Do not run per-module late lints if they can be all skipped

[Kobzol]

We run ~70 late lints for all dependencies even if they use --cap-lints=allow, which seems wasteful. It looks like these lints are super fast (unlike early lints), but still.

r? @ghost

[Kobzol]

@bors try @rust-timer queue

This won't actually have any improvements in rustc-perf, I want to get a compiler build, and see if there are any regressions for the new check.

[bors]

⌛ Trying commit 3c4fb44 with merge 0d5b1b6...

[Kobzol]

Actually, locally it looks like the early lints are much more expensive, even though there are less of them. It saved 3% of time locally while building dependencies (modulo parallelism though).

[Kobzol]

Calling tcx.lints_that_dont_need_to_run in check_ast_node seems to introduce some hang, CPU at 100% :/

[bors]

☀️ Try build successful - checks-actions
Build commit: 0d5b1b6 (0d5b1b628a87bacb4d18bdd2a01048efdf15d319)

[rust-timer]

Finished benchmarking commit (0d5b1b6): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.3%	[0.2%, 0.4%]	2
Improvements ✅ (primary)	-0.4%	[-0.6%, -0.2%]	6
Improvements ✅ (secondary)	-0.8%	[-1.6%, -0.2%]	8
All ❌✅ (primary)	-0.4%	[-0.6%, -0.2%]	6

Max RSS (memory usage)

Results (primary -2.5%, secondary 1.4%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.4%	[1.4%, 1.4%]	1
Improvements ✅ (primary)	-2.5%	[-2.5%, -2.5%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-2.5%	[-2.5%, -2.5%]	1

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 780.203s -> 781.223s (0.13%)
Artifact size: 366.14 MiB -> 365.99 MiB (-0.04%)

[Kobzol]

@rustbot ready

r? compiler

[BoxyUwU]

Suggested change

	builtin_lints.get_lints().iter().all(|l| dont_need_to_run.contains(&LintId::of(l)));
	builtin_lints.get_lints().iter().all(|lint| dont_need_to_run.contains(&LintId::of(lint)));

i misread l as 1 and was like wow its kinda weird to hardcode LintId::of(1) lmao

[BoxyUwU]

why not do the same thing here that we do in late_lint_crate and filter out passes that only contain stuff that doesn't need to be run?

[Kobzol]

We could do that, but I think that would be slower. When we run everything, it is built in the compile-time.prepared structure, which merged all lints together. If we filtered individual lints, they need to be built using a runtime-prepared structure. For the Clippy lints that makes sense, but here I essentially only want to distinguish between "everything is disabled due to --cap-lints=allow" and "something might be disabled, but let's still run in the merged mode".

[BoxyUwU]

i meant in the else branch (but github doesnt let me comment on the right line ☹️) we make a RuntimeCombinedLateLintPass but don't do any filtering out of dyn LateLintPasss that are just only allowed lints. e.g.

        let mut binding = store
            .late_module_passes
            .iter()
            .map(|mk_pass| (mk_pass)(tcx))
            .chain(std::iter::once(builtin_lints))
            .filter(|pass| { ... }) // new, filter out passes that only contain allowed lints
            .collect::<Vec<_>>();

[Kobzol]

But builtin_lints is a single LintPass that combines all the builtin late lints together. In other words, in its e.g. check_mod implementation, it calls:

LINT_1.check_mod();
LINT_2.check_mod();

etc. So we cannot filter these lints from the outside. In theory, we could filter them inside these lint visitor functions, but that would mean doing that check for every call of each visitor function times the number of lints we have, which... doesn't sound fast 😅