duckdb 0.7.1-dev90.0 → 0.7.2-dev0.0

package/README.md

@@ -1,6 +1,6 @@
 # DuckDB Node Bindings
 
-This package provides a node.js API for [DuckDB](https://github.com/cwida/duckdb), the "SQLite for Analytics". The API for this client is somewhat compliant to the SQLite node.js client for easier transition (and transition you must eventually).
+This package provides a node.js API for [DuckDB](https://github.com/duckdb/duckdb), the "SQLite for Analytics". The API for this client is somewhat compliant to the SQLite node.js client for easier transition (and transition you must eventually).
 
 Load the package and create a database object:
 

package/binding.gyp

@@ -222,16 +222,16 @@
                 "src/duckdb/third_party/zstd/compress/zstd_lazy.cpp", 
                 "src/duckdb/third_party/zstd/compress/zstd_ldm.cpp", 
                 "src/duckdb/third_party/zstd/compress/zstd_opt.cpp", 
-                "src/duckdb/extension/icu/./icu-timezone.cpp", 
-                "src/duckdb/extension/icu/./icu-makedate.cpp", 
-                "src/duckdb/extension/icu/./icu-datepart.cpp", 
-                "src/duckdb/extension/icu/./icu-datesub.cpp", 
+                "src/duckdb/extension/icu/./icu-dateadd.cpp", 
                 "src/duckdb/extension/icu/./icu-datetrunc.cpp", 
-                "src/duckdb/extension/icu/./icu-timebucket.cpp", 
                 "src/duckdb/extension/icu/./icu-strptime.cpp", 
-                "src/duckdb/extension/icu/./icu-extension.cpp", 
-                "src/duckdb/extension/icu/./icu-dateadd.cpp", 
                 "src/duckdb/extension/icu/./icu-datefunc.cpp", 
+                "src/duckdb/extension/icu/./icu-extension.cpp", 
+                "src/duckdb/extension/icu/./icu-makedate.cpp", 
+                "src/duckdb/extension/icu/./icu-timezone.cpp", 
+                "src/duckdb/extension/icu/./icu-datesub.cpp", 
+                "src/duckdb/extension/icu/./icu-timebucket.cpp", 
+                "src/duckdb/extension/icu/./icu-datepart.cpp", 
                 "src/duckdb/ub_extension_icu_third_party_icu_common.cpp", 
                 "src/duckdb/ub_extension_icu_third_party_icu_i18n.cpp", 
                 "src/duckdb/extension/icu/third_party/icu/stubdata/stubdata.cpp", 

package/package.json

@@ -2,7 +2,7 @@
   "name": "duckdb",
   "main": "./lib/duckdb.js",
   "types": "./lib/duckdb.d.ts",
-  "version": "0.7.1-dev90.0",
+  "version": "0.7.2-dev0.0",
   "description": "DuckDB node.js API",
   "gypfile": true,
   "dependencies": {
@@ -41,7 +41,7 @@
   },
   "repository": {
     "type": "git",
-    "url": "git+https://github.com/cwida/duckdb.git"
+    "url": "git+https://github.com/duckdb/duckdb.git"
   },
   "ts-node": {
     "require": [
@@ -56,7 +56,7 @@
   "author": "Hannes Mühleisen",
   "license": "MPL-2.0",
   "bugs": {
-    "url": "https://github.com/cwida/duckdb/issues"
+    "url": "https://github.com/duckdb/duckdb/issues"
   },
   "homepage": "https://www.duckdb.org"
 }

package/src/duckdb/extension/json/buffered_json_reader.cpp

@@ -25,7 +25,12 @@ JSONBufferHandle::JSONBufferHandle(idx_t buffer_index_p, idx_t readers_p, Alloca
 JSONFileHandle::JSONFileHandle(unique_ptr<FileHandle> file_handle_p, Allocator &allocator_p)
     : file_handle(std::move(file_handle_p)), allocator(allocator_p), can_seek(file_handle->CanSeek()),
       plain_file_source(file_handle->OnDiskFile() && can_seek), file_size(file_handle->GetFileSize()), read_position(0),
-      cached_size(0) {
+      requested_reads(0), actual_reads(0), cached_size(0) {
+}
+
+void JSONFileHandle::Close() {
+	file_handle->Close();
+	cached_buffers.clear();
 }
 
 idx_t JSONFileHandle::FileSize() const {
@@ -36,10 +41,6 @@ idx_t JSONFileHandle::Remaining() const {
 	return file_size - read_position;
 }
 
