predicate) : GetMatchFunction<false>(type, predicate);
+}
+
+template <bool NO_MATCH_SEL>
+MatchFunction RowMatcher::GetMatchFunction(const LogicalType &type, const ExpressionType predicate) {
+	switch (type.InternalType()) {
+	case PhysicalType::BOOL:
+		return GetMatchFunction<NO_MATCH_SEL, bool>(predicate);
+	case PhysicalType::INT8:
+		return GetMatchFunction<NO_MATCH_SEL, int8_t>(predicate);
+	case PhysicalType::INT16:
+		return GetMatchFunction<NO_MATCH_SEL, int16_t>(predicate);
+	case PhysicalType::INT32:
+		return GetMatchFunction<NO_MATCH_SEL, int32_t>(predicate);
+	case PhysicalType::INT64:
+		return GetMatchFunction<NO_MATCH_SEL, int64_t>(predicate);
+	case PhysicalType::INT128:
+		return GetMatchFunction<NO_MATCH_SEL, hugeint_t>(predicate);
+	case PhysicalType::UINT8:
+		return GetMatchFunction<NO_MATCH_SEL, uint8_t>(predicate);
+	case PhysicalType::UINT16:
+		return GetMatchFunction<NO_MATCH_SEL, uint16_t>(predicate);
+	case PhysicalType::UINT32:
+		return GetMatchFunction<NO_MATCH_SEL, uint32_t>(predicate);
+	case PhysicalType::UINT64:
+		return GetMatchFunction<NO_MATCH_SEL, uint64_t>(predicate);
+	case PhysicalType::FLOAT:
+		return GetMatchFunction<NO_MATCH_SEL, float>(predicate);
+	case PhysicalType::DOUBLE:
+		return GetMatchFunction<NO_MATCH_SEL, double>(predicate);
+	case PhysicalType::INTERVAL:
+		return GetMatchFunction<NO_MATCH_SEL, interval_t>(predicate);
+	case PhysicalType::VARCHAR:
+		return GetMatchFunction<NO_MATCH_SEL, string_t>(predicate);
+	case PhysicalType::STRUCT:
+		return GetStructMatchFunction<NO_MATCH_SEL>(type, predicate);
+	case PhysicalType::LIST:
+		return GetListMatchFunction<NO_MATCH_SEL>(predicate);
+	default:
+		throw InternalException("Unsupported PhysicalType for RowMatcher::GetMatchFunction: %s",
+		                        EnumUtil::ToString(type.InternalType()));
+	}
+}
+
+template <bool NO_MATCH_SEL, class T>
+MatchFunction RowMatcher::GetMatchFunction(const ExpressionType predicate) {
+	MatchFunction result;
+	switch (predicate) {
+	case ExpressionType::COMPARE_EQUAL:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, Equals>;
+		break;
+	case ExpressionType::COMPARE_NOTEQUAL:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, NotEquals>;
+		break;
+	case ExpressionType::COMPARE_DISTINCT_FROM:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, DistinctFrom>;
+		break;
+	case ExpressionType::COMPARE_NOT_DISTINCT_FROM:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, NotDistinctFrom>;
+		break;
+	case ExpressionType::COMPARE_GREATERTHAN:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, GreaterThan>;
+		break;
+	case ExpressionType::COMPARE_GREATERTHANOREQUALTO:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, GreaterThanEquals>;
+		break;
+	case ExpressionType::COMPARE_LESSTHAN:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, LessThan>;
+		break;
+	case ExpressionType::COMPARE_LESSTHANOREQUALTO:
+		result.function = TemplatedMatch<NO_MATCH_SEL, T, LessThanEquals>;
+		break;
+	default:
+		throw InternalException("Unsupported ExpressionType for RowMatcher::GetMatchFunction: %s",
+		                        EnumUtil::ToString(predicate));
+	}
+	return result;
+}
+
+template <bool NO_MATCH_SEL>
+MatchFunction RowMatcher::GetStructMatchFunction(const LogicalType &type, const ExpressionType predicate) {
+	// We perform equality conditions like it's just a row, but we cannot perform inequality conditions like a row,
+	// because for equality conditions we need to always loop through all columns, but for inequality conditions,
+	// we need to find the first inequality, so the loop looks very different
+	MatchFunction result;
+	ExpressionType child_predicate = predicate;
+	switch (predicate) {
+	case ExpressionType::COMPARE_EQUAL:
+		result.function = StructMatchEquality<NO_MATCH_SEL, Equals>;
+		child_predicate = ExpressionType::COMPARE_NOT_DISTINCT_FROM;
+		break;
+	case ExpressionType::COMPARE_NOTEQUAL:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, NotEquals>;
+		return result;
+	case ExpressionType::COMPARE_DISTINCT_FROM:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, DistinctFrom>;
+		return result;
+	case ExpressionType::COMPARE_NOT_DISTINCT_FROM:
+		result.function = StructMatchEquality<NO_MATCH_SEL, NotDistinctFrom>;
+		break;
+	case ExpressionType::COMPARE_GREATERTHAN:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, GreaterThan>;
+		return result;
+	case ExpressionType::COMPARE_GREATERTHANOREQUALTO:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, GreaterThanEquals>;
+		return result;
+	case ExpressionType::COMPARE_LESSTHAN:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, LessThan>;
+		return result;
+	case ExpressionType::COMPARE_LESSTHANOREQUALTO:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, LessThanEquals>;
+		return result;
+	default:
+		throw InternalException("Unsupported ExpressionType for RowMatcher::GetStructMatchFunction: %s",
+		                        EnumUtil::ToString(predicate));
+	}
+
+	result.child_functions.reserve(StructType::GetChildCount(type));
+	for (const auto &child_type : StructType::GetChildTypes(type)) {
+		result.child_functions.push_back(GetMatchFunction<NO_MATCH_SEL>(child_type.second, child_predicate));
+	}
+
+	return result;
+}
+
+template <bool NO_MATCH_SEL>
+MatchFunction RowMatcher::GetListMatchFunction(const ExpressionType predicate) {
+	MatchFunction result;
+	switch (predicate) {
+	case ExpressionType::COMPARE_EQUAL:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, Equals>;
+		break;
+	case ExpressionType::COMPARE_NOTEQUAL:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, NotEquals>;
+		break;
+	case ExpressionType::COMPARE_DISTINCT_FROM:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, DistinctFrom>;
+		break;
+	case ExpressionType::COMPARE_NOT_DISTINCT_FROM:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, NotDistinctFrom>;
+		break;
+	case ExpressionType::COMPARE_GREATERTHAN:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, GreaterThan>;
+		break;
+	case ExpressionType::COMPARE_GREATERTHANOREQUALTO:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, GreaterThanEquals>;
+		break;
+	case ExpressionType::COMPARE_LESSTHAN:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, LessThan>;
+		break;
+	case ExpressionType::COMPARE_LESSTHANOREQUALTO:
+		result.function = GenericNestedMatch<NO_MATCH_SEL, LessThanEquals>;
+		break;
+	default:
+		throw InternalException("Unsupported ExpressionType for RowMatcher::GetListMatchFunction: %s",
+		                        EnumUtil::ToString(predicate));
+	}
+	return result;
+}
+
+} // namespace duckdb

