Winnow specialized impls during selection in new solver
We need to be able to winnow impls that are specialized by more specific impls in order for codegen to be able to proceed.
r? ``@lcnr``
Move `ty::ConstKind` to `rustc_type_ir`
Needed this in another PR for custom debug impls, and this will also be required to move the new solver into a separate crate that does not use `TyCtxt` so that r-a and friends can depend on the trait solver.
Rebased on top of #113325, only the second and third commits needs reviewing
Add some extra information to opaque type cycle errors
Plus a bunch of cleanups.
This should help users debug query cycles due to auto trait checking. We'll probably want to fix cycle errors in most (or all?) cases by looking at the current item's hidden types (new solver does this), and by delaying the auto trait checks to after typeck.
`TypeParameterDefinition` always require a `DefId`
the `None` case never actually reaches diagnostics so it feels better for diagnostics to be able to rely on the `DefId` being there, cc #113310
add `ecx.probe_candidate`
Not yet changing the candidate source to an enum because that would be more involved, but this by itself should already be a significant improvement imo
r? `@BoxyUwU`
Rollup of 5 pull requests
Successful merges:
- #113192 (`assemble_candidates_after_normalizing_self_ty` docs)
- #113251 (Use scoped-tls for SMIR to map between TyCtxt and SMIR datastructures)
- #113282 (Update platform-support.md to improve ARM target descriptions)
- #113296 (add flag for enabling global cache usage for proof trees and printing proof trees on error)
- #113324 (implement `ConstEvaluatable` goals in new solver)
r? `@ghost`
`@rustbot` modify labels: rollup
implement `ConstEvaluatable` goals in new solver
this only supports stable const generics. `feature(generic_const_exprs)` needs to extend that function is non-trivial ways. Leaving this for someone else or some later date.
r? `@BoxyUwU`
add flag for enabling global cache usage for proof trees and printing proof trees on error
This adds a few new things:
- `-Zdump-solver-proof-tree=always/never/on-error`
- `always`/`never` were previosuly specifiable by whether the flag exists or not, th new flag is `on_error` which reruns obligations of fulfillment and selection errors with proof tree generation enabled and prints them out
- `-Zdump-solver-proof-tree-uses-cache`
- allows forcing global cache to be used or unused for all generated proof trees, global cache is enabled by default for `always` so that it accurately represents what happend. This flag currently would affect misc uses of `GenerateProofTree::Yes` which will be added in the future for things like diagnostics logic and rustdoc's auto_trait file. We can fix this when we start using proof tree generation for those use cases if it's desirable.
I also changed the output to go straight to stdout instead of going through `debug!` so that `-Zdump-solver-proof-tree` can be adequately used on `nightly` not just a locally built toolchain.
The idea for `on-error` is that it should hopefully make it easier to quickly figure out "why doesnt this code compile"- you just pass in `-Zdump-solver-proof-tree=on-error` and you'll only get proof trees you care about.
---
r? `@lcnr` `@compiler-errors`
`assemble_candidates_after_normalizing_self_ty` docs
I already explained that in different places a few times, should have added that explanation as a doc comment the first time I did so :3
r? `@BoxyUwU`
Normalize opaques with late-bound vars again
We have a hack in the compiler where if an opaque has escaping late-bound vars, we skip revealing it even though we *could* reveal it from a technical perspective. First of all, this is weird, since we really should be revealing all opaques in `Reveal::All` mode. Second of all, it causes subtle bugs (linked below).
I attempted to fix this in #100980, which was unfortunately reverted due to perf regressions on codebases that used really deeply nested futures in some interesting ways. The worst of which was #103423, which caused the project to hang on build. Another one was #104842, which was just a slow-down, but not a hang. I took some time afterwards to investigate how to rework `normalize_erasing_regions` to take advantage of better caching, but that effort kinda fizzled out (#104133).
However, recently, I was made aware of more bugs whose root cause is not revealing opaques during codegen. That made me want to fix this again -- in the process, interestingly, I took the the minimized example from https://github.com/rust-lang/rust/issues/103423#issuecomment-1292947043, and it doesn't seem to hang any more...
Thinking about this harder, there have been some changes to the way we lower and typecheck async futures that may have reduced the pathologically large number of outlives obligations (see description of #103423) that we were encountering when normalizing opaques with bound vars the last time around:
* #104321 (lower `async { .. }` directly as a generator that implements `Future`, removing the `from_generator` shim)
* #104833 (removing an `identity_future` fn that was wrapping desugared future generators)
... so given that I can see:
* No significant regression on rust perf bot (https://github.com/rust-lang/rust/pull/107620#issuecomment-1600070317)
* No timeouts in crater run I did (https://github.com/rust-lang/rust/pull/107620#issuecomment-1605428952, rechecked failing crates in https://github.com/rust-lang/rust/pull/107620#issuecomment-1605973434)
... and given that this PR:
* Fixes#104601
* Fixes#107557
* Fixes#109464
* Allows us to remove a `DefiningAnchor::Bubble` from codegen (75a8f68183)
I'm inclined to give this another shot at landing this. Best case, it just works -- worst case, we get more examples to study how we need to improve the compiler to make this work.
r? types
Don't emit same goal as input during `wf::unnormalized_obligations`
r? `@aliemjay` cc `@lcnr`
I accidentally pruned the logic to handle `WF(?0)` when writing `wf::unnormalized_obligations`.
idk if you wanted to construct a test first, but this is an obvious fix. Copied the comment from above.
Fixesrust-lang/trait-system-refactor-initiative#36
Migrate `item_bounds` to `ty::Clause`
Should be simpler than the next PR that's coming up. Last three commits are the relevant ones.
r? ``@oli-obk`` or ``@lcnr``
When a trait is used without specifying the implementation (e.g. calling
a non-member associated function without fully-qualified syntax) and
there are multiple implementations available, use a placeholder comment
for the implementation type in the suggestion instead of picking a
random implementation.
