`-Znext-solver` Allow method calls on chains of assoc types of not-yet defined opaque types by ShoyuVanilla · Pull Request #161414 · rust-lang/rust

ShoyuVanilla · 2026-08-20T17:33:49Z

Fixes rust-lang/trait-system-refactor-initiative#248

Still lacks of tons of comments and some tests (such as recursive projections or assoc item of supertraits) 😅

And it have some temporary FIXME comments as I'm leaving for a vacation until next Monday and will fix them as soon as I can

r? lcnr

ShoyuVanilla · 2026-08-20T17:41:08Z


 pub type PredefinedOpaques<'tcx> = &'tcx ty::List<(ty::OpaqueTypeKey<'tcx>, Ty<'tcx>)>;
+pub type HiddenTypesOfOpaques<'tcx> =
+    &'tcx ty::List<(Ty<'tcx>, Option<ty::OpaqueHiddenTyBound<'tcx>>)>;


This is the flattend version of { (?x, [p1, p2]), (?y, []) } into [(?x, Some(p1)), (?x, Some(p2)), (?y, None) ], to make it slice-internable.

I'm not sure whether we should use a single representation 🤔
And I'm not sure whether we really need the empty bounds, as we still have <?x: MetaSized> bounds for impl ?Sized (would we have some real empty one someday?)

View changes since the review

ShoyuVanilla · 2026-08-20T17:47:17Z

@bors try @rust-timer queue

`-Znext-solver` Allow method calls on (recursive) assoc types of not-yet defined opaque types

rust-bors · 2026-08-20T18:52:56Z

☀️ Try build successful (CI)
Build commit: 9b6316b (9b6316b23f8c267f599a254241d3b4bc0bac90c9)
Base parent: f7d782a (f7d782a3be46d6bb4b9792fe69a61db389ba1769)

rust-timer · 2026-08-20T19:35:42Z

Finished benchmarking commit (9b6316b): comparison URL.

Overall result: ❌ regressions - please read:

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.

Next, please: If you can, justify the regressions found in this try perf run in writing along with @rustbot label: +perf-regression-triaged. If not, fix the regressions and do another perf run. Neutral or positive results will clear the label automatically.

@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)	0.2%	[0.1%, 0.3%]	6
Regressions ❌ (secondary)	2.9%	[0.2%, 15.7%]	29
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.0%	[-1.0%, -1.0%]	1
All ❌✅ (primary)	0.2%	[0.1%, 0.3%]	6

Max RSS (memory usage)

Results (primary -1.1%, secondary 1.8%)

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)	3.9%	[0.7%, 7.6%]	14
Improvements ✅ (primary)	-1.1%	[-1.1%, -1.1%]	1
Improvements ✅ (secondary)	-4.1%	[-7.9%, -2.1%]	5
All ❌✅ (primary)	-1.1%	[-1.1%, -1.1%]	1

Cycles

Results (secondary 6.5%)

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)	6.5%	[3.2%, 11.4%]	9
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

Results (primary -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
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.1%	[-0.1%, -0.1%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.1%	[-0.1%, -0.1%]	1

Bootstrap: 456.256s -> 457.797s (0.34%)
Artifact size: 398.95 MiB -> 399.18 MiB (0.06%)

lcnr · 2026-08-25T16:22:03Z

+                self.cx
+            }
+
+            fn fold_ty(&mut self, ty: I::Ty) -> I::Ty {


this needs to fold_binder to because we need to shift the index of the anon-bound inwards. Otherwise, use could use https://doc.rust-lang.org/nightly/nightly-rustc/rustc_middle/ty/enum.BoundVarIndexKind.html#variant.Canonical and change this to BoundVarIndexKind::UniqueOuterScope

View changes since the review

The former sounds more preferable, as the later might not work well with the canonicalization 😄

lcnr

really cool. Pretty much what I expected from looking at the main changes :3

View changes since this review

lcnr · 2026-08-25T17:56:09Z

    opaque_types: usize,
    duplicate_entries: usize,
+    hidden_types_of_opaques: FxIndexMap<Ty<'tcx>, usize>,


that seems quite expensive 🤔

I guess this and more importantly, canonicalization of opaque types now has to also canonicalize hidden_types_of_opaques

Do you mean the canonicalizations in the rustc_infer? (as the canonicaliztions in the solver handles them)

For most cases, hidden_types_of_opaques doesn't grow too large by now, so it might make sense to just flatten this in the storage and track the whole len() for it

lcnr · 2026-08-25T18:12:21Z

I feel like one big change is to change the way we handle opaque types in inputs and external constraints in the response outside of this PR. The fact that we intern and then canonicalize and then reintern seems clearly bad 🤣