package/src/duckdb/src/common/types/row/tuple_data_allocator.cpp

@@ -294,7 +294,7 @@ static inline void VerifyStrings(const LogicalTypeId type_id, const data_ptr_t r
 	for (idx_t i = 0; i < count; i++) {
 		const auto &row_location = row_locations[offset + i] + base_col_offset;
 		ValidityBytes row_mask(row_location);
-		if (row_mask.RowIsValid(row_mask.GetValidityEntry(entry_idx), idx_in_entry)) {
+		if (row_mask.RowIsValid(row_mask.GetValidityEntryUnsafe(entry_idx), idx_in_entry)) {
 			auto recomputed_string = Load<string_t>(row_location + col_offset);
 			recomputed_string.Verify();
 		}
@@ -328,7 +328,7 @@ void TupleDataAllocator::RecomputeHeapPointers(Vector &old_heap_ptrs, const Sele
 				const auto idx = offset + i;
 				const auto &row_location = row_locations[idx] + base_col_offset;
 				ValidityBytes row_mask(row_location);
-				if (!row_mask.RowIsValid(row_mask.GetValidityEntry(entry_idx), idx_in_entry)) {
+				if (!row_mask.RowIsValid(row_mask.GetValidityEntryUnsafe(entry_idx), idx_in_entry)) {
 					continue;
 				}
 
@@ -352,7 +352,7 @@ void TupleDataAllocator::RecomputeHeapPointers(Vector &old_heap_ptrs, const Sele
 				const auto idx = offset + i;
 				const auto &row_location = row_locations[idx] + base_col_offset;
 				ValidityBytes row_mask(row_location);
-				if (!row_mask.RowIsValid(row_mask.GetValidityEntry(entry_idx), idx_in_entry)) {
+				if (!row_mask.RowIsValid(row_mask.GetValidityEntryUnsafe(entry_idx), idx_in_entry)) {
 					continue;
 				}
 

package/src/duckdb/src/common/types/row/tuple_data_collection.cpp

@@ -37,13 +37,17 @@ void TupleDataCollection::Initialize() {
 	}
 }
 
-void TupleDataCollection::GetAllColumnIDs(vector<column_t> &column_ids) {
-	column_ids.reserve(layout.ColumnCount());
-	for (idx_t col_idx = 0; col_idx < layout.ColumnCount(); col_idx++) {
+void GetAllColumnIDsInternal(vector<column_t> &column_ids, const idx_t column_count) {
+	column_ids.reserve(column_count);
+	for (idx_t col_idx = 0; col_idx < column_count; col_idx++) {
 		column_ids.emplace_back(col_idx);
 	}
 }
 
