Prevent rustdoc auto-trait ICEs with `-Znext-solver=globally` by Jamesbarford · Pull Request #158346 · rust-lang/rust

Jamesbarford · 2026-06-24T08:34:06Z

Ensure rustdoc with -Znext-solver=globally does not ICE.

lcnr · 2026-06-25T08:04:01Z

+
+        let (infcx, orig_env) = tcx.infer_ctxt().build_with_typing_env(typing_env);
+        let mut selcx = SelectionContext::new(&infcx);
+        for polarity in [ty::PredicatePolarity::Positive, ty::PredicatePolarity::Negative] {


can you instead copy the behavior of

rust/compiler/rustc_next_trait_solver/src/solve/trait_goals.rs

Line 1247 in 8e25b41

fn disqualify_auto_trait_candidate_due_to_possible_impl(

here.

rn this doesn't consider there to be an explicit impl for MyType<T> if there's just impl Send for MyType<u32>

View changes since the review

ab62f15 👍

lcnr · 2026-06-25T08:04:09Z

+        let field_clauses = adt_def
+            .all_fields()
+            .map(|field| field.ty(tcx, args).skip_norm_wip())
+            .filter(|field_ty| field_ty.has_non_region_param())


why this?

View changes since the review

lcnr · 2026-06-25T08:04:56Z

+        let full_user_env = ty::ParamEnv::new(
+            tcx.mk_clauses_from_iter(orig_env.caller_bounds().iter().chain(field_clauses)),
+        );


hmm? 🤔 what are you trying to do/please provide a comment

View changes since the review

generally, can you describe the algorithm you're trying to implement in words?

This is currently somewhat wrong and I would like to know what it is that you are trying to do. Might also be useful to show the intended behavior on one or two examples

It is meant to be deliberately coarse and simply not ICE. Not a faithful port of the old algorithm. If I'm not mistaken the old path minimises the where-clauses by repeatedly selecting and folding unimplemented param-bounds back into the ParamEnv using the old-solver internals which ICEs under the new solver.

The new path just does one rather simple pass:

If there is a user-written impl then ExplicitImpl.

If ty isn't an ADT NoImpl.

Otherwise assume each generic field implements the trait (field_ty: Trait for every field mentioning a type/const param) and add those as the where-clauses.

Verify ty: Trait under those assumptions

true error NegativeImpl

ambiguity NoImpl

clean PositiveImpl

Example:

// foo.rs pub struct Foo<T> { pub data: Box<T>, } // previously impl<T> Send for Foo<T> where T: Send, // This PR impl<T> Send for Foo<T> where Box<T>: Send,

So I'm emitting the raw field bound rather than minimising to T: Send. But I understand that this is not what we are wanting to do? Instead I should be following fn disqualify_auto_trait_candidate_due_to_possible_impl(...)?

That makes sense 👍

Treating ambiguity as NoImpl should be wrong though 🤔

struct Foo<'a, 'b, T: 'a + 'b>(&'a T, &'b T);

proving Foo<'a, 'b, T>: Send given &'a T: Send and &'b T: Send results in an ambiguity error.

I think what step 4 is trying to handle is types which don't implement auto traits because of one of the non-generic fields 🤔

what we can do here is to instead replace all generic parameters with inference variables and then check whether Foo<'?0, '?1, ?t>: Send holds. If this returns an error, we emit a NegativeImpl. We never emit NoImpl for now

I have some ideas on how to refine this in the future, but I think as an initial impl this is good enough :>

The way to replace generic params with infer vars is e.g.

rust/compiler/rustc_hir_typeck/src/demand.rs

Lines 981 to 983 in c397dae

let generic_args = ty::GenericArgs::for_item(self.tcx, m.def_id, |param, _| {

self.var_for_def(deref.span, param)

});

and then you take the Adt type, wrap it in an EarlyBinder and instantiate it

b6fdb1d 👍 I found the following helper which does what you say under the hood so used it instead.

rust/compiler/rustc_infer/src/infer/mod.rs

Lines 1049 to 1053 in c397dae

/// Given a set of generics defined on a type or impl, returns the generic parameters mapping

/// each type/region parameter to a fresh inference variable.

pub fn fresh_args_for_item(&self, span: Span, def_id: DefId) -> GenericArgsRef<'tcx> {

GenericArgs::for_item(self.tcx, def_id, |param, _| self.var_for_def(span, param))

}

I've also added a couple of tests, not sure if you want them but they were fairly useful for making sure I captured the behaviour you described.

…t-bounds, r=fmease Fix: auto trait, const trait bound Split from rust-lang#158346. I was fiddling about with the code to understand the behaviour of the snippet below, got confused, and somehow fixed something 🤷 ```rust #![feature(const_trait_impl)] pub struct Outer<T>(Inner<T>); struct Inner<T>(T); impl<T> Unpin for Inner<T> where T: const std::default::Default, {} ``` Fixes rust-lang#149281

Rollup merge of #158390 - Jamesbarford:fix/trait-solver-const-bounds, r=fmease Fix: auto trait, const trait bound Split from #158346. I was fiddling about with the code to understand the behaviour of the snippet below, got confused, and somehow fixed something 🤷 ```rust #![feature(const_trait_impl)] pub struct Outer<T>(Inner<T>); struct Inner<T>(T); impl<T> Unpin for Inner<T> where T: const std::default::Default, {} ``` Fixes #149281

lcnr · 2026-07-06T12:03:45Z

good stuff, sgtm after fixing CI

rustbot · 2026-07-06T13:09:53Z

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.

lcnr · 2026-07-06T14:38:37Z

+        let errors = ocx.evaluate_obligations_error_on_ambiguity();
+        if errors.iter().any(|error| error.is_true_error()) {
+            return AutoTraitResult::NegativeImpl;
+        }


wait, that's just try_evaluate_obligations and then errors.is_empty() 😅

View changes since the review

I've tidied up the commits and made that change

lcnr · 2026-07-06T19:07:38Z

@bors r+ rollup

rust-bors · 2026-07-06T19:07:41Z

📌 Commit 84aa27d has been approved by lcnr

It is now in the queue for this repository.