Example:
```
fn main() {
let _ = Default::default();
}
```
Previous output:
```
error[E0790]: cannot call associated function on trait without specifying the corresponding `impl` type
--> test.rs:2:13
|
2 | let _ = Default::default();
| ^^^^^^^^^^^^^^^^ cannot call associated function of trait
|
help: use a fully-qualified path to a specific available implementation (273 found)
|
2 | let _ = <FileTimes as Default>::default();
| +++++++++++++ +
```
New output:
```
error[E0790]: cannot call associated function on trait without specifying the corresponding `impl` type
--> test.rs:2:13
|
2 | let _ = Default::default();
| ^^^^^^^^^^^^^^^^ cannot call associated function of trait
|
help: use a fully-qualified path to a specific available implementation (273 found)
|
2 | let _ = </* self type */ as Default>::default();
| +++++++++++++++++++ +
```
Account for sealed traits in privacy and trait bound errors
On trait bound errors caused by super-traits, identify if the super-trait is publicly accessibly and if not, explain "sealed traits".
```
error[E0277]: the trait bound `S: Hidden` is not satisfied
--> $DIR/sealed-trait-local.rs:17:20
|
LL | impl a::Sealed for S {}
| ^ the trait `Hidden` is not implemented for `S`
|
note: required by a bound in `Sealed`
--> $DIR/sealed-trait-local.rs:3:23
|
LL | pub trait Sealed: self:🅱️:Hidden {
| ^^^^^^^^^^^^^^^ required by this bound in `Sealed`
= note: `Sealed` is a "sealed trait", because to implement it you also need to implelement `a:🅱️:Hidden`, which is not accessible; this is usually done to force you to use one of the provided types that already implement it
```
Deduplicate privacy errors that point to the same path segment even if their deduplication span are different.
When encountering a path that is not reachable due to privacy constraints path segments other than the last, keep metadata for the last path segment's `Res` in order to look for alternative import paths for that item to suggest. If there are none, be explicit that the item is not accessible.
```
error[E0603]: module `b` is private
--> $DIR/re-exported-trait.rs:11:9
|
LL | impl a:🅱️:Trait for S {}
| ^ private module
|
note: the module `b` is defined here
--> $DIR/re-exported-trait.rs:5:5
|
LL | mod b {
| ^^^^^
help: consider importing this trait through its public re-export instead
|
LL | impl a::Trait for S {}
| ~~~~~~~~
```
```
error[E0603]: module `b` is private
--> $DIR/private-trait.rs:8:9
|
LL | impl a:🅱️:Hidden for S {}
| ^ ------ trait `b` is not publicly reachable
| |
| private module
|
note: the module `b` is defined here
--> $DIR/private-trait.rs:2:5
|
LL | mod b {
| ^^^^^
```
When implementing a public trait with a private super-trait, we now emit
a note that the missing bound is not going to be able to be satisfied,
and we explain the concept of a sealed trait.
Add a fully fledged `Clause` type, rename old `Clause` to `ClauseKind`
Does two basic things before I put up a more delicate set of PRs (along the lines of #112714, but hopefully much cleaner) that migrate existing usages of `ty::Predicate` to `ty::Clause` (`predicates_of`/`item_bounds`/`ParamEnv::caller_bounds`).
1. Rename `Clause` to `ClauseKind`, so it's parallel with `PredicateKind`.
2. Add a new `Clause` type which is parallel to `Predicate`.
* This type exposes `Clause::kind(self) -> Binder<'tcx, ClauseKind<'tcx>>` which is parallel to `Predicate::kind` 😸
The new `Clause` type essentially acts as a newtype wrapper around `Predicate` that asserts that it is specifically a `PredicateKind::Clause`. Turns out from experimentation[^1] that this is not negative performance-wise, which is wonderful, since this a much simpler design than something that requires encoding the discriminant into the alignment bits of a predicate kind, or something else like that...
r? ``@lcnr`` or ``@oli-obk``
[^1]: https://github.com/rust-lang/rust/pull/112714#issuecomment-1595653910
The only regression is one ambiguity in the new trait solver, having to
do with two param-env candidates that may apply. I think this is fine,
since the error message already kinda sucks.
Add `implement_via_object` to `rustc_deny_explicit_impl` to control object candidate assembly
Some built-in traits are special, since they are used to prove facts about the program that are important for later phases of compilation such as codegen and CTFE. For example, the `Unsize` trait is used to assert to the compiler that we are able to unsize a type into another type. It doesn't have any methods because it doesn't actually *instruct* the compiler how to do this unsizing, but this is later used (alongside an exhaustive match of combinations of unsizeable types) during codegen to generate unsize coercion code.
Due to this, these built-in traits are incompatible with the type erasure provided by object types. For example, the existence of `dyn Unsize<T>` does not mean that the compiler is able to unsize `Box<dyn Unsize<T>>` into `Box<T>`, since `Unsize` is a *witness* to the fact that a type can be unsized, and it doesn't actually encode that unsizing operation in its vtable as mentioned above.
The old trait solver gets around this fact by having complex control flow that never considers object bounds for certain built-in traits:
2f896da247/compiler/rustc_trait_selection/src/traits/select/candidate_assembly.rs (L61-L132)
However, candidate assembly in the new solver is much more lovely, and I'd hate to add this list of opt-out cases into the new solver. Instead of maintaining this complex and hard-coded control flow, instead we can make this a property of the trait via a built-in attribute. We already have such a build attribute that's applied to every single trait that we care about: `rustc_deny_explicit_impl`. This PR adds `implement_via_object` as a meta-item to that attribute that allows us to opt a trait out of object-bound candidate assembly as well.
r? `@lcnr`
Don't consider TAIT normalizable to hidden ty if it would result in impossible item bounds
See test for example where we shouldn't consider it possible to alias-relate a TAIT and hidden type.
r? `@lcnr`
Don't ICE on bound var in `reject_fn_ptr_impls`
We may try to use an impl like `impl<T: FnPtr> PartialEq {}` to satisfy a predicate like `for<T> T: PartialEq` -- don't ICE in that case.
Fixes#112735
Continue folding in query normalizer on weak aliases
Fixes#112752Fixes#112731 (same root cause, so didn't make a test for it)
fixes#112776
r? ```@oli-obk```
Add `AliasKind::Weak` for type aliases.
`type Foo<T: Debug> = Bar<T>;` does not check `T: Debug` at use sites of `Foo<NotDebug>`, because in contrast to a
```rust
trait Identity {
type Identity;
}
impl<T: Debug> Identity for T {
type Identity = T;
}
<NotDebug as Identity>::Identity
```
type aliases do not exist in the type system, but are expanded to their aliased type immediately when going from HIR to the type layer.
Similarly:
* a private type alias for a public type is a completely fine thing, even though it makes it a bit hard to write out complex times sometimes
* rustdoc expands the type alias, even though often times users use them for documentation purposes
* diagnostics show the expanded type, which is confusing if the user wrote a type alias and the diagnostic talks about another type that they don't know about.
