[STACKED] GCI: Imply outlives-bounds on free (generic) const items by fmease · Pull Request #143029 · rust-lang/rust

fmease · 2025-06-25T21:08:45Z

View all comments

Caution

Stacked on top of PR #164012. The empty "branch separator" commit is the boundary.

Part of #113521.

_{(No LLM was or will be used by me during the entire creation process of this PR)}

fmease · 2025-06-25T21:09:02Z

@bors2 try @rust-timer queue

GCI: Imply outlives-bounds on free (generic) const items

rust-bors · 2025-06-25T21:09:07Z

⌛ Trying commit 5aa3d2a with merge eb427c5…

To cancel the try build, run the command @bors2 try cancel.

rust-bors · 2025-06-25T23:39:56Z

☀️ Try build successful (CI)
Build commit: eb427c5 (eb427c5414bee321329c635f289f46816100d758, parent: d14d202262d13df896b0c624b0cec6e4bfde631a)

rust-timer · 2025-06-27T06:55:56Z

Finished benchmarking commit (eb427c5): comparison URL.

Overall result: ❌ regressions - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@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)	0.2%	[0.2%, 0.2%]	3
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[0.2%, 0.2%]	3

Max RSS (memory usage)

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

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: 689.033s -> 689.561s (0.08%)
Artifact size: 372.09 MiB -> 372.11 MiB (0.01%)

fmease · 2025-06-27T20:12:37Z

(All of these failures are cycle errors that now occur when we try to recover from the user using _ in / omitting the type of consts, this "niceness" really is the bane of my existence)

fmease · 2025-06-27T20:29:18Z

@bors2 try parent=last @rust-timer queue

rust-bors · 2025-06-27T20:29:22Z

⌛ Trying commit c2b8a0b with merge 25444d3…

To cancel the try build, run the command @bors2 try cancel.

GCI: Imply outlives-bounds on free (generic) const items Part of #113521.

rust-bors · 2025-06-27T22:59:43Z

☀️ Try build successful (CI)
Build commit: 25444d3 (25444d3ddb1285d7ea04f0b89fdf4318a474d5df, parent: d14d202262d13df896b0c624b0cec6e4bfde631a)

rust-timer · 2025-06-28T00:40:08Z

Finished benchmarking commit (25444d3): comparison URL.

Overall result: no relevant changes - no action needed

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

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

Instruction count

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

Max RSS (memory usage)

Results (secondary 2.6%)

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%	[2.5%, 2.7%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

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: 689.033s -> 690.033s (0.15%)
Artifact size: 372.09 MiB -> 372.08 MiB (-0.00%)

fmease · 2026-10-18T11:15:21Z

This is now ready for review. However, it's still stacked on top of cleanup / refactoring PR #164012 which needs to be reviewed & merged beforehand.

What has changed since the last review?

We now infer outlives bounds from const aliases (in ADT / CTA / const item signatures & const item RHSs (if direct)) addressing [STACKED] GCI: Imply outlives-bounds on free (generic) const items #143029 (comment) / [STACKED] GCI: Imply outlives-bounds on free (generic) const items #143029 (comment).
- We don't "yet" try to infer outlives bounds from other kinds of type-system consts / direct const args like "struct literals" (ConstKind::Value where the Ty is an Adt). Last week I couldn't come up with any snippet that could observe this, nothing of that sort compiled, maybe (m)GCA isn't far enough along in this regard. In any case, I'd like to postpone it & only ship what's implemented right now
Greatly improved the test coverage & quality
Replied to two or your review comments about perf & soundness/robustness: [STACKED] GCI: Imply outlives-bounds on free (generic) const items #143029 (comment), [STACKED] GCI: Imply outlives-bounds on free (generic) const items #143029 (comment)
- UPDATE: The fast path is no longer / not necessary. see most recent perf run: [perf-only] "[STACKED] GCI: Imply outlives-bounds on free (generic) const items" without the supposed fastpath #164515 (comment). I've removed it :)

