Compute the result of a projection type with region errors by makai410 · Pull Request #153105 · rust-lang/rust

makai410 · 2026-02-25T20:10:24Z

With the old trait solver, type_known_to_meet_bound_modulo_regions() isn't really operating "modulo regions" if there are any region errors, since normalize will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions.

So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation.

That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

rustbot · 2026-02-25T20:10:29Z

r? @petrochenkov

rustbot has assigned @petrochenkov.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

Owners of files modified in this PR: compiler
compiler expanded to 68 candidates
Random selection from 14 candidates

makai410 · 2026-02-25T20:16:57Z

tests/ui/consts/const-eval/ice-unsized-struct-const-eval-123154.stderr

+error[E0277]: the size for values of type `[&usize]` cannot be known at compilation time
+  --> $DIR/ice-unsized-struct-const-eval-123154.rs:14:1
+   |
+LL | static ST: AA = AA::new();
+   | ^^^^^^^^^^^^^ doesn't have a size known at compile-time
+   |
+   = help: within `AA`, the trait `Sized` is not implemented for `[&usize]`
+note: required because it appears within the type `AA`
+  --> $DIR/ice-unsized-struct-const-eval-123154.rs:3:8
+   |
+LL | struct AA {
+   |        ^^
+   = note: statics and constants must have a statically known size
+
+error[E0277]: the size for values of type `[&usize]` cannot be known at compilation time
+  --> $DIR/ice-unsized-struct-const-eval-123154.rs:9:23
+   |
+LL |     const fn new() -> Self { }
+   |                       ^^^^ doesn't have a size known at compile-time
+   |
+   = help: within `AA`, the trait `Sized` is not implemented for `[&usize]`
+note: required because it appears within the type `AA`
+  --> $DIR/ice-unsized-struct-const-eval-123154.rs:3:8
+   |
+LL | struct AA {
+   |        ^^
+   = note: the return type of a function must have a statically known size
+


Anyway these errors will be triggered if the code don't have the lifetime error, so

petrochenkov · 2026-02-26T14:13:44Z

r? types

lcnr

so what's happening here is that type_is_sized_modulo_regions encounters a field of type ty::Error as normalizing <&'re_error [fn()] as core::ops::Deref>::Target results in a type error,

but then during const eval we normalize differently and don't get a type error as the output?

I feel like erasing error regions this way seems... unfortunate and maybe we should instead change the trait solver/type system to check for HAS_NON_REGION_ERROR when creating type errors 🤔

View changes since this review

makai410 · 2026-02-27T17:00:34Z

but then during const eval we normalize differently and don't get a type error as the output?

Yeah since they get the ty using normalize_erasing_regions which wont return a type error in this case.

makai410 · 2026-02-27T17:06:23Z

I feel like erasing error regions this way seems... unfortunate and maybe we should instead change the trait solver/type system to check for HAS_NON_REGION_ERROR when creating type errors 🤔

cool I'm going to try this to see if it would cause any regressions or something.

makai410 · 2026-03-08T17:49:30Z

@rustbot ready

lcnr · 2026-03-11T16:34:39Z

compiler/rustc_type_ir/src/visit.rs

@@ -352,6 +352,11 @@ pub trait TypeVisitableExt<I: Interner>: TypeVisitable<I> {
    fn still_further_specializable(&self) -> bool {
        self.has_type_flags(TypeFlags::STILL_FURTHER_SPECIALIZABLE)
    }
+
+    /// True if there is no region error
+    fn has_non_region_error(&self) -> bool {


that name is very confusing I would expect that to be TypeFlags::HAS_TY_OR_CONST_ERROR same as has_non_region_infer

right now this is "has_no_region_error", which also still breaks if you have both a region error and a proper type error

I split TypeFlags::HAS_ERROR into TypeFlags::HAS_TY_OR_CONST_ERROR and TypeFlags::HAS_RE_ERROR, which should cover the case.

lcnr

nits, then r=me

View changes since this review

lcnr · 2026-03-30T11:52:54Z

compiler/rustc_trait_selection/src/traits/project.rs

-    if let Err(guar) = obligation.predicate.error_reported() {
+    // We can still compute a projection type when there are only region errors,
+    // but type/const errors require early return.
+    if obligation.predicate.has_type_or_const_error()


we tend to use has_non_region_X instead of type_or_const

lcnr · 2026-03-30T11:53:21Z

compiler/rustc_trait_selection/src/traits/project.rs

+    // We can still compute a projection type when there are only region errors,
+    // but type/const errors require early return.
+    if obligation.predicate.has_type_or_const_error()
+        && let Err(guar) = obligation.predicate.error_reported()


could just be Err(guar) = obligation.predicate.non_region_error_reported()

lcnr · 2026-03-30T11:53:39Z

compiler/rustc_type_ir/src/flags.rs

@@ -91,27 +91,32 @@ bitflags::bitflags! {
                                          | TypeFlags::HAS_TY_INHERENT.bits()
                                          | TypeFlags::HAS_CT_PROJECTION.bits();

