Skip to content

fix(aggregate): preserve listagg result ownership #281

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lxy-9602 merged 4 commits into from
Sep 4, 2026
Merged

fix(aggregate): preserve listagg result ownership #281

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a7630e5
fix(aggregate): preserve listagg result ownership
ChaomingZhangCN Sep 4, 2026
1d08a4e
Merge branch 'main' into codex/fix-listagg-result-lifetime
lxy-9602 Sep 4, 2026
3e3407e
fix(tests): use portable bigint literals in core tests
ChaomingZhangCN Sep 4, 2026
76aee1a
test(aggregate): add PK listagg integration coverage
ChaomingZhangCN Sep 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 7 additions & 10 deletions src/paimon/core/mergetree/compact/aggregate/field_listagg_agg.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,18 +77,16 @@ class FieldListaggAgg : public FieldAggregator {
return input_field;
}

std::string result;
if (distinct_) {
result_ = AggDistinctImpl(acc_str, in_str);
result = AggDistinctImpl(acc_str, in_str);
} else {
// Build into a local string to avoid aliasing when acc_str points into result_
std::string new_result;
new_result.reserve(acc_str.size() + delimiter_.size() + in_str.size());
new_result.append(acc_str);
new_result.append(delimiter_);
new_result.append(in_str);
result_ = std::move(new_result);
result.reserve(acc_str.size() + delimiter_.size() + in_str.size());
result.append(acc_str);
result.append(delimiter_);
result.append(in_str);
}
return VariantType(std::string_view{result_});
return VariantType(BinaryString::FromString(result, pool_.get()));
}

private:
Expand Down Expand Up @@ -136,6 +134,5 @@ class FieldListaggAgg : public FieldAggregator {

std::string delimiter_;
bool distinct_;
std::string result_;
};
} // namespace paimon
46 changes: 29 additions & 17 deletions src/paimon/core/mergetree/compact/aggregate/field_listagg_agg_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,13 +47,13 @@ class FieldListaggAggTest : public testing::Test {
TEST_F(FieldListaggAggTest, TestSimple) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg());
auto ret = agg->Agg(std::string_view("hello"), std::string_view(" world")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "hello, world");
ASSERT_EQ(DataDefine::GetStringView(ret), "hello, world");
}

TEST_F(FieldListaggAggTest, TestDelimiter) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg("-"));
auto ret = agg->Agg(std::string_view("user1"), std::string_view("user2")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "user1-user2");
ASSERT_EQ(DataDefine::GetStringView(ret), "user1-user2");
}

TEST_F(FieldListaggAggTest, TestNull) {
Expand All @@ -62,12 +62,12 @@ TEST_F(FieldListaggAggTest, TestNull) {
// input null -> return accumulator
{
auto ret = agg->Agg(std::string_view("hello"), NullType()).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "hello");
ASSERT_EQ(DataDefine::GetStringView(ret), "hello");
}
// accumulator null -> return input
{
auto ret = agg->Agg(NullType(), std::string_view("world")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "world");
ASSERT_EQ(DataDefine::GetStringView(ret), "world");
}
// both null -> return null
{
Expand All @@ -82,17 +82,17 @@ TEST_F(FieldListaggAggTest, TestEmptyString) {
// empty input -> return accumulator
{
auto ret = agg->Agg(std::string_view("hello"), std::string_view("")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "hello");
ASSERT_EQ(DataDefine::GetStringView(ret), "hello");
}
// empty accumulator -> return input
{
auto ret = agg->Agg(std::string_view(""), std::string_view("world")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "world");
ASSERT_EQ(DataDefine::GetStringView(ret), "world");
}
// blank input -> return accumulator (which is empty)
{
auto ret = agg->Agg(std::string_view(""), std::string_view("")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "");
ASSERT_EQ(DataDefine::GetStringView(ret), "");
}
}

Expand All @@ -112,12 +112,12 @@ TEST_F(FieldListaggAggTest, TestBlankStrings) {
u8" \t\u3000\u2000\n"};
for (const std::string& blank : blank_strings) {
auto ret = agg->Agg(std::string_view("user1"), std::string_view(blank)).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "user1");
ASSERT_EQ(DataDefine::GetStringView(ret), "user1");
}

// A blank accumulator must not add a leading delimiter.
auto ret = agg->Agg(std::string_view(u8"\u3000\t"), std::string_view("user1")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "user1");
ASSERT_EQ(DataDefine::GetStringView(ret), "user1");