-bool JSONFileHandle::PlainFileSource() const {
-	return plain_file_source;
-}
-
 bool JSONFileHandle::CanSeek() const {
 	return can_seek;
 }
@@ -53,6 +54,9 @@ idx_t JSONFileHandle::GetPositionAndSize(idx_t &position, idx_t requested_size)
 	position = read_position;
 	auto actual_size = MinValue<idx_t>(requested_size, Remaining());
 	read_position += actual_size;
+	if (actual_size != 0) {
+		requested_reads++;
+	}
 	return actual_size;
 }
 
@@ -60,11 +64,13 @@ void JSONFileHandle::ReadAtPosition(const char *pointer, idx_t size, idx_t posit
 	D_ASSERT(size != 0);
 	if (plain_file_source) {
 		file_handle->Read((void *)pointer, size, position);
+		actual_reads++;
 		return;
 	}
 
 	if (sample_run) { // Cache the buffer
 		file_handle->Read((void *)pointer, size, position);
+		actual_reads++;
 		cached_buffers.emplace_back(allocator.Allocate(size));
 		memcpy(cached_buffers.back().get(), pointer, size);
 		cached_size += size;
@@ -73,22 +79,24 @@ void JSONFileHandle::ReadAtPosition(const char *pointer, idx_t size, idx_t posit
 
 	if (!cached_buffers.empty() || position < cached_size) {
 		ReadFromCache(pointer, size, position);
+		actual_reads++;
 	}
 	if (size != 0) {
 		file_handle->Read((void *)pointer, size, position);
+		actual_reads++;
 	}
 }
 
 idx_t JSONFileHandle::Read(const char *pointer, idx_t requested_size, bool sample_run) {
 	D_ASSERT(requested_size != 0);
 	if (plain_file_source) {
-		auto actual_size = file_handle->Read((void *)pointer, requested_size);
+		auto actual_size = ReadInternal(pointer, requested_size);
 		read_position += actual_size;
 		return actual_size;
 	}
 
 	if (sample_run) { // Cache the buffer
-		auto actual_size = file_handle->Read((void *)pointer, requested_size);
+		auto actual_size = ReadInternal(pointer, requested_size);
 		if (actual_size > 0) {
 			cached_buffers.emplace_back(allocator.Allocate(actual_size));
 			memcpy(cached_buffers.back().get(), pointer, actual_size);
@@ -103,7 +111,7 @@ idx_t JSONFileHandle::Read(const char *pointer, idx_t requested_size, bool sampl
 		actual_size += ReadFromCache(pointer, requested_size, read_position);
 	}
 	if (requested_size != 0) {
-		actual_size += file_handle->Read((void *)pointer, requested_size);
+		actual_size += ReadInternal(pointer, requested_size);
 	}
 	return actual_size;
 }
@@ -111,7 +119,10 @@ idx_t JSONFileHandle::Read(const char *pointer, idx_t requested_size, bool sampl
 idx_t JSONFileHandle::ReadFromCache(const char *&pointer, idx_t &size, idx_t &position) {
 	idx_t read_size = 0;
 	idx_t total_offset = 0;
-	for (auto &cached_buffer : cached_buffers) {
+
+	idx_t cached_buffer_idx;
+	for (cached_buffer_idx = 0; cached_buffer_idx < cached_buffers.size(); cached_buffer_idx++) {
+		auto &cached_buffer = cached_buffers[cached_buffer_idx];
 		if (size == 0) {
 			break;
 		}
@@ -127,9 +138,23 @@ idx_t JSONFileHandle::ReadFromCache(const char *&pointer, idx_t &size, idx_t &po
 		}
 		total_offset += cached_buffer.GetSize();
 	}
+
 	return read_size;
 }
 
+idx_t JSONFileHandle::ReadInternal(const char *pointer, const idx_t requested_size) {
+	// Deal with reading from pipes
+	idx_t total_read_size = 0;
+	while (total_read_size < requested_size) {
+		auto read_size = file_handle->Read((void *)(pointer + total_read_size), requested_size - total_read_size);
+		if (read_size == 0) {
+			break;
+		}
+		total_read_size += read_size;
+	}
+	return total_read_size;
+}
+
 BufferedJSONReader::BufferedJSONReader(ClientContext &context, BufferedJSONReaderOptions options_p, string file_path_p)
     : file_path(std::move(file_path_p)), context(context), options(std::move(options_p)), buffer_index(0) {
 }
@@ -143,6 +168,16 @@ void BufferedJSONReader::OpenJSONFile() {
 	file_handle = make_unique<JSONFileHandle>(std::move(regular_file_handle), BufferAllocator::Get(context));
 }
 