+        /// Is a type or const error reachable?
+        const HAS_TY_OR_CT_ERROR          = 1 << 15;


Suggested change

const HAS_TY_OR_CT_ERROR = 1 << 15;

const HAS_NON_REGION_ERROR = 1 << 15;

rustbot · 2026-04-01T08:44:01Z

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.

makai410 · 2026-04-01T10:19:34Z

@bors r=lcnr

rust-bors · 2026-04-01T10:19:37Z

📌 Commit 339fb64 has been approved by lcnr

It is now in the queue for this repository.

Compute the result of a projection type with region errors Fixes: rust-lang#152682 With the old trait solver, `type_known_to_meet_bound_modulo_regions()` isn't really operating "modulo regions" if there are any region errors, since `normalize` will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions. So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation. That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

Rollup of 7 pull requests Successful merges: - #153105 (Compute the result of a projection type with region errors) - #153960 (Make `layout_of` cycles fatal errors) - #154666 (Remove `StableHashContext` impls) - #154669 (Introduce #[diagnostic::on_move] on `Arc`) - #154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - #154660 (Avoid creating async return opaques for foreign async fns) - #154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible)

Compute the result of a projection type with region errors Fixes: rust-lang#152682 With the old trait solver, `type_known_to_meet_bound_modulo_regions()` isn't really operating "modulo regions" if there are any region errors, since `normalize` will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions. So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation. That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

Rollup of 11 pull requests Successful merges: - #153105 (Compute the result of a projection type with region errors) - #153286 (various fixes for scalable vectors) - #153960 (Make `layout_of` cycles fatal errors) - #154527 (Emit pre-expansion feature gate warnings for negative impls and specialization) - #154666 (Remove `StableHashContext` impls) - #154669 (Introduce #[diagnostic::on_move] on `Arc`) - #154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - #154644 (rustdoc: seperate methods and associated functions in sidebar) - #154660 (Avoid creating async return opaques for foreign async fns) - #154671 (Add a test for a past ICE when calling a const fn of an unresolved type with the wrong number of args) - #154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible)

Compute the result of a projection type with region errors Fixes: rust-lang#152682 With the old trait solver, `type_known_to_meet_bound_modulo_regions()` isn't really operating "modulo regions" if there are any region errors, since `normalize` will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions. So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation. That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

Rollup of 11 pull requests Successful merges: - #153105 (Compute the result of a projection type with region errors) - #153960 (Make `layout_of` cycles fatal errors) - #154527 (Emit pre-expansion feature gate warnings for negative impls and specialization) - #154666 (Remove `StableHashContext` impls) - #154669 (Introduce #[diagnostic::on_move] on `Arc`) - #154213 (tidy-alphabetical: fix line number in error message) - #154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - #154644 (rustdoc: seperate methods and associated functions in sidebar) - #154660 (Avoid creating async return opaques for foreign async fns) - #154671 (Add a test for a past ICE when calling a const fn of an unresolved type with the wrong number of args) - #154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible)

Compute the result of a projection type with region errors Fixes: rust-lang#152682 With the old trait solver, `type_known_to_meet_bound_modulo_regions()` isn't really operating "modulo regions" if there are any region errors, since `normalize` will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions. So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation. That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

…uwer Rollup of 21 pull requests Successful merges: - #153105 (Compute the result of a projection type with region errors) - #153286 (various fixes for scalable vectors) - #153532 (Attributes containing rustc) - #153960 (Make `layout_of` cycles fatal errors) - #154527 (Emit pre-expansion feature gate warnings for negative impls and specialization) - #154666 (Remove `StableHashContext` impls) - #154669 (Introduce #[diagnostic::on_move] on `Arc`) - #154710 (opaque_generic_const_args -> generic_const_args) - #154712 (Revert "`-Znext-solver` Remove the forced ambiguity hack from search graph") - #154713 (Stop compiling when we get resolving crate failure) - #154213 (tidy-alphabetical: fix line number in error message) - #154425 (Migrate transmute tests) - #154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - #154469 (mGCA: Lower spans for literal const args) - #154578 (Rename `probe_ty_var` to `try_resolve_ty_var`) - #154615 (Moving issues) - #154644 (rustdoc: seperate methods and associated functions in sidebar) - #154660 (Avoid creating async return opaques for foreign async fns) - #154671 (Add a test for a past ICE when calling a const fn of an unresolved type with the wrong number of args) - #154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible) - #154709 (Revert `Ty` type alias in `rustc_type_ir`)

