Move attributes out of rustc_hir by mejrs · Pull Request #160336 · rust-lang/rust

mejrs · 2026-08-01T13:27:31Z

Previous graph:

Graph with this pr:

So while it doesn't really look like much (might be nicer for pipeling or if you're editing rustc_hir), it unblocks followup work to split up this part of the crate graph further, like:

moving SanitizerSet to a (new?) crate below rustc_target so that rustc_target and rustc_attr_ir no longer depend on each other and can compile in parallel
rustc_session will only depend on rustc_hir for rustc_hir::definitions in its cstore module, there may be something there but I haven't looked in depth.

There's some work left wrt cleaning up imports everywhere, but that's for followups.

r? @JonathanBrouwer

rustbot · 2026-08-01T13:27:35Z

These commits modify the Cargo.lock file. Unintentional changes to Cargo.lock can be introduced when switching branches and rebasing PRs.

If this was unintentional then you should revert the changes before this PR is merged.
Otherwise, you can ignore this comment.

Some changes occurred in compiler/rustc_hir/src/attrs

cc @jdonszelmann, @JonathanBrouwer

mejrs · 2026-08-01T13:28:39Z

@rustbot try @rust-timer queue

JonathanBrouwer · 2026-08-01T20:39:47Z

@bors try

Move attributes out of rustc_hir

mejrs · 2026-08-01T20:40:36Z

Oops

mejrs · 2026-08-01T20:42:39Z

By the way I did a perf run on the other PR as well, see https://perf.rust-lang.org/compare.html?start=29e68fe2295f8fc2feb52b8cb0b61a055842fdcf&end=5bfd1598fde1f19d34aa3199461963dcce69bb0a&stat=instructions%3Au&tab=bootstrap

But the rustc_symbol_mangling result is weird, let's see if that happens again.

joshtriplett · 2026-08-01T21:13:52Z

@mejrs What did you use to graph the dependencies?

mejrs · 2026-08-01T21:43:30Z

@mejrs What did you use to graph the dependencies?

cargo depgraph --workspace-only --dedup-transitive-deps --root rustc-main | dot -Tpng > graph.png

Also take a look at the --timings option (https://rustc-dev-guide.rust-lang.org/profiling.html#profiling-the-compiler) - that's a good way to tell what's actually bottlenecking what.

rust-bors · 2026-08-01T22:46:58Z

☀️ Try build successful (CI)
Build commit: 8ff9bde (8ff9bdec9d19be2195089ff2c4d871a9dfe437fe)
Base parent: b430378 (b430378746a835fd5240e2c7169795d7de0547e8)

rust-timer · 2026-08-02T00:07:58Z

Finished benchmarking commit (8ff9bde): comparison URL.

Overall result: no relevant changes - no action needed

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

@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

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

Max RSS (memory usage)

Results (primary -6.2%, 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)	-	-	0
Regressions ❌ (secondary)	0.8%	[0.8%, 0.8%]	1
Improvements ✅ (primary)	-6.2%	[-6.2%, -6.2%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-6.2%	[-6.2%, -6.2%]	1

Cycles

Results (primary 0.5%, secondary -0.2%)

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

	mean	range	count
Regressions ❌ (primary)	1.4%	[0.5%, 3.4%]	7
Regressions ❌ (secondary)	3.0%	[0.5%, 8.5%]	4
Improvements ✅ (primary)	-1.0%	[-1.4%, -0.4%]	4
Improvements ✅ (secondary)	-1.3%	[-3.5%, -0.4%]	11
All ❌✅ (primary)	0.5%	[-1.4%, 3.4%]	11

Binary size

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

Bootstrap: 489.711s -> 487.73s (-0.40%)
Artifact size: 390.35 MiB -> 390.34 MiB (-0.00%)

joshtriplett · 2026-08-02T00:45:49Z

@mejrs wrote:

cargo depgraph --workspace-only --dedup-transitive-deps --root rustc-main | dot -Tpng > graph.png

Thanks!