+void BufferedJSONReader::CloseJSONFile() {
+	while (true) {
+		lock_guard<mutex> guard(lock);
+		if (file_handle->RequestedReadsComplete()) {
+			file_handle->Close();
+			break;
+		}
+	}
+}
+
 bool BufferedJSONReader::IsOpen() {
 	return file_handle != nullptr;
 }
@@ -246,9 +281,15 @@ void BufferedJSONReader::Reset() {
 
 void JSONFileHandle::Reset() {
 	read_position = 0;
+	requested_reads = 0;
+	actual_reads = 0;
 	if (plain_file_source) {
 		file_handle->Reset();
 	}
 }
 
+bool JSONFileHandle::RequestedReadsComplete() {
+	return requested_reads == actual_reads;
+}
+
 } // namespace duckdb

package/src/duckdb/extension/json/include/buffered_json_reader.hpp

@@ -21,7 +21,7 @@ enum class JSONFormat : uint8_t {
 	AUTO_DETECT = 0,
 	//! One object after another, newlines can be anywhere
 	UNSTRUCTURED = 1,
-	//! Objects are separated by newlines, newlines do not occur within objects (NDJSON)
+	//! Objects are separated by newlines, newlines do not occur within values (NDJSON)
 	NEWLINE_DELIMITED = 2,
 };
 
@@ -58,11 +58,11 @@ public:
 struct JSONFileHandle {
 public:
 	JSONFileHandle(unique_ptr<FileHandle> file_handle, Allocator &allocator);
+	void Close();
 
 	idx_t FileSize() const;
 	idx_t Remaining() const;
 
-	bool PlainFileSource() const;
 	bool CanSeek() const;
 	void Seek(idx_t position);
 
@@ -71,9 +71,11 @@ public:
 	idx_t Read(const char *pointer, idx_t requested_size, bool sample_run);
 
 	void Reset();
+	bool RequestedReadsComplete();
 
 private:
 	idx_t ReadFromCache(const char *&pointer, idx_t &size, idx_t &position);
+	idx_t ReadInternal(const char *pointer, const idx_t requested_size);
 
 private:
 	//! The JSON file handle
@@ -87,6 +89,8 @@ private:
 
 	//! Read properties
 	idx_t read_position;
+	idx_t requested_reads;
+	atomic<idx_t> actual_reads;
 
 	//! Cached buffers for resetting when reading stream
 	vector<AllocatedData> cached_buffers;
@@ -98,6 +102,7 @@ public:
 	BufferedJSONReader(ClientContext &context, BufferedJSONReaderOptions options, string file_path);
 
 	void OpenJSONFile();
+	void CloseJSONFile();
 	bool IsOpen();
 
 	BufferedJSONReaderOptions &GetOptions();

package/src/duckdb/extension/json/include/json_scan.hpp

@@ -20,12 +20,25 @@ enum class JSONScanType : uint8_t {
 	INVALID = 0,
 	//! Read JSON straight to columnar data
 	READ_JSON = 1,
-	//! Read JSON objects as strings
+	//! Read JSON values as strings
 	READ_JSON_OBJECTS = 2,
 	//! Sample run for schema detection
 	SAMPLE = 3,
 };
 
