Refactor `AliasTy` & `AliasTerm` to use `Alias` by Jamesbarford · Pull Request #156538 · rust-lang/rust

Jamesbarford · 2026-05-13T10:54:26Z

Refactors AliasTy & AliasTerm to use Alias. To move around errors I had to manually implement Lift, TypeVisitable and TypeFoldable

Part of #156181

r? @lcnr

lcnr · 2026-05-13T13:03:48Z

can you say more about the errors you had to work around? this feels odd to me and something we should fix in the derives instead of requiring manual impls here imo

lcnr · 2026-05-13T12:57:13Z

 )]
-pub struct AliasTy<I: Interner> {
-    /// The parameters of the associated or opaque type.
+pub struct Alias<I: Interner, K> {


can you move Alias and AliasTerm in a separate rustc_type_ir/ty/alias module?

View changes since the review

667b095 👍, Only moved Alias & AliasTerm, not AliasTy

Jamesbarford · 2026-05-13T14:07:25Z

Edit: I've been experimenting and think I've expanded Lift_Generic to be able to handle more use cases so I've collapsed my original message as it is a wall of text. But does contextualise what Lift_Generic can now do for us.

Original Message

I've whittled it down to only a manual Lift implementation for Alias as below.

(comments omitted for brevity)

#[derive_where(Clone, Copy, Hash, PartialEq, Debug; I: Interner, K)]
#[derive(TypeVisitable_Generic, GenericTypeVisitable, TypeFoldable_Generic, Lift_Generic)]
#[cfg_attr(
    feature = "nightly",
    derive(Decodable_NoContext, Encodable_NoContext, StableHash_NoContext)
)]
pub struct Alias<I: Interner, K> {
    pub args: I::GenericArgs,

    pub kind: K,

    #[derive_where(skip(Debug))]
    #[type_visitable(ignore)]
    #[type_foldable(identity)]
    pub(crate) _use_alias_new_instead: (),
}

I get 2000+ errors of which fall into the following;

error[E0308]: mismatched types
error[E0521]: borrowed data escapes outside of associated function
error[E0521]: borrowed data escapes outside of function
error[E0521]: borrowed data escapes outside of method

A sample of errors being;

error[E0521]: borrowed data escapes outside of method
   --> compiler/rustc_middle/src/ty/trait_def.rs:229:13
    |
181 | impl<'tcx> TyCtxt<'tcx> {
    |      ---- lifetime `'tcx` defined here
...
223 |         self,
    |         ---- `self` is a reference that is only valid in the method body
...
229 |             fast_reject::simplify_type(self, self_ty, TreatParams::InstantiateWithInfer)
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    |             |
    |             `self` escapes the method body here
    |             argument requires that `'tcx` must outlive `'static`
    |
    = note: requirement occurs because of the type `context::TyCtxt<'_>`, which makes the generic argument `'_` invariant
    = note: the struct `context::TyCtxt<'tcx>` is invariant over the parameter `'tcx`
    = help: see <https://doc.rust-lang.org/nomicon/subtyping.html> for more information about variance
note: due to a current limitation of the type system, this implies a `'static` lifetime
   --> compiler/rustc_middle/src/ty/print/mod.rs:416:15
    |
416 |     T: Copy + for<'a, 'tcx> Lift<TyCtxt<'tcx>, Lifted: Print<FmtPrinter<'a, 'tcx>>>,
    |               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

...

error[E0308]: mismatched types
   --> compiler/rustc_middle/src/ty/trait_def.rs:229:13
    |
229 |             fast_reject::simplify_type(self, self_ty, TreatParams::InstantiateWithInfer)
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ one type is more general than the other
    |
    = note: expected enum `rustc_type_ir::AliasTyKind<context::TyCtxt<'tcx>>`
               found enum `rustc_type_ir::AliasTyKind<context::TyCtxt<'_>>`
note: the lifetime requirement is introduced here
   --> compiler/rustc_type_ir/src/interner.rs:25:7
    |
 25 |     + IrPrint<ty::AliasTy<Self>>
    |       ^^^^^^^^^^^^^^^^^^^^^^^^^^

error: implementation of `Interner` is not general enough
   --> compiler/rustc_middle/src/ty/trait_def.rs:229:13
    |
229 |             fast_reject::simplify_type(self, self_ty, TreatParams::InstantiateWithInfer)
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ implementation of `Interner` is not general enough
    |
    = note: `Interner` would have to be implemented for the type `context::TyCtxt<'tcx>`
    = note: ...but `Interner` is actually implemented for the type `context::TyCtxt<'0>`, for some specific lifetime `'0`


