GH-45664: [C++] Allow LargeString,LargeBinary,FixedSizeBinary,StringView and BinaryView for RecordBatch::MakeStatisticsArray()

[andishgar]

Rationale for this change

MakeStatisticsArray in RecordBatch does not support StringView,BinaryView,LargeString,LargeBinary,FixedSizeBinary type

What changes are included in this PR?

the correction of MakeStatisticsArray in RecordBatch to support all arrow string types and relevant tests. Additionally, some changes applied to MakeStatisticsArray in record_batch_test.cc

Are these changes tested?

Yes, I run all relevant unit tests

Are there any user-facing changes?

No.( Add support for large_utf8, large_binary,fixed_size_binary, StringView, and BinaryView to RecordBatch::MakeStatisticsArray()).

• GitHub Issue: [C++] Allow LargeString,LargeBinary,FixedSizeBinary,StringView and BinaryView for RecordBatch::MakeStatisticsArray() #45664

[kou]

The PR title and associated issue focus on ArrayStatistics and StringView/BinaryView.
But the PR description and changes focus on RecordBatch::MakeStatisticsArray().
How about updating the PR title and associated issue to focus on RecordBatch::MakeStatisticsArray()?

The PR description says that this is a "Critical Fix". Why do you think that this is a "Critical Fix"?

[andishgar]

The PR description says that this is a "Critical Fix". Why do you think that this is a "Critical Fix"?

"The issue with the initial implementation (the one in the main branch) is that RecordBatch::MakeStatisticsArray() cannot handle large_utf8 and large_binary. I’m referring to the following code in the main branch

TEST_F(TestRecordBatch, MakeStatisticsArrayString) {
  auto schema =
      ::arrow::schema({field("no-statistics", boolean()), field("string", large_utf8())});
  auto no_statistics_array = ArrayFromJSON(boolean(), "[true, false, true]");
  auto string_array_data = ArrayFromJSON(large_utf8(), "[\"a\", null, \"c\"]")->data()->Copy();
  string_array_data->statistics = std::make_shared<ArrayStatistics>();
  string_array_data->statistics->is_max_exact = true;
  string_array_data->statistics->max = "c";
  auto string_array = MakeArray(std::move(string_array_data));
  auto batch = RecordBatch::Make(schema, string_array->length(),
                                 {no_statistics_array, string_array});

  ASSERT_OK_AND_ASSIGN(auto statistics_array, batch->MakeStatisticsArray());
  ARROW_LOGGER_INFO("",statistics_array->ToString());
}

print a massage like

/home/arashandishgar/Desktop/arrow/cpp/src/arrow/record_batch_test.cc:1440: -- is_valid: all not null
Running main() from /home/arashandishgar/Desktop/arrow/cpp/cmake-build-ninja-debug-1/_deps/googletest-src/googletest/src/gtest_main.cc
-- child 0 type: int32
  [
    null,
    1
  ]
-- child 1 type: map<dictionary<values=string, indices=int32, ordered=0>, dense_union<int64: int64=0, large_utf8: large_string=1>>
<Invalid array: List child array invalid: Invalid: Struct child array #1 invalid: Invalid: Union child array #1 invalid: Invalid: Offsets buffer size (bytes): 12 isn't large enough for length: 1 and offset: 0
/home/arashandishgar/Desktop/arrow/cpp/src/arrow/array/validate.cc:597  ValidateOffsetsAndSizes(type, values.size())

/home/arashandishgar/Desktop/arrow/cpp/src/arrow/array/validate.cc:176  ValidateBinaryLike(type)
/home/arashandishgar/Desktop/arrow/cpp/src/arrow/array/validate.cc:277  ValidateListLike(type)>

[kou]

OK. I think that it's just "not supported yet" not a bug.

[andishgar]

OK. I think that it's just "not supported yet" not a bug.

I apologize for using 'critical fix' in my description—I’ve updated it. (I used it because I thought it tied to this: (b) a bug causing incorrect or invalid data,)