Compute the result of a projection type with region errors Fixes: rust-lang#152682 With the old trait solver, `type_known_to_meet_bound_modulo_regions()` isn't really operating "modulo regions" if there are any region errors, since `normalize` will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions. So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation. That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

…uwer Rollup of 21 pull requests Successful merges: - #153105 (Compute the result of a projection type with region errors) - #153286 (various fixes for scalable vectors) - #153532 (Attributes containing rustc) - #153960 (Make `layout_of` cycles fatal errors) - #154527 (Emit pre-expansion feature gate warnings for negative impls and specialization) - #154666 (Remove `StableHashContext` impls) - #154669 (Introduce #[diagnostic::on_move] on `Arc`) - #154710 (opaque_generic_const_args -> generic_const_args) - #154712 (Revert "`-Znext-solver` Remove the forced ambiguity hack from search graph") - #153614 (`FindParamInClause` handle edge-cases) - #154213 (tidy-alphabetical: fix line number in error message) - #154425 (Migrate transmute tests) - #154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - #154469 (mGCA: Lower spans for literal const args) - #154578 (Rename `probe_ty_var` to `try_resolve_ty_var`) - #154615 (Moving issues) - #154644 (rustdoc: seperate methods and associated functions in sidebar) - #154660 (Avoid creating async return opaques for foreign async fns) - #154671 (Add a test for a past ICE when calling a const fn of an unresolved type with the wrong number of args) - #154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible) - #154709 (Revert `Ty` type alias in `rustc_type_ir`)

…uwer Rollup of 20 pull requests Successful merges: - #153105 (Compute the result of a projection type with region errors) - #153532 (Attributes containing rustc) - #153960 (Make `layout_of` cycles fatal errors) - #154527 (Emit pre-expansion feature gate warnings for negative impls and specialization) - #154666 (Remove `StableHashContext` impls) - #154669 (Introduce #[diagnostic::on_move] on `Arc`) - #154710 (opaque_generic_const_args -> generic_const_args) - #154712 (Revert "`-Znext-solver` Remove the forced ambiguity hack from search graph") - #153614 (`FindParamInClause` handle edge-cases) - #154213 (tidy-alphabetical: fix line number in error message) - #154425 (Migrate transmute tests) - #154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - #154469 (mGCA: Lower spans for literal const args) - #154578 (Rename `probe_ty_var` to `try_resolve_ty_var`) - #154615 (Moving issues) - #154644 (rustdoc: seperate methods and associated functions in sidebar) - #154660 (Avoid creating async return opaques for foreign async fns) - #154671 (Add a test for a past ICE when calling a const fn of an unresolved type with the wrong number of args) - #154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible) - #154709 (Revert `Ty` type alias in `rustc_type_ir`)

Rollup merge of #153105 - makai410:erase-if-error, r=lcnr Compute the result of a projection type with region errors Fixes: #152682 With the old trait solver, `type_known_to_meet_bound_modulo_regions()` isn't really operating "modulo regions" if there are any region errors, since `normalize` will just return a type error to the trait solver if given a ty with a region error, which then starts cascading when there are so many assumptions. So I think it would be good to erase regions if there are any region errors before we normalize the type when collecting predicates for confirmation. That said, I somehow feel like this is kind of ad-hoc... I'd really appreciate if someone more familiar with this code could take a closer look :3