// A blank input must not turn a null accumulator into a non-null value.
ret = agg->Agg(NullType(), std::string_view(u8" \t\u3000")).value();
Expand All @@ -129,17 +129,29 @@ TEST_F(FieldListaggAggTest, TestMultipleAccumulation) {

// "a" + "," + "b" = "a,b", then "a,b" + "," + "c" = "a,b,c"
auto ret = agg->Agg(std::string_view("a"), std::string_view("b")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a,b");
ASSERT_EQ(DataDefine::GetStringView(ret), "a,b");
ret = agg->Agg(std::move(ret), std::string_view("c")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a,b,c");
ASSERT_EQ(DataDefine::GetStringView(ret), "a,b,c");
}

TEST_F(FieldListaggAggTest, TestResultOwnershipAcrossAggregations) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg());

ASSERT_OK_AND_ASSIGN(VariantType first,
agg->Agg(std::string_view("alpha"), std::string_view("beta")));
ASSERT_OK_AND_ASSIGN(VariantType second,
agg->Agg(std::string_view("one"), std::string_view("two")));

ASSERT_EQ(DataDefine::GetStringView(first), "alpha,beta");
ASSERT_EQ(DataDefine::GetStringView(second), "one,two");
}

TEST_F(FieldListaggAggTest, TestDistinct) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg(";", true));

// "a;b" + "b;c" -> "a;b;c" (deduplicate "b")
auto ret = agg->Agg(std::string_view("a;b"), std::string_view("b;c")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a;b;c");
ASSERT_EQ(DataDefine::GetStringView(ret), "a;b;c");
}

TEST_F(FieldListaggAggTest, TestDistinctIgnoresBlankTokens) {
Expand All @@ -148,39 +160,39 @@ TEST_F(FieldListaggAggTest, TestDistinctIgnoresBlankTokens) {
auto ret =
agg->Agg(std::string_view("user1"), std::string_view(u8" ,user2,\t,\u3000,user1,\u2000"))
.value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "user1,user2");
ASSERT_EQ(DataDefine::GetStringView(ret), "user1,user2");
}

TEST_F(FieldListaggAggTest, TestDistinctNoDuplicates) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg(" ", true));

// "a b" + "c d" -> "a b c d" (no dups to remove)
auto ret = agg->Agg(std::string_view("a b"), std::string_view("c d")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a b c d");
ASSERT_EQ(DataDefine::GetStringView(ret), "a b c d");
}

TEST_F(FieldListaggAggTest, TestDistinctWithEmptyDelimiterFallsBackToWhitespace) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg("", true));

// Empty delimiter falls back to whitespace, so the repeated "b" is removed.
auto ret = agg->Agg(std::string_view("a b"), std::string_view("b c")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a b c");
ASSERT_EQ(DataDefine::GetStringView(ret), "a b c");
}

TEST_F(FieldListaggAggTest, TestDistinctEmptyInput) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg(";", true));

// empty input -> return accumulator
auto ret = agg->Agg(std::string_view("a;b"), std::string_view("")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a;b");
ASSERT_EQ(DataDefine::GetStringView(ret), "a;b");
}

TEST_F(FieldListaggAggTest, TestDistinctFalse) {
ASSERT_OK_AND_ASSIGN(auto agg, MakeAgg(";", false));

// "a;b" + "b;c" -> "a;b;b;c" (no dedup)
auto ret = agg->Agg(std::string_view("a;b"), std::string_view("b;c")).value();
ASSERT_EQ(DataDefine::GetVariantValue<std::string_view>(ret), "a;b;b;c");
ASSERT_EQ(DataDefine::GetStringView(ret), "a;b;b;c");
}
Comment thread
ChaomingZhangCN marked this conversation as resolved.

TEST_F(FieldListaggAggTest, TestInvalidType) {
Expand Down
2 changes: 1 addition & 1 deletion src/paimon/core/operation/data_evolution_split_read_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,7 +111,7 @@ TEST_F(DataEvolutionSplitReadTest, TestCreatePushDownPredicate) {
auto f1_predicate =
PredicateBuilder::Equal(/*field_index=*/1, /*field_name=*/"f1", FieldType::INT, Literal(2));
auto row_id_predicate = PredicateBuilder::Equal(
/*field_index=*/2, SpecialFields::RowId().Name(), FieldType::BIGINT, Literal(3l));
/*field_index=*/2, SpecialFields::RowId().Name(), FieldType::BIGINT, Literal(int64_t{3}));
ASSERT_OK_AND_ASSIGN(std::shared_ptr<Predicate> predicate,
PredicateBuilder::And({f0_predicate, f1_predicate, row_id_predicate}));