Also take a look at the --timings option (https://rustc-dev-guide.rust-lang.org/profiling.html#profiling-the-compiler) - that's a good way to tell what's actually bottlenecking what.

I'm familiar; I added the support for doing ./x build compiler --timings. :)

(I just sent a PR to rustc-dev-guide to document that.)

rustbot · 2026-08-08T09:43:37Z

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 · 2026-08-08T10:59:21Z

Ok then I'm convinced, r=me after fixing CI
@bors p=1
@bors rollup=iffy
@bors note=likely to conflict

rust-bors · 2026-08-08T10:59:23Z

Unknown command "note". Run @bors help or go to https://bors.rust-lang.org/help to see available commands.

JonathanBrouwer · 2026-08-08T10:59:41Z

@bors rollup=iffy note=likely to conflict

JonathanBrouwer · 2026-08-08T13:15:51Z

@bors r+

rust-bors · 2026-08-08T13:15:53Z

📌 Commit bd11f25 has been approved by JonathanBrouwer

It is now in the queue for this repository.

@JonathanBrouwer

…ouwer Move attributes out of rustc_hir Previous graph: <img width="397" height="503" alt="image" src="https://github.com/user-attachments/assets/f1c101b1-eb2f-4032-a8e7-6b4f6aef39d5" /> Graph with this pr: <img width="505" height="524" alt="image" src="https://github.com/user-attachments/assets/489be58e-41c5-4c08-a25a-9dc04ebe00e2" /> So while it doesn't really look like much (might be nicer for pipeling or if you're editing rustc_hir), it unblocks followup work to split up this part of the crate graph further, like: - moving `SanitizerSet` to a (new?) crate below `rustc_target` so that `rustc_target` and `rustc_attr_ir` no longer depend on each other and can compile in parallel - `rustc_session` will only depend on `rustc_hir` for `rustc_hir::definitions` in its `cstore` module, there may be something there but I haven't looked in depth. There's some work left wrt cleaning up imports everywhere, but that's for followups. r? @JonathanBrouwer

…uwer Rollup of 12 pull requests Successful merges: - #160336 (Move attributes out of rustc_hir) - #160715 (ignore tests with the GCC backend if we can't find `libgccjit.so` for the target) - #157609 (fix: build-std on 32 bit arm with 64 bit time) - #160613 (Add regression test for unsized non-last struct field with overlapping impls) - #160658 (miri: implement more restrictive trivial-ABI checks) - #160704 (sort lint names in lint pass declarations) - #160707 (Add regression test for higher ranked fn pointer impl not general enough) - #160713 (Add regression test for #135287) - #160720 (triagebot: add ubiratan to infra-ci) - #160747 (rustc_errors: remove unused code) - #160751 (Add regression test for incremental borrowck ICE with generic const exprs) - #160753 (rustc_lint: remove unused rustc_attrs feature) Failed merges: - #158835 (rustc_passes: lint unused `#[path]` attributes on inline modules)

@JonathanBrouwer

Rollup merge of #160336 - mejrs:move_attribute, r=JonathanBrouwer Move attributes out of rustc_hir Previous graph: <img width="397" height="503" alt="image" src="https://github.com/user-attachments/assets/f1c101b1-eb2f-4032-a8e7-6b4f6aef39d5" /> Graph with this pr: <img width="505" height="524" alt="image" src="https://github.com/user-attachments/assets/489be58e-41c5-4c08-a25a-9dc04ebe00e2" /> So while it doesn't really look like much (might be nicer for pipeling or if you're editing rustc_hir), it unblocks followup work to split up this part of the crate graph further, like: - moving `SanitizerSet` to a (new?) crate below `rustc_target` so that `rustc_target` and `rustc_attr_ir` no longer depend on each other and can compile in parallel - `rustc_session` will only depend on `rustc_hir` for `rustc_hir::definitions` in its `cstore` module, there may be something there but I haven't looked in depth. There's some work left wrt cleaning up imports everywhere, but that's for followups. r? @JonathanBrouwer