For type alias impl trait, these issues do not actually apply in most cases, but sometimes you have a type alias impl trait like `type Foo<T: Debug> = (impl Debug, Bar<T>);`, which only really checks it for `impl Debug`, but by accident prevents `Bar<T>` from only being instantiated after proving `T: Debug`. This PR makes sure that we always check these bounds explicitly and don't rely on an implementation accident.
To not break all the type aliases out there, we only use it when the type alias contains an opaque type. We can decide to do this for all type aliases over an edition.
Or we can later extend this to more types if we figure out the back-compat concerns with suddenly checking such bounds.
As a side effect, easily allows fixing https://github.com/rust-lang/rust/issues/108617, which I did.
fixes https://github.com/rust-lang/rust/issues/108617
Make assumption functions in new solver take `Binder<'tcx, Clause<'tcx>>`
We just use an if-let to match on an optional clause at all the places where we transition from `Predicate` -> `Clause`, but I assume that when things like item-bounds and param-env start to only store `Clause`s then those can just be trivially dropped.
r? ``@lcnr``
Opportunistically resolve regions in new solver
Use `opportunistic_resolve_var` during canonicalization to collapse some regions.
We have to start using `CanonicalVarValues::is_identity_modulo_regions`. We also have to modify that function to consider responses like `['static, ^0, '^1, ^2]` to be an "identity" response, since because we opportunistically resolve regions, there's no longer a 1:1 mapping between canonical var values and bound var indices in the response...
There's one nasty side-effect -- one test (`tests/ui/dyn-star/param-env-infer.rs`) starts to ICE because the certainty goes from `Yes` to `Maybe(Overflow)`... Not exactly sure why, though? Putting this up for discussion/investigation.
r? ```@lcnr```
Improve docs/clean up negative overlap functions
Clean up some functions in ways that should not affect behavior, change some names to be clearer (`negative_impl` and `implicit_negative` are not really clear imo), and add some documentation examples.
r? `@spastorino`
Collect VTable stats & add `-Zprint-vtable-sizes`
This is a bit hacky/buggy, but I'm not entirely sure how to fix it, so I want to ask reviewers for help...
To try this, use either of those:
- `cargo clean && RUSTFLAGS="-Zprint-vtable-sizes" cargo +toolchain b`
- `cargo clean && cargo rustc +toolchain -Zprint-vtable-sizes`
- `rustc +toolchain -Zprint-vtable-sizes ./file.rs`
Safe Transmute: Enable handling references
This patch enables support for references in Safe Transmute, by generating nested obligations during trait selection. Specifically, when we call `confirm_transmutability_candidate(...)`, we now recursively traverse the `rustc_transmute::Answer` tree and create obligations for all the `Answer` variants, some of which include multiple nested `Answer`s.
- Create `Answer` type that is not just a type alias of `Result`
- Remove a usage of `map_layouts` to make the code easier to read
- Don't hide errors related to Unknown Layout when computing transmutability
Add `-Ztrait-solver=next-coherence`
Flag that conditionally uses the trait solver *only* during coherence, for more testing and/or eventual partial-migration onto the trait solver (in the medium- to long-term).
* This still uses the selection context in some of the coherence methods I think, so it's not "complete". Putting this up for review and/or for further work in-tree.
* I probably need to spend a bit more time making sure that we don't sneakily create any other infcx's during coherence that also need the new solver enabled.
r? `@lcnr`
Fall back to bidirectional normalizes-to if no subst-relate candidate in alias-relate goal
Sometimes we get into the case where the choice of normalizes-to branch in alias-relate are both valid, but we cannot make a choice of which one to take because they are different -- either returning equivalent but permuted region constraints, or equivalent opaque type definitions but differing modulo normalization.
In this case, we can make progress by considering a fourth candidate where we compute both normalizes-to branches together and canonicalize that as a response. This is essentially the AND intersection of both normalizes-to branches. In an ideal world, we'd be returning something more like the OR intersection of both branches, but we have no way of representing that either for regions (maybe eventually) or opaques (don't see that happening ever).
This is incomplete, so like the subst-relate fallback it's only considered outside of coherence. But it doesn't seem like a dramatic strengthening of inference or anything, and is useful for helping opaque type inference succeed when the hidden type is a projection.
## Example
Consider the goal - `AliasRelate(Tait, <[i32; 32] as IntoIterator>::IntoIter)`.
We have three ways of currently solving this goal:
1. SubstRelate - fails because we can't directly equate the substs of different alias kinds.
2. NormalizesToRhs - `Tait normalizes-to <[i32; 32] as IntoIterator>::IntoIter`
* Ends up infering opaque definition - `Tait := <[i32; 32] as IntoIterator>::IntoIter`
3. NormalizesToLhs - `<[i32; 32] as IntoIterator>::IntoIter normalizes-to Tait`
* Find impl candidate, substitute the associated type - `std::array::IntoIter<i32, 32>`
* Equate `std::array::IntoIter<i32, 32>` and `Tait`
* Ends up infering opaque definition - `Tait := std::array::IntoIter<i32, 32>`
The problem here is that 2 and 3 are essentially both valid, since we have aliases that normalize on both sides, but due to lazy norm, they end up inferring different opaque type definitions that are only equal *after* normalizing them further.
---
r? `@lcnr`
Merge method, type and const object safety checks
cc `@spastorino` and `@compiler-errors` on the first commit. I believe it to be correct, as the field is only `Some` for assoc types, so just checking the field without checking the assoc kind to be `Type` is fine.
The second commit avoids going through all associated items thrice and just goes over all of them once, running the object safety checks per assoc item kind.
Normalize in infcx instead of globally for `Option::as_deref` suggestion
fixes#112293
The projection may contain inference variables. These inference variables are local to the local inference context. Using `tcx.normalize_erasing_regions` doesn't work here because this method is global and does not have access to the inference context. It's therefore unable to deal with the inference variables. We normalize in the local inference context instead, which knowns about the inference variables.
The test looks a little different than the issue example, I made it more minimal and verified that it still ICEs on nightly.
Also contains a drive-by fix to properly compare the types.
r? `@compiler-errors`
The projection may contain inference variables. These inference
variables are local to the local inference context. Using
`tcx.normalize_erasing_regions` doesn't work here because this method is
global and does not have access to the inference context. It's therefore
unable to deal with the inference variables. We normalize in the local
inference context instead, which knowns about the inference variables.
suggest `Option::as_deref(_mut)` on type mismatch in option combinator if it passes typeck
Fixes#106342.
