mgca: Register `ConstArgHasType` when normalizing projection consts by lapla-cogito · Pull Request #154853 · rust-lang/rust

lapla-cogito · 2026-04-05T14:52:39Z

Fixes #152962
Fixes #154748
Fixes #154750

When running CTFE on a MIR body, normalizing a type const within that body can change the type of the resulting value, causing the MIR to become ill-formed. Since no prior error has been reported at that point, MIR validation fires a span_bug!. The existing ConstArgHasType check in wfcheck::check_type_const does catch this at the definition site, but due to query evaluation ordering, the normalization path can reach MIR validation before that check has run.

Fix this by registering a ConstArgHasType(ct, expected_ty) obligation/goal when normalizing projection consts (both trait and inherent), in both the old and new trait solvers. This ensures the type mismatch is reported as an error during normalization itself, which taints the MIR before validation runs.

The first commit fixes the original case reported in the issue. The second commit fixes a different ICE pattern reported within the issue (see #152962 (comment)).

r? BoxyUwU

rustbot · 2026-04-05T14:52:45Z

BoxyUwU is currently at their maximum review capacity.
They may take a while to respond.

lapla-cogito · 2026-04-05T15:04:01Z

        assoc_term_own_obligations(selcx, obligation, &mut nested);
-        Progress { term: term.instantiate(tcx, args), obligations: nested }
+        let instantiated_term: Term<'tcx> = term.instantiate(tcx, args);
+        if let Some(ct) = instantiated_term.as_const() {


Boxy and I discussed the fix approach and settled on registering a nested goal when normalizing type consts to verify that the normalized term has the type of the type const item. However, this implementation applies the same nested goal to all associated consts, not just type consts. This better expresses the invariant that the normalized result of a const should match its declared type (I think only type consts currently go through this path, but this also guards against future changes).

BoxyUwU · 2026-04-07T11:56:19Z

can you add an equivalent test for free type consts, e.g:

type const N: usize = "this isn't a usize";
fn f() -> [u8; const { N }] {}

actuall;y it seems like this doesn't even ICE on nightly rn 🤔 can you look into why this doesn't ICE but the other stuff does. would like to properly understand why we need this for projections but not free type consts.

When a type const has a value whose type doesn't match the declared type, normalizing a reference to it can trigger CTFE, which in turn runs MIR validation on the const body

the type const has no associated const body. so i think what's happening here is that we normalize everything in a MIR body and then do MIR validation on it afterwards, rather than normalization causing validation to happen 🤔

lapla-cogito · 2026-04-07T13:58:13Z

can you add an equivalent test for free type consts

actuall;y it seems like this doesn't even ICE on nightly rn 🤔

Ah, I didn't consider that. Indeed, in such cases, it seems ICE doesn't occur.

IIUC, the key is the ordering of check_type_wf(). For free type consts, check_type_const() runs during par_items(), which catches the type mismatch and emits an error. CTFE of inline consts like const { N } happens later (during MIR_borrow_checking phase, after check_crate completes), so has_errors() is already Some by then and MIR validation suppresses the span_bug!.

For trait/inherent impl type consts, the impl block's check_well_formed() runs during par_items() and can trigger CTFE via compare_impl_item (trait) or signature normalization (inherent). But check_type_const() for the associated item doesn't run until par_impl_items(), which hasn't started yet. So no error has been reported when MIR validation fires. Therefore, I added a test to verify that ICE doesn't occur for free type consts cases as well, thanks.

BoxyUwU · 2026-04-14T15:17:07Z

IIUC, the key is the ordering of check_type_wf().

oooh okay I think I get it. this test case relies on evaluating the anon const during wfcheck of free items and results in an ICE. does this PR fix this variant: ?

//@ compile-flags: -Zvalidate-mir
#![feature(min_generic_const_args)]

type const X: usize = const { N };
type const N: usize = "this isn't a usize";

BoxyUwU · 2026-04-14T15:25:31Z

wrt your PR description it's a bit wrong right now:

When a type const has a value whose type doesn't match the declared type, normalizing a reference to it can trigger CTFE, which in turn runs MIR validation on the const body. The MIR body contains a type mismatch, and since no prior error has been reported at that point, MIR validation fires a span_bug!.

it's not that normalizing a usage of the type const triggers CTFE. rather that when running CTFE on a MIR body we may wind up normalizing a type const in the body, which can change the type of the value causing the MIR to become illformed.

can you update the description

BoxyUwU · 2026-04-14T15:29:47Z

        tcx.const_of_item(alias_term.def_id).instantiate(tcx, args).into()
    };

