When encountering `-> Trait`, suggest `-> Box<dyn Trait>` (instead of `-> Box<Trait>`.
If there's a single returned type within the `fn`, suggest `-> impl Trait`.
interpret: add sanity check in dyn upcast to double-check what codegen does
For dyn receiver calls, we already have two codepaths: look up the function to call by indexing into the vtable, or alternatively resolve the DefId given the dynamic type of the receiver. With debug assertions enabled, the interpreter does both and compares the results. (Without debug assertions we always use the vtable as it is simpler.)
This PR does the same for dyn trait upcasts. However, for casts *not* using the vtable is the easier thing to do, so now the vtable path is the debug-assertion-only path. In particular, there are cases where the vtable does not contain a pointer for upcasts but instead reuses the old pointer: when the supertrait vtable is a prefix of the larger vtable. We don't want to expose this optimization and detect UB if people do a transmute assuming this optimization, so we cannot in general use the vtable indexing path.
r? ``@oli-obk``
Represent type-level consts with new-and-improved `hir::ConstArg`
### Summary
This is a step toward `min_generic_const_exprs`. We now represent all const
generic arguments using an enum that differentiates between const *paths*
(temporarily just bare const params) and arbitrary anon consts that may perform
computations. This will enable us to cleanly implement the `min_generic_const_args`
plan of allowing the use of generics in paths used as const args, while
disallowing their use in arbitrary anon consts. Here is a summary of the salient
aspects of this change:
- Add `current_def_id_parent` to `LoweringContext`
This is needed to track anon const parents properly once we implement
`ConstArgKind::Path` (which requires moving anon const def-creation
outside of `DefCollector`).
- Create `hir::ConstArgKind` enum with `Path` and `Anon` variants. Use it in the
existing `hir::ConstArg` struct, replacing the previous `hir::AnonConst` field.
- Use `ConstArg` for all instances of const args. Specifically, use it instead
of `AnonConst` for assoc item constraints, array lengths, and const param
defaults.
- Some `ast::AnonConst`s now have their `DefId`s created in
rustc_ast_lowering rather than `DefCollector`. This is because in some
cases they will end up becoming a `ConstArgKind::Path` instead, which
has no `DefId`. We have to solve this in a hacky way where we guess
whether the `AnonConst` could end up as a path const since we can't
know for sure until after name resolution (`N` could refer to a free
const or a nullary struct). If it has no chance as being a const
param, then we create a `DefId` in `DefCollector` -- otherwise we
decide during ast_lowering. This will have to be updated once all path
consts use `ConstArgKind::Path`.
- We explicitly use `ConstArgHasType` for array lengths, rather than
implicitly relying on anon const type feeding -- this is due to the
addition of `ConstArgKind::Path`.
- Some tests have their outputs changed, but the changes are for the
most part minor (including removing duplicate or almost-duplicate
errors). One test now ICEs, but it is for an incomplete, unstable
feature and is now tracked at https://github.com/rust-lang/rust/issues/127009.
### Followup items post-merge
- Use `ConstArgKind::Path` for all const paths, not just const params.
- Fix (no github dont close this issue) #127009
- If a path in generic args doesn't resolve as a type, try to resolve as a const
instead (do this in rustc_resolve). Then remove the special-casing from
`rustc_ast_lowering`, so that all params will automatically be lowered as
`ConstArgKind::Path`.
- (?) Consider making `const_evaluatable_unchecked` a hard error, or at least
trying it in crater
r? `@BoxyUwU`
This is a very large commit since a lot needs to be changed in order to
make the tests pass. The salient changes are:
- `ConstArgKind` gets a new `Path` variant, and all const params are now
represented using it. Non-param paths still use `ConstArgKind::Anon`
to prevent this change from getting too large, but they will soon use
the `Path` variant too.
- `ConstArg` gets a distinct `hir_id` field and its own variant in
`hir::Node`. This affected many parts of the compiler that expected
the parent of an `AnonConst` to be the containing context (e.g., an
array repeat expression). They have been changed to check the
"grandparent" where necessary.
- Some `ast::AnonConst`s now have their `DefId`s created in
rustc_ast_lowering rather than `DefCollector`. This is because in some
cases they will end up becoming a `ConstArgKind::Path` instead, which
has no `DefId`. We have to solve this in a hacky way where we guess
whether the `AnonConst` could end up as a path const since we can't
know for sure until after name resolution (`N` could refer to a free
const or a nullary struct). If it has no chance as being a const
param, then we create a `DefId` in `DefCollector` -- otherwise we
decide during ast_lowering. This will have to be updated once all path
consts use `ConstArgKind::Path`.
- We explicitly use `ConstArgHasType` for array lengths, rather than
implicitly relying on anon const type feeding -- this is due to the
addition of `ConstArgKind::Path`.
- Some tests have their outputs changed, but the changes are for the
most part minor (including removing duplicate or almost-duplicate
errors). One test now ICEs, but it is for an incomplete, unstable
feature and is now tracked at #127009.
