GH-37378: [C++] Add A Dictionary Compaction Function For DictionaryArray

[R-JunmingChen]

Rationale for this change

A Dictionary Compaction Function for DictionaryArray is supported.

What changes are included in this PR?

Add a Function for Dictionary Compaction

Are these changes tested?

Yes

Are there any user-facing changes?
No

• Closes: [C++] Add a Dictionary Compaction Function For DictionaryArray #37378

[github-actions]

⚠️ GitHub issue #37378 has been automatically assigned in GitHub to PR creator.

[bkietz]

Checking here enables skipping the rest of the dictionary, which is good. However I think it'd also be useful to detect usage of only a slice of the dictionary. If you'd prefer not to handle that in this PR, please write a follow up issue

[R-JunmingChen]

Hi, @bkietz, do you mean if we find it just use only a slice of dictionay, we use slice() instead of Take? I prefer to leave it as an new issue. Since we have another PR which is wating for this PR to be merged.

[mapleFU]

Can we just use Slice outside the Compact to handling this?

[R-JunmingChen]

I write a follow up issue #38247 to handle the optimization

[mapleFU]

Rest LGTM, thanks!

[mapleFU]

Should we add a test for invalid input type?

[R-JunmingChen]

No. We can't create a DictionaryArray with an invalid index type.

[mapleFU]

Oh I see a

  if (data->type->id() != Type::DICTIONARY) {
    return Status::TypeError("Expected dictionary type");
  }

Here, but seems it cannot be called

[bkietz]

LGTM, thanks for working on this!

[R-JunmingChen]

Strange. All C++ tests are canceled in CI.

[bkietz]

CI failures look unrelated. Thanks!

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 6 benchmarking runs that have been run so far on merge-commit 73454b7.

There were 4 benchmark results indicating a performance regression:

• Commit Run on ursa-i9-9960x at 2023-10-12 15:57:42Z
  • tpch (R) with engine=arrow, format=parquet, language=R, memory_map=False, query_id=TPCH-01, scale_factor=1
  • tpch (R) with engine=arrow, format=native, language=R, memory_map=False, query_id=TPCH-06, scale_factor=1
• and 2 more (see the report linked below)

The full Conbench report has more details. It also includes information about 4 possible false positives for unstable benchmarks that are known to sometimes produce them.