+void TupleDataCollection::GetAllColumnIDs(vector<column_t> &column_ids) {
+	GetAllColumnIDsInternal(column_ids, layout.ColumnCount());
+}
+
 const TupleDataLayout &TupleDataCollection::GetLayout() const {
 	return layout;
 }
@@ -108,7 +112,7 @@ void TupleDataCollection::InitializeAppend(TupleDataAppendState &append_state, v
                                            TupleDataPinProperties properties) {
 	VerifyAppendColumns(layout, column_ids);
 	InitializeAppend(append_state.pin_state, properties);
-	InitializeAppend(append_state.chunk_state, std::move(column_ids));
+	InitializeChunkState(append_state.chunk_state, std::move(column_ids));
 }
 
 void TupleDataCollection::InitializeAppend(TupleDataPinState &pin_state, TupleDataPinProperties properties) {
@@ -130,11 +134,11 @@ static void InitializeVectorFormat(vector<TupleDataVectorFormat> &vector_data, c
 			for (const auto &child_entry : child_list) {
 				child_types.emplace_back(child_entry.second);
 			}
-			InitializeVectorFormat(vector_data[col_idx].child_formats, child_types);
+			InitializeVectorFormat(vector_data[col_idx].children, child_types);
 			break;
 		}
 		case PhysicalType::LIST:
-			InitializeVectorFormat(vector_data[col_idx].child_formats, {ListType::GetChildType(type)});
+			InitializeVectorFormat(vector_data[col_idx].children, {ListType::GetChildType(type)});
 			break;
 		default:
 			break;
@@ -142,11 +146,16 @@ static void InitializeVectorFormat(vector<TupleDataVectorFormat> &vector_data, c
 	}
 }
 
-void TupleDataCollection::InitializeAppend(TupleDataChunkState &chunk_state, vector<column_t> column_ids) {
+void TupleDataCollection::InitializeChunkState(TupleDataChunkState &chunk_state, vector<column_t> column_ids) {
+	TupleDataCollection::InitializeChunkState(chunk_state, layout.GetTypes(), std::move(column_ids));
+}
+
+void TupleDataCollection::InitializeChunkState(TupleDataChunkState &chunk_state, const vector<LogicalType> &types,
+                                               vector<column_t> column_ids) {
 	if (column_ids.empty()) {
-		GetAllColumnIDs(column_ids);
+		GetAllColumnIDsInternal(column_ids, types.size());
 	}
-	InitializeVectorFormat(chunk_state.vector_data, layout.GetTypes());
+	InitializeVectorFormat(chunk_state.vector_data, types);
 	chunk_state.column_ids = std::move(column_ids);
 }
 
@@ -211,21 +220,23 @@ void TupleDataCollection::AppendUnified(TupleDataPinState &pin_state, TupleDataC
 }
 
 static inline void ToUnifiedFormatInternal(TupleDataVectorFormat &format, Vector &vector, const idx_t count) {
-	vector.ToUnifiedFormat(count, format.data);
-	format.original_sel = format.data.sel;
-	format.original_owned_sel.Initialize(format.data.owned_sel);
+	vector.ToUnifiedFormat(count, format.unified);
+	format.original_sel = format.unified.sel;
+	format.original_owned_sel.Initialize(format.unified.owned_sel);
 	switch (vector.GetType().InternalType()) {
 	case PhysicalType::STRUCT: {
 		auto &entries = StructVector::GetEntries(vector);
-		D_ASSERT(format.child_formats.size() == entries.size());
+		D_ASSERT(format.children.size() == entries.size());
 		for (idx_t struct_col_idx = 0; struct_col_idx < entries.size(); struct_col_idx++) {
-			ToUnifiedFormatInternal(format.child_formats[struct_col_idx], *entries[struct_col_idx], count);
+			ToUnifiedFormatInternal(reinterpret_cast<TupleDataVectorFormat &>(format.children[struct_col_idx]),
+			                        *entries[struct_col_idx], count);
 		}
 		break;
 	}
 	case PhysicalType::LIST:
-		D_ASSERT(format.child_formats.size() == 1);
-		ToUnifiedFormatInternal(format.child_formats[0], ListVector::GetEntry(vector), ListVector::GetListSize(vector));
+		D_ASSERT(format.children.size() == 1);
+		ToUnifiedFormatInternal(reinterpret_cast<TupleDataVectorFormat &>(format.children[0]),
+		                        ListVector::GetEntry(vector), ListVector::GetListSize(vector));
 		break;
 	default:
 		break;
@@ -242,7 +253,7 @@ void TupleDataCollection::ToUnifiedFormat(TupleDataChunkState &chunk_state, Data
 void TupleDataCollection::GetVectorData(const TupleDataChunkState &chunk_state, UnifiedVectorFormat result[]) {
 	const auto &vector_data = chunk_state.vector_data;
 	for (idx_t i = 0; i < vector_data.size(); i++) {
-		const auto &source = vector_data[i].data;
+		const auto &source = vector_data[i].unified;
 		auto &target = result[i];
 		target.sel = source.sel;
 		target.data = source.data;