Invert infer `error_reporting` mod struture
Parallel change to #127493, which moves `rustc_infer::infer::error_reporting` to `rustc_infer::error_reporting::infer`. After this, we should just be able to merge this into `rustc_trait_selection::error_reporting::infer`, and pull down `TypeErrCtxt` into that crate. 👍
r? lcnr
This commit changes the error reporting mechanism for not implemented
traits to skip impl marked as `#[diagnostic::do_not_recommend]` in the
help part of the error message ("the following other types implement
trait `Foo`:"). The main use case here is to allow crate authors to skip
non-meaningful confusing suggestions. A common example for this are
fully generic impls on tuples.
Uplift elaboration into `rustc_type_ir`
Allows us to deduplicate and consolidate elaboration (including these stupid elaboration duplicate fns i added for pretty printing like 3 years ago) so I'm pretty hyped about this change :3
r? lcnr
Make `can_eq` process obligations (almost) everywhere
Move `can_eq` to an extension trait on `InferCtxt` in `rustc_trait_selection`, and change it so that it processes obligations. This should strengthen it to be more accurate in some cases, but is most important for the new trait solver which delays relating aliases to `AliasRelate` goals. Without this, we always basically just return true when passing aliases to `can_eq`, which can lead to weird errors, for example #127149.
I'm not actually certain if we should *have* `can_eq` be called on the good path. In cases where we need `can_eq`, we probably should just be using a regular probe.
Fixes#127149
r? lcnr
Don't try to label `ObligationCauseCode::CompareImplItem` for an RPITIT, since it has no name
The old (current) trait solver has a limitation that when a where clause in param-env must be normalized using the same where clause, then we get spurious errors in `normalize_param_env_or_error`. I don't think there's an issue tracking it, but it's the root cause for many of the "fixed-by-next-solver" labeled issues.
Specifically, these errors may occur when checking predicate entailment of the GAT that comes out of desugaring RPITITs. Since we use `ObligationCauseCode::CompareImplItem` for these predicates, we try calling `item_name` on an RPITIT which fails, since the RPITIT has no name.
We simply suppress this logic when we're reporting a predicate entailment error for an RPITIT. RPITITs should never have predicate entailment errors, *by construction*, but they may due to this bug in the old solver.
Addresses the ICE in #127331, though doesn't fix the underlying issue (which is fundamental to the old solver).
r? types
Use `ControlFlow` results for visitors that are only looking for a single value
These visitors all had a `Option<Value>` or `bool` field, that, once set, was never unset or modified again. They have been refactored by removing the field and returning `ControlFlow` directly from the visitor
Re-implement a type-size based limit
r? lcnr
This PR reintroduces the type length limit added in #37789, which was accidentally made practically useless by the caching changes to `Ty::walk` in #72412, which caused the `walk` function to no longer walk over identical elements.
Hitting this length limit is not fatal unless we are in codegen -- so it shouldn't affect passes like the mir inliner which creates potentially very large types (which we observed, for example, when the new trait solver compiles `itertools` in `--release` mode).
This also increases the type length limit from `1048576 == 2 ** 20` to `2 ** 24`, which covers all of the code that can be reached with craterbot-check. Individual crates can increase the length limit further if desired.
Perf regression is mild and I think we should accept it -- reinstating this limit is important for the new trait solver and to make sure we don't accidentally hit more type-size related regressions in the future.
Fixes#125460
Actually report normalization-based type errors correctly for alias-relate obligations in new solver
We have some special casing to report type mismatch errors that come from projection predicates, but we don't do that for alias-relate obligations. This PR implements that. There's a bit of code duplication, but 🤷
Best reviewed without whitespace.
r? lcnr
Check alias args for WF even if they have escaping bound vars
#### What
This PR stops skipping arguments of aliases if they have escaping bound vars, instead recursing into them and only discarding the resulting obligations referencing bounds vars.
#### An example:
From the test:
```
trait Trait {
type Gat<U: ?Sized>;
}
fn test<T>(f: for<'a> fn(<&'a T as Trait>::Gat<&'a [str]>)) where for<'a> &'a T: Trait {}
//~^ ERROR the size for values of type `[()]` cannot be known at compilation time
fn main() {}
```
We now prove that `str: Sized` in order for `&'a [str]` to be well-formed. We were previously unconditionally skipping over `&'a [str]` as it referenced a buond variable. We now recurse into it and instead only discard the `[str]: 'a` obligation because of the escaping bound vars.
#### Why?
This is a change that improves consistency about proving well-formedness earlier in the pipeline, which is necessary for future work on where-bounds in binders and correctly handling higher-ranked implied bounds. I don't expect this to fix any unsoundness.
#### What doesn't it fix?
Specifically, this doesn't check projection predicates' components are well-formed, because there are too many regressions: https://github.com/rust-lang/rust/pull/123737#issuecomment-2052198478
Automatically taint InferCtxt when errors are emitted
r? `@nnethercote`
Basically `InferCtxt::dcx` now returns a `DiagCtxt` that refers back to the `Cell<Option<ErrorGuaranteed>>` of the `InferCtxt` and thus when invoking `Diag::emit`, and the diagnostic is an error, we taint the `InferCtxt` directly.