…uwer Rollup of 12 pull requests Successful merges: - rust-lang/rust#160336 (Move attributes out of rustc_hir) - rust-lang/rust#160715 (ignore tests with the GCC backend if we can't find `libgccjit.so` for the target) - rust-lang/rust#157609 (fix: build-std on 32 bit arm with 64 bit time) - rust-lang/rust#160613 (Add regression test for unsized non-last struct field with overlapping impls) - rust-lang/rust#160658 (miri: implement more restrictive trivial-ABI checks) - rust-lang/rust#160704 (sort lint names in lint pass declarations) - rust-lang/rust#160707 (Add regression test for higher ranked fn pointer impl not general enough) - rust-lang/rust#160713 (Add regression test for rust-lang/rust#135287) - rust-lang/rust#160720 (triagebot: add ubiratan to infra-ci) - rust-lang/rust#160747 (rustc_errors: remove unused code) - rust-lang/rust#160751 (Add regression test for incremental borrowck ICE with generic const exprs) - rust-lang/rust#160753 (rustc_lint: remove unused rustc_attrs feature) Failed merges: - rust-lang/rust#158835 (rustc_passes: lint unused `#[path]` attributes on inline modules)

Remove rustc_middle dependency on rustc_hir_pretty There is a `impl PpAnn for TyCtxt` that is unneeded. None of the big crates (middle, trait_selection) actually do any hir pretty printing so it can be removed and can either be implemented for local structs elsewhere or done by casting to `&dyn PpAnn` instead. This probably doesn't have any perf effects (its not bottlenecking in timings) but I expect it to compose with rust-lang/rust#160336 and its followup works quite well. Old graph: <img width="1999" height="2171" alt="graph" src="https://github.com/user-attachments/assets/756e6a50-a413-4a8d-a0f3-7bbf4de07933" /> New graph: <img width="1853" height="2171" alt="graph_hir_pretty" src="https://github.com/user-attachments/assets/f9607e5a-d00b-4a86-8f6a-cefb45b75ad7" />

@JonathanBrouwer

split up `rustc_session` Followup on rust-lang#160336 cc @JonathanBrouwer (feel free to review if you want) The first commit splits out the `cstore` module from `rustc_session` into its own crate. `rustc_session` actually never used it anywhere internally, and there are also several crates that only depend on `rustc_session` for it. So it is a natural candidate to split off. The second commit is similar; `rustc_hir::definitions` is moved to `rustc_hir_id`; it is used nowhere in `rustc_hir`, except to implement an inherent method: ```rust impl DefKind { pub fn def_path_data(self, name: Option<Symbol>) -> DefPathData { } } ``` which is used (once) in rustc_middle, so the items in `rustc_hir::definitions` and this conversion function could be in many places. `rustc_hir_id` is the earliest and the easiest, so that's what I went with. Together these allow `rustc_crate_store`, `rustc_hir` and `rustc_session` to compile in parallel. Current graph: <img width="825" height="617" alt="image" src="https://github.com/user-attachments/assets/aa954265-0057-4596-a7c4-38290937fb74" /> Graph after this pr: <img width="853" height="572" alt="image" src="https://github.com/user-attachments/assets/9bcd16f8-897c-4990-920b-249500b2e78e" />

@JonathanBrouwer

split up `rustc_session` Followup on rust-lang#160336 cc @JonathanBrouwer (feel free to review if you want) The first commit splits out the `cstore` module from `rustc_session` into its own crate. `rustc_session` actually never used it anywhere internally, and there are also several crates that only depend on `rustc_session` for it. So it is a natural candidate to split off. The second commit is similar; `rustc_hir::definitions` is moved to `rustc_hir_id`; it is used nowhere in `rustc_hir`, except to implement an inherent method: ```rust impl DefKind { pub fn def_path_data(self, name: Option<Symbol>) -> DefPathData { } } ``` which is used (once) in rustc_middle, so the items in `rustc_hir::definitions` and this conversion function could be in many places. `rustc_hir_id` is the earliest and the easiest, so that's what I went with. Together these allow `rustc_crate_store`, `rustc_hir` and `rustc_session` to compile in parallel. Current graph: <img width="825" height="617" alt="image" src="https://github.com/user-attachments/assets/aa954265-0057-4596-a7c4-38290937fb74" /> Graph after this pr: <img width="853" height="572" alt="image" src="https://github.com/user-attachments/assets/9bcd16f8-897c-4990-920b-249500b2e78e" />