...pub struct Placeholder

error[E0308]: mismatched types
   --> compiler/rustc_middle/src/ty/trait_def.rs:229:13
    |
229 |             fast_reject::simplify_type(self, self_ty, TreatParams::InstantiateWithInfer)
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ one type is more general than the other
    |
    = note: expected enum `rustc_type_ir::AliasTermKind<context::TyCtxt<'tcx>>`
               found enum `rustc_type_ir::AliasTermKind<context::TyCtxt<'_>>`
note: the lifetime requirement is introduced here
   --> compiler/rustc_type_ir/src/interner.rs:33:7
    |
 33 |     + IrPrint<ty::NormalizesTo<Self>>
    |       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

...

error[E0308]: mismatched types
   --> compiler/rustc_middle/src/ty/relate.rs:55:12
    |
 55 | impl<'tcx> Relate<TyCtxt<'tcx>> for &'tcx ty::List<ty::PolyExistentialPredicate<'tcx>> {
    |            ^^^^^^^^^^^^^^^^^^^^ lifetime mismatch
    |
    = note: expected enum `rustc_type_ir::AliasTermKind<context::TyCtxt<'tcx>>`
               found enum `rustc_type_ir::AliasTermKind<context::TyCtxt<'tcx>>`
note: the required lifetime does not necessarily outlive the lifetime `'tcx` as defined here
   --> compiler/rustc_middle/src/ty/relate.rs:55:6
    |
 55 | impl<'tcx> Relate<TyCtxt<'tcx>> for &'tcx ty::List<ty::PolyExistentialPredicate<'tcx>> {
    |      ^^^^
note: the lifetime requirement is introduced here
   --> compiler/rustc_type_ir/src/relate.rs:116:21
    |
116 | pub trait Relate<I: Interner>: TypeFoldable<I> + PartialEq + Copy {
    |                     ^^^^^^^^

And so on... This was similar to pub struct Placeholder<I: Interner, T> where I think the the generic parameter T required manually implementing Lift. Though this is a superficial understanding

Jamesbarford · 2026-05-21T12:04:42Z

The latest commits expand Lift_Generic to be able to handle:

pub struct Binder<I: Interner, T> {
    //
}

and then

pub struct Binder<I: Interner, T = SomeKind<I>> {
    //
}

e83546d, fbf009a Not sure if these would be best in a separate PR that also clean up binder.rs then this PR can be solely focused on Alias refactoring?

Which I then used to clean up pub struct Binder {} which is out of scope for these changes but was satisfying to remove a FIXME: e2ba945

I can't get Eq to work for Alias, error bellow;

Eq Error

error[E0277]: the trait bound `AliasTyKind<rustc_middle::ty::TyCtxt<'tcx>>: std::cmp::Eq` is not satisfied
   --> compiler/rustc_infer/src/infer/region_constraints/mod.rs:191:11
    |
187 | #[derive(Copy, Clone, PartialEq, Eq, Hash, TypeFoldable, TypeVisitable)]
    |                                  -- in this derive macro expansion
...
191 |     Alias(ty::AliasTy<'tcx>),
    |           ^^^^^^^^^^^^^^^^^ the trait `std::cmp::Eq` is not implemented for `AliasTyKind<rustc_middle::ty::TyCtxt<'tcx>>`
    |
   --> /rustc/ef0fb8a2563200e322fa4419f09f65a63742038c/library/core/src/cmp.rs:364:0
    |
    = note: in this expansion of `#[derive(Eq)]`
help: the trait `std::cmp::Eq` is implemented for `rustc_type_ir::Alias<I, K>`
   --> compiler/rustc_type_ir/src/ty/alias.rs:15:1

edit: Actually we can remove the empty implementation if we make AliasTyKind derive Eq

@lcnr

…ric-capabilities, r=lcnr Support generic params in `Lift_Generic` Handle generic type parameters, including nested occurrences, when deriving `Lift_Generic`. This lets types like `Binder<I, T>` use `#[derive(Lift_Generic)]` instead of requiring a manual `Lift` impl. Concretely it can now handle structs as follows; ```rs // Generic type parameter struct Binder<I: Interner, T> { // body } // Nested generic type parameter pub struct Binder<I: Interner, T = SomeKind<I>> { // } ``` Split off from the `Alias` refactor work; rust-lang#156538 r? @lcnr

