Unnormalized migration: assert_fully_normalized and struct_tail by khyperia · Pull Request #156224 · rust-lang/rust

khyperia · 2026-05-06T07:13:34Z

tracking issue: #155345 (first checkbox, and partial second checkbox, of that issue)

I'm going a bit slower than expected (less free time than I'd hope, lots of GCA work that I'm doing instead), and figured I'd just submit what I have now rather than building up a big batch of changes. Slow and steady!

r? @lcnr

rustbot · 2026-05-06T07:13:38Z

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

rustbot · 2026-05-06T07:35:18Z

The GCC codegen subtree was changed

cc @antoyo, @GuillaumeGomez

lcnr · 2026-05-06T07:25:31Z

        let mut normalize = |ty| self.normalize(ty, location);
        let tail = tcx.struct_tail_raw(ty, &cause, &mut normalize, || {});
-        normalize(tail)
+        normalize(Unnormalized::new_wip(tail))


that normalize call can be removed or struct_tail_raw has to return Unnormalized 🤔

View changes since the review

lcnr · 2026-05-06T07:26:46Z

@@ -461,7 +461,7 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
            let projected_ty = curr_projected_ty.projection_ty_core(


projection_ty_core should also have its normalization closure take Unnormalized<Ty>, unless u want to do that in a later PR

View changes since the review

lcnr · 2026-05-06T07:31:40Z

        // Need to renormalize `a` as typecheck may have failed to normalize
        // higher-ranked aliases if normalization was ambiguous due to inference.
-        let a = self.normalize(a, locations);
-        let ty = self.normalize(curr_projected_ty.ty, locations);
+        let a = self.normalize(ty::Unnormalized::new_wip(a), locations);
+        let ty = self.normalize(ty::Unnormalized::new_wip(curr_projected_ty.ty), locations);


Suggested change

// Need to renormalize `a` as typecheck may have failed to normalize

// higher-ranked aliases if normalization was ambiguous due to inference.

let a = self.normalize(a, locations);

let ty = self.normalize(curr_projected_ty.ty, locations);

let a = self.normalize(ty::Unnormalized::new_wip(a), locations);

let ty = self.normalize(ty::Unnormalized::new_wip(curr_projected_ty.ty), locations);

// Need to renormalize `a` in the old solver as typecheck may have failed

// to normalize higher-ranked aliases if normalization was ambiguous due

// to inference.

//

// We properly normalize higher-ranked aliases during writeback with the

// new solver, so this is no longer necessary.

let mut a = a;

let mut ty = curr_projected_ty.ty;

if !self.infcx.next_trait_solver() {

a = self.normalize(ty::Unnormalized::new_wip(a), locations);

ty = self.normalize(ty::Unnormalized::new_wip(curr_projected_ty.ty), locations);

}

View changes since the review

lcnr · 2026-05-06T07:35:27Z

        layout.ty,
        &ObligationCause::dummy(),
-        |ty| ty,
+        |ty| ty.skip_normalization(),


what are we doing here 😭 that'a broken, please add a "FIXME(#155345): Missing normalization call" here

View changes since the review

lcnr · 2026-05-06T07:37:02Z

                // Even if `ty` is normalized, the search for the unsized tail will project
                // to fields, which can yield non-normalized types. So we need to provide a
                // normalization function.
-                let normalize = |ty| {
-                    self.tcx.normalize_erasing_regions(self.typing_env, Unnormalized::new_wip(ty))
-                };
+                let normalize = |ty| self.tcx.normalize_erasing_regions(self.typing_env, ty);
                ty.ptr_metadata_ty(*self.tcx, normalize)


remove that comment and maybe inline the comment. I don't think that comment is useful as this holds for all uses of ptr_metadata_ty, so it should be on that function if anywhere

View changes since the review

"inline the closure" i mean

I added the comment because I can never remember when one has to normalize what and we had a bunch of bugs due to that.

lcnr · 2026-05-06T07:37:56Z

@@ -81,7 +81,7 @@ impl<'a, 'tcx> Expectation<'tcx> {
        let cause = ObligationCause::misc(span, fcx.body_id);

        // FIXME: This is not right, even in the old solver...


Suggested change

// FIXME: This is not right, even in the old solver...

// FIXME(#155345): Missing normalization call

View changes since the review

lcnr · 2026-05-06T07:40:21Z

            tcx.types.unit
        } else {
-            match pointee_ty.ptr_metadata_ty_or_tail(tcx, |x| x) {
+            match pointee_ty.ptr_metadata_ty_or_tail(tcx, |x| x.skip_normalization()) {


aaaaaaaaaaaaaaaa xd

Suggested change

match pointee_ty.ptr_metadata_ty_or_tail(tcx, |x| x.skip_normalization()) {

// FIXME(#155345): Missing normalization call

match pointee_ty.ptr_metadata_ty_or_tail(tcx, |x| x.skip_normalization()) {

View changes since the review

lcnr · 2026-05-06T07:41:15Z


                ty::Alias(..) => {
-                    let normalized = normalize(ty);
+                    let normalized = normalize(Unnormalized::new(ty));


Suggested change

let normalized = normalize(Unnormalized::new(ty));

let normalized = normalize(Unnormalized::new_wip(ty));

we should remove this as all the inputs should already be normalized

View changes since the review

lcnr · 2026-05-06T07:43:24Z

                                pointee,
                                &ObligationCause::dummy(),
-                                |ty| ty,
+                                |ty| ty.skip_norm_wip(),


that...

why are we not matching on try_normalize_erasing_regions inside of struct_tail_raw

ah well, not necessarily for your PR. would keep that as is for now

View changes since the review

lcnr

a few more suggestions, then r=me 😊 thank you for working on this

View changes since this review

Unnormalized migration: introduce assert_fully_normalized

9b28f31

rustbot assigned lcnr May 6, 2026

This comment has been minimized.

Sign in to view

Unnormalized migration: struct_tail takes fn taking Unnormalized

d618d7a

khyperia force-pushed the unnormalized-migration branch from 906e38e to d618d7a Compare May 6, 2026 07:35

lcnr reviewed May 6, 2026

View reviewed changes

		@@ -461,7 +461,7 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
		let projected_ty = curr_projected_ty.projection_ty_core(

		@@ -81,7 +81,7 @@ impl<'a, 'tcx> Expectation<'tcx> {
		let cause = ObligationCause::misc(span, fcx.body_id);

		// FIXME: This is not right, even in the old solver...

	// FIXME: This is not right, even in the old solver...
	// FIXME(#155345): Missing normalization call

	match pointee_ty.ptr_metadata_ty_or_tail(tcx, \|x\| x.skip_normalization()) {
	// FIXME(#155345): Missing normalization call
	match pointee_ty.ptr_metadata_ty_or_tail(tcx, \|x\| x.skip_normalization()) {

	let normalized = normalize(Unnormalized::new(ty));
	let normalized = normalize(Unnormalized::new_wip(ty));

Uh oh!

Conversation

khyperia commented May 6, 2026

Uh oh!

rustbot commented May 6, 2026

Uh oh!

This comment has been minimized.

rustbot commented May 6, 2026

Uh oh!

lcnr May 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!

lcnr May 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!

lcnr May 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!

lcnr May 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!

lcnr May 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!

lcnr May 6, 2026

Choose a reason for hiding this comment

Uh oh!

RalfJung May 6, 2026

Choose a reason for hiding this comment

Uh oh!

lcnr May 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!

lcnr May 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!

lcnr May 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!

lcnr May 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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr May 6, 2026 •

edited by rustbot

Loading

lcnr left a comment •

edited by rustbot

Loading