+enum class JSONRecordType : uint8_t {
+	//! Sequential values
+	RECORDS = 0,
+	//! Array of values
+	ARRAY_OF_RECORDS = 1,
+	//! Sequential non-object JSON
+	JSON = 2,
+	//! Array of non-object JSON
+	ARRAY_OF_JSON = 3,
+	//! Auto-detect
+	AUTO = 4,
+};
+
 //! Even though LogicalTypeId is just a uint8_t, this is still needed ...
 struct LogicalTypeIdHash {
 	inline std::size_t operator()(const LogicalTypeId &id) const {
@@ -104,8 +117,8 @@ public:
 	vector<idx_t> valid_cols;
 	//! Max depth we go to detect nested JSON schema (defaults to unlimited)
 	idx_t max_depth = NumericLimits<idx_t>::Maximum();
-	//! Whether we're parsing objects (usually), or something else like arrays
-	bool objects = true;
+	//! Whether we're parsing values (usually), or something else
+	JSONRecordType record_type = JSONRecordType::RECORDS;
 	//! Forced date/timestamp formats
 	string date_format;
 	string timestamp_format;
@@ -119,12 +132,13 @@ public:
 struct JSONScanInfo : public TableFunctionInfo {
 public:
 	explicit JSONScanInfo(JSONScanType type_p = JSONScanType::INVALID, JSONFormat format_p = JSONFormat::AUTO_DETECT,
-	                      bool auto_detect_p = false)
-	    : type(type_p), format(format_p), auto_detect(auto_detect_p) {
+	                      JSONRecordType record_type_p = JSONRecordType::AUTO, bool auto_detect_p = false)
+	    : type(type_p), format(format_p), record_type(record_type_p), auto_detect(auto_detect_p) {
 	}
 
 	JSONScanType type;
 	JSONFormat format;
+	JSONRecordType record_type;
 	bool auto_detect;
 };
 
@@ -179,10 +193,15 @@ public:
 public:
 	idx_t ReadNext(JSONScanGlobalState &gstate);
 	yyjson_alc *GetAllocator();
-	void ThrowTransformError(idx_t count, idx_t object_index, const string &error_message);
+	void ThrowTransformError(idx_t object_index, const string &error_message);
 
+	idx_t scan_count;
 	JSONLine lines[STANDARD_VECTOR_SIZE];
-	yyjson_val *objects[STANDARD_VECTOR_SIZE];
+	yyjson_val *values[STANDARD_VECTOR_SIZE];
+
+	idx_t array_idx;
+	idx_t array_offset;
+	yyjson_val *array_values[STANDARD_VECTOR_SIZE];
 
 	idx_t batch_index;
 
@@ -192,6 +211,7 @@ public:
 
 private:
 	yyjson_val *ParseLine(char *line_start, idx_t line_size, idx_t remaining, JSONLine &line);
+	idx_t GetObjectsFromArray(JSONScanGlobalState &gstate);
 
 private:
 	//! Bind data
@@ -212,7 +232,7 @@ private:
 	idx_t prev_buffer_remainder;
 	idx_t lines_or_objects_in_buffer;
 
-	//! Buffer to reconstruct split objects
+	//! Buffer to reconstruct split values
 	AllocatedData reconstruct_buffer;
 	//! Copy of current buffer for YYJSON_READ_INSITU
 	AllocatedData current_buffer_copy;
@@ -276,6 +296,21 @@ public:
 		return lstate.GetBatchIndex();
 	}
 
+	static unique_ptr<NodeStatistics> JSONScanCardinality(ClientContext &context, const FunctionData *bind_data) {
+		auto &data = (JSONScanData &)*bind_data;
+		idx_t per_file_cardinality;
+		if (data.stored_readers.empty()) {
+			// The cardinality of an unknown JSON file is the almighty number 42 except when it's not
+			per_file_cardinality = 42;
+		} else {
+			// If we multiply the almighty number 42 by 10, we get the exact average size of a JSON
+			// Not really, but the average size of a lineitem row in JSON is around 360 bytes
+			per_file_cardinality = data.stored_readers[0]->GetFileHandle().FileSize() / 420;
+		}
+		// Obviously this can be improved but this is better than defaulting to 0
+		return make_unique<NodeStatistics>(per_file_cardinality * data.file_paths.size());
+	}
+
 	static void JSONScanSerialize(FieldWriter &writer, const FunctionData *bind_data_p, const TableFunction &function) {
 		auto &bind_data = (JSONScanData &)*bind_data_p;
 		bind_data.Serialize(writer);
@@ -291,16 +326,16 @@ public:
 	static void TableFunctionDefaults(TableFunction &table_function) {
 		table_function.named_parameters["maximum_object_size"] = LogicalType::UINTEGER;
 		table_function.named_parameters["ignore_errors"] = LogicalType::BOOLEAN;
-		table_function.named_parameters["format"] = LogicalType::VARCHAR;
+		table_function.named_parameters["lines"] = LogicalType::VARCHAR;
 		table_function.named_parameters["compression"] = LogicalType::VARCHAR;
 
 		table_function.table_scan_progress = JSONScanProgress;
 		table_function.get_batch_index = JSONScanGetBatchIndex;
+		table_function.cardinality = JSONScanCardinality;
 
 		table_function.serialize = JSONScanSerialize;
 		table_function.deserialize = JSONScanDeserialize;
 