+    if let Some(ct) = term.as_const() {


View changes since the review

same for the old solver's normalization routine

BoxyUwU · 2026-04-14T15:30:05Z

thanks for working on this ✨

BoxyUwU · 2026-04-14T15:30:45Z

@rustbot author

rustbot · 2026-04-14T15:30:51Z

Reminder, once the PR becomes ready for a review, use @rustbot ready.

lapla-cogito · 2026-04-16T11:35:38Z

IIUC, the key is the ordering of check_type_wf().

oooh okay I think I get it. this test case relies on evaluating the anon const during wfcheck of free items and results in an ICE. does this PR fix this variant: ?
//@ compile-flags: -Zvalidate-mir
#![feature(min_generic_const_args)]

type const X: usize = const { N };
type const N: usize = "this isn't a usize";

I tested this variant and adding ConstArgHasType to normalize_free_alias does not fix it. The obligation is registered in the outer wfcheck context, but the ICE happens inside the AnonConst's CTFE -> MIR validation, where has_errors() hasn't been updated yet. This would need a different approach to fix, so I think it's out of scope for this PR 🤔

edit: Of course, even if this PR doesn't address such cases, it would be possible for me to handle them later.

lapla-cogito · 2026-04-16T11:55:51Z

@rustbot ready

…to_term`

lapla-cogito · 2026-06-07T23:38:17Z

@rustbot ready

reddevilmidzy

Could you also add close: #154748, close: #154750 to the PR description?

View changes since this review

BoxyUwU · 2026-06-11T00:48:02Z

@bors r+

thanks ashley & reddy for the reviews :3 thanks lapla for working on this for so long despite my inconsistent and long review times 😅

rust-bors · 2026-06-11T00:48:05Z

📌 Commit 277dccc has been approved by BoxyUwU

It is now in the queue for this repository.

…const, r=BoxyUwU mgca: Register `ConstArgHasType` when normalizing projection consts Fixes rust-lang#152962 Fixes rust-lang#154748 Fixes rust-lang#154750 When running CTFE on a MIR body, normalizing a type const within that body can change the type of the resulting value, causing the MIR to become ill-formed. Since no prior error has been reported at that point, MIR validation fires a `span_bug!`. The existing `ConstArgHasType` check in `wfcheck::check_type_const` does catch this at the definition site, but due to query evaluation ordering, the normalization path can reach MIR validation before that check has run. Fix this by registering a `ConstArgHasType(ct, expected_ty)` obligation/goal when normalizing projection consts (both trait and inherent), in both the old and new trait solvers. This ensures the type mismatch is reported as an error during normalization itself, which taints the MIR before validation runs. The first commit fixes the original case reported in the issue. The second commit fixes a different ICE pattern reported within the issue (see rust-lang#152962 (comment)). r? BoxyUwU

Rollup of 31 pull requests Successful merges: - #141030 (Expand free alias types during variance computation) - #154853 (mgca: Register `ConstArgHasType` when normalizing projection consts) - #155527 (Replace printables table with `unicode_data.rs` tables) - #156629 (Stabilize `core::range::{legacy, RangeFull, RangeTo}`) - #157280 (traits: Allow escaping self types in ExistentialTraitRef::with_self_ty) - #157282 (Fix post-monomorphization error note race in the parallel frontend) - #157352 (Make the retained dep graph deterministic under the parallel frontend) - #157601 (Emit error for unused target expression in glob and list delegations) - #157611 (Update `browser-ui-test` version to `0.24.0`) - #157620 (Add a strategy FnMut to inject behavior into the flush cycle) - #157645 (Windows TLS - Only register the `atexit` hook when `cleanup` can be unloaded) - #157646 (Keep rename-imported main alive in dead-code analysis under `--test`) - #157647 (Start using comptime for reflection intrinsics and their wrapper functions) - #157719 (resolve: Partially revert "Remove a special case for dummy imports") - #155153 (Ensure Send/Sync is not implemented for std::env::Vars{,Os}) - #155198 (fix(mgca): Allow specifying generic args (of enum) on enum itself in unit & tuple variant constructions in (direct) const args) - #155323 (refactor `TypeRelativePath::AssocItem` to use `AliasTerm`) - #156497 (fix-155516: Don't suggest wrong unwrap expect) - #156583 (Support defaults for static EIIs) - #157013 (Ensure inferred let pattern types are well-formed) - #157196 (Only suggest reborrowing a moved value where the reborrow is valid) - #157230 (borrowck: avoid ICE describing fields on generic params) - #157288 (platform support: add SNaN erratum to MIPS targets) - #157330 (remove `IsTypeConst` from `hir::TraitItemKind`) - #157350 (compiletest: ignore SVG `y` offset in by-lines comparison) - #157355 (Add `or_try_*` variants for `HashMap` and `BTreeMap` Entry APIs) - #157577 (Fix parser error recovery treating 'dyn' as a strict keyword in Rust 2015 when used in `dyn + dyn`) - #157670 (Rename `errors.rs` file to `diagnostics.rs` (4/N)) - #157691 (Move symbol hiding code to a new file) - #157700 (Rename `errors.rs` file to `diagnostics.rs` (5/N)) - #157703 (Fix doc link to Instant sub in saturating caveat) Failed merges: - #157699 (Arg splat experiment - hir FnDecl impl)

