[JonathanBrouwer]

This PR makes the compiler no longer emit any errors on pre-expansion lint attributes.
That is, the following code currently doesn't compile, but compiles after this PR:

#[deny(dead_code(arguments_not_allowed_here))]
#[cfg(false)]
fn test() { }

This was historically hard to fix, but since we ported the lint attributes to the attribute parsing system, this is now relatively easy.

Blocked on #164985 and #164895. This PR is stacked on top of those two changes.

Reviewable commit by commit, every commit passes tests. The first 3 commits are from the stacked PRs.

This PR introduces the AttrErrorGuaranteed type, which is similar to ErrorGuaranteed. AttrErrorGuaranteed represents that ShouldEmit::emit_err was called, which may or may not represent an actual diagnostic being emitted, because attributes parsed pre-expansion should not emit diagnostics.

r? @mejrs

[JonathanBrouwer]

@bors try @rust-timer queue

[rust-bors]

☀️ Try build successful (CI)
Build commit: 9509e76 (9509e763f32f4738ee0877b99b28a3f18d47d7e7)
Base parent: dc44e7f (dc44e7f2c436b6041742e138b9c2176f093196d5)

[rust-timer]

Finished benchmarking commit (9509e76): comparison URL.

Overall result: ✅ improvements - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

@bors rollup=iffy rustc-perf
@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.2%	[-0.2%, -0.2%]	3
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.2%	[-0.2%, -0.2%]	3

Max RSS (memory usage)

Results (primary 0.3%, secondary -0.8%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	1.2%	[0.5%, 1.9%]	4
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.5%	[-0.6%, -0.4%]	5
Improvements ✅ (secondary)	-0.8%	[-0.8%, -0.8%]	1
All ❌✅ (primary)	0.3%	[-0.6%, 1.9%]	9

Cycles

Results (primary 0.4%, secondary 0.5%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	0.7%	[0.4%, 1.5%]	11
Regressions ❌ (secondary)	0.7%	[0.4%, 1.5%]	6
Improvements ✅ (primary)	-0.5%	[-0.5%, -0.4%]	5
Improvements ✅ (secondary)	-0.7%	[-0.7%, -0.7%]	1
All ❌✅ (primary)	0.4%	[-0.5%, 1.5%]	16

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 487.351s -> 487.992s (0.13%)
Artifact size: 405.87 MiB -> 405.97 MiB (0.02%)

[rustbot]

Some changes occurred to diagnostic attributes.

cc @mejrs

Some changes occurred in compiler/rustc_attr_ir

cc @jdonszelmann

Some changes occurred in compiler/rustc_attr_parsing

cc @jdonszelmann

The parser was modified, potentially altering the grammar of (stable) Rust
which would be a breaking change.

cc @fmease

[JonathanBrouwer]

This needs a @rust-lang/lang FCP.

[JonathanBrouwer]

@bors try @rust-timer queue

[rust-bors]

☀️ Try build successful (CI)
Build commit: bf3b3ba (bf3b3ba03ef9e7aeb49f7f0f21f46e6fc2a15b72)
Base parent: 019b59d (019b59d8101bb3e141f1b880684747b4217beeb6)

[JonathanBrouwer]

This test needs to be changed

View changes since the review

[mejrs]

Personally I don't like assigning semantics to these attributes pre-expansion, perhaps we could get rid of that so we don't need to parse them at all.

Perhaps this could be done by getting rid of pre expansion lints entirely (for rustc, it's just the keywords in new edition lint). Or we could make these lints crate level only, so we need not think about lint attributes on such items.

See also #162813 (comment) (cc @petrochenkov)

[JonathanBrouwer]

I agree, I think the easiest thing to do is to make them crate-level only, as that's trivial to implement and preserves the behaviour of the lint which is slightly better pre expansion (tho debatable if that's even worth the implementation complexity)

If lang decides to do that I can change this PR to implement that instead (or do it as a follow-up). It would also make the code a lot nicer, as the entire last commit is not needed then :)

[rust-bors]

☔ The latest upstream changes (presumably #165002) made this pull request unmergeable. Please resolve the merge conflicts by rebasing.