@lcnr

Rollup merge of #156956 - Jamesbarford:feat/extend-lift-generic-capabilities, r=lcnr Support generic params in `Lift_Generic` Handle generic type parameters, including nested occurrences, when deriving `Lift_Generic`. This lets types like `Binder<I, T>` use `#[derive(Lift_Generic)]` instead of requiring a manual `Lift` impl. Concretely it can now handle structs as follows; ```rs // Generic type parameter struct Binder<I: Interner, T> { // body } // Nested generic type parameter pub struct Binder<I: Interner, T = SomeKind<I>> { // } ``` Split off from the `Alias` refactor work; #156538 r? @lcnr

rustbot · 2026-06-05T08:17:01Z

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.

lcnr · 2026-06-08T09:11:27Z

+    /// The parameters of the associated, opaque, or constant alias.
+    ///
+    /// For a projection, these are the generic parameters for the trait and the
+    /// GAT parameters, if there are any.
+    ///
+    /// For an inherent projection, they consist of the self type and the GAT parameters,
+    /// if there are any.
+    ///
+    /// For RPIT the generic parameters are for the generics of the function,
+    /// while for TAIT it is used for the generic parameters of the alias.
+    pub args: I::GenericArgs,
+
+    pub kind: K,


Suggested change

/// The parameters of the associated, opaque, or constant alias.

///

/// For a projection, these are the generic parameters for the trait and the

/// GAT parameters, if there are any.

///

/// For an inherent projection, they consist of the self type and the GAT parameters,

/// if there are any.

///

/// For RPIT the generic parameters are for the generics of the function,

/// while for TAIT it is used for the generic parameters of the alias.

pub args: I::GenericArgs,

pub kind: K,

pub kind: K,

/// The parameters of the associated, opaque, or constant alias.

///

/// For a projection, these are the generic parameters for the trait and the

/// GAT parameters, if there are any.

///

/// For an inherent projection, they consist of the self type and the GAT parameters,

/// if there are any.

///

/// For RPIT the generic parameters are for the generics of the function,

/// while for TAIT it is used for the generic parameters of the alias.

pub args: I::GenericArgs,

personal style preference

View changes since the review

lcnr · 2026-06-08T09:16:02Z

+        interner.debug_assert_args_compatible(kind.into(), args);
+        Alias { kind, args, _use_alias_new_instead: () }
+    }
+}


hmm, could we change this to

trait AliasKind<I: Interner>: Copy { fn assert_args_compatible(self, cx: I, args: I::GenericArgs); #[inline] fn debug_assert_args_compatible(self, cx: I, args: I::GenericArgs) { if cfg!(debug_assertions) { self.assert... } } } impl<I: Interner, K: AliasKind> Alias<I, K> { pub fn new( cx: I, kind: K, args: I::GenericArgs, ) -> Self { kind.debug_assert_args_compatible(interner, args); Alias { kind, args, _use_alias_new_instead: () } } pub fn new_from_iter( cx: I, kind: K, args: impl IntoIterator<...>, ) -> Self { ... } }

View changes since the review

lcnr · 2026-06-08T09:17:54Z

+    fn fold_with<F: TypeFolder<I>>(self, folder: &mut F) -> Self {
+        NormalizesTo { alias: self.alias.fold_with(folder), term: self.term.fold_with(folder) }
+    }
+}


why do the derives not work?

View changes since the review

In essence a similar problem that hampered Lift_Generic and the work needs to be done to get it working is similar to what we did for Lift_Generic, I can open another PR for this?

khyperia · 2026-06-10T08:07:25Z

+            goal.predicate.alias.args,
+        );
+        let alias = ty::AliasTerm::from(uv);
+        let goal = goal.with(cx, ty::NormalizesTo { alias, term: goal.predicate.term });


I'm extremely on board with the style of thing that lcnr said with

with this we can now update NormalizesTo to also be generic over the kind, and e.g. for normalize_anon_const have the argument be Goal<I, ty::NormalizesTo<I, I::UnevaluatedConstId>>

there's a lot of places in the code that could hugely benefit from this (e.g. there's a large number of unwraps in #157653 that could be avoided), but this particular case with normalize_anon_const is a bit gross, and I would prefer keeping this as NormalizesTo<I, K = AliasTermKind<I>> (at least until additional refactorings happen that would e.g. allow evaluate_const_and_instantiate_normalizes_to_term to be more generic)