Expand Down
5 changes: 3 additions & 2 deletions src/paimon/core/table/source/table_read_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@

#include "paimon/table/source/table_read.h"

#include <cstdint>
#include <map>
#include <memory>
#include <string>
Expand Down Expand Up @@ -52,7 +53,7 @@ TEST(TableReadTest, TestReadWithInvalidContext) {
{
// field type and literal type mismatch
auto predicate = PredicateBuilder::Equal(/*field_index=*/3, /*field_name=*/"f3",
FieldType::DOUBLE, Literal(15l));
FieldType::DOUBLE, Literal(int64_t{15}));
ReadContextBuilder context_builder(path);
context_builder.SetPredicate(predicate);
ASSERT_OK_AND_ASSIGN(auto read_context, context_builder.Finish());
Expand All @@ -63,7 +64,7 @@ TEST(TableReadTest, TestReadWithInvalidContext) {
{
// field type in predicate mismatch schema
auto predicate = PredicateBuilder::Equal(/*field_index=*/3, /*field_name=*/"f3",
FieldType::BIGINT, Literal(15l));
FieldType::BIGINT, Literal(int64_t{15}));
ReadContextBuilder context_builder(path);
context_builder.SetPredicate(predicate);
ASSERT_OK_AND_ASSIGN(auto read_context, context_builder.Finish());
Expand Down
13 changes: 7 additions & 6 deletions src/paimon/core/utils/field_mapping_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@
#include "paimon/core/utils/field_mapping.h"

#include <algorithm>
#include <cstdint>

#include "arrow/type_fwd.h"
#include "gtest/gtest.h"
Expand Down Expand Up @@ -444,7 +445,7 @@ TEST_F(FieldMappingTest, TestSchemaEvolutionWithPredicate) {
auto greater_or_equal = PredicateBuilder::GreaterOrEqual(
/*field_index=*/0, /*field_name=*/"key0", FieldType::INT, Literal(4));
auto equal = PredicateBuilder::Equal(/*field_index=*/1, /*field_name=*/"key1",
FieldType::BIGINT, Literal(3l));
FieldType::BIGINT, Literal(int64_t{3}));
auto less_or_equal = PredicateBuilder::LessOrEqual(/*field_index=*/2, /*field_name=*/"k",
FieldType::INT, Literal(10));
// greater_than will not be pushed down, as with casting, only integer predicates can be pushed
Expand All @@ -455,7 +456,7 @@ TEST_F(FieldMappingTest, TestSchemaEvolutionWithPredicate) {
FieldType::INT, Literal(40));
// in can be pushed down
auto in = PredicateBuilder::In(/*field_index=*/5, /*field_name=*/"a", FieldType::BIGINT,
{Literal(100l)});
{Literal(int64_t{100})});
auto not_in =
PredicateBuilder::In(/*field_index=*/6, /*field_name=*/"e", FieldType::INT, {Literal(50)});

Expand Down Expand Up @@ -545,7 +546,7 @@ TEST_F(FieldMappingTest, TestSchemaEvolutionWithPredicate2) {
auto greater_or_equal = PredicateBuilder::GreaterOrEqual(
/*field_index=*/6, /*field_name=*/"key0", FieldType::INT, Literal(4));
auto equal = PredicateBuilder::Equal(/*field_index=*/1, /*field_name=*/"key1",
FieldType::BIGINT, Literal(3l));
FieldType::BIGINT, Literal(int64_t{3}));
auto less_or_equal = PredicateBuilder::LessOrEqual(/*field_index=*/4, /*field_name=*/"k",
FieldType::INT, Literal(10));
// greater_than will not be pushed down, as with casting, only integer predicates can be pushed
Expand All @@ -556,7 +557,7 @@ TEST_F(FieldMappingTest, TestSchemaEvolutionWithPredicate2) {
FieldType::INT, Literal(40));
// in will not be pushed down, as with casting, literal from BIGINT to INT is overflow
auto in = PredicateBuilder::In(/*field_index=*/2, /*field_name=*/"a", FieldType::BIGINT,
{Literal(9223372036854775807l)});
{Literal(int64_t{9223372036854775807LL})});
auto not_in =
PredicateBuilder::In(/*field_index=*/3, /*field_name=*/"e", FieldType::INT, {Literal(50)});

