Optimizing Clippy & linting

[nikomatsakis]

Metadata
Point of contact	@blyxyas
Team(s)	clippy
Goal document	2025h1/optimize-clippy-linting-2

Summary

This is the formalization and documentation of the Clippy Performance Project, a project first talked about on Zulip, July 2023. As the project consists of several points and is ever-changing, this document also has a dynamic structure and the team can add points.

In short, this is an effort to optimize Clippy, and Rust's linting infrastructure with a point of view of making Clippy faster (both on CI/CD pipelines, and on devs' machines)

Tasks and status

• Implementation (@blyxyas, @Alexendoo)
• Standard reviews (clippy )

[blyxyas]

✔️ New medium-sized objective completed

rust-lang/rust-clippy#12830 Use ControlFlow in more places, closing rust-lang/rust-clippy#12829 Use ControlFlow in more visitors

[nikomatsakis]

This issue is intended for status updates only.

For general questions or comments, please contact the owner(s) directly.

[blyxyas]

Key developments:

• 1 Issue open with performance-project label ([Project Goal, Help Wanted] Optimizing MSRV attribute collection rust-clippy#13169), we've already integrated it into the project. @Alexendoo will be taking a look at it for the time being.
• 1 performance PR merged (see above comment)
• We're remaking our benchmarking tool (becnhv2)

  Our last benchmarking tool (becnh) wasn't actually benchmarking the way we would've liked. I'm currently remaking it from scratch, this time in Rust instead of Bash. See here and here for more details

What's to come:

• After becnhv2 is done, we'll continue working on (Big performance change) Do not run lints that cannot emit rust#125116 (which currently needs rebasing).

Blockers: There are no blockers, although we need some help figuring out the main bug on rust-lang/rust#125116, explained here. Someone from the compiler team taking a look at it would be greatly appreciated

[blyxyas]

Key developments:

• (Big performance change) Do not run lints that cannot emit rust#125116 is now all green! 🎉 We're currently doing benchmarks.
• (Clippy compilation) I'm working on moving our declare_clippy_lint macro from procedural to a macro_rules one.
• 1 Issue open (proposal) with the performance project label ([CI] Use rust-cache@v2 when building on CI rust-clippy#13033)

Blockers:
There are no blockers.

[blyxyas]

Key developments:
- GnomedDev pushed rust-lang/rust#130553, which replaced an old Clippy infrastructure with a faster one (string matching into symbol matching)
- Inspections into Clippy's type sizes and cache alignment are being started. Nothing fruitful yet, but I'm checking if there's something to be improved.
- The WIP benchmarking tool for Clippy has changed the algorithm from git based to using static source code distributions (like rustup), I'm experimenting with more possibilities.

Blockers:
There are no blockers

[blyxyas]

✔️ New large objective completed

rust-lang/rust#125116 has been merged, marking half of our goal formally done!

[blyxyas]

Key developments:

• Discussions on [CI] Use rust-cache@v2 when building on CI rust-clippy#13033 (comment) are beginning to take form.
• (Big performance change) Do not run lints that cannot emit rust#125116 may be contested in results. The impact may not be as large as expected, even on Clippy.
• We've been experimenting with Clippy using rustc_driver as a static library, instead of dynamic linking. This would be us both a way to check the performance impact of rustc_driver as a shared library, and a way to profile Clippy without filtering between dl_* calls.

Blockers: None known.
Help wanted: Help on any issue marked with performance-project would be greatly appreciated!

[blyxyas]

Key developments:

• Check for MSRV attributes in late passes using the HIR rust-clippy#13821 is open, has been benchmarked and is in the reviewing process. This PR addresses one of the sub-goals.

Blockers:
Help wanted: Help on any issue marked with performance-project would be greatly appreciated! Specially rust-lang/rust-clippy#13714

[nikomatsakis]

This is a continuing project goal, and the updates below this comment will be for the new period 2025h1

[blyxyas]

As a monthly update (previously posted on Zulip): We have some big progress!

• rust-clippy#13821 has been open and is currently being reviewed. It moves the MSRV logic out of lint-individualistic attribute extraction and into Clippy-wide MSRV (with a very good optimization, taking into account that only a small percentage of crates use.

• A Clippy-exclusive benchmarker has arrived, powered by the existing lintcheck infrastructure and perf. So it's compatible with flamegraph and other such tools. rust-clippy#14194 We can later expand this into CI or a dedicated bot.

• As you probably know, rust#125116 has been merged, just as a reminder of how that big goal was slayed like a dragon 🐉.

We now know what functions to optimize (or, at least have a basic knowledge of where Clippy spends its time).
As some future functionality, I'd love to have a functionality to build cargo and rust with debug symbols and hook it up to Clippy, but that may be harder. It's not perfect, but it's a good start!