This adds a suggestion to call `.as_deref()` (or `.as_deref_mut()` resp.) if typeck fails due to a type mismatch in the function passed to an `Option` combinator such as `.map()` or `.and_then()`.
For example:
```rs
fn foo(_: &str) {}
Some(String::new()).map(foo);
```
The `.map()` method requires its argument to satisfy `F: FnOnce(String)`, but it received `fn(&str)`, which won't pass. However, placing a `.as_deref()` before the `.map()` call fixes this since `&str == &<String as Deref>::Target`
refactor and cleanup the leak check, add it to new solver
ended up being a bit more involved than I wanted but is hopefully still easy enough to review as a single PR, can split it into separate ones otherwise.
this can be reviewed commit by commit:
a473d55cdb9284aa2b01282d1b529a2a4d26547b 31a686646534ca006d906ec757ece4e771d6f973 949039c107852a5e36361c08b62821a0613656f5 242917bf5170d9a723c6c8e23e9d9d0c2fa8dc9d ed2b25a7aa28be3184be9e3022c2796a30eaad87 are all pretty straightforward.
03dd83b4c3f4ff27558f5c8ab859bd9f83db1d04 makes it easier to refactor coherence in a later commit, see the commit description, cc `@oli-obk`
4fe311d807a77b6270f384e41689bf5d58f46aec I don't quite remember what we wanted to test here, this definitely doesn't test that the occurs check doesn't cause incorrect errors in coherence, also cc `@oli-obk` here. I may end up writing a new test for this myself later.
5c200d88a91b75bd0875b973150655bd581ef97a is the main refactor of the leak check, changing it to take the `outer_universe` instead of getting it from a snapshot. Using a snapshot requires us to be in a probe which we aren't in the new solver, it also just feels dirty as snapshots don't really have anything to do with universes.
with all of this cfc230d54188d9c7ed867a9a0d1f51be77b485f9 is now kind of trivial.
r? `@nikomatsakis`
`EarlyBinder::new` -> `EarlyBinder::bind`
for consistency with `Binder::bind`. it may make sense to also add `EarlyBinder::dummy` in places where we know that no parameters exist, but I left that out of this PR.
r? `@jackh726` `@kylematsuda`
Each of `{D,Subd}iagnosticMessage::{Str,Eager}` has a comment:
```
// FIXME(davidtwco): can a `Cow<'static, str>` be used here?
```
This commit answers that question in the affirmative. It's not the most
compelling change ever, but it might be worth merging.
This requires changing the `impl<'a> From<&'a str>` impls to `impl
From<&'static str>`, which involves a bunch of knock-on changes that
require/result in call sites being a little more precise about exactly
what kind of string they use to create errors, and not just `&str`. This
will result in fewer unnecessary allocations, though this will not have
any notable perf effects given that these are error paths.
Note that I was lazy within Clippy, using `to_string` in a few places to
preserve the existing string imprecision. I could have used `impl
Into<{D,Subd}iagnosticMessage>` in various places as is done in the
compiler, but that would have required changes to *many* call sites
(mostly changing `&format("...")` to `format!("...")`) which didn't seem
worthwhile.
Enable MatchBranchSimplification
This pass is one of the small number of benefits from `-Zmir-opt-level=3` that has motivated rustc_codegen_cranelift to use it:
19ed0aade6/compiler/rustc_codegen_cranelift/build_system/build_sysroot.rs (L244-L246)
Cranelift's motivation for this is _runtime_ performance improvements in debug builds. Lifting this pass all the way to `-Zmir-opt-level=1` seems to come without significant perf overhead, so that's what I'm suggesting here.
Add warn-by-default lint when local binding shadows exported glob re-export item
This PR introduces a warn-by-default rustc lint for when a local binding (a use statement, or a type declaration) produces a name which shadows an exported glob re-export item, causing the name from the exported glob re-export to be hidden (see #111336).
### Unresolved Questions
- [x] ~~Is this approach correct? While it passes the UI tests, I'm not entirely convinced it is correct.~~ Seems to be ok now.
- [x] ~~What should the lint be called / how should it be worded? I don't like calling `use x::*;` or `struct Foo;` a "local binding" but they are `NameBinding`s internally if I'm not mistaken.~~ ~~The lint is called `local_binding_shadows_glob_reexport` for now, unless a better name is suggested.~~ `hidden_glob_reexports`.
Fixes#111336.
- Change `Condition` to not contain `Answer`s but instead just contain other
`Condition`s directly.
- Also improve error reporting for `DstHasStricterAlignment`
Use `Option::is_some_and` and `Result::is_ok_and` in the compiler
`.is_some_and(..)`/`.is_ok_and(..)` replace `.map_or(false, ..)` and `.map(..).unwrap_or(false)`, making the code more readable.
This PR is a sibling of https://github.com/rust-lang/rust/pull/111873#issuecomment-1561316515
- Only create dst -> src obligation if Dst is mutable
- Add some long comments to explain parts of the transmutability code that were
unclear to me when reading
- Update/add tests
- Make sure that the most specific Reason is the one that bubbles up when we
are folding over the `Answer` tree. `Reason::DstIsBitIncompatible` is the
least specific, so that should be used only when there isn't anything else
available.
- Small fixes where we used the wrong Reason variant.
- Tiny cleanups
This patch updates the `Answer` type from `rustc_transmute` so that it just a
type alias to `Result`. This makes it so that the standard methods for `Result`
can be used to process the `Answer` tree, including being able to make use of
the `?` operator on `Answer`s.
Also, remove some unused functions
This patch enables support for references in Safe Transmute, by generating
nested obligations during trait selection. Specifically, when we call
`confirm_transmutability_candidate(...)`, we now recursively traverse the
`rustc_transmute::Answer` tree and create obligations for all the `Answer`
variants, some of which include multiple nested `Answer`s.
Also, to handle recursive types, enable support for coinduction for the Safe
Transmute trait (`BikeshedIntrinsicFrom`) by adding the `#[rustc_coinduction]`
annotation.
Also fix some small logic issues when reducing the `or` and `and` combinations
in `rustc_transmute`, so that we don't end up with additional redundant
`Answer`s in the tree.
