GH-47167: [C++][Dev] Update clang-format dependency

[AntoinePrv]

Rationale for this change

Update clang-format to better match modern IDEs.

What changes are included in this PR?

• Update clang-format
• Reformat code base with newer version

Are these changes tested?

Yes

Are there any user-facing changes?

No

• GitHub Issue: [C++][Dev] Update clang-format dependency #47167

[github-actions]

⚠️ GitHub issue #47167 has been automatically assigned in GitHub to PR creator.

[raulcd]

The Python failures are unrelated, I just merged the fix on main, if you rebase they will go away.
It seems the newer clang-format pre-commit (from 16.0.0) fails on Ubuntu 22.04 because textproto was added to identify 2.5.20 but Ubuntu seems to be using indentify 2.4.10-1 (https://packages.ubuntu.com/jammy/python3-identify).
@kou is using the system pre-commit a requirement? I am unsure what the discussion was when we moved from archery lint to pre-commit. Some info on when this was added:

• [CI][Dev] Add support for pre-commit 2.17.0 provided by Ubuntu 22.04 #46798

[AntoinePrv]

Without it, that sends back pre-commit to 15.*

[kou]

Supporting old pre-commit (system pre-commit) is for new contributors on Ubuntu 22.04.

If we have enough documentation (or something) for new contributors, we can require more newer pre-commit (and identify).

[AntoinePrv]

pre-commit is very easy to install via either conda or pip. Are there other dependencies we are getting from these sources?

[kou]

apache/arrow has many language implementations. We can't assume that all new contributors are familiar with Python, conda and/or pip.

For example, this thread #46686 (comment) considered about new R contributors.

[AntoinePrv]

A possible direction to help both could be to leverage pixi.

I'm using it locally and in many other projects (C++/Rust/Python): it combines conda dependencies and run commands in managed environments.

Once installed (which is fairly easy IMHO), a one command pixi run lint, pixi run test etc. could download and install all required dependencies as needed, execute intermediary steps (with caching), and run the commands.
What's great is that all type of dependencies (Python itself, R, compilers...) can be pulled, on all platforms.
In the case of pre-commit, I've already set it up to lint/format with the same executable used for both one-stop command, and through pre-commit.

I'm not claiming it could handle all Arrow development cases, but it certainly make for a great on-boarding.
Using it also does not prevent from not using it.
I've been meaning to write a blog post about how I use it in development. If that seems like an interesting direction, I could come back when done, and will be willing to contribute a configuration.

[kou]

I don't have a strong opinion how to care about new contributors. I don't object the Pixi (or pip) approach if we have the official document how to prepare pre-commit with Pixi or something.
I chose the supporting system pre-commit approach just because we don't need to write additional document how to prepare pre-commit.

It may be better that we start a discussion thread on [email protected] https://lists.apache.org/[email protected] to collect more opinions.

[zanmato1984]

I think it becomes more reasonable now to move forward this PR, since:

1. We have officially upgraded to C++20, which already requires a more recent clang (GH-45885: [C++] Require C++20 #48414). Actually we are using v18 in our CI ([1] [2]).
2. I added several concepts in GH-46063: [C++][Compute] Fix the issue that MinMax kernel emits -inf/inf for all-NaN input #48459 ([3]), which is formatted differently between current v14 (in precommit) and v15+ (in my local). This is quite annoying and I had to manually modify it to match the required v14 format. The clang-format behavior change wrt C++20 concepts can be found in [4] [5].

I've made a commit in this PR with an update to the clang-format version change from v20 to a less aggressive v18, and related code formatting. @kou @pitrou @AntoinePrv what do you think?

Thanks.

[1]

arrow/.github/workflows/cpp.yml

Line 84 in 7fe39d4

clang-tools: 18

[2]

arrow/.env

Line 60 in 7fe39d4

CLANG_TOOLS=18

[3]

arrow/cpp/src/arrow/compute/kernels/hash_aggregate.cc

Lines 283 to 285 in 3fb84ff

	template <typename T>
	concept CDecimalConcept = std::same_as<T, Decimal32> || std::same_as<T, Decimal64> ||
	std::same_as<T, Decimal128> || std::same_as<T, Decimal256>;

[4] llvm/llvm-project#56283
[5] https://releases.llvm.org/15.0.0/tools/clang/docs/ReleaseNotes.html#clang-format

[pitrou]

I've made a commit in this PR with an update to the clang-format version change from v20 to a less aggressive v18, and related code formatting.

That's fine with me on the principle. The lint CI job is failing, though.

[kou]

How about requiring (recommending?) Ubuntu 24.04 or later (not 22.04) for development in our documentation?