That change on its own has no effect at all, because `InferCtxt` already tracks whether errors have been emitted by recording the global error count when it gets opened, and checking at the end whether the count changed. So I removed that error count check, which had a bit of fallout that I immediately fixed by invoking `InferCtxt::dcx` instead of `TyCtxt::dcx` in a bunch of places.
The remaining new errors are because an error was reported in another query, and never bubbled up. I think they are minor enough for this to be ok, and sometimes it actually improves diagnostics, by not silencing useful diagnostics anymore.
fixes#126485 (cc `@olafes)`
There are more improvements we can do (like tainting in hir ty lowering), but I would rather do that in follow up PRs, because it requires some refactorings.
Remove unused `rustc_trait_selection` dependencies
Found using `cargo-machete`. The `bitflags` and `derivative` crates were added for the new trait solver, but weren't removed when the next trait solver code was uplifted to a separate crate.
Implement new effects desugaring
cc `@rust-lang/project-const-traits.` Will write down notes once I have finished.
* [x] See if we want `T: Tr` to desugar into `T: Tr, T::Effects: Compat<true>`
* [x] Fix ICEs on `type Assoc: ~const Tr` and `type Assoc<T: ~const Tr>`
* [ ] add types and traits to minicore test
* [ ] update rustc-dev-guide
Fixes#119717Fixes#123664Fixes#124857Fixes#126148
Allow constraining opaque types during various unsizing casts
allows unsizing of tuples, arrays and Adts to constraint opaque types in their generic parameters to concrete types on either side of the unsizing cast.
Also allows constraining opaque types during trait object casts that only differ in auto traits or lifetimes.
cc #116652
Add `SliceLike` to `rustc_type_ir`, use it in the generic solver code (+ some other changes)
First, we split out `TraitRef::new_from_args` which takes *just* `ty::GenericArgsRef` from `TraitRef::new` which takes `impl IntoIterator<Item: Into<GenericArg>>`. I will explain in a minute why.
Second, we introduce `SliceLike`, which allows us to be generic over `List<T>` and `[T]`. This trait has an `as_slice()` and `into_iter()` method, and some other convenience functions. However, importantly, since types like `I::GenericArgs` now implement `SliceLike` rather than `IntoIter<Item = I::GenericArg>`, we can't use `TraitRef::new` on this directly. That's where `new_from_args` comes in.
Finally, we adjust all the code to use these slice operators. Some things get simpler, some things get a bit more annoying since we need to use `as_slice()` in a few places. 🤷
r? lcnr
Clean up some comments near `use` declarations
#125443 will reformat all `use` declarations in the repository. There are a few edge cases involving comments on `use` declarations that require care. This PR cleans up some clumsy comment cases, taking us a step closer to #125443 being able to merge.
r? ``@lqd``
Actually taint InferCtxt when a fulfillment error is emitted
And avoid checking the global error counter
fixes#122044fixes#123255fixes#123276fixes#125799
Most modules have such a blank line, but some don't. Inserting the blank
line makes it clearer that the `//!` comments are describing the entire
module, rather than the `use` declaration(s) that immediately follows.
Uplift next trait solver to `rustc_next_trait_solver`
🎉
There's so many FIXMEs! Sorry! Ideally this merges with the FIXMEs and we track and squash them over the near future.
Also, this still doesn't build on anything other than rustc. I still need to fix `feature = "nightly"` in `rustc_type_ir`, and remove and fix all the nightly feature usage in the new trait solver (notably: let-chains).
Also, sorry `@lcnr` I know you asked for me to separate the commit where we `mv rustc_trait_selection/solve/... rustc_next_trait_solver/solve/...`, but I had already done all the work by that point. Luckily, `git` understands the file moves so it should still be relatively reviewable.
If this is still very difficult to review, then I can do some rebasing magic to try to separate this out. Please let me know!
r? lcnr
Spell out other trait diagnostic
I recently saw somebody confused about the diagnostic thinking it was suggesting to add an `as` cast. This change is longer but I think it's clearer
Only compute vtable information during codegen
This PR removes vtable information from the `Object` and `TraitUpcasting` candidate sources in the trait solvers, and defers the computation of relevant information to `Instance::resolve`. This is because vtables really aren't a thing in the trait world -- they're an implementation detail in codegen.
Previously it was just easiest to tangle this information together since we were already doing the work of looking at all the supertraits in the trait solver, and specifically because we use traits to represent when it's possible to call a method via a vtable (`Object` candidate) and do upcasting (`Unsize` candidate). but I am somewhat suspicious we're doing a *lot* of extra work, especially in polymorphic contexts, so let's see what perf says.
Make proof tree probing and `Candidate`/`CandidateSource` generic over interner
`<TyCtxt<'tcx>>` is ugly, but will become `<I>` when things actually become generic.
r? lcnr
Check that alias-relate terms are WF if reporting an error in alias-relate
Check that each of the left/right term is WF when deriving a best error obligation for an alias-relate goal. This will make sure that given `<i32 as NotImplemented>::Assoc = ()` will drill down into `i32: NotImplemented` since we currently treat the projection as rigid.
r? lcnr
Use a consistent way to filter out bounds instead of splitting it into three places
just a small cleanup, no logic change.