Co-authored-by: Jack Wrenn <jack@wrenn.fyi>
Remove return type sized check hack from hir typeck
Remove a bunch of special-cased suggestions when someone returns `-> dyn Trait` that checks for type equality, etc.
This was a pretty complex piece of code that also relied on a hack in hir typeck (see changes to `compiler/rustc_hir_typeck/src/check.rs`), and I'm not convinced that it's necessary to maintain, when all we really need to tell the user is that they should return `-> impl Trait` or `-> Box<dyn Trait>`, depending on their specific use-case.
This is necessary because we may need to move the "return type is sized" check from hir typeck to wfcheck, which does not have access to typeck results. This is a prerequisite for that, and I'm fairly confident that the diagnostics "regressions" here are not a big deal.
Deal with unnormalized projections when structurally resolving types with new solver
1. Normalize types in `structurally_resolved_type` when the new solver is enabled
2. Normalize built-in autoderef targets in `Autoderef` when the new solver is enabled
3. Normalize-erasing-regions in `resolve_type` in writeback
This is motivated by the UI test provided, which currently fails with:
```
error[E0609]: no field `x` on type `<usize as SliceIndex<[Foo]>>::Output`
--> <source>:9:11
|
9 | xs[0].x = 1;
| ^
```
I'm pretty happy with the approach in (1.) and (2.) and think we'll inevitably need something like this in the long-term, but (3.) seems like a hack to me. It's a *lot* of work to add tons of new calls to every user of these typeck results though (mir build, late lints, etc). Happy to discuss further.
r? `@lcnr`
Give better error when collecting into `&[T]`
The detection of slice reference of `{integral}` in `rustc_on_unimplemented` is hacky, but a proper solution requires changing `FmtPrinter` to add a parameter to print integers as `{integral}` and I didn't want to change it just for `rustc_on_unimplemented`. I can do that if requested, though.
I'm open to better wording; this is the best I could come up with.
do not allow inference in `predicate_must_hold` (alternative approach)
See the FCP description for more info, but tl;dr is that we should not return `EvaluatedToOkModuloRegions` if an obligation may hold only with some choice of inference vars being constrained.
Attempts to solve this in the approach laid out by lcnr here: https://github.com/rust-lang/rust/pull/109558#discussion_r1147318134, rather than by eagerly replacing infer vars with placeholders which is a bit too restrictive.
r? `@ghost`
Exclude inherent projections from some alias type `match`es
Updating (hopefully) all remaining `match`es which I overlooked to update when adding `AliasKind::Inherent` in #109410.
Fixes#111399.
Sadly the regression test is a clippy test instead of a rustc one as I don't know of another way to test that a trait bound like `Ty::InhProj: Trait` doesn't cause a crash without reaching a cycle error first (this is getting old ^^').
`@rustbot` label F-inherent_associated_types
r? `@compiler-errors`
Handle error body in generator layout
Fixes#111468
I feel like making this query return `Option<GeneratorLayout>` might be better but had some issues with that approach
Suppress "erroneous constant used" for constants tainted by errors
When constant evaluation fails because its MIR is tainted by errors,
suppress note indicating that erroneous constant was used, since those
errors have to be fixed regardless of the constant being used or not.
Fixes#110891.
Error message all end up passing into a function as an `impl
Into<{D,Subd}iagnosticMessage>`. If an error message is creatd as
`&format("...")` that means we allocate a string (in the `format!`
call), then take a reference, and then clone (allocating again) the
reference to produce the `{D,Subd}iagnosticMessage`, which is silly.
This commit removes the leading `&` from a lot of these cases. This
means the original `String` is moved into the
`{D,Subd}iagnosticMessage`, avoiding the double allocations. This
requires changing some function argument types from `&str` to `String`
(when all arguments are `String`) or `impl
Into<{D,Subd}iagnosticMessage>` (when some arguments are `String` and
some are `&str`).
When constant evaluation fails because its MIR is tainted by errors,
suppress note indicating that erroneous constant was used, since those
errors have to be fixed regardless of the constant being used or not.
Shrink `SelectionError` a lot
`SelectionError` used to be 80 bytes (on 64 bit). That's quite big. Especially because the selection cache contained `Result<_, SelectionError>. The Ok type is only 32 bytes, so the 80 bytes significantly inflate the size of the cache.
Most variants of the `SelectionError` seem to be hard errors, only `Unimplemented` shows up in practice (for cranelift-codegen, it occupies 23.4% of all cache entries). We can just box away the biggest variant, `OutputTypeParameterMismatch`, to get the size down to 16 bytes, well within the size of the Ok type inside the cache.
This trait ref is derived from the self type and then equated to the
trait ref from the obligation.
For example, for `fn(): Fn(u32)`, `self_ty_trait_ref` is `Fn()`, which
is then equated to `Fn(u32)` (which will fail, causing the obligation to
fail).
`SelectionError` used to be 80 bytes (on 64 bit). That's quite big.
Especially because the selection cache contained `Result<_,
SelectionError>. The Ok type is only 32 bytes, so the 80 bytes
significantly inflate the size of the cache.
Most variants of the `SelectionError` seem to be hard errors, only
`Unimplemented` shows up in practice (for cranelift-codegen, it occupies
23.4% of all cache entries). We can just box away the biggest variant,
`OutputTypeParameterMismatch`, to get the size down to 16 bytes, well
within the size of the Ok type inside the cache.
Min specialization improvements
- Don't allow specialization impls with no items, such implementations are probably not correct and only occur as mistakes in the compiler and standard library
- Fix a missing normalization call
- Adds spans for lifetime errors from overly general specializations
Closes#79457Closes#109815
Introduce `AliasKind::Inherent` for inherent associated types
Allows us to check (possibly generic) inherent associated types for well-formedness.
Type inference now also works properly.
Follow-up to #105961. Supersedes #108430.
Fixes#106722.
Fixes#108957.
Fixes#109768.
Fixes#109789.
Fixes#109790.
~Not to be merged before #108860 (`AliasKind::Weak`).~
CC `@jackh726`
r? `@compiler-errors`
`@rustbot` label T-types F-inherent_associated_types
Support return-type bounds on associated methods from supertraits
Support `T: Trait<method(): Bound>` when `method` comes from a supertrait, aligning it with the behavior of associated type bounds (both equality and trait bounds).
The only wrinkle is that I have to extend `super_predicates_that_define_assoc_type` to look for *all* items, not just `AssocKind::Ty`. This will also be needed to support `feature(associated_const_equality)` as well, which is subtly broken when it comes to supertraits, though this PR does not fix those yet. There's a slight chance there's a perf regression here, in which case I guess I could split it out into a separate query.
