GH-35785: [C++] Support null type non-key columns for join operation

[llama90]

Rationale for this change

Supports join on null types for non-key columns.

• Issue raised: [Python] Support joinining tables with null columns #35785

What changes are included in this PR?

• Separated the check of supported types for key and non-key columns during join execution
• Updated the logic associated with Null types for non-key columns.
• Modified the random tests for Hash Join (TEST(HashJoin, Random)) to allow the creation of Null as non-key column values.

Are these changes tested?

Yes

Are there any user-facing changes?

For users utilizing the Join operation

• Closes: [Python] Support joinining tables with null columns #35785

[github-actions]

⚠️ GitHub issue #35785 has been automatically assigned in GitHub to PR creator.

[llama90]

There were policy considerations and a personal lack of understanding regarding the code when it came to supporting Null types for key field columns.

Therefore, although there was an intent to support key columns as well, the initial improvements were made to enable support for non-key columns.

    // Constraints
    RandomDataTypeConstraints key_type_constraints;
    key_type_constraints.Default();
    key_type_constraints.withoutNullColumn();

In the code, when the // key_type_constraints.withoutNullColumn(); part is commented out, the following error can be encountered.

error message

Test 55: RIGHT_OUTER EQ_IS parallel = true bloom_filter = false
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/hash_join_node_test.cc:1231: Failure
Failed
'_error_or_value45.status()' failed with NotImplemented: Function 'equal' has no kernel matching input types (null, null)
/Users/lama/workspace/arrow-2/cpp/src/arrow/compute/expression.cc:551 call.function->DispatchBest(&types)
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/hash_join_node.cc:386 filter.Bind(filter_schema, exec_context)
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/hash_join_node.cc:737 schema_mgr->BindFilter(join_options.filter, left_schema, right_schema, plan->query_context()->exec_context())
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/exec_plan.cc:587 MakeExecNode(this->factory_name, plan, std::move(inputs), *this->options, registry)
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/exec_plan.cc:582 std::get(input).AddToPlan(plan, registry)
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/exec_plan.cc:686 with_sink.AddToPlan(exec_plan.get())
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/hash_join_node_test.cc:945 DeclarationToExecBatches(std::move(join), parallel)
Google Test trace:
/Users/lama/workspace/arrow-2/cpp/src/arrow/acero/hash_join_node_test.cc:1120: RIGHT_OUTER EQ_IS parallel = true bloom_filter = false

It mentions that an "equal" operator for null, null is needed, which seems to require further investigation (however, I'm actually not sure why a join operation on null key columns would be necessary)

[github-actions]

⚠️ GitHub issue #35785 has been automatically assigned in GitHub to PR creator.