Initially the code had me looking for why anything was special here, only to realize there's nothing interesting going on
Use `Variance` glob imported variants everywhere
Fully commit to using the globbed variance. Could be convinced the other way, and change this PR to not use the globbed variants anywhere, but I'd rather we do one or the other.
r? lcnr
Harmonize using root or leaf obligation in trait error reporting
When #121826 changed the error reporting to use root obligation and not the leafmost obligation, it didn't actually make sure that all the other diagnostics helper functions used the right obligation.
Specifically, when reporting similar impl candidates we are looking for impls of the root obligation, but trying to match them against the trait ref of the leaf obligation.
This does a few other miscellaneous changes. There's a lot more clean-up that could be done here, but working with this code is really grief-inducing due to how messy it has become over the years. Someone really needs to show it love. 😓
r? ``@estebank``
Fixes#126129
Use `tidy` to sort crate attributes for all compiler crates.
We already do this for a number of crates, e.g. `rustc_middle`, `rustc_span`, `rustc_metadata`, `rustc_span`, `rustc_errors`.
For the ones we don't, in many cases the attributes are a mess.
- There is no consistency about order of attribute kinds (e.g. `allow`/`deny`/`feature`).
- Within attribute kind groups (e.g. the `feature` attributes), sometimes the order is alphabetical, and sometimes there is no particular order.
- Sometimes the attributes of a particular kind aren't even grouped all together, e.g. there might be a `feature`, then an `allow`, then another `feature`.
This commit extends the existing sorting to all compiler crates, increasing consistency. If any new attribute line is added there is now only one place it can go -- no need for arbitrary decisions.
Exceptions:
- `rustc_log`, `rustc_next_trait_solver` and `rustc_type_ir_macros`, because they have no crate attributes.
- `rustc_codegen_gcc`, because it's quasi-external to rustc (e.g. it's ignored in `rustfmt.toml`).
r? `@davidtwco`
For E0277 suggest adding `Result` return type for function when using QuestionMark `?` in the body.
Adding suggestions for following function in E0277.
```rust
fn main() {
let mut _file = File::create("foo.txt")?;
}
```
to
```rust
fn main() -> Result<(), Box<dyn std::error::Error>> {
let mut _file = File::create("foo.txt")?;
return Ok(());
}
```
According to the issue #125997, only the code examples in the issue are targeted, but the issue covers a wider range of situations.
<!--
If this PR is related to an unstable feature or an otherwise tracked effort,
please link to the relevant tracking issue here. If you don't know of a related
tracking issue or there are none, feel free to ignore this.
This PR will get automatically assigned to a reviewer. In case you would like
a specific user to review your work, you can assign it to them by using
r? <reviewer name>
-->
We already do this for a number of crates, e.g. `rustc_middle`,
`rustc_span`, `rustc_metadata`, `rustc_span`, `rustc_errors`.
For the ones we don't, in many cases the attributes are a mess.
- There is no consistency about order of attribute kinds (e.g.
`allow`/`deny`/`feature`).
- Within attribute kind groups (e.g. the `feature` attributes),
sometimes the order is alphabetical, and sometimes there is no
particular order.
- Sometimes the attributes of a particular kind aren't even grouped
all together, e.g. there might be a `feature`, then an `allow`, then
another `feature`.
This commit extends the existing sorting to all compiler crates,
increasing consistency. If any new attribute line is added there is now
only one place it can go -- no need for arbitrary decisions.
Exceptions:
- `rustc_log`, `rustc_next_trait_solver` and `rustc_type_ir_macros`,
because they have no crate attributes.
- `rustc_codegen_gcc`, because it's quasi-external to rustc (e.g. it's
ignored in `rustfmt.toml`).
Make `ObligationEmittingRelation`s emit `Goal` rather than `Obligation`
Helps avoid needing to uplift `Obligation` into the solver. We still can't get rid of `ObligationCause`, but we can keep it as an associated type for `InferCtxtLike` and just give it a `dummy` function.
There's some shuttling between `Goal` and `Obligation` that may be perf-sensitive... Let's see what rust-timer says.
r? lcnr
Only compute `specializes` query if (min)specialization is enabled in the crate of the specializing impl
Fixes (after backport) https://github.com/rust-lang/rust/issues/125197
### What
https://github.com/rust-lang/rust/pull/122791 makes it so that inductive cycles are no longer hard errors. That means that when we are testing, for example, whether these impls overlap:
```rust
impl PartialEq<Self> for AnyId {
fn eq(&self, _: &Self) -> bool {
todo!()
}
}
impl<T: Identifier> PartialEq<T> for AnyId {
fn eq(&self, _: &T) -> bool {
todo!()
}
}
```
...given...
```rust
pub trait Identifier: Display + 'static {}
impl<T> Identifier for T where T: PartialEq + Display + 'static {}
```
Then we try to see if the second impl holds given `T = AnyId`. That requires `AnyId: Identifier`, which requires that `AnyId: PartialEq`, which is satisfied by these two impl candidates... The `PartialEq<T>` impl is a cycle, and we used to winnow it when we used to treat inductive cycles as errors.