Use fulfillment to check `Drop` impl compatibility
Use an `ObligationCtxt` to ensure that a `Drop` impl does not have stricter requirements than the ADT that it's implemented for, rather than using a `SimpleEqRelation` to (more or less) syntactically equate predicates on an ADT with predicates on an impl.
r? types
### Some background
The old code reads:
```rust
// An earlier version of this code attempted to do this checking
// via the traits::fulfill machinery. However, it ran into trouble
// since the fulfill machinery merely turns outlives-predicates
// 'a:'b and T:'b into region inference constraints. It is simpler
// just to look for all the predicates directly.
```
I'm not sure what this means, but perhaps in the 8 years since that this comment was written (cc #23638) it's gotten easier to process region constraints after doing fulfillment? I don't know how this logic differs from anything we do in the `compare_impl_item` module. Ironically, later on it says:
```rust
// However, it may be more efficient in the future to batch
// the analysis together via the fulfill (see comment above regarding
// the usage of the fulfill machinery), rather than the
// repeated `.iter().any(..)` calls.
```
Also:
* Removes `SimpleEqRelation` which was far too syntactical in its relation.
* Fixes#110557
Implement negative bounds for internal testing purposes
Implements partial support the `!` negative polarity on trait bounds. This is incomplete, but should allow us to at least be able to play with the feature.
Not even gonna consider them as a public-facing feature, but I'm implementing them because would've been nice to have in UI tests, for example in #110671.
Currently a `{D,Subd}iagnosticMessage` can be created from any type that
impls `Into<String>`. That includes `&str`, `String`, and `Cow<'static,
str>`, which are reasonable. It also includes `&String`, which is pretty
weird, and results in many places making unnecessary allocations for
patterns like this:
```
self.fatal(&format!(...))
```
This creates a string with `format!`, takes a reference, passes the
reference to `fatal`, which does an `into()`, which clones the
reference, doing a second allocation. Two allocations for a single
string, bleh.
This commit changes the `From` impls so that you can only create a
`{D,Subd}iagnosticMessage` from `&str`, `String`, or `Cow<'static,
str>`. This requires changing all the places that currently create one
from a `&String`. Most of these are of the `&format!(...)` form
described above; each one removes an unnecessary static `&`, plus an
allocation when executed. There are also a few places where the existing
use of `&String` was more reasonable; these now just use `clone()` at
the call site.
As well as making the code nicer and more efficient, this is a step
towards possibly using `Cow<'static, str>` in
`{D,Subd}iagnosticMessage::{Str,Eager}`. That would require changing
the `From<&'a str>` impls to `From<&'static str>`, which is doable, but
I'm not yet sure if it's worthwhile.
Add `ConstParamTy` trait
This is a bit sketch, but idk.
r? `@BoxyUwU`
Yet to be done:
- [x] ~~Figure out if it's okay to implement `StructuralEq` for primitives / possibly remove their special casing~~ (it should be okay, but maybe not in this PR...)
- [ ] Maybe refactor the code a little bit
- [x] Use a macro to make impls a bit nicer
Future work:
- [ ] Actually™ use the trait when checking if a `const` generic type is allowed
- [ ] _Really_ refactor the surrounding code
- [ ] Refactor `marker.rs` into multiple modules for each "theme" of markers
Tweak await span to not contain dot
Fixes a discrepancy between method calls and await expressions where the latter are desugared to have a span that *contains* the dot (i.e. `.await`) but method call identifiers don't contain the dot. This leads to weird suggestions suggestions in borrowck -- see linked issue.
Fixes#110761
This mostly touches a bunch of tests to tighten their `await` span.
Clear response values for overflow in new solver
When we have an overflow, return a trivial query response. This fixes an ICE with the code described in #110544:
```rust
trait Trait {}
struct W<T>(T);
impl<T, U> Trait for W<(W<T>, W<U>)>
where
W<T>: Trait,
W<U>: Trait,
{}
fn impls<T: Trait>() {}
fn main() {
impls::<W<_>>()
}
```
Where, while proving `W<?0>: Trait`, we overflow but still apply the query response of `?0 = (W<?1>, W<?2>)`. Then while re-processing the query to validate that our evaluation result was stable, we get a different query response that looks like `?1 = (W<?3>, W<?4>), ?2 = (W<?5>, W<?6>)`, and so we trigger the ICE.
Also, by returning a trivial query response we also avoid the infinite-loop/OOM behavior of the old solver.
r? ``@lcnr``
Consider polarity in new solver
It's kinda ugly to have a polarity check in all of the builtin impls -- I guess I could consider the polarity at the top of assemble-builtin but that would require adding a polarity fn to `GoalKind`...
🤷 putting this up just so i dont forget, since it's needed to bootstrap core during coherence (this alone does not allow core to bootstrap though, additional work is needed!)
r? ``@lcnr``
Switch to `EarlyBinder` for `explicit_item_bounds`
Part of the work to finish https://github.com/rust-lang/rust/issues/105779.
This PR adds `EarlyBinder` to the return type of the `explicit_item_bounds` query and removes `bound_explicit_item_bounds`.
r? `@compiler-errors` (hope it's okay to request you, since you reviewed #110299 and #110498😃)
Break up long function in trait selection error reporting + clean up nearby code
- Move blocks of code into their own functions
- Replace a few function argument types with their type aliases
- Create "AppendConstMessage" enum to replace a nested `Option`.
Allow to feed a value in another query's cache and remove `WithOptConstParam`
I used it to remove `WithOptConstParam` queries, as an example.
The idea is that a query (here `typeck(function)`) can write into another query's cache (here `type_of(anon const)`). The dependency node for `type_of` would depend on all the current dependencies of `typeck`.
There is still an issue with cycles: if `type_of(anon const)` is accessed before `typeck(function)`, we will still have the usual cycle. The way around this issue is to `ensure` that `typeck(function)` is called before accessing `type_of(anon const)`.
When replayed, we may the following cases:
- `typeck` is green, in that case `type_of` is green too, and all is right;
- `type_of` is green, `typeck` may still be marked as red (it depends on strictly more things than `type_of`) -> we verify that the saved value and the re-computed value of `type_of` have the same hash;
- `type_of` is red, then `typeck` is red -> it's the caller responsibility to ensure `typeck` is recomputed *before* `type_of`.