🌲 The tree is currently closed for pull requests below priority 6. This pull request will be tested once the tree is reopened.

@lcnr

…r=lcnr Prevent rustdoc auto-trait ICEs with `-Znext-solver=globally` - Ensure rustdoc with `-Znext-solver=globally` does not ICE. r? @lcnr

Rollup of 13 pull requests Successful merges: - #158478 (Use correct typing mode in mir building for the next solver) - #158796 (Distinguish null reference production from null pointer dereference in UB checks) - #158821 (add regression test for late-bound type method probe ICE) - #158245 (Update the rustc_perf submodule) - #158346 (Prevent rustdoc auto-trait ICEs with `-Znext-solver=globally`) - #158536 (Instatiate binder instead of skipping when suggesting function arg error) - #158553 (Avoid infinite recursion when trying to build valtrees for self-referential consts) - #158679 (Document blocking guarantees for `std::sync`) - #158826 (Reorganize `tests/ui/issues` [19/N]) - #158860 (Delete `impl_def_id` field from `ImplHeader`) - #158867 (rewrite `Align::max_for_target` in a more obvious way) - #158878 (borrowck: Inline `free_region_constraint_info()` to simplify the code) - #158881 (Update mdbook to 0.5.4)

@lcnr

Rollup merge of #158346 - Jamesbarford:fix/auto-trait-impl, r=lcnr Prevent rustdoc auto-trait ICEs with `-Znext-solver=globally` - Ensure rustdoc with `-Znext-solver=globally` does not ICE. r? @lcnr

rustbot assigned lcnr Jun 24, 2026

rustbot added 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. T-rustdoc Relevant to the rustdoc team, which will review and decide on the PR/issue. labels Jun 24, 2026

Jamesbarford force-pushed the fix/auto-trait-impl branch from 8806252 to 52e368f Compare June 25, 2026 07:38

Jamesbarford mentioned this pull request Jun 25, 2026

Fix: auto trait, const trait bound #158390

Merged

lcnr reviewed Jun 25, 2026

View reviewed changes

rustbot added the T-rustdoc-frontend Relevant to the rustdoc-frontend team, which will review and decide on the web UI/UX output. label Jul 6, 2026

This comment has been minimized.

Sign in to view

Jamesbarford force-pushed the fix/auto-trait-impl branch from 81e10c1 to 5592d3f Compare July 6, 2026 13:09

lcnr reviewed Jul 6, 2026

View reviewed changes

Jamesbarford added 2 commits July 6, 2026 16:09

Create simple non ICE-ing path for new trait solver

2740e11

Tests for find_auto_trait_generics_next_solver path

84aa27d

Jamesbarford force-pushed the fix/auto-trait-impl branch from 5592d3f to 84aa27d Compare July 6, 2026 16:02

Jamesbarford changed the title ~~Avoid rustdoc auto-trait ICEs with the global next solver~~ Prevent rustdoc auto-trait ICEs with -Znext-solver=globally Jul 6, 2026

rust-bors Bot added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Jul 6, 2026

JonathanBrouwer mentioned this pull request Jul 6, 2026

Rollup of 7 pull requests #158884

Closed

jhpratt mentioned this pull request Jul 7, 2026

Rollup of 13 pull requests #158891

Merged

rust-bors Bot merged commit 1c4fe1a into rust-lang:main Jul 7, 2026
13 checks passed

rustbot added this to the 1.99.0 milestone Jul 7, 2026

Jamesbarford deleted the fix/auto-trait-impl branch July 7, 2026 11:42

	let generic_args = ty::GenericArgs::for_item(self.tcx, m.def_id, \|param, _\| {
	self.var_for_def(deref.span, param)
	});

	/// Given a set of generics defined on a type or impl, returns the generic parameters mapping
	/// each type/region parameter to a fresh inference variable.
	pub fn fresh_args_for_item(&self, span: Span, def_id: DefId) -> GenericArgsRef<'tcx> {
	GenericArgs::for_item(self.tcx, def_id, \|param, _\| self.var_for_def(span, param))
	}

Uh oh!

Uh oh!

Conversation

Jamesbarford commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lcnr Jun 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!

Jamesbarford Jul 1, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr Jun 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!

lcnr Jun 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!

lcnr Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

Jamesbarford Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

Jamesbarford Jul 6, 2026

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

lcnr commented Jul 6, 2026

Uh oh!

This comment has been minimized.

rustbot commented Jul 6, 2026

Uh oh!

lcnr Jul 6, 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!

Jamesbarford Jul 6, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr commented Jul 6, 2026

Uh oh!

rust-bors Bot commented Jul 6, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Jamesbarford commented Jun 24, 2026 •

edited

Loading

lcnr Jun 25, 2026 •

edited by rustbot

Loading

lcnr Jun 25, 2026 •

edited by rustbot

Loading

lcnr Jun 25, 2026 •

edited by rustbot

Loading

lcnr Jul 6, 2026 •

edited by rustbot

Loading