GH-40126: [C++] Decimal types with different precisions and scales bind failed in resolve type when call arithmetic function

[ZhangHuiGui]

Rationale for this change

Fix decimal types with different precisions and scales bind failed in resolve type when call arithmetic function.

What changes are included in this PR?

Add IsNeedDispatchBest function to check the decimal types and arithmetic functions, if success we will
go into the dispatchBest path and do the implicit cast correctly.

Are these changes tested?

Yes

Are there any user-facing changes?

Yes, user needn't do their own cast for decimal related logic.

• GitHub Issue: [C++] DCHECK_EQ failed in ResolveDecimalAdditionOrSubtractionOutput function when execute two different decimal scale arrays' add  #40126

[github-actions]

⚠️ GitHub issue #40126 has been automatically assigned in GitHub to PR creator.

[ZhangHuiGui]

@bkietz PTAL, the decimal types with arithmetic function need the user cast by their own in the past, for these changes, we could go into a correct path and do implicit cast internal. Any suggestions?

[mapleFU]

I'm not sure handling it like this is prefered. cc @felipecrv @pitrou

[pitrou]

@bkietz Would you like to take a look at this?

[bkietz]

Please rebase to pick up the fix for https://github.com/apache/arrow/actions/runs/8092590090/job/22113575436?pr=40223#step:6:6328

[bkietz]

LGTM, I'll merge when CI completes

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 7 benchmarking runs that have been run so far on merge-commit e295a7e.

There was 1 benchmark result with an error:

• Commit Run on ursa-i9-9960x at 2024-02-29 22:54:01Z
  • tpch (R) with engine=arrow, format=native, language=R, memory_map=False, query_id=TPCH-12, scale_factor=10

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 6 possible false positives for unstable benchmarks that are known to sometimes produce them.

[mapleFU]

(so, as mentioned: #40223 (comment) . This pr:

1. Change ResolveDecimalBinaryOperationOutput to return false when precision match failed
2. When DispatchExact failed, FinishBind would fail, and code will falls to cast and dispatch best? )

[ZhangHuiGui]

2. When DispatchExact failed, FinishBind would fail, and code will falls to cast and dispatch best? )

Actually, the failure of FinishBind has nothing to do with DispatchExact, they are independent logics. We want FinishBind fail if resolve type failed when scale and precision not correct, so the code will falls dispatch best and cast to correct precision and scale as expected rules.