As `anon consts` have their own `DefPathData`, it's not possible to have the def-id of the anon-const point to something outside the original function, but the general case may have to be resolved before using this device more broadly.
There is an open question about loading from the on-disk cache. If `typeck` is loaded from the on-disk cache, the side-effect does not happen. The regular `type_of` implementation can go and fetch the correct value from the decoded `typeck` results, and the dep-graph will check that the hashes match, but I'm not sure we want to rely on this behaviour.
I specifically allowed to feed the value to `type_of` from inside a call to `type_of`. In that case, the dep-graph will check that the fingerprints of both values match.
This implementation is still very sensitive to cycles, and requires that we call `typeck(function)` before `typeck(anon const)`. The reason is that `typeck(anon const)` calls `type_of(anon const)`, which calls `typeck(function)`, which feeds `type_of(anon const)`, and needs to build the MIR so needs `typeck(anon const)`. The latter call would not cycle, since `type_of(anon const)` has been set, but I'd rather not remove the cycle check.
Add `rustc_fluent_macro` to decouple fluent from `rustc_macros`
Fluent, with all the icu4x it brings in, takes quite some time to compile. `fluent_messages!` is only needed in further downstream rustc crates, but is blocking more upstream crates like `rustc_index`. By splitting it out, we allow `rustc_macros` to be compiled earlier, which speeds up `x check compiler` by about 5 seconds (and even more after the needless dependency on `serde_json` is removed from `rustc_data_structures`).
Switch to `EarlyBinder` for `collect_return_position_impl_trait_in_trait_tys`
Part of the work to finish https://github.com/rust-lang/rust/issues/105779.
This PR adds `EarlyBinder` to the return type of the `collect_return_position_impl_trait_in_trait_tys` query and removes `bound_return_position_impl_trait_in_trait_tys`.
r? `@lcnr`
Fluent, with all the icu4x it brings in, takes quite some time to
compile. `fluent_messages!` is only needed in further downstream rustc
crates, but is blocking more upstream crates like `rustc_index`. By
splitting it out, we allow `rustc_macros` to be compiled earlier, which
speeds up `x check compiler` by about 5 seconds (and even more after the
needless dependency on `serde_json` is removed from
`rustc_data_structures`).
Various minor Idx-related tweaks
Nothing particularly exciting here, but a couple of things I noticed as I was looking for more index conversions to simplify.
cc https://github.com/rust-lang/compiler-team/issues/606
r? `@WaffleLapkin`
Remove `TypeSuper{Foldable,Visitable}` impls for `Region`.
These traits exist so that folders/visitors can recurse into types of interest: binders, types, regions, predicates, and consts. But `Region` is non-recursive and cannot contain other types of interest, so its methods in these traits are trivial.
This commit inlines and removes those trivial methods.
r? `@compiler-errors`
Remove `remap_env_constness` in queries
This removes some of the complexities with const traits. #88119 used to be caused by this but was fixed by `param_env = param_env.without_const()`.
These traits exist so that folders/visitors can recurse into types of
interest: binders, types, regions, predicates, and consts. But `Region`
is non-recursive and cannot contain other types of interest, so its
methods in these traits are trivial.
This commit inlines and removes those trivial methods.
Switch to `EarlyBinder` for `impl_subject` query
Part of the work to finish https://github.com/rust-lang/rust/issues/105779.
Several queries `X` have a `bound_X` variant that wraps the output in `EarlyBinder`. This adds `EarlyBinder` to the return type of the `impl_subject` query and removes `bound_impl_subject`.
r? ```@lcnr```
don't uniquify regions when canonicalizing
uniquifying causes a bunch of issues, most notably it causes `AliasEq(<?x as Trait<'a>>::Assoc, <?x as Trait<'a>>::Assoc)` to result in ambiguity because both `normalizes-to` paths result in ambiguity and substs equate should trivially succeed but doesn't because we uniquified `'a` to two different regions.
I originally added uniquification to make it easier to deal with requirement 6 from the dev-guide: https://rustc-dev-guide.rust-lang.org/solve/trait-solving.html#requirements
> ### 6. Trait solving must be (free) lifetime agnostic
>
> Trait solving during codegen should have the same result as during typeck. As we erase
> all free regions during codegen we must not rely on them during typeck. A noteworthy example
> is special behavior for `'static`.
cc https://github.com/rust-lang/rustc-dev-guide/pull/1671
Relying on regions being identical may cause ICE during MIR typeck, but even without this PR we can end up relying on that as type inference vars can resolve to types which contain an identical region. Let's land this and deal with any ICE that crop up as we go. Will look at this issue again before stabilization.
r? ```@compiler-errors```
Improve safe transmute error reporting
This patch updates the error reporting when Safe Transmute is not possible between 2 types by including the reason.
Also, fix some small bugs that occur when computing the `Answer` for transmutability.
This patch updates the error reporting when Safe Transmute is not
possible between 2 types by including the reason.
Also, fix some small bugs that occur when computing the `Answer` for
transmutability.
Check for body owner fallibly in error reporting
Sometimes the "body id" we use for an obligation cause is not actually a body owner, like when we're doing WF checking on items.
Fixes#110157
cleanup our region error API
- require `TypeErrCtxt` to always result in an error, closing #108810
- move `resolve_regions_and_report_errors` to the `ObligationCtxt`
- call `process_registered_region_obligations` in `resolve_regions`
- move `resolve_regions` into the `outlives` submodule
- add `#[must_use]` to functions returning lists of errors
r? types
I'm surprised the compiler doesn't warn about these. It appears having
an `impl` on a struct is enough to avoid a warning about it never being
constructed.
Split implied and super predicate queries, then allow elaborator to filter only supertraits
Split the `super_predicates_of` query into a new `implied_predicates_of` query. The former now only returns the *real* supertraits of a trait alias, and the latter now returns the implied predicates (which include all of the `where` clauses of the trait alias). The behavior of these queries is identical for regular traits.
Now that the two queries are split, we can add a new filter method to the elaborator, `filter_only_self()`, which can be used in instances that we need only the *supertrait* predicates, such as during the elaboration used in closure signature deduction. This toggles the usage of `super_predicates_of` instead of `implied_predicates_of` during elaboration of a trait predicate.
This supersedes #104745, and fixes the four independent bugs identified in that PR.
