[C++] Add Compute Kernel for Casting from String to Dictionary

[llama90]

Describe the enhancement requested

This is a sub-issue of the issue mentioned below.

• [C++] Remove Scalar legacy cast implementation #35560

As we can see in the parent issue, we can find tables organizing the related functions.

file name	code snippet
partition_test.cc	ASSERT_OK_AND_ASSIGN(auto dict_hello, MakeScalar("hello")->CastTo(DictStr("")->type()));
partition_test.cc	ASSERT_OK_AND_ASSIGN(auto dict_hello, MakeScalar("hello")->CastTo(dict_type));
scalar_test.cc	ASSERT_OK_AND_ASSIGN(auto cast_alpha, alpha->CastTo(ty));

The functions result in "Unsupported cast from string to dictionary using function cast_dictionary" error.

We need to implement casting from String to Dictionary.

partition_test.cc

Snippet to reproduce

TEST_F(TestPartitioning, DirectoryPartitioningFormatDictionary) {

  ...

- ASSERT_OK_AND_ASSIGN(auto dict_hello, MakeScalar("hello")->CastTo(DictStr("")->type()));
+ ASSERT_OK_AND_ASSIGN(auto dict_hello, Cast(MakeScalar("hello"), DictStr("")->type()));
}

TEST_F(TestPartitioning, DirectoryPartitioningFormatDictionaryCustomIndex) {

  ...

- ASSERT_OK_AND_ASSIGN(auto dict_hello, MakeScalar("hello")->CastTo(dict_type));
+ ASSERT_OK_AND_ASSIGN(auto dict_hello, Cast(MakeScalar("hello"), dict_type));

scalar_test.cc

Snippet to reproduce

TEST(TestDictionaryScalar, Cast) {

  ...

      // Cast string to dict(..., string)
-     ASSERT_OK_AND_ASSIGN(auto cast_alpha, alpha->CastTo(ty));
-     ASSERT_OK(cast_alpha->ValidateFull());
+     ASSERT_OK_AND_ASSIGN(auto cast_alpha, Cast(alpha, ty));
+     const auto& scalar = cast_alpha.scalar();
+     ASSERT_OK(scalar->ValidateFull());
      ASSERT_OK_AND_ASSIGN(
          auto roundtripped_alpha,
-         checked_cast<const DictionaryScalar&>(*cast_alpha).GetEncodedValue());
+         checked_cast<const DictionaryScalar&>(*scalar).GetEncodedValue());

      ...

      // dictionaries differ, though encoded values are identical
-     ASSERT_FALSE(alpha_dict.Equals(*cast_alpha));
+     ASSERT_FALSE(alpha_dict.Equals(*scalar));

Component(s)

C++

[llama90]

Progress Update

What is the casting from String to Dictionary?

arrow/cpp/src/arrow/scalar_test.cc

Lines 1468 to 1498 in f7947cc

	TEST(TestDictionaryScalar, Cast) {
	for (auto index_ty : all_dictionary_index_types()) {
	auto ty = dictionary(index_ty, utf8());
	auto dict = checked_pointer_cast<StringArray>(
	ArrayFromJSON(utf8(), R"(["alpha", null, "gamma"])"));
	for (int64_t i = 0; i < dict->length(); ++i) {
	auto alpha =
	dict->IsValid(i) ? MakeScalar(dict->GetString(i)) : MakeNullScalar(utf8());
	// Cast string to dict(..., string)
	ASSERT_OK_AND_ASSIGN(auto cast_alpha, alpha->CastTo(ty));
	ASSERT_OK(cast_alpha->ValidateFull());
	ASSERT_OK_AND_ASSIGN(
	auto roundtripped_alpha,
	checked_cast<const DictionaryScalar&>(*cast_alpha).GetEncodedValue());
	ASSERT_OK_AND_ASSIGN(auto i_scalar, MakeScalar(index_ty, i));
	auto alpha_dict = DictionaryScalar({i_scalar, dict}, ty);
	ASSERT_OK(alpha_dict.ValidateFull());
	ASSERT_OK_AND_ASSIGN(
	auto encoded_alpha,
	checked_cast<const DictionaryScalar&>(alpha_dict).GetEncodedValue());
	AssertScalarsEqual(*alpha, *roundtripped_alpha);
	AssertScalarsEqual(*encoded_alpha, *roundtripped_alpha);
	// dictionaries differ, though encoded values are identical
	ASSERT_FALSE(alpha_dict.Equals(*cast_alpha));
	}
	}
	}

When printing casted value (cast_alpha->ToString())

• "alpha" -> [ "alpha" ][0]
• null -> null
• "gamma" -> [ "gamma" ][0]

Legacy CastTo logic

arrow/cpp/src/arrow/scalar.cc

Lines 1264 to 1269 in f7947cc

	Status Visit(const DictionaryType& dict_type) {
	auto& out = checked_cast<DictionaryScalar*>(out_)->value;
	ARROW_ASSIGN_OR_RAISE(auto cast_value, from_.CastTo(dict_type.value_type()));
	ARROW_ASSIGN_OR_RAISE(out.dictionary, MakeArrayFromScalar(*cast_value, 1));
	return Int32Scalar(0).CastTo(dict_type.index_type()).Value(&out.index);
	}