create fast path for attributes created in builtin macro expansions by mejrs · Pull Request #164260 · rust-lang/rust

mejrs · 2026-10-13T13:14:00Z

Emitting attributes in rustc_builtin_macros involves creating an ast::Attribute and token trees, just to parse that into a parsed attribute almost immediately. This PR sidesteps that by creating a new ast::Attribute variant that does not do this. I've taken care to only do this for the derive and test macros, so that they never re-enter the parser.

This results in some moderate performance gains. It'll also take away almost all of the regressions in #164217 (comment) (see #164281 (comment) for a perf run of this and that PR stacked)

r? @nnethercote

mejrs · 2026-10-13T13:14:18Z

@bors try @rust-timer queue

create fast path for attributes created in derive expansions

mejrs · 2026-10-13T13:25:59Z

@bors cancel

rust-bors · 2026-10-13T13:26:02Z

❗ There is currently no auto build in progress on this PR.

Hint: There is a pending try build on this PR. Maybe you meant to cancel it? You can do that using @bors try cancel.

mejrs · 2026-10-13T13:26:14Z

@bors try cancel

rust-bors · 2026-10-13T13:26:18Z

Try build cancelled. Cancelled workflows:

https://github.com/rust-lang/rust/actions/runs/38403374473

Hint: if you want to run another try build, you do not need to manually cancel the previous one. Just run @bors try and bors will cancel the previous build automatically.

mejrs · 2026-10-13T13:54:55Z

@bors try @rust-timer queue

create fast path for attributes created in derive expansions

rust-bors · 2026-10-13T14:54:37Z

☀️ Try build successful (CI)
Build commit: 88e073e (88e073e9b84eab1e63c7cd966216841f2284fa1f)
Base parent: 91e426a (91e426a0518d5bb3672330c5238ccda6fd26fb54)

rust-timer · 2026-10-13T17:10:38Z

Finished benchmarking commit (88e073e): comparison URL.

Overall result: ✅ improvements - no action needed

Benchmarking means the PR may be perf-sensitive. It's automatically marked not fit for rolling up. Overriding is possible but disadvised: it risks changing compiler perf.

@bors rollup=never 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)	-0.4%	[-1.1%, -0.1%]	74
Improvements ✅ (secondary)	-1.3%	[-3.8%, -0.2%]	26
All ❌✅ (primary)	-0.4%	[-1.1%, -0.1%]	74

Max RSS (memory usage)

Results (primary -0.8%, secondary 1.0%)

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

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.6%	[2.1%, 3.1%]	2
Improvements ✅ (primary)	-0.8%	[-0.8%, -0.8%]	1
Improvements ✅ (secondary)	-2.3%	[-2.3%, -2.3%]	1
All ❌✅ (primary)	-0.8%	[-0.8%, -0.8%]	1

Cycles

Results (primary -1.5%, secondary -3.9%)

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

	mean	range	count
Regressions ❌ (primary)	2.3%	[2.3%, 2.3%]	1
Regressions ❌ (secondary)	3.8%	[3.8%, 3.8%]	1
Improvements ✅ (primary)	-2.5%	[-3.2%, -1.6%]	4
Improvements ✅ (secondary)	-5.0%	[-6.9%, -3.2%]	7
All ❌✅ (primary)	-1.5%	[-3.2%, 2.3%]	5

Binary size

Results (primary 0.0%, secondary 0.1%)

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

	mean	range	count
Regressions ❌ (primary)	0.0%	[0.0%, 0.1%]	9
Regressions ❌ (secondary)	0.1%	[0.0%, 0.1%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.0%	[0.0%, 0.1%]	9

Bootstrap: 489.812s -> 488.427s (-0.28%)
Artifact size: 406.88 MiB -> 406.97 MiB (0.02%)

mejrs · 2026-10-13T18:57:32Z

@bors try @rust-timer queue

create fast path for attributes created in derive expansions

JonathanBrouwer

Like the idea, love the performance.
Some comments tho :)

View changes since this review

JonathanBrouwer · 2026-10-25T11:08:47Z

    DocComment(CommentKind, Symbol),
+
+    /// A macro generated attribute inserted by a builtin macro.
+    Generated(GeneratedAttr),


I'd really like to see the AttrKind::Synthetic and AttrKind::Generated be merged.
They're basically the same concept. I have no strong preferences about the naming

I saw @nnethercote proposed this in #164260 (comment) as well, but I didn't see this addressed.

They're a little different: synthetic ones cannot be written by hand and cannot be pretty-printed, while generated ones can (in both cases). IMO this is probably enough to justify keeping them separate.

I'd say the way they're generated is indeed different, and they need a different path in pretty printing (every variant is already handled individually there anyways so that's fine), but they basically behave the same. I don't really see the advantage to keeping them separate, and the disadvantage is more code.

I'd prefer the compiler to crash if they re-enter the parser somehow, rather than be silently swallowed like the trace attributes are.

I agree that that's desirable. I think you can still relatively easily make it work, it would be one extra match where the trace attributes are allowed, and a wildcard arm is unreachable

I think there's enough difference between them to justify having different enum variants. I'd rather this isn't an implicit thing, and having different variants/types makes for a good place to put this documentation

Since you both seem to agree that keeping them separate is better lets do that.
I'm not entirely convinced it's worth it, but don't feel strongly enough to block this PR on this.

rustbot · 2026-10-25T14:39:57Z

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

JonathanBrouwer

@bors r=JonathanBrouwer,nnethercote

Thanks! :3
The perf improvement makes me very happy

View changes since this review

rust-bors · 2026-10-25T14:44:31Z

📌 Commit db2155b has been tentatively approved by JonathanBrouwer,nnethercote

It will be put into the queue for this repository once PR CI succeeds.