However, now that we don't winnow it, this means that we *now* try calling `candidate_should_be_dropped_in_favor_of`, which tries to check whether one of the impls specializes the other: the `specializes` query. In that query, we currently bail early if the impl is local.
However, in a foreign crate, we try to compute if the two impls specialize each other by doing trait solving. This may itself lead to the same situation where we call `specializes`, which will lead to a query cycle.
### How does this fix the problem
We now record whether specialization is enabled in foreign crates, and extend this early-return behavior to foreign impls too. This means that we can only encounter these cycles if we truly have a specializing impl from a crate with specialization enabled.
-----
r? `@oli-obk` or `@lcnr`
Remove the `ty` field from type system `Const`s
Fixes#125556Fixes#122908
Part of the work on `adt_const_params`/`generic_const_param_types`/`min_generic_const_exprs`/generally making the compiler nicer. cc rust-lang/project-const-generics#44
Please review commit-by-commit otherwise I wasted a lot of time not just squashing this into a giant mess (and also it'll be SO much nicer because theres a lot of fluff changes mixed in with other more careful changes if looking via File Changes
---
Why do this?
- The `ty` field keeps causing ICEs and weird behaviour due to it either being treated as "part of the const" or it being forgotten about leading to ICEs.
- As we move forward with `adt_const_params` and a potential `min_generic_const_exprs` it's going to become more complex to actually lower the correct `Ty<'tcx>`
- It muddles the idea behind how we check `Const` arguments have the correct type. By having the `ty` field it may seem like we ought to be relating it when we relate two types, or that its generally important information about the `Const`.
- Brings the compiler more in line with `a-mir-formality` as that also tracks the type of type system `Const`s via `ConstArgHasType` bounds in the env instead of on the `Const` itself.
- A lot of stuff is a lot nicer when you dont have to pass around the type of a const lol. Everywhere we construct `Const` is now significantly nicer 😅
See #125671's description for some more information about the `ty` field
---
General summary of changes in this PR:
- Add `Ty` to `ConstKind::Value` as otherwise there is no way to implement `ConstArgHasType` to ensure that const arguments are correctly typed for the parameter when we stop creating anon consts for all const args. It's also just incredibly difficult/annoying to thread the correct `Ty` around to a bunch of ctfe functions otherwise.
- Fully implement `ConstArgHasType` in both the old and new solver. Since it now has no reliance on the `ty` field it serves its originally intended purpose of being able to act as a double check that trait vs impls have correctly typed const parameters. It also will now be able to be responsible for checking types of const arguments to parameters under `min_generic_const_exprs`.
- Add `Ty` to `mir::Const::Ty`. I dont have a great understanding of why mir constants are setup like this to be honest. Regardless they need to be able to determine the type of the const and the easiest way to make this happen was to simply store the `Ty` along side the `ty::Const`. Maybe we can do better here in the future but I'd have to spend way more time looking at everywhere we use `mir::Const`.
- rustdoc has its own `Const` which also has a `ty` field. It was relatively easy to remove this.
---
r? `@lcnr` `@compiler-errors`
Rollup of 9 pull requests
Successful merges:
- #124840 (resolve: mark it undetermined if single import is not has any bindings)
- #125622 (Winnow private method candidates instead of assuming any candidate of the right name will apply)
- #125648 (Remove unused(?) `~/rustsrc` folder from docker script)
- #125672 (Add more ABI test cases to miri (RFC 3391))
- #125800 (Fix `mut` static task queue in SGX target)
- #125871 (Orphanck[old solver]: Consider opaque types to never cover type parameters)
- #125893 (Handle all GVN binops in a single place.)
- #126008 (Port `tests/run-make-fulldeps/issue-19371` to ui-fulldeps)
- #126032 (Update description of the `IsTerminal` example)
r? `@ghost`
`@rustbot` modify labels: rollup
Orphanck[old solver]: Consider opaque types to never cover type parameters
This fixes an oversight of mine in #117164. The change itself has already been FCP'ed.
This only affects the old solver, the next solver already correctly rejects the added test since #117164.
r? ``@lcnr``
Refactor `#[diagnostic::do_not_recommend]` support
This commit refactors the `#[do_not_recommend]` support in the old parser to also apply to projection errors and not only to selection errors. This allows the attribute to be used more widely.
Part of #51992
r? `@compiler-errors`
<!--
If this PR is related to an unstable feature or an otherwise tracked effort,
please link to the relevant tracking issue here. If you don't know of a related
tracking issue or there are none, feel free to ignore this.
This PR will get automatically assigned to a reviewer. In case you would like
a specific user to review your work, you can assign it to them by using
r? <reviewer name>
-->
Align `Term` methods with `GenericArg` methods, add `Term::expect_*`
* `Term::ty` -> `Term::as_type`.
* `Term::ct` -> `Term::as_const`.
* Adds `Term::expect_type` and `Term::expect_const`, and uses them in favor of `.ty().unwrap()`, etc.
