Replace infers and non-rigid aliases with `Ty/Const::Error` if param env normalization fails by adwinwhite · Pull Request #160856 · rust-lang/rust

adwinwhite · 2026-08-10T12:48:28Z

Previously we use unnormalized param env if we fail to normalize it. This causes problem for places which expect normalized param env, like in lexical region solving.

I also sneak in a one-line change that we resolve regions with normalized param env in param env normalization now.
This fixes ICEs in polkadot and zerocopy which had non-rigid const aliases in type outlive env before. I can split it into another PR if that's preferred.

r? lcnr

adwinwhite · 2026-08-10T12:49:07Z

@bors try @rust-timer queue

Replace infers and non-rigid aliases with `Ty/Const::Error` if param env normalization fails

rust-bors · 2026-08-10T13:47:43Z

☀️ Try build successful (CI)
Build commit: 63c3ea0 (63c3ea08d1963147f00c41c604e746c02103f144)
Base parent: 7088e4b (7088e4b63a9516ebfbfe2ab2d999cf01a528ac14)

rust-timer · 2026-08-10T14:30:05Z

Finished benchmarking commit (63c3ea0): comparison URL.

Overall result: no relevant changes - 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

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

Max RSS (memory usage)

Results (primary 1.4%, secondary 0.3%)

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

	mean	range	count
Regressions ❌ (primary)	1.4%	[1.4%, 1.4%]	1
Regressions ❌ (secondary)	2.5%	[0.5%, 6.1%]	3
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.6%	[-0.8%, -0.4%]	8
All ❌✅ (primary)	1.4%	[1.4%, 1.4%]	1

Cycles

Results (secondary 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)	0.8%	[0.4%, 1.5%]	10
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.8%	[-1.6%, -0.4%]	7
All ❌✅ (primary)	-	-	0

Binary size

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

Bootstrap: 455.179s -> 456.527s (0.30%)
Artifact size: 398.37 MiB -> 398.42 MiB (0.01%)

lcnr · 2026-08-10T15:15:15Z

+                        _ => ty,
+                    }
+                },
+                lt_op: |lt| lt,


we have to deal with infer regions here, don't we? 🤔

/ does fully_resolve even make sense as an abstraction? it's a version of resolve_vars_if_possible which also looks at the region graph?

View changes since the review

We already replace unresolved region infers with Region::Error in fully_resolve.

it's a version of resolve_vars_if_possible which also looks at the region graph?

Yap, another difference is that it eagerly returns with error.

doesn't that mean this incorrectly handles having both unresolved regions and an unresolved ty var? because it keeps the unresolved region

unsure when exactly fully_resolve errors here

also, can you move these folders into a separate replace_infer_and_non_rigid_alias_with_error function or sth 🤔

doesn't that mean this incorrectly handles having both unresolved regions and an unresolved ty var? because it keeps the unresolved region

What do you mean? fully_resolve itself would replace unresolved region and return Ok with delayed error.

unsure when exactly fully_resolve errors here

Yeah, it's not obvious.
It errors when the clauses contain ty vars or const vars which would happen if we don't have normalization failure but have ill-formed impls like impl<T> Trait for () where the generic parameter is unused. I have a comment about this above but it doesn't mention what the impl would look like.
See tests/ui/traits/normalize/self-referential-param-env-normalization.rs

also, can you move these folders into a separate replace_infer_and_non_rigid_alias_with_error function or sth 🤔

The two folders are slightly different. Unsure we should fold non-rigid aliases to error if we don't have normalization failure. Those non-rigid aliases can come from opaques. Maybe we can still do that since we're on a failed code path anyway. Could cause some spurious failures? 🤔

What do you mean? fully_resolve itself would replace unresolved region and return Ok with delayed error.

if fully_resolve has both unresolved regions (which it replaced) and an unresolved type (which causes it to return an Err), then wouldn't we fold over the type without the unresolved region replaced?

lcnr

one slightly confusing thing, otherwise r=me

View changes since this review

rustbot · 2026-08-11T04:44:29Z

This PR changes a file inside tests/crashes. If a crash was fixed, please move into the corresponding ui subdir and add 'Fixes #' to the PR description to autoclose the issue upon merge.

lcnr · 2026-08-12T10:02:25Z

    })
 }

+enum ReplaceRegion {


Suggested change

enum ReplaceRegion {

enum ReplaceRegions {

View changes since the review

lcnr · 2026-08-12T10:02:59Z

+                // We can't resolve regions using lexical resolution here since that's
+                // private. It probably doesn't matter since we already got more severe error.


vibe, want to move this above the ReplaceRegions::Yes as that should make the formatting slightly nicer here

View changes since the review

lcnr · 2026-08-12T10:04:00Z

+    // FIXME: We should avoid interning clauses both here and at the caller sites. We should also
+    // avoid cloning if possible.


Suggested change

// FIXME: We should avoid interning clauses both here and at the caller sites. We should also

// avoid cloning if possible.

// FIXME: We should avoid interning clauses both here and at the

// caller sites. We should also avoid cloning if possible.

minor formatting nit, the lines are very long here in general :<

View changes since the review

lcnr

r=me after style nits

View changes since this review

adwinwhite · 2026-08-13T03:52:32Z

@bors r=lcnr

rust-bors · 2026-08-13T03:52:34Z

📌 Commit 51ef2f2 has been approved by lcnr

It is now in the queue for this repository.

Rollup of 8 pull requests Successful merges: - #159593 (merge ambiguity errors that blame the same inference variable) - #160687 (Experiment: Add `core::cmp::smallest` and `core::cmp::largest`) - #160856 (Replace infers and non-rigid aliases with `Ty/Const::Error` if param env normalization fails) - #160961 (bootstrap: Overhaul matching of command-line selectors to steps) - #160975 (Remove target argument from get_proc_macros) - #161023 (bootstrap: Replace the `exit!` macro with a function `helpers::exit_process`) - #160932 (Make tidy::Version public) - #161029 (mailmap: Update my default email)

Rollup merge of #160856 - adwinwhite:param-env-norm-fail-with-err, r=lcnr Replace infers and non-rigid aliases with `Ty/Const::Error` if param env normalization fails Fixes #160196 Fixes #148630 Previously we use unnormalized param env if we fail to normalize it. This causes problem for places which expect normalized param env, like in lexical region solving. I also sneak in a one-line change that we resolve regions with normalized param env in param env normalization now. This fixes ICEs in `polkadot` and `zerocopy` which had non-rigid const aliases in type outlive env before. I can split it into another PR if that's preferred. r? lcnr

replace infers and non-rigid aliases with `Ty/Const::Error`

rustbot assigned lcnr Aug 10, 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. WG-trait-system-refactor The Rustc Trait System Refactor Initiative (-Znext-solver) labels Aug 10, 2026