chore: Update vendored sources to duckdb/duckdb@f9e96b1 (#438)

Bugfixes (duckdb/duckdb#13982)

Co-authored-by: krlmlr <[email protected]>

src/duckdb/src/common/types/column/column_data_allocator.cpp

@@ -2,6 +2,7 @@
 
 #include "duckdb/common/types/column/column_data_collection_segment.hpp"
 #include "duckdb/storage/buffer/block_handle.hpp"
+#include "duckdb/storage/buffer/buffer_pool.hpp"
 #include "duckdb/storage/buffer_manager.hpp"
 
 namespace duckdb {
@@ -45,6 +46,21 @@ ColumnDataAllocator::ColumnDataAllocator(ColumnDataAllocator &other) {
 	}
 }
 
+ColumnDataAllocator::~ColumnDataAllocator() {
+	if (type == ColumnDataAllocatorType::IN_MEMORY_ALLOCATOR) {
+		return;
+	}
+	for (auto &block : blocks) {
+		block.handle->SetDestroyBufferUpon(DestroyBufferUpon::UNPIN);
+	}
+	const auto data_size = SizeInBytes();
+	blocks.clear();
+	if (Allocator::SupportsFlush() &&
+	    data_size > alloc.buffer_manager->GetBufferPool().GetAllocatorBulkDeallocationFlushThreshold()) {
+		Allocator::FlushAll();
+	}
+}
+
 BufferHandle ColumnDataAllocator::Pin(uint32_t block_id) {
 	D_ASSERT(type == ColumnDataAllocatorType::BUFFER_MANAGER_ALLOCATOR || type == ColumnDataAllocatorType::HYBRID);
 	shared_ptr<BlockHandle> handle;

src/duckdb/src/common/types/row/tuple_data_segment.cpp

@@ -1,6 +1,7 @@
 #include "duckdb/common/types/row/tuple_data_segment.hpp"
 
 #include "duckdb/common/types/row/tuple_data_allocator.hpp"
+#include "duckdb/storage/buffer/buffer_pool.hpp"
 
 namespace duckdb {
 
@@ -118,6 +119,10 @@ TupleDataSegment::~TupleDataSegment() {
 	}
 	pinned_row_handles.clear();
 	pinned_heap_handles.clear();
+	if (Allocator::SupportsFlush() && allocator &&
+	    data_size > allocator->GetBufferManager().GetBufferPool().GetAllocatorBulkDeallocationFlushThreshold()) {
+		Allocator::FlushAll();
+	}
 	allocator.reset();
 }
 

src/duckdb/src/execution/operator/join/physical_piecewise_merge_join.cpp

@@ -618,7 +618,12 @@ OperatorResultType PhysicalPiecewiseMergeJoin::ResolveComplexJoin(ExecutionConte
 
 				if (tail_count < result_count) {
 					result_count = tail_count;
-					chunk.Slice(*sel, result_count);
+					if (result_count == 0) {
+						// Need to reset here otherwise we may use the non-flat chunk when constructing LEFT/OUTER
+						chunk.Reset();
+					} else {
+						chunk.Slice(*sel, result_count);
+					}
 				}
 			}
 

src/duckdb/src/function/table/version/pragma_version.cpp

@@ -1,5 +1,5 @@
 #ifndef DUCKDB_PATCH_VERSION
-#define DUCKDB_PATCH_VERSION "1-dev223"
+#define DUCKDB_PATCH_VERSION "1-dev236"
 #endif
 #ifndef DUCKDB_MINOR_VERSION
 #define DUCKDB_MINOR_VERSION 1
@@ -8,10 +8,10 @@
 #define DUCKDB_MAJOR_VERSION 1
 #endif
 #ifndef DUCKDB_VERSION
-#define DUCKDB_VERSION "v1.1.1-dev223"
+#define DUCKDB_VERSION "v1.1.1-dev236"
 #endif
 #ifndef DUCKDB_SOURCE_ID
-#define DUCKDB_SOURCE_ID "2ff9c687e2"
+#define DUCKDB_SOURCE_ID "f9e96b1910"
 #endif
 #include "duckdb/function/table/system_functions.hpp"
 #include "duckdb/main/database.hpp"

src/duckdb/src/include/duckdb/common/types/column/column_data_allocator.hpp

@@ -32,6 +32,7 @@ class ColumnDataAllocator {
 	explicit ColumnDataAllocator(BufferManager &buffer_manager);
 	ColumnDataAllocator(ClientContext &context, ColumnDataAllocatorType allocator_type);
 	ColumnDataAllocator(ColumnDataAllocator &allocator);
+	~ColumnDataAllocator();
 
 	//! Returns an allocator object to allocate with. This returns the allocator in IN_MEMORY_ALLOCATOR, and a buffer
 	//! allocator in case of BUFFER_MANAGER_ALLOCATOR.