I could also shorten these to `as_ty` and then do `GenericArg::as_ty` as well, but I do think the `as_` is important to signal that this is a conversion method, and not a getter, like `Const::ty` is.
r? types
Make `WHERE_CLAUSES_OBJECT_SAFETY` a regular object safety violation
#### The issue
In #50781, we have known about unsound `where` clauses in function arguments:
```rust
trait Impossible {}
trait Foo {
fn impossible(&self)
where
Self: Impossible;
}
impl Foo for &() {
fn impossible(&self)
where
Self: Impossible,
{}
}
// `where` clause satisfied for the object, meaning that the function now *looks* callable.
impl Impossible for dyn Foo {}
fn main() {
let x: &dyn Foo = &&();
x.impossible();
}
```
... which currently segfaults at runtime because we try to call a method in the vtable that doesn't exist. :(
#### What did u change
This PR removes the `WHERE_CLAUSES_OBJECT_SAFETY` lint and instead makes it a regular object safety violation. I choose to make this into a hard error immediately rather than a `deny` because of the time that has passed since this lint was authored, and the single (1) regression (see below).
That means that it's OK to mention `where Self: Trait` where clauses in your trait, but making such a trait into a `dyn Trait` object will report an object safety violation just like `where Self: Sized`, etc.
```rust
trait Impossible {}
trait Foo {
fn impossible(&self)
where
Self: Impossible; // <~ This definition is valid, just not object-safe.
}
impl Foo for &() {
fn impossible(&self)
where
Self: Impossible,
{}
}
fn main() {
let x: &dyn Foo = &&(); // <~ THIS is where we emit an error.
}
```
#### Regressions
From a recent crater run, there's only one crate that relies on this behavior: https://github.com/rust-lang/rust/pull/124305#issuecomment-2122381740. The crate looks unmaintained and there seems to be no dependents.
#### Further
We may later choose to relax this (e.g. when the where clause is implied by the supertraits of the trait or something), but this is not something I propose to do in this FCP.
For example, given:
```
trait Tr {
fn f(&self) where Self: Blanket;
}
impl<T: ?Sized> Blanket for T {}
```
Proving that some placeholder `S` implements `S: Blanket` would be sufficient to prove that the same (blanket) impl applies for both `Concerete: Blanket` and `dyn Trait: Blanket`.
Repeating here that I don't think we need to implement this behavior right now.
----
r? lcnr
Use parenthetical notation for `Fn` traits
Always use the `Fn(T) -> R` format when printing closure traits instead of `Fn<(T,), Output = R>`.
Address #67100:
```
error[E0277]: expected a `Fn()` closure, found `F`
--> file.rs:6:13
|
6 | call_fn(f)
| ------- ^ expected an `Fn()` closure, found `F`
| |
| required by a bound introduced by this call
|
= note: wrap the `F` in a closure with no arguments: `|| { /* code */ }`
note: required by a bound in `call_fn`
--> file.rs:1:15
|
1 | fn call_fn<F: Fn() -> ()>(f: &F) {
| ^^^^^^^^^^ required by this bound in `call_fn`
help: consider further restricting this bound
|
5 | fn call_any<F: std::any::Any + Fn()>(f: &F) {
| ++++++
```
Uplift `{Closure,Coroutine,CoroutineClosure}Args` and friends to `rustc_type_ir`
Part of converting the new solver's `structural_traits.rs` to be interner-agnostic.
I decided against aliasing `ClosureArgs<TyCtxt<'tcx>>` to `ClosureArgs<'tcx>` because it seemed so rare. I could do so if desired, though.
r? lcnr
Implement `needs_async_drop` in rustc and optimize async drop glue
This PR expands on #121801 and implements `Ty::needs_async_drop` which works almost exactly the same as `Ty::needs_drop`, which is needed for #123948.
Also made compiler's async drop code to look more like compiler's regular drop code, which enabled me to write an optimization where types which do not use `AsyncDrop` can simply forward async drop glue to `drop_in_place`. This made size of the async block from the [async_drop test](67980dd6fb/tests/ui/async-await/async-drop.rs) to decrease by 12%.
Rename HIR `TypeBinding` to `AssocItemConstraint` and related cleanup
Rename `hir::TypeBinding` and `ast::AssocConstraint` to `AssocItemConstraint` and update all items and locals using the old terminology.
Motivation: The terminology *type binding* is extremely outdated. "Type bindings" not only include constraints on associated *types* but also on associated *constants* (feature `associated_const_equality`) and on RPITITs of associated *functions* (feature `return_type_notation`). Hence the word *item* in the new name. Furthermore, the word *binding* commonly refers to a mapping from a binder/identifier to a "value" for some definition of "value". Its use in "type binding" made sense when equality constraints (e.g., `AssocTy = Ty`) were the only kind of associated item constraint. Nowadays however, we also have *associated type bounds* (e.g., `AssocTy: Bound`) for which the term *binding* doesn't make sense.
---
Old terminology (HIR, rustdoc):
```
`TypeBinding`: (associated) type binding
├── `Constraint`: associated type bound
└── `Equality`: (associated) equality constraint (?)