@JonathanBrouwer

split up `rustc_session` Followup on rust-lang#160336 cc @JonathanBrouwer (feel free to review if you want) The first commit splits out the `cstore` module from `rustc_session` into its own crate. `rustc_session` actually never used it anywhere internally, and there are also several crates that only depend on `rustc_session` for it. So it is a natural candidate to split off. The second commit is similar; `rustc_hir::definitions` is moved to `rustc_hir_id`; it is used nowhere in `rustc_hir`, except to implement an inherent method: ```rust impl DefKind { pub fn def_path_data(self, name: Option<Symbol>) -> DefPathData { } } ``` which is used (once) in rustc_middle, so the items in `rustc_hir::definitions` and this conversion function could be in many places. `rustc_hir_id` is the earliest and the easiest, so that's what I went with. Together these allow `rustc_crate_store`, `rustc_hir` and `rustc_session` to compile in parallel. Current graph: <img width="825" height="617" alt="image" src="https://github.com/user-attachments/assets/aa954265-0057-4596-a7c4-38290937fb74" /> Graph after this pr: <img width="853" height="572" alt="image" src="https://github.com/user-attachments/assets/9bcd16f8-897c-4990-920b-249500b2e78e" />

@JonathanBrouwer

Rollup merge of #160924 - mejrs:split_cstore, r=JonathanBrouwer split up `rustc_session` Followup on #160336 cc @JonathanBrouwer (feel free to review if you want) The first commit splits out the `cstore` module from `rustc_session` into its own crate. `rustc_session` actually never used it anywhere internally, and there are also several crates that only depend on `rustc_session` for it. So it is a natural candidate to split off. The second commit is similar; `rustc_hir::definitions` is moved to `rustc_hir_id`; it is used nowhere in `rustc_hir`, except to implement an inherent method: ```rust impl DefKind { pub fn def_path_data(self, name: Option<Symbol>) -> DefPathData { } } ``` which is used (once) in rustc_middle, so the items in `rustc_hir::definitions` and this conversion function could be in many places. `rustc_hir_id` is the earliest and the easiest, so that's what I went with. Together these allow `rustc_crate_store`, `rustc_hir` and `rustc_session` to compile in parallel. Current graph: <img width="825" height="617" alt="image" src="https://github.com/user-attachments/assets/aa954265-0057-4596-a7c4-38290937fb74" /> Graph after this pr: <img width="853" height="572" alt="image" src="https://github.com/user-attachments/assets/9bcd16f8-897c-4990-920b-249500b2e78e" />

Remove rustc_middle dependency on rustc_hir_pretty There is a `impl PpAnn for TyCtxt` that is unneeded. None of the big crates (middle, trait_selection) actually do any hir pretty printing so it can be removed and can either be implemented for local structs elsewhere or done by casting to `&dyn PpAnn` instead. This probably doesn't have any perf effects (its not bottlenecking in timings) but I expect it to compose with rust-lang/rust#160336 and its followup works quite well. Old graph: <img width="1999" height="2171" alt="graph" src="https://github.com/user-attachments/assets/756e6a50-a413-4a8d-a0f3-7bbf4de07933" /> New graph: <img width="1853" height="2171" alt="graph_hir_pretty" src="https://github.com/user-attachments/assets/f9607e5a-d00b-4a86-8f6a-cefb45b75ad7" />

Update deprecated rustc_hir imports Followup to rust-lang#160336 That's ~half of them, I'm not doing it all in one go to reduce conflicts.

rustbot added A-meta Area: Issues & PRs about the rust-lang/rust repository itself S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Aug 1, 2026

rustbot assigned JonathanBrouwer Aug 1, 2026

mejrs added the A-attributes Area: Attributes (`#[…]`, `#![…]`) label Aug 1, 2026