You have a FIXME in the code somewhere for that already :>

ShoyuVanilla · 2026-08-26T05:51:15Z

I feel like one big change is to change the way we handle opaque types in inputs and external constraints in the response outside of this PR. The fact that we intern and then canonicalize and then reintern seems clearly bad 🤣

You have a FIXME in the code somewhere for that already :>

Yeah, I really should refactor and optimize them. Currently I'm trying to make things correct wrt the above my comment (has_no_inference_or_external_constraints) It won't take too long though 😅

adwinwhite · 2026-08-26T07:35:48Z

            assert!(delegate.clone_opaque_types_lookup_table().is_empty());
        }

+        for chunk in input.hidden_types_of_opaques_in_body.as_slice().chunk_by(|a, b| a.0 == b.0) {


why do we flatten the bounds outside the method_autoderef query and gather them by hidden type here again?

View changes since the review

It's kinda mess how I pass hidden_types_of_opaques in this PR yet, basically in two ways: flattened vec and the nested one keyed by hidden type.
I gathered them by the hidden type here just to prevent pushing to many undo logs. I'll fix the hidden_types_of_opaques eventually and I hope it would be look much nicer by then 😅

adwinwhite · 2026-08-26T08:05:25Z

    let self_ty_is_opaque = |ty: Ty<'_>| {
        if let &ty::Infer(ty::TyVar(vid)) = ty.kind() {
-            infcx.has_opaques_with_sub_unified_hidden_type(vid)
+            infcx.has_hidden_types_of_opaques_modulo_sub_unification(vid)


This might be obvious but I'm struggling to understand why we can detect the projection term via hidden infer var. 😢
Issue 248 is about <hidden as Trait>::Projection: ProjectionItemBound. The self_ty in method lookup is an fresh infer var from normalizing <hidden as Trait>::Projection, I presume? So it's not the hidden infer var. However, item bounds we gather from the projection item are keyed by hidden 🤔

View changes since the review

The self_ty in method lookup is an fresh infer var from normalizing <hidden as Trait>::Projection, I presume?

Yes, it is. So has_hidden_types_of_opaques_modulo_sub_unification checks both cases for ?fresh_infer_var_for_opaque and ?fresh_infer_var_for_possibly_multiple_projection_on_opaque), the key for those bounds.

It is necessary to check the later as we have the cases with <<{opaque} as TraitA>::Assoc> as TraitB>::Assoc like the following, which is compiled with the old solver (and this PR) but not with the next-solver:

trait Foo { fn foo(&self) {} } trait Bar { type Assoc: Foo; fn bar(&self) -> Self::Assoc { loop {} } } trait Baz { type Assoc: Bar; fn baz(&self) -> Self::Assoc { loop {} } } impl Foo for () {} impl Bar for () { type Assoc = (); } impl Baz for () { type Assoc = (); } fn heck() -> impl Baz { heck().baz().bar().foo() } fn main() {}

So has_hidden_types_of_opaques_modulo_sub_unification checks both cases for ?fresh_infer_var_for_opaque and ?fresh_infer_var_for_possibly_multiple_projection_on_opaque), the key for those bounds.

In the method body, has_hidden_types_of_opaques_modulo_sub_unification only checks the keys of hidden_types_of_opaques so you mean those keys also contains fresh infer vars from normalizing projections 🤔
But I don't find where we add bounds for these fresh infer vars? They're all for hidden infers of opaques?

It's here:

https://github.com/rust-lang/rust/pull/161414/changes#diff-d8a295bd5782c31d629fd6936ca4d1874550fe422f1d7f798dac8998c3844524R1101-R1118

Finally got it. Thank you for the explanation!
We could use better naming for these things now that they handle more than hidden types of opaques?
Ofc we should fix correctness issues first :>

Yeah, I really should fix those namings 😄 I hope most of the correctness things(except folding binders and some canonicalization things) might be fixed by now so I'll try the perf and the naming/comments sides soon

ShoyuVanilla · 2026-08-26T21:14:34Z

        // FIXME: Explain this hack. Why this is needed and why should be done here
+        // FIXME: Maybe we need probing for whole this call as the following lines
+        // directly add `hidden_types_of_opaques` to the context without probing or
+        // instantiating the response.


Hmm, thinking twice, this feel might be okay to me, as the num_entries are taken from the delegate before the canonicalization of the goal and compared to the evaluation result for both function-like evaluation of normalized-to and the outer projection goal

View changes since the review

ShoyuVanilla · 2026-08-27T02:42:55Z

Hmm, the CI failure looks unfortunate

fn features_ok1() -> impl Iterator<Item = ()> {
    None.into_iter()
}

fn features_ok2<'a>() -> impl Iterator<Item = &'a ()> {
    None.into_iter()
}

// This emits an error :<
fn features_err() -> impl Iterator<Item = &'static ()> {
    None.into_iter()
}

'static lifetime in the bound of hidden_types_of_opaques from the input is canonicalized and instantiated into a placeholder region in the nested context, while the same lifetime in the response is canonicalized as-is.