Expand All @@ -577,7 +578,7 @@ TEST_F(FieldMappingTest, TestSchemaEvolutionWithPredicate2) {
auto greater_or_equal_new = PredicateBuilder::GreaterOrEqual(
/*field_index=*/0, /*field_name=*/"key0", FieldType::INT, Literal(4));
auto equal_new = PredicateBuilder::Equal(/*field_index=*/1, /*field_name=*/"key1",
FieldType::BIGINT, Literal(3l));
FieldType::BIGINT, Literal(int64_t{3}));
expected_part_info.partition_filter =
PredicateBuilder::And({greater_or_equal_new, equal_new}).value_or(nullptr);
CheckPartitionInfo(mapping->partition_info.value(), expected_part_info);
Expand Down Expand Up @@ -623,7 +624,7 @@ TEST_F(FieldMappingTest, TestCompoundPredicateWithoutPushDown) {
auto equal = PredicateBuilder::Equal(/*field_index=*/0, /*field_name=*/"f0", FieldType::INT,
Literal(55));
auto greater_than = PredicateBuilder::GreaterThan(/*field_index=*/2, /*field_name=*/"f2",
FieldType::BIGINT, Literal(30l));
FieldType::BIGINT, Literal(int64_t{30}));
auto less_than = PredicateBuilder::LessThan(/*field_index=*/3, /*field_name=*/"f3",
FieldType::INT, Literal(55));
ASSERT_OK_AND_ASSIGN(auto or_predicate, PredicateBuilder::Or({greater_than, less_than}));
Expand Down
44 changes: 44 additions & 0 deletions test/inte/write_and_read_inte_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -679,6 +679,50 @@ TEST_P(WriteAndReadInteTest, TestPKSimple) {
ASSERT_TRUE(success);
}

TEST_P(WriteAndReadInteTest, TestPKListAggPreservesResultsAcrossKeys) {
arrow::FieldVector fields = {arrow::field("pk", arrow::utf8()),
arrow::field("value", arrow::utf8())};
auto [file_format, file_system] = GetParam();
std::map<std::string, std::string> options = {
{Options::FILE_FORMAT, file_format},
{Options::TARGET_FILE_SIZE, "1024"},
{Options::BUCKET, "1"},
{Options::FILE_SYSTEM, file_system},
{Options::MERGE_ENGINE, "aggregation"},
{"fields.value.aggregate-function", "listagg"},
};
if (file_system == "jindo") {
options = AddOptionsForJindo(options);
}
ASSERT_OK_AND_ASSIGN(
std::unique_ptr<TestHelper> helper,
TestHelper::Create(test_dir_, arrow::schema(fields), /*partition_keys=*/{},
/*primary_keys=*/{"pk"}, options, /*is_streaming_mode=*/true));

ASSERT_OK_AND_ASSIGN(std::unique_ptr<RecordBatch> first_batch,
TestHelper::MakeRecordBatch(arrow::struct_(fields),
R"([["first", "alpha"], ["second", "one"]])",
/*partition_map=*/{}, /*bucket=*/0, {}));
ASSERT_OK(helper->WriteAndCommit(std::move(first_batch), /*commit_identifier=*/0,
/*expected_commit_messages=*/std::nullopt));
ASSERT_OK_AND_ASSIGN(std::unique_ptr<RecordBatch> second_batch,
TestHelper::MakeRecordBatch(arrow::struct_(fields),
R"([["first", "beta"], ["second", "two"]])",
/*partition_map=*/{}, /*bucket=*/0, {}));
ASSERT_OK(helper->WriteAndCommit(std::move(second_batch), /*commit_identifier=*/1,
/*expected_commit_messages=*/std::nullopt));

arrow::FieldVector result_fields = fields;
result_fields.insert(result_fields.begin(), arrow::field("_VALUE_KIND", arrow::int8()));
ASSERT_OK_AND_ASSIGN(std::vector<std::shared_ptr<Split>> data_splits,
helper->NewScan(StartupMode::LatestFull(), /*snapshot_id=*/std::nullopt));
ASSERT_OK_AND_ASSIGN(bool success,
helper->ReadAndCheckResult(arrow::struct_(result_fields), data_splits,
R"([[0, "first", "alpha,beta"],
[0, "second", "one,two"]])"));
ASSERT_TRUE(success);
}

TEST_P(WriteAndReadInteTest, TestInputChangelogStreamRead) {
arrow::FieldVector fields = {
arrow::field("pk", arrow::utf8()),
Expand Down
Loading