…uwer Rollup of 20 pull requests Successful merges: - rust-lang/rust#153105 (Compute the result of a projection type with region errors) - rust-lang/rust#153532 (Attributes containing rustc) - rust-lang/rust#153960 (Make `layout_of` cycles fatal errors) - rust-lang/rust#154527 (Emit pre-expansion feature gate warnings for negative impls and specialization) - rust-lang/rust#154666 (Remove `StableHashContext` impls) - rust-lang/rust#154669 (Introduce #[diagnostic::on_move] on `Arc`) - rust-lang/rust#154710 (opaque_generic_const_args -> generic_const_args) - rust-lang/rust#154712 (Revert "`-Znext-solver` Remove the forced ambiguity hack from search graph") - rust-lang/rust#153614 (`FindParamInClause` handle edge-cases) - rust-lang/rust#154213 (tidy-alphabetical: fix line number in error message) - rust-lang/rust#154425 (Migrate transmute tests) - rust-lang/rust#154442 (Export `derive` at the crate root: `core::derive` and `std::derive`) - rust-lang/rust#154469 (mGCA: Lower spans for literal const args) - rust-lang/rust#154578 (Rename `probe_ty_var` to `try_resolve_ty_var`) - rust-lang/rust#154615 (Moving issues) - rust-lang/rust#154644 (rustdoc: seperate methods and associated functions in sidebar) - rust-lang/rust#154660 (Avoid creating async return opaques for foreign async fns) - rust-lang/rust#154671 (Add a test for a past ICE when calling a const fn of an unresolved type with the wrong number of args) - rust-lang/rust#154680 ([rustdoc] Replace `DocContext` with `TyCtxt` wherever possible) - rust-lang/rust#154709 (Revert `Ty` type alias in `rustc_type_ir`)

rustbot assigned petrochenkov Feb 25, 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. labels Feb 25, 2026

makai410 commented Feb 25, 2026

View reviewed changes

rustbot added the T-types Relevant to the types team, which will review and decide on the PR/issue. label Feb 26, 2026

rustbot assigned lcnr and unassigned petrochenkov Feb 26, 2026

lcnr reviewed Feb 27, 2026

View reviewed changes

makai410 force-pushed the erase-if-error branch from 260d016 to 94c3811 Compare March 8, 2026 17:36

makai410 changed the title ~~Erase regions if there are any region errors when collecting predicates~~ Compute the result of a projection type with region errors Mar 8, 2026

lcnr reviewed Mar 11, 2026

View reviewed changes

makai410 force-pushed the erase-if-error branch from 94c3811 to 7840805 Compare March 16, 2026 14:15

lcnr reviewed Mar 30, 2026

View reviewed changes

makai410 force-pushed the erase-if-error branch from 7840805 to c499ec1 Compare April 1, 2026 08:34

This comment has been minimized.

Sign in to view

makai410 added 2 commits April 1, 2026 16:40

add TypeFlags::HAS_NON_REGION_ERROR and TypeFlags::HAS_RE_ERROR

e18dd4a

skip early return for region-only errors in projection types

339fb64

makai410 force-pushed the erase-if-error branch from c499ec1 to 339fb64 Compare April 1, 2026 08:43

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 Apr 1, 2026

JonathanBrouwer mentioned this pull request Apr 1, 2026

Rollup of 6 pull requests #154688

Closed

GuillaumeGomez mentioned this pull request Apr 1, 2026

Rollup of 7 pull requests #154693

Closed

jhpratt mentioned this pull request Apr 2, 2026

Rollup of 11 pull requests #154701

Closed

jhpratt mentioned this pull request Apr 2, 2026

Rollup of 11 pull requests #154705

Closed

JonathanBrouwer mentioned this pull request Apr 2, 2026

Rollup of 21 pull requests #154716

Closed

JonathanBrouwer mentioned this pull request Apr 2, 2026

Rollup of 21 pull requests #154718

Closed

JonathanBrouwer mentioned this pull request Apr 2, 2026

Rollup of 20 pull requests #154727

Merged

rust-bors bot merged commit 828e302 into rust-lang:main Apr 2, 2026
11 checks passed

rustbot added this to the 1.96.0 milestone Apr 2, 2026

	const HAS_TY_OR_CT_ERROR = 1 << 15;
	const HAS_NON_REGION_ERROR = 1 << 15;

Uh oh!

Conversation

makai410 commented Feb 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Feb 25, 2026

Uh oh!

makai410 Feb 25, 2026

Choose a reason for hiding this comment

Uh oh!

petrochenkov commented Feb 26, 2026

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!

makai410 commented Feb 27, 2026

Uh oh!

makai410 commented Feb 27, 2026

Uh oh!

makai410 commented Mar 8, 2026

Uh oh!

lcnr Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

makai410 Mar 16, 2026

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!

lcnr Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

rustbot commented Apr 1, 2026

Uh oh!

makai410 commented Apr 1, 2026

Uh oh!

rust-bors bot commented Apr 1, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

makai410 commented Feb 25, 2026 •

edited

Loading

lcnr left a comment •

edited by rustbot

Loading

lcnr left a comment •

edited by rustbot

Loading