So we have two different hidden_types_of_opaques in the nested context, one from the input <{opaque} as Iterator>::Item = &'!1 () and one we make trying to normalize the opaque inside the query <{opaque} as Iterator>::Item = &'static (). They are actually unified into the single one from the outer context, but inside the nested context, the later doesn't exist from the input so it keeps goal evaluation progressed forever with no actual progress.

So, basically I'm fixing such infinite progress with no fruitful result at all over multiple times 😅 I think I would be able to fix this one as well.

Edit) Fixed by checking whether the instantiation of the response actually changes the num_entries of the opaque storage

ShoyuVanilla · 2026-08-30T16:09:43Z

I'd like to do a perf run before trying some optimizations

@bors try @rust-timer queue

`-Znext-solver` Allow method calls on chains of assoc types of not-yet defined opaque types

rust-bors · 2026-08-30T17:08:59Z

☀️ Try build successful (CI)
Build commit: 3d55cff (3d55cff0498709d6bb560ae69670c9eb17280a97)
Base parent: 2e071b2 (2e071b28ef7e8a066b49a179e1da753c53500c62)

rust-timer · 2026-08-30T17:52:21Z

Finished benchmarking commit (3d55cff): comparison URL.

Overall result: ❌✅ regressions and improvements - please read:

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.

Next, please: If you can, justify the regressions found in this try perf run in writing along with @rustbot label: +perf-regression-triaged. If not, fix the regressions and do another perf run. Neutral or positive results will clear the label automatically.

@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)	0.2%	[0.1%, 0.3%]	8
Regressions ❌ (secondary)	2.9%	[0.4%, 13.6%]	27
Improvements ✅ (primary)	-0.2%	[-0.2%, -0.2%]	2
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	1
All ❌✅ (primary)	0.1%	[-0.2%, 0.3%]	10

Max RSS (memory usage)

Results (primary -0.4%, secondary 1.8%)

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

	mean	range	count
Regressions ❌ (primary)	1.1%	[0.6%, 1.6%]	2
Regressions ❌ (secondary)	3.8%	[0.6%, 11.2%]	17
Improvements ✅ (primary)	-3.4%	[-3.4%, -3.4%]	1
Improvements ✅ (secondary)	-3.8%	[-5.8%, -2.5%]	6
All ❌✅ (primary)	-0.4%	[-3.4%, 1.6%]	3

Cycles

Results (secondary 5.0%)

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)	5.0%	[2.3%, 9.4%]	10
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

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

Bootstrap: 475.013s -> 478.006s (0.63%)
Artifact size: 402.89 MiB -> 403.05 MiB (0.04%)

…yet defined opaque types

ShoyuVanilla · 2026-09-06T06:14:13Z

I'd like to do a perf run before trying some optimizations, again 😅

@bors try @rust-timer queue

`-Znext-solver` Allow method calls on chains of assoc types of not-yet defined opaque types

rust-bors · 2026-09-06T07:15:36Z

☀️ Try build successful (CI)
Build commit: 3f30747 (3f3074761460f29e6fd9b22330231bf7778f903d)
Base parent: dbad1ba (dbad1bab60868d49978b716635ac2c3203e327cf)

rust-timer · 2026-09-06T08:28:38Z

Finished benchmarking commit (3f30747): comparison URL.

Overall result: ❌✅ regressions and improvements - please read:

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.

Next, please: If you can, justify the regressions found in this try perf run in writing along with @rustbot label: +perf-regression-triaged. If not, fix the regressions and do another perf run. Neutral or positive results will clear the label automatically.

@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)	2.6%	[0.1%, 13.7%]	30
Improvements ✅ (primary)	-0.2%	[-0.2%, -0.2%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.2%	[-0.2%, -0.2%]	2

Max RSS (memory usage)

Results (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)	3.0%	[0.7%, 4.6%]	4
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-6.6%	[-11.5%, -2.2%]	6
All ❌✅ (primary)	-	-	0

Cycles

Results (primary -2.5%, secondary 4.4%)

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)	5.5%	[2.7%, 7.9%]	7
Improvements ✅ (primary)	-2.5%	[-2.6%, -2.5%]	2
Improvements ✅ (secondary)	-3.3%	[-3.3%, -3.3%]	1
All ❌✅ (primary)	-2.5%	[-2.6%, -2.5%]	2

Binary size

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

Bootstrap: 478.423s -> 479.616s (0.25%)
Artifact size: 403.41 MiB -> 403.61 MiB (0.05%)

rustbot assigned lcnr Aug 20, 2026

ShoyuVanilla commented Aug 20, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

ShoyuVanilla commented Aug 20, 2026

View reviewed changes

Comment thread compiler/rustc_next_trait_solver/src/solve/mod.rs Outdated

ShoyuVanilla force-pushed the opaque-hidden-bounds branch from 67742fd to a99368e Compare August 20, 2026 17:46

This comment has been minimized.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Aug 20, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Aug 20, 2026

Auto merge of #161414 - ShoyuVanilla:opaque-hidden-bounds, r=<try>

9b6316b

`-Znext-solver` Allow method calls on (recursive) assoc types of not-yet defined opaque types

This comment has been minimized.