(this is just my code style opinion though, up to you!)

edit: just want to clarify, that PR has a lot of unwraps of the form:

let def_id = match inherent.kind { ty::AliasTermKind::InherentTy { def_id } => def_id.into(), ty::AliasTermKind::InherentConst { def_id } => def_id.into(), kind => panic!("expected inherent alias, found {kind:?}"), };

having the Alias have a more specific Kind in all the cases that PR does that style of unwrap (it's so common that projections already have a helper method for it, expect_projection_def_id) would be super nice, but yeah

@lcnr

…able-generic, r=lcnr Extend capabilities of `TypeFoldable_Generic` Split from rust-lang#156538. - Lets `TypeFoldable_Generic` derive structural folding for types with extra generic parameters by adding the necessary `T: TypeFoldable<I>` bounds automatically - Means in rust-lang#156538 we can remove the manual `TypeFoldable` implementation for `NormalizesTo` - Refactors shared traversal logic between `Lift_Generic` and `TypeFoldable_Generic` into a shared helper with a callback. r? @lcnr

rust-bors · 2026-06-11T11:41:28Z

☔ The latest upstream changes (presumably #157739) made this pull request unmergeable. Please resolve the merge conflicts.

rustbot assigned lcnr May 13, 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 May 13, 2026

This comment has been minimized.

Sign in to view

Jamesbarford force-pushed the chore/merge-AliasTy-AliasTerm branch from a630681 to 89cd50d Compare May 13, 2026 12:56

lcnr reviewed May 13, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

Jamesbarford force-pushed the chore/merge-AliasTy-AliasTerm branch from 667b095 to b7965ca Compare May 21, 2026 08:30

This comment has been minimized.

Sign in to view

Jamesbarford mentioned this pull request May 26, 2026

Support generic params in Lift_Generic #156956

Merged

This comment has been minimized.

Sign in to view

Jamesbarford added 2 commits June 5, 2026 09:16

Refactor AliasTy & AliasTerm to use Alias

627d73f

Update tests to reflect Alias changes

c3bdf72

Jamesbarford force-pushed the chore/merge-AliasTy-AliasTerm branch from e2ba945 to c3bdf72 Compare June 5, 2026 08:16

Move AliasTy and AliasTerm types to ty/alias.rs

bf409af

Jamesbarford force-pushed the chore/merge-AliasTy-AliasTerm branch from 71c7efb to bf409af Compare June 5, 2026 08:42

lcnr reviewed Jun 8, 2026

View reviewed changes

khyperia mentioned this pull request Jun 10, 2026

[Work Item]: Rename UnevaluatedConst to AliasConst rust-lang/project-const-generics#115

Open

Jamesbarford mentioned this pull request Jun 10, 2026

move rustc_type_ir Term things to term_kind.rs #157419

Open

khyperia reviewed Jun 10, 2026

View reviewed changes

Jamesbarford mentioned this pull request Jun 10, 2026

Extend capabilities of TypeFoldable_Generic #157695

Open

khyperia mentioned this pull request Jun 11, 2026

remove AliasTerm::def_id() #157653

Open

Uh oh!

Conversation

Jamesbarford commented May 13, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

lcnr commented May 13, 2026

Uh oh!

lcnr May 13, 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 May 14, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Jamesbarford commented May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

Jamesbarford commented May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

rustbot commented Jun 5, 2026

Uh oh!

lcnr Jun 8, 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 8, 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 8, 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 Jun 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

khyperia Jun 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rust-bors Bot commented Jun 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

lcnr May 13, 2026 •

edited by rustbot

Loading

Jamesbarford commented May 13, 2026 •

edited

Loading

Jamesbarford commented May 21, 2026 •

edited

Loading

lcnr Jun 8, 2026 •

edited by rustbot

Loading

lcnr Jun 8, 2026 •

edited by rustbot

Loading

lcnr Jun 8, 2026 •

edited by rustbot

Loading

Jamesbarford Jun 9, 2026 •

edited

Loading

khyperia Jun 10, 2026 •

edited

Loading