-		// TODO: might be able to do some of these
 		table_function.projection_pushdown = false;
 		table_function.filter_pushdown = false;
 		table_function.filter_prune = false;

package/src/duckdb/extension/json/json_functions/copy_json.cpp

@@ -1,7 +1,9 @@
 #include "duckdb/function/copy_function.hpp"
 #include "duckdb/parser/expression/constant_expression.hpp"
 #include "duckdb/parser/expression/function_expression.hpp"
+#include "duckdb/parser/expression/positional_reference_expression.hpp"
 #include "duckdb/parser/query_node/select_node.hpp"
+#include "duckdb/parser/tableref/subqueryref.hpp"
 #include "duckdb/planner/binder.hpp"
 #include "json_functions.hpp"
 #include "json_scan.hpp"
@@ -12,42 +14,52 @@ namespace duckdb {
 static BoundStatement CopyToJSONPlan(Binder &binder, CopyStatement &stmt) {
 	auto stmt_copy = stmt.Copy();
 	auto &copy = (CopyStatement &)*stmt_copy;
-	auto &select_stmt = (SelectNode &)*copy.select_statement;
 	auto &info = *copy.info;
 
-	// strftime if the user specified a format TODO: deal with date/timestamp within nested types
-	auto date_it = info.options.find("dateformat");
-	auto timestamp_it = info.options.find("timestampformat");
-
 	// Bind the select statement of the original to resolve the types
 	auto dummy_binder = Binder::CreateBinder(binder.context, &binder, true);
 	auto bound_original = dummy_binder->Bind(*stmt.select_statement);
-	D_ASSERT(bound_original.types.size() == select_stmt.select_list.size());
-	const idx_t num_cols = bound_original.types.size();
 
-	// This loop also makes sure the columns have an alias (needed for struct_pack)
+	// Create new SelectNode with the original SelectNode as a subquery in the FROM clause
+	auto select_stmt = make_unique<SelectStatement>();
+	select_stmt->node = std::move(copy.select_statement);
+	auto subquery_ref = make_unique<SubqueryRef>(std::move(select_stmt));
+	copy.select_statement = make_unique_base<QueryNode, SelectNode>();
+	auto &new_select_node = (SelectNode &)*copy.select_statement;
+	new_select_node.from_table = std::move(subquery_ref);
+
+	// Create new select list
+	vector<unique_ptr<ParsedExpression>> select_list;
+	select_list.reserve(bound_original.types.size());
+
+	// strftime if the user specified a format (loop also gives columns a name, needed for struct_pack)
+	// TODO: deal with date/timestamp within nested types
+	const auto date_it = info.options.find("dateformat");
+	const auto timestamp_it = info.options.find("timestampformat");
 	vector<unique_ptr<ParsedExpression>> strftime_children;
-	for (idx_t i = 0; i < num_cols; i++) {
+	for (idx_t col_idx = 0; col_idx < bound_original.types.size(); col_idx++) {
+		auto column = make_unique_base<ParsedExpression, PositionalReferenceExpression>(col_idx + 1);
 		strftime_children.clear();
-		auto &col = select_stmt.select_list[i];
-		auto name = col->GetName();
-		if (bound_original.types[i] == LogicalTypeId::DATE && date_it != info.options.end()) {
-			strftime_children.emplace_back(std::move(col));
+		const auto &type = bound_original.types[col_idx];
+		const auto &name = bound_original.names[col_idx];
+		if (date_it != info.options.end() && type == LogicalTypeId::DATE) {
+			strftime_children.emplace_back(std::move(column));
 			strftime_children.emplace_back(make_unique<ConstantExpression>(date_it->second.back()));
-			col = make_unique<FunctionExpression>("strftime", std::move(strftime_children));
-		} else if (bound_original.types[i] == LogicalTypeId::TIMESTAMP && timestamp_it != info.options.end()) {
-			strftime_children.emplace_back(std::move(col));
+			column = make_unique<FunctionExpression>("strftime", std::move(strftime_children));
+		} else if (timestamp_it != info.options.end() && type == LogicalTypeId::TIMESTAMP) {
+			strftime_children.emplace_back(std::move(column));
 			strftime_children.emplace_back(make_unique<ConstantExpression>(timestamp_it->second.back()));
-			col = make_unique<FunctionExpression>("strftime", std::move(strftime_children));
+			column = make_unique<FunctionExpression>("strftime", std::move(strftime_children));
 		}
-		col->alias = name;
+		column->alias = name;
+		select_list.emplace_back(std::move(column));
 	}
 