@rustbot review

fmease · 2026-10-18T11:26:41Z

@bors try @rust-timer queue

[STACKED] GCI: Imply outlives-bounds on free (generic) const items

rust-bors · 2026-10-18T12:31:17Z

☀️ Try build successful (CI)
Build commit: ef27a0e (ef27a0e2fd3476ff2405e22d7d3b1072d6ca1bcf)
Base parent: 33295d7 (33295d734447c7a2c22c44190f11bce6164a035a)

rust-timer · 2026-10-18T13:26:16Z

Finished benchmarking commit (ef27a0e): comparison URL.

Overall result: ❌ regressions - 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

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)	0.2%	[0.2%, 0.2%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary -3.9%, secondary 2.8%)

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)	2.8%	[0.9%, 4.8%]	6
Improvements ✅ (primary)	-3.9%	[-3.9%, -3.9%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-3.9%	[-3.9%, -3.9%]	1

Cycles

Results (secondary -1.4%)

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)	1.7%	[1.7%, 1.7%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-4.5%	[-4.5%, -4.5%]	1
All ❌✅ (primary)	-	-	0

Binary size

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

Bootstrap: 488.147s -> 489.834s (0.35%)
Artifact size: 408.98 MiB -> 409.08 MiB (0.02%)

rustbot · 2026-10-23T19:47:40Z

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.

In summary, rename various things & update a lot of comments. 1. Rename some types, fns that have "clause" in their name to contain "outlives clause" instead. While we are in a module called `outlives`, it's not immediately obvious & it certainly doesn't imply that all of these things only represent / act on outlives-clauses. Requested by Boxy. 2. Both `global_inferred_outlives` and `explicit_map` store a "global" mapping from each crate-local item to a set of outlives-clauses collected so far (one containing the inferred ones, the other the explicit ones). So name both consistently: `global_inferred_clauses` and `global_explicit_clauses`, respectively. Similarly for their types which are now prefixed with `Global`. 3. Remove a bunch of really obvious & distracting comments. 4. Reduce rightward drift in the `ty::Dynamic` match arm by using an if-let guard. 5. Deduplicate code shared between ADTs & CTAs in `inferred_outlives_of` 6. `bug!()` in a `GenericArgKind::Const(_)` branch which we know is currently unreachable and which should explode if that were to change. 7. Rename `CrateClausesMap` to `CrateInferredOutlivesClausesMap`. The type is public & lives in `rustc_middle` since `inferred_outlives_crate` is a query. Considering that, its name is way too general. Moreover, remove all of its docs. This type can only be constructed via query `inferred_outlives_crate` (well, and via the provider) which has good docs already that also warn about the fact that query `inferred_outlives_of` should always be used instead. 8. Remove comment "Typically there won't be many of these [explicit outlives clauses], except in older code where they were mandatory" is unnecessary and factually incorrect since there are various places where we don't imply (early-bound) outlives-bounds. 9. Re. comment "Visit all the crates [...]" and "[...] we walk the crates again [...]" is no longer correct: We only (re)visit the items in the local crate; the inferred outlives clauses for cross-crate items have already been computed and stored in the relevant crate metadata files. 10. Fix the example given in `check_explicit_clauses`'s docs, it has several issues: Struct `Outer` gets used as a trait bound, it uses `...: 'static` bounds and states they get processed but as a matter of fact we don't process static bounds, we specifically only infer clauses where the LHS is early-bound! 11. Various minor code tweaks.

…HS if direct

fmease assigned BoxyUwU Jun 25, 2025

fmease added the F-generic_const_items `#![feature(generic_const_items)]` label Jun 25, 2025

fmease added this to Generic Const Items (GCI) Jun 25, 2025

rustbot added A-rustc-dev-guide Area: rustc-dev-guide T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Jun 25, 2025

fmease moved this to In Progress in Generic Const Items (GCI) Jun 25, 2025