Sign in to view

rustbot added perf-regression Performance regression. and removed S-waiting-on-perf Status: Waiting on a perf run to be completed. labels Aug 20, 2026

lcnr reviewed Aug 21, 2026

View reviewed changes

Comment thread compiler/rustc_infer/src/infer/opaque_types/table.rs Outdated

lcnr mentioned this pull request Aug 23, 2026

New trait solver can't infer types inferable by the old solver #161560

Open

lcnr reviewed Aug 25, 2026

View reviewed changes

adwinwhite reviewed Aug 26, 2026

View reviewed changes

ShoyuVanilla force-pushed the opaque-hidden-bounds branch from a99368e to dba8782 Compare August 26, 2026 20:48

ShoyuVanilla commented Aug 26, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

ShoyuVanilla changed the title ~~-Znext-solver Allow method calls on (recursive) assoc types of not-yet defined opaque types~~ -Znext-solver Allow method calls on chains of assoc types of not-yet defined opaque types Aug 27, 2026

This comment has been minimized.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Aug 30, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Aug 30, 2026

Auto merge of #161414 - ShoyuVanilla:opaque-hidden-bounds, r=<try>

3d55cff

`-Znext-solver` Allow method calls on chains of assoc types of not-yet defined opaque types

ShoyuVanilla force-pushed the opaque-hidden-bounds branch from 99f86af to 81691be Compare August 30, 2026 16:48

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Aug 30, 2026

BoxyUwU mentioned this pull request Sep 1, 2026

Type inference regression on nightly (E0282) #162150

Open

This comment has been minimized.

Sign in to view

ShoyuVanilla added 8 commits September 6, 2026 15:04

-Znext-solver Allow method calls on (recursive) assoc types of not-…

0acf19a

…yet defined opaque types

Fix incorrect things

6299b71

Add another test

0ae7231

Fix some comments

5a930e6

Check has_changed of response after instantiating the opaques

1abf02e

Remove Option wrapper of hidden bounds and rename bunch of things

287a874

Improve and fix constructions of OpaqueHiddenTyBound

1df5105

Less surprise in hidden types of opaques dedup

cf9f807

ShoyuVanilla force-pushed the opaque-hidden-bounds branch from 81691be to cf9f807 Compare September 6, 2026 06:13

This comment has been minimized.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Sep 6, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Sep 6, 2026

Auto merge of #161414 - ShoyuVanilla:opaque-hidden-bounds, r=<try>

3f30747

`-Znext-solver` Allow method calls on chains of assoc types of not-yet defined opaque types

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Sep 6, 2026

Uh oh!

Uh oh!

Conversation

ShoyuVanilla commented Aug 20, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ShoyuVanilla Aug 20, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

Uh oh!

ShoyuVanilla commented Aug 20, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Aug 20, 2026

Uh oh!

This comment has been minimized.

rust-timer commented Aug 20, 2026

Overall result: ❌ regressions - please read:

Uh oh!

Uh oh!

lcnr Aug 25, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lcnr left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lcnr commented Aug 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ShoyuVanilla commented Aug 26, 2026

Uh oh!

adwinwhite Aug 26, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

adwinwhite Aug 26, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

adwinwhite Aug 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ShoyuVanilla Aug 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ShoyuVanilla Aug 26, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

ShoyuVanilla commented Aug 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ShoyuVanilla commented Aug 30, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

ShoyuVanilla commented Aug 20, 2026 •

edited by rustbot

Loading

ShoyuVanilla Aug 20, 2026 •

edited by rustbot

Loading

lcnr Aug 25, 2026 •

edited by rustbot

Loading

lcnr left a comment •

edited by rustbot

Loading

lcnr commented Aug 25, 2026 •

edited

Loading

adwinwhite Aug 26, 2026 •

edited by rustbot

Loading

adwinwhite Aug 26, 2026 •

edited by rustbot

Loading

adwinwhite Aug 26, 2026 •

edited

Loading

ShoyuVanilla Aug 27, 2026 •

edited

Loading

ShoyuVanilla Aug 26, 2026 •

edited by rustbot

Loading

ShoyuVanilla commented Aug 27, 2026 •

edited

Loading