JonathanBrouwer · 2026-10-25T15:03:34Z

@bors r=JonathanBrouwer,nnethercote rollup=iffy force
Creating a rollup of perf-sensitive PRs

rust-bors · 2026-10-25T15:03:37Z

📌 Commit db2155b has been approved by JonathanBrouwer,nnethercote

It is now in the queue for this repository.

@nnethercote

…rouwer,nnethercote create fast path for attributes created in builtin macro expansions Emitting attributes in rustc_builtin_macros involves creating an `ast::Attribute` and token trees, just to parse that into a parsed attribute almost immediately. This PR sidesteps that by creating a new `ast::Attribute` variant that does not do this. I've taken care to only do this for the derive and test macros, so that they never re-enter the parser. This results in some moderate performance gains. It'll also take away almost all of the regressions in rust-lang#164217 (comment) (see rust-lang#164281 (comment) for a perf run of this and that PR stacked) r? @nnethercote

…uwer Rollup of 3 perf-sensitive pull requests Successful merges: - #163845 (Add various `#[inline]` annotations for simple functions on hot paths that don't get inlined) - #164122 (Don't emit allow attributes in try/contract desugaring) - #164260 (create fast path for attributes created in builtin macro expansions) Failed merges: - #164790 (Split `LintLevelsBuilder::add` into `add_ast` and `add_hir`)

@nnethercote

…rouwer,nnethercote create fast path for attributes created in builtin macro expansions Emitting attributes in rustc_builtin_macros involves creating an `ast::Attribute` and token trees, just to parse that into a parsed attribute almost immediately. This PR sidesteps that by creating a new `ast::Attribute` variant that does not do this. I've taken care to only do this for the derive and test macros, so that they never re-enter the parser. This results in some moderate performance gains. It'll also take away almost all of the regressions in rust-lang#164217 (comment) (see rust-lang#164281 (comment) for a perf run of this and that PR stacked) r? @nnethercote

…uwer Rollup of 4 perf-sensitive pull requests Successful merges: - #163845 (Add various `#[inline]` annotations for simple functions on hot paths that don't get inlined) - #164122 (Don't emit allow attributes in try/contract desugaring) - #164260 (create fast path for attributes created in builtin macro expansions) - #164300 (Deduplicate DefId encodings in the query cache) Failed merges: - #164790 (Split `LintLevelsBuilder::add` into `add_ast` and `add_hir`)

…uwer Rollup of 5 pull requests Successful merges: - #163845 (Add various `#[inline]` annotations for simple functions on hot paths that don't get inlined) - #164122 (Don't emit allow attributes in try/contract desugaring) - #164260 (create fast path for attributes created in builtin macro expansions) - #164300 (Deduplicate DefId encodings in the query cache) - #164929 (Revert "rustc_codegen_gcc subtree update") Failed merges: - #164790 (Split `LintLevelsBuilder::add` into `add_ast` and `add_hir`)

@nnethercote

Rollup merge of #164260 - mejrs:generated_attrs, r=JonathanBrouwer,nnethercote create fast path for attributes created in builtin macro expansions Emitting attributes in rustc_builtin_macros involves creating an `ast::Attribute` and token trees, just to parse that into a parsed attribute almost immediately. This PR sidesteps that by creating a new `ast::Attribute` variant that does not do this. I've taken care to only do this for the derive and test macros, so that they never re-enter the parser. This results in some moderate performance gains. It'll also take away almost all of the regressions in #164217 (comment) (see #164281 (comment) for a perf run of this and that PR stacked) r? @nnethercote

rust-timer · 2026-10-26T04:23:27Z

Note

This PR was benchmarked as part of triage of its containing rollup: triage URL.

Finished benchmarking commit (5d7c38d): comparison URL.

Overall result: ✅ improvements - no action needed

@rustbot label: -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)	-0.4%	[-1.3%, -0.1%]	76
Improvements ✅ (secondary)	-1.3%	[-3.9%, -0.2%]	26
All ❌✅ (primary)	-0.4%	[-1.3%, -0.1%]	76

Max RSS (memory usage)

Results (primary 0.4%, secondary 1.1%)

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

	mean	range	count
Regressions ❌ (primary)	0.9%	[0.4%, 2.0%]	12
Regressions ❌ (secondary)	1.3%	[0.5%, 4.2%]	8
Improvements ✅ (primary)	-1.3%	[-2.9%, -0.5%]	4
Improvements ✅ (secondary)	-0.7%	[-0.7%, -0.7%]	1
All ❌✅ (primary)	0.4%	[-2.9%, 2.0%]	16

Cycles

Results (primary -0.6%, secondary -1.9%)

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

	mean	range	count
Regressions ❌ (primary)	0.5%	[0.4%, 0.5%]	5
Regressions ❌ (secondary)	1.2%	[0.5%, 2.7%]	4
Improvements ✅ (primary)	-1.0%	[-2.1%, -0.4%]	17
Improvements ✅ (secondary)	-3.6%	[-6.8%, -0.5%]	7
All ❌✅ (primary)	-0.6%	[-2.1%, 0.5%]	22

Binary size

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

Artifact size: 405.91 MiB -> 406.05 MiB (0.03%)

rustbot added A-attributes Area: Attributes (`#[…]`, `#![…]`) S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Oct 13, 2026