 	// Now create the struct_pack/to_json to create a JSON object per row
+	auto &select_node = (SelectNode &)*copy.select_statement;
 	vector<unique_ptr<ParsedExpression>> struct_pack_child;
-	struct_pack_child.emplace_back(make_unique<FunctionExpression>("struct_pack", std::move(select_stmt.select_list)));
-	select_stmt.select_list.clear();
-	select_stmt.select_list.emplace_back(make_unique<FunctionExpression>("to_json", std::move(struct_pack_child)));
+	struct_pack_child.emplace_back(make_unique<FunctionExpression>("struct_pack", std::move(select_list)));
+	select_node.select_list.emplace_back(make_unique<FunctionExpression>("to_json", std::move(struct_pack_child)));
 
 	// Now we can just use the CSV writer
 	info.format = "csv";
@@ -101,7 +113,8 @@ CreateCopyFunctionInfo JSONFunctions::GetJSONCopyFunction() {
 
 	function.copy_from_bind = CopyFromJSONBind;
 	function.copy_from_function = JSONFunctions::GetReadJSONTableFunction(
-	    false, make_shared<JSONScanInfo>(JSONScanType::READ_JSON, JSONFormat::AUTO_DETECT, false));
+	    false,
+	    make_shared<JSONScanInfo>(JSONScanType::READ_JSON, JSONFormat::AUTO_DETECT, JSONRecordType::RECORDS, false));
 
 	return CreateCopyFunctionInfo(function);
 }

package/src/duckdb/extension/json/json_functions/json_create.cpp

@@ -56,7 +56,7 @@ static LogicalType GetJSONType(unordered_map<string, unique_ptr<Vector>> &const_
 	// The nested types need to conform as well
 	case LogicalTypeId::LIST:
 		return LogicalType::LIST(GetJSONType(const_struct_names, ListType::GetChildType(type)));
-	// Struct and MAP are treated as JSON objects
+	// Struct and MAP are treated as JSON values
 	case LogicalTypeId::STRUCT: {
 		child_list_t<LogicalType> child_types;
 		for (const auto &child_type : StructType::GetChildTypes(type)) {
@@ -247,14 +247,14 @@ static void TemplatedCreateValues(yyjson_mut_doc *doc, yyjson_mut_val *vals[], V
 
 static void CreateValuesStruct(const JSONCreateFunctionData &info, yyjson_mut_doc *doc, yyjson_mut_val *vals[],
                                Vector &value_v, idx_t count) {
-	// Structs become objects, therefore we initialize vals to JSON objects
+	// Structs become values, therefore we initialize vals to JSON values
 	for (idx_t i = 0; i < count; i++) {
 		vals[i] = yyjson_mut_obj(doc);
 	}
 	// Initialize re-usable array for the nested values
 	auto nested_vals = (yyjson_mut_val **)doc->alc.malloc(doc->alc.ctx, sizeof(yyjson_mut_val *) * count);
 
-	// Add the key/value pairs to the objects
+	// Add the key/value pairs to the values
 	auto &entries = StructVector::GetEntries(value_v);
 	for (idx_t entry_i = 0; entry_i < entries.size(); entry_i++) {
 		auto &struct_key_v = *info.const_struct_names.at(StructType::GetChildName(value_v.GetType(), entry_i));
@@ -284,7 +284,7 @@ static void CreateValuesMap(const JSONCreateFunctionData &info, yyjson_mut_doc *
 	auto map_val_count = ListVector::GetListSize(value_v);
 	auto nested_vals = (yyjson_mut_val **)doc->alc.malloc(doc->alc.ctx, sizeof(yyjson_mut_val *) * map_val_count);
 	CreateValues(info, doc, nested_vals, map_val_v, map_val_count);
-	// Add the key/value pairs to the objects
+	// Add the key/value pairs to the values
 	UnifiedVectorFormat map_data;
 	value_v.ToUnifiedFormat(count, map_data);
 	auto map_key_list_entries = (list_entry_t *)map_data.data;
@@ -308,7 +308,7 @@ static void CreateValuesMap(const JSONCreateFunctionData &info, yyjson_mut_doc *
 