src/duckdb/src/include/duckdb/storage/buffer/buffer_pool.hpp

@@ -50,6 +50,7 @@ class BufferPool {
 
 	//! If bulk deallocation larger than this occurs, flush outstanding allocations
 	void SetAllocatorBulkDeallocationFlushThreshold(idx_t threshold);
+	idx_t GetAllocatorBulkDeallocationFlushThreshold();
 
 	void UpdateUsedMemory(MemoryTag tag, int64_t size);
 

src/duckdb/src/optimizer/filter_pushdown.cpp

@@ -93,9 +93,14 @@ unique_ptr<LogicalOperator> FilterPushdown::Rewrite(unique_ptr<LogicalOperator>
 		// we can just push directly through these operations without any rewriting
 		op->children[0] = Rewrite(std::move(op->children[0]));
 		return op;
-	case LogicalOperatorType::LOGICAL_MATERIALIZED_CTE:
+	case LogicalOperatorType::LOGICAL_MATERIALIZED_CTE: {
+		// we can't push filters into the materialized CTE (LHS), but we do want to recurse into it
+		FilterPushdown pushdown(optimizer, convert_mark_joins);
+		op->children[0] = pushdown.Rewrite(std::move(op->children[0]));
+		// we can push filters into the rest of the query plan (RHS)
 		op->children[1] = Rewrite(std::move(op->children[1]));
 		return op;
+	}
 	case LogicalOperatorType::LOGICAL_GET:
 		return PushdownGet(std::move(op));
 	case LogicalOperatorType::LOGICAL_LIMIT:

src/duckdb/src/optimizer/join_order/cardinality_estimator.cpp

@@ -2,10 +2,10 @@
 #include "duckdb/common/enums/join_type.hpp"
 #include "duckdb/common/limits.hpp"
 #include "duckdb/common/printer.hpp"
-#include "duckdb/planner/expression_iterator.hpp"
 #include "duckdb/function/table/table_scan.hpp"
 #include "duckdb/optimizer/join_order/join_node.hpp"
 #include "duckdb/optimizer/join_order/query_graph_manager.hpp"
+#include "duckdb/planner/expression_iterator.hpp"
 #include "duckdb/planner/operator/logical_comparison_join.hpp"
 #include "duckdb/storage/data_table.hpp"
 
@@ -291,10 +291,18 @@ DenomInfo CardinalityEstimator::GetDenominator(JoinRelationSet &set) {
 	// and we start to choose the filters that join relations in the set.
 
 	// edges are guaranteed to be in order of largest tdom to smallest tdom.
+	unordered_set<idx_t> unused_edge_tdoms;
 	auto edges = GetEdges(relations_to_tdoms, set);
 	for (auto &edge : edges) {
-		auto subgraph_connections = SubgraphsConnectedByEdge(edge, subgraphs);
+		if (subgraphs.size() == 1 && subgraphs.at(0).relations->ToString() == set.ToString()) {
+			// the first subgraph has connected all the desired relations, just skip the rest of the edges
+			if (edge.has_tdom_hll) {
+				unused_edge_tdoms.insert(edge.tdom_hll);
+			}
+			continue;
+		}
 
+		auto subgraph_connections = SubgraphsConnectedByEdge(edge, subgraphs);
 		if (subgraph_connections.empty()) {
 			// create a subgraph out of left and right, then merge right into left and add left to subgraphs.
 			// this helps cover a case where there are no subgraphs yet, and the only join filter is a SEMI JOIN
@@ -342,13 +350,11 @@ DenomInfo CardinalityEstimator::GetDenominator(JoinRelationSet &set) {
 			                                   [](Subgraph2Denominator &s) { return !s.relations; });
 			subgraphs.erase(remove_start, subgraphs.end());
 		}
-		if (subgraphs.size() == 1 && subgraphs.at(0).relations->ToString() == set.ToString()) {
-			// the first subgraph has connected all the desired relations, no need to iterate
-			// through the rest of the edges.
-			break;
-		}
 	}
 
+	// Slight penalty to cardinality for unused edges
+	auto denom_multiplier = 1.0 + static_cast<double>(unused_edge_tdoms.size());
+
 	// It's possible cross-products were added and are not present in the filters in the relation_2_tdom
 	// structures. When that's the case, merge all remaining subgraphs.
 	if (subgraphs.size() > 1) {
@@ -367,7 +373,7 @@ DenomInfo CardinalityEstimator::GetDenominator(JoinRelationSet &set) {
 		// denominator is 1 and numerators are a cross product of cardinalities.
 		return DenomInfo(set, 1, 1);
 	}
-	return DenomInfo(*subgraphs.at(0).numerator_relations, 1, subgraphs.at(0).denom);
+	return DenomInfo(*subgraphs.at(0).numerator_relations, 1, subgraphs.at(0).denom * denom_multiplier);
 }
 
 template <>

src/duckdb/src/planner/binder.cpp

@@ -342,7 +342,8 @@ unique_ptr<BoundQueryNode> Binder::BindNode(QueryNode &node) {
 
 BoundStatement Binder::Bind(QueryNode &node) {
 	BoundStatement result;
-	if (context.db->config.options.disabled_optimizers.find(OptimizerType::MATERIALIZED_CTE) ==
+	if (node.type != QueryNodeType::CTE_NODE && // Issue #13850 - Don't auto-materialize if users materialize (for now)
+	    context.db->config.options.disabled_optimizers.find(OptimizerType::MATERIALIZED_CTE) ==
 	        context.db->config.options.disabled_optimizers.end() &&
 	    context.config.enable_optimizer && OptimizeCTEs(node)) {
 		switch (node.type) {

src/duckdb/src/storage/buffer/buffer_pool.cpp

@@ -413,6 +413,10 @@ void BufferPool::SetAllocatorBulkDeallocationFlushThreshold(idx_t threshold) {
 	allocator_bulk_deallocation_flush_threshold = threshold;
 }
 
+idx_t BufferPool::GetAllocatorBulkDeallocationFlushThreshold() {
+	return allocator_bulk_deallocation_flush_threshold;
+}
+
 BufferPool::MemoryUsage::MemoryUsage() {
 	for (auto &v : memory_usage) {
 		v = 0;