Rollup merge of #154853 - lapla-cogito:normlize_projecttion_const, r=BoxyUwU mgca: Register `ConstArgHasType` when normalizing projection consts Fixes #152962 Fixes #154748 Fixes #154750 When running CTFE on a MIR body, normalizing a type const within that body can change the type of the resulting value, causing the MIR to become ill-formed. Since no prior error has been reported at that point, MIR validation fires a `span_bug!`. The existing `ConstArgHasType` check in `wfcheck::check_type_const` does catch this at the definition site, but due to query evaluation ordering, the normalization path can reach MIR validation before that check has run. Fix this by registering a `ConstArgHasType(ct, expected_ty)` obligation/goal when normalizing projection consts (both trait and inherent), in both the old and new trait solvers. This ensures the type mismatch is reported as an error during normalization itself, which taints the MIR before validation runs. The first commit fixes the original case reported in the issue. The second commit fixes a different ICE pattern reported within the issue (see #152962 (comment)). r? BoxyUwU

Rollup of 31 pull requests Successful merges: - rust-lang#141030 (Expand free alias types during variance computation) - rust-lang#154853 (mgca: Register `ConstArgHasType` when normalizing projection consts) - rust-lang#155527 (Replace printables table with `unicode_data.rs` tables) - rust-lang#156629 (Stabilize `core::range::{legacy, RangeFull, RangeTo}`) - rust-lang#157280 (traits: Allow escaping self types in ExistentialTraitRef::with_self_ty) - rust-lang#157282 (Fix post-monomorphization error note race in the parallel frontend) - rust-lang#157352 (Make the retained dep graph deterministic under the parallel frontend) - rust-lang#157601 (Emit error for unused target expression in glob and list delegations) - rust-lang#157611 (Update `browser-ui-test` version to `0.24.0`) - rust-lang#157620 (Add a strategy FnMut to inject behavior into the flush cycle) - rust-lang#157645 (Windows TLS - Only register the `atexit` hook when `cleanup` can be unloaded) - rust-lang#157646 (Keep rename-imported main alive in dead-code analysis under `--test`) - rust-lang#157647 (Start using comptime for reflection intrinsics and their wrapper functions) - rust-lang#157719 (resolve: Partially revert "Remove a special case for dummy imports") - rust-lang#155153 (Ensure Send/Sync is not implemented for std::env::Vars{,Os}) - rust-lang#155198 (fix(mgca): Allow specifying generic args (of enum) on enum itself in unit & tuple variant constructions in (direct) const args) - rust-lang#155323 (refactor `TypeRelativePath::AssocItem` to use `AliasTerm`) - rust-lang#156497 (fix-155516: Don't suggest wrong unwrap expect) - rust-lang#156583 (Support defaults for static EIIs) - rust-lang#157013 (Ensure inferred let pattern types are well-formed) - rust-lang#157196 (Only suggest reborrowing a moved value where the reborrow is valid) - rust-lang#157230 (borrowck: avoid ICE describing fields on generic params) - rust-lang#157288 (platform support: add SNaN erratum to MIPS targets) - rust-lang#157330 (remove `IsTypeConst` from `hir::TraitItemKind`) - rust-lang#157350 (compiletest: ignore SVG `y` offset in by-lines comparison) - rust-lang#157355 (Add `or_try_*` variants for `HashMap` and `BTreeMap` Entry APIs) - rust-lang#157577 (Fix parser error recovery treating 'dyn' as a strict keyword in Rust 2015 when used in `dyn + dyn`) - rust-lang#157670 (Rename `errors.rs` file to `diagnostics.rs` (4/N)) - rust-lang#157691 (Move symbol hiding code to a new file) - rust-lang#157700 (Rename `errors.rs` file to `diagnostics.rs` (5/N)) - rust-lang#157703 (Fix doc link to Instant sub in saturating caveat) Failed merges: - rust-lang#157699 (Arg splat experiment - hir FnDecl impl)

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. WG-trait-system-refactor The Rustc Trait System Refactor Initiative (-Znext-solver) labels Apr 5, 2026

rustbot assigned BoxyUwU Apr 5, 2026