├── `Ty`: (associated) type binding
└── `Const`: associated const equality (constraint)
```
Old terminology (AST, abbrev.):
```
`AssocConstraint`
├── `Bound`
└── `Equality`
├── `Ty`
└── `Const`
```
New terminology (AST, HIR, rustdoc):
```
`AssocItemConstraint`: associated item constraint
├── `Bound`: associated type bound
└── `Equality`: associated item equality constraint OR associated item binding (for short)
├── `Ty`: associated type equality constraint OR associated type binding (for short)
└── `Const`: associated const equality constraint OR associated const binding (for short)
```
r? compiler-errors
Make `body_owned_by` return the `Body` instead of just the `BodyId`
fixes#125677
Almost all `body_owned_by` callers immediately called `body`, too, so just return `Body` directly.
This makes the inline-const query feeding more robust, as all calls to `body_owned_by` will now yield a body for inline consts, too.
I have not yet figured out a good way to make `tcx.hir().body()` return an inline-const body, but that can be done as a follow-up
Do not equate `Const`'s ty in `super_combine_const`
Fixes#114456
In #125451 we started relating the `Const`'s tys outside of a probe so it was no longer simply an assertion to catch bugs.
This was done so that when we _do_ provide a wrongly typed const argument to an item if we wind up relating it with some other instantiation we'll have a `TypeError` we can bubble up and taint the resulting mir allowing const eval to skip evaluation.
In this PR I instead change `ConstArgHasType` to correctly handle checking the types of const inference variables. Previously if we had something like `impl<const N: u32> Trait for [(); N]`, when using the impl we would instantiate it with infer vars and then check that `?x: u32` is of type `u32` and succeed. Then later we would infer `?x` to some `Const` of type `usize`.
We now stall on `?x` in `ConstArgHasType` until it has a concrete value that we can determine the type of. This allows us to fail using the erroneous implementation of `Trait` which allows us to taint the mir.
Long term we intend to remove the `ty` field on `Const` so we would have no way of accessing the `ty` of a const inference variable anyway and would have to do this. I did not fully update `ConstArgHasType` to avoid using the `ty` field as it's not entirely possible right now- we would need to lookup `ConstArgHasType` candidates in the env.
---
As for _why_ I think we should do this, relating the types of const's is not necessary for soundness of the type system. Originally this check started off as a plain `==` in `super_relate_consts` and gradually has been growing in complexity as we support more complicated types. It was never actually required to ensure that const arguments are correctly typed for their parameters however.
The way we currently check that a const argument has the correct type is a little convoluted and confusing (and will hopefully be less weird as time goes on). Every const argument has an anon const with its return type set to type of the const parameter it is an argument to. When type checking the anon const regular type checking rules require that the expression is the same type as the return type. This effectively ensure that no matter what every const argument _always_ has the correct type.
An extra bit of complexity is that during `hir_ty_lowering` we do not represent everything as a `ConstKind::Unevaluated` corresponding to the anon const. For generic parameters i.e. `[(); N]` we simply represent them as `ConstKind::Param` as we do not want `ConstKind::Unevaluated` with generic substs on stable under min const generics. The anon const still gets type checked resulting in errors about type mismatches.
Eventually we intend to not create anon consts for all const arguments (for example for `ConstKind::Param`) and instead check that the argument type is correct via `ConstArgHasType` obligations (these effectively also act as a check that the anon consts have the correctly set return type).
What this all means is that the the only time we should ever have mismatched types when relating two `Const`s is if we have messed up our logic for ensuring that const arguments are of the correct type. Having this not be an assert is:
- Confusing as it may incorrectly lead people to believe this is an important check that is actually required
- Opens the possibility for bugs or behaviour reliant on this (unnecessary) check existing
---
This PR makes two tests go from pass->ICE (`generic_const_exprs/ice-125520-layout-mismatch-mulwithoverflow.rs` and `tests/crashes/121858.rs`). This is caused by the fact that we evaluate anon consts even if their where clauses do not hold and is a pre-existing issue and only affects `generic_const_exprs`. I am comfortable exposing the brokenness of `generic_const_exprs` more with this PR
This PR makes a test go from ICE->pass (`const-generics/issues/issue-105821.rs`). I have no idea why this PR affects that but I believe that ICE is an unrelated issue to do with the fact that under `generic_const_exprs`/`adt_const_params` we do not handle lifetimes in const parameter types correctly. This PR is likely just masking this bug.
Note: this PR doesn't re-introduce the assertion that the two consts' tys are equal. I'm not really sure how I feel about this but tbh it has caused more ICEs than its found lately so 🤷♀️
r? `@oli-obk` `@compiler-errors`
Always use the `Fn(T) -> R` format when printing closure traits instead of `Fn<(T,), Output = R>`.