 static void CreateValuesUnion(const JSONCreateFunctionData &info, yyjson_mut_doc *doc, yyjson_mut_val *vals[],
                               Vector &value_v, idx_t count) {
-	// Structs become objects, therefore we initialize vals to JSON objects
+	// Structs become values, therefore we initialize vals to JSON values
 	for (idx_t i = 0; i < count; i++) {
 		vals[i] = yyjson_mut_obj(doc);
 	}
@@ -320,7 +320,7 @@ static void CreateValuesUnion(const JSONCreateFunctionData &info, yyjson_mut_doc
 	UnifiedVectorFormat tag_data;
 	tag_v.ToUnifiedFormat(count, tag_data);
 
-	// Add the key/value pairs to the objects
+	// Add the key/value pairs to the values
 	for (idx_t member_idx = 0; member_idx < UnionType::GetMemberCount(value_v.GetType()); member_idx++) {
 		auto &member_val_v = UnionVector::GetMember(value_v, member_idx);
 		auto &member_key_v = *info.const_struct_names.at(UnionType::GetMemberName(value_v.GetType(), member_idx));
@@ -425,7 +425,7 @@ static void ObjectFunction(DataChunk &args, ExpressionState &state, Vector &resu
 	auto &lstate = JSONFunctionLocalState::ResetAndGet(state);
 	auto alc = lstate.json_allocator.GetYYJSONAllocator();
 
-	// Initialize objects
+	// Initialize values
 	const idx_t count = args.size();
 	auto doc = JSONCommon::CreateDocument(alc);
 	yyjson_mut_val *objs[STANDARD_VECTOR_SIZE];
@@ -440,7 +440,7 @@ static void ObjectFunction(DataChunk &args, ExpressionState &state, Vector &resu
 		Vector &value_v = args.data[pair_idx * 2 + 1];
 		CreateKeyValuePairs(info, doc, objs, vals, key_v, value_v, count);
 	}
-	// Write JSON objects to string
+	// Write JSON values to string
 	auto objects = FlatVector::GetData<string_t>(result);
 	for (idx_t i = 0; i < count; i++) {
 		objects[i] = JSONCommon::WriteVal<yyjson_mut_val>(objs[i], alc);

package/src/duckdb/extension/json/json_functions/json_structure.cpp

@@ -214,9 +214,6 @@ void JSONStructureNode::RefineCandidateTypesObject(yyjson_val *vals[], idx_t cou
 		}
 	}
 
-	if (count > STANDARD_VECTOR_SIZE) {
-		string_vector.Initialize(false, count);
-	}
 	for (idx_t child_idx = 0; child_idx < child_count; child_idx++) {
 		desc.children[child_idx].RefineCandidateTypes(child_vals[child_idx], count, string_vector, allocator,
 		                                              date_format_map);
@@ -431,6 +428,10 @@ static inline yyjson_mut_val *ConvertStructureArray(const JSONStructureNode &nod
 static inline yyjson_mut_val *ConvertStructureObject(const JSONStructureNode &node, yyjson_mut_doc *doc) {
 	D_ASSERT(node.descriptions.size() == 1 && node.descriptions[0].type == LogicalTypeId::STRUCT);
 	auto &desc = node.descriptions[0];
+	if (desc.children.empty()) {
+		// Empty struct - let's do JSON instead
+		return yyjson_mut_str(doc, JSONCommon::JSON_TYPE_NAME);
+	}
 
 	auto obj = yyjson_mut_obj(doc);
 	for (auto &child : desc.children) {
@@ -495,6 +496,10 @@ static LogicalType StructureToTypeObject(ClientContext &context, const JSONStruc
                                          idx_t depth) {
 	D_ASSERT(node.descriptions.size() == 1 && node.descriptions[0].type == LogicalTypeId::STRUCT);
 	auto &desc = node.descriptions[0];
+	if (desc.children.empty()) {
+		// Empty struct - let's do JSON instead
+		return JSONCommon::JSONType();
+	}
 
 	child_list_t<LogicalType> child_types;
 	child_types.reserve(desc.children.size());