Fix#67100:
```
error[E0277]: expected a `Fn()` closure, found `F`
--> file.rs:6:13
|
6 | call_fn(f)
| ------- ^ expected an `Fn()` closure, found `F`
| |
| required by a bound introduced by this call
|
= note: wrap the `F` in a closure with no arguments: `|| { /* code */ }`
note: required by a bound in `call_fn`
--> file.rs:1:15
|
1 | fn call_fn<F: Fn() -> ()>(f: &F) {
| ^^^^^^^^^^ required by this bound in `call_fn`
help: consider further restricting this bound
|
5 | fn call_any<F: std::any::Any + Fn()>(f: &F) {
| ++++++
```
This commit refactors the `#[do_not_recommend]` support in the old
parser to also apply to projection errors and not only to selection
errors. This allows the attribute to be used more widely.
Tweak relations to no longer rely on `TypeTrace`
Remove `At::trace`, and inline all of the `Trace::equate`,etc methods into `At`.
The only nontrivial change is that we use `AliasTerm` to relate two unevaluated consts in the old-solver impl of `ConstEquate`, since `AliasTerm` does implement `ToTrace` and will relate the args structurally (shallowly).
r? lcnr
drop region constraints for ambiguous goals
See the comment in `compute_external_query_constraints`. While the underlying issue is preexisting, this fixes a bug introduced by #125343.
It slightly weakens the leak chec, even if we didn't have any test which was affected. I want to write such a test before merging this PR.
r? `@compiler-errors`
Uplift `EarlyBinder` into `rustc_type_ir`
We also need to give `EarlyBinder` a `'tcx` param, so that we can carry the `Interner` in the `EarlyBinder` too. This is necessary because otherwise we have an unconstrained `I: Interner` parameter in many of the `EarlyBinder`'s inherent impls.
I also generally think that this is desirable to have, in case we later want to track some state in the `EarlyBinder`.
r? lcnr
[perf] Delay the construction of early lint diag structs
Attacks some of the perf regressions from https://github.com/rust-lang/rust/pull/124417#issuecomment-2123700666.
See individual commits for details. The first three commits are not strictly necessary.
However, the 2nd one (06bc4fc671, *Remove `LintDiagnostic::msg`*) makes the main change way nicer to implement.
It's also pretty sweet on its own if I may say so myself.
Remove `DefId` from `EarlyParamRegion`
Currently we represent usages of `Region` parameters via the `ReEarlyParam` or `ReLateParam` variants. The `ReEarlyParam` is effectively equivalent to `TyKind::Param` and `ConstKind::Param` (i.e. it stores a `Symbol` and a `u32` index) however it also stores a `DefId` for the definition of the lifetime parameter.
This was used in roughly two places:
- Borrowck diagnostics instead of threading the appropriate `body_id` down to relevant locations. Interestingly there were already some places that had to pass down a `DefId` manually.
- Some opaque type checking logic was using the `DefId` field to track captured lifetimes
I've split this PR up into a commit for generate rote changes to diagnostics code to pass around a `DefId` manually everywhere, and another commit for the opaque type related changes which likely require more careful review as they might change the semantics of lints/errors.
Instead of manually passing the `DefId` around everywhere I previously tried to bundle it in with `TypeErrCtxt` but ran into issues with some call sites of `infcx.err_ctxt` being unable to provide a `DefId`, particularly places involved with trait solving and normalization. It might be worth investigating adding some new wrapper type to pass this around everywhere but I think this might be acceptable for now.
This pr also has the effect of reducing the size of `EarlyParamRegion` from 16 bytes -> 8 bytes. I wouldn't expect this to have any direct performance improvement however, other variants of `RegionKind` over `8` bytes are all because they contain a `BoundRegionKind` which is, as far as I know, mostly there for diagnostics. If we're ever able to remove this it would shrink the `RegionKind` type from `24` bytes to `12` (and with clever bit packing we might be able to get it to `8` bytes). I am curious what the performance impact would be of removing interning of `Region`'s if we ever manage to shrink `RegionKind` that much.
Sidenote: by removing the `DefId` the `Debug` output for `Region` has gotten significantly nicer. As an example see this opaque type debug print before vs after this PR:
`Opaque(DefId(0:13 ~ impl_trait_captures[aeb9]::foo::{opaque#0}), [DefId(0:9 ~ impl_trait_captures[aeb9]::foo::'a)_'a/#0, T, DefId(0:9 ~ impl_trait_captures[aeb9]::foo::'a)_'a/#0])`
`Opaque(DefId(0:13 ~ impl_trait_captures[aeb9]::foo::{opaque#0}), ['a/#0, T, 'a/#0])`
r? `@compiler-errors` (I would like someone who understands the opaque type setup to atleast review the type system commit, but the rest is likely reviewable by anyone)
remove proof tree formatting, make em shallow
Debugging via tracing `RUSTC_LOG=rustc_trait_selection::solve=debug` is now imo slightly more readable then the actual proof tree formatter. Removing everything that's not needed for the `analyse` visitor allows us to remove a bunch of code.
I personally believe that we should continue to use tracing over proof trees for debugging:
- it eagerly prints, allowing us to debug ICEs
- the proof tree builder ends up going out of sync with the actual runtime behavior, which is confusing
- using shallow proof trees is a lot more performant as we frequently do not recurse into all nested goals when using an analyse visitor
- this allows us to clean up the implementation and remove some code
r? ```@compiler-errors```