Fixes#104719Fixes#106238Fixes#110023Fixes#109514
r? types
- require `TypeErrCtxt` to always result in an error
- move `resolve_regions_and_report_errors` to the `ObligationCtxt`
- merge `process_registered_region_obligations` into `resolve_regions`
Support safe transmute in new solver
Basically copies the same implementation as the old solver, but instead of looking for param types, we look for type or const placeholders.
Enforce that `PointerLike` requires a pointer-like ABI
At least temporarily, let's ban coercing things that are pointer-sized and pointer-aligned but *not* `Abi::Scalar(..)` into `dyn*`. See: https://github.com/rust-lang/rust/pull/104694#discussion_r1142522073
This can be lifted in the future if we decie that we *want* to be able to coerce something `repr(C)` into a `dyn*`, but we'll have to figure out what to do with Miri and codegen...
r? compiler
Make elaboration generic over input
Combines all the `elaborate_*` family of functions into just one, which is an iterator over the same type that you pass in (e.g. elaborating `Predicate` gives `Predicate`s, elaborating `Obligation`s gives `Obligation`s, etc.)
Implement support for `GeneratorWitnessMIR` in new solver
r? ```@cjgillot```
I mostly want this to cut down the number of failing UI tests when running the UI test suite with `--compare-mode=next-solver`, but there doesn't seem like much reason to block implementing this since it adds minimal complexity to the existing structural traits impl in the new solver.
If others are against adding this for some reason, then maybe we should just make `GeneratorWitnessMIR` return `NoSolution` for these traits. Anything but an ICE please 😸🧊
Tweak debug outputs to make debugging new solver easier
1. Move the fields that are "most important" (I know this is subjective) to the beginning of the structs.
For goals, I typically care more about the predicate than the param-env (which is significantly longer in debug output).
For canonicalized things, I typically care more about what is *being* canonicalized.
For a canonical response, I typically care about the response -- or at least, it's typically useful to put it first since it's short and affects the whether the solver recurses or not...
2. Add some more debug and instrument calls to functions to add more structure to tracing lines.
r? `@oli-obk` or `@BoxyUwU` (since I think `@lcnr` is on holiday)
Remove `intercrate` and `mark_ambiguous` from `TypeRelation`
Fixes#109863
Pulls this logic into `super_combine_tys`, which has access to `InferCtxt` and takes a `ObligationEmittingRelation` -- both of which simplify the logic here.
r? `@oli-obk` `@aliemjay`
This solves a regression where `0.0.cmp()` was ambiguous when a custom
trait with a `cmp` method was in scope.
FOr integers it shouldn't be a problem in practice so I wasn't able to
add a test.
Update `ty::VariantDef` to use `IndexVec<FieldIdx, FieldDef>`
And while doing the updates for that, also uses `FieldIdx` in `ProjectionKind::Field` and `TypeckResults::field_indices`.
There's more places that could use it (like `rustc_const_eval` and `LayoutS`), but I tried to keep this PR from exploding to *even more* places.
Part 2/? of https://github.com/rust-lang/compiler-team/issues/606
Closures always implement `FnOnce` in new solver
We should process `[closure]: FnOnce(Tys...) -> Ty` obligations *before* fallback and closure analysis. We can do this by taking advantage of the fact that `FnOnce` is always implemented by closures, even before we definitely know the closure kind.
Fixescompiler-errors/next-solver-hir-issues#15
r? ``@oli-obk`` (trying to spread the reviewer load for new trait solver prs, and this one is pretty self-contained, though feel free to reassign 😸)
Freshen normalizes-to hack goal RHS in the evaluate loop
Ensure that we repeatedly equate the unconstrained RHS of the normalizes-to hack goal with the *actual* RHS of the goal, even if the normalizes-to goal loops several times and thus we replace the unconstrained RHS var repeatedly.
Alternative to #109583.
And while doing the updates for that, also uses `FieldIdx` in `ProjectionKind::Field` and `TypeckResults::field_indices`.
There's more places that could use it (like `rustc_const_eval` and `LayoutS`), but I tried to keep this PR from exploding to *even more* places.
Part 2/? of https://github.com/rust-lang/compiler-team/issues/606
Canonicalize float var as float in new solver
Typo in new canonicalizer -- we should be canonicalizing float vars as `CanonicalTyVarKind::Float`, not `CanonicalTyVarKind::Int`.
Fixescompiler-errors/next-solver-hir-issues#9
Don't ICE on `DiscriminantKind` projection in new solver
As title says, since we now actually call `Ty::discriminant_kind` on placeholder types 😃
Also drive-by simplify `Pointee::Metadata` projection logic, and fix the UI test because the `<T as Pointee>::Metadata` tests weren't actually exercising the new projection logic, since we still eagerly normalize (which hits `project.rs` in the old solver) in HIR typeck.
r? `@lcnr` tho feel free to re-roll, this pr is very low-priority and not super specific to the new trait solver.
Fixescompiler-errors/next-solver-hir-issues#14
Do not consider elaborated projection predicates for objects in new solver
Object types have projection bounds which are elaborated during astconv. There's no need to do it again for projection goals, since that'll give us duplicate projection candidatesd that are distinct up to regions due to the fact that we canonicalize every region to a separate variable. See quick example below the break for a better explanation.
Discussed this with lcnr, and adding a stop-gap until we get something like intersection region constraints (or modify canonicalization to canonicalize identical regions to the same canonical regions) -- after which, this will hopefully not matter and may be removed.
r? `@lcnr`
---
See `tests/ui/traits/new-solver/more-object-bound.rs`:
Consider a goal: `<dyn Iter<'a, ()> as Iterator>::Item = &'a ()`.
After canonicalization: `<dyn Iter<'!0r, (), Item = '!1r ()> as Iterator>::Item == &!'2r ()`
* First object candidate comes from the item bound in the dyn's bounds itself, giving us `<dyn Iter<'!0r, (), Item = '?!r ()> as Iterator>::Item == &!'1r ()`. This gives us one region constraint: `!'1r == !'2r`.
* Second object candidate comes from elaborating the principal trait ref, gives us `<dyn Iter<'!0r, (), Item = '!1r ()> as Iterator>::Item == &!'0r ()`. This gives us one region constraint: `!'0r == !'2r`.
* Oops! Ambiguity!
Rename `IndexVec::last` → `last_index`
As I've been trying to replace a `Vec` with an `IndexVec`, having `last` exist on both but returning very different types makes the transition a bit awkward -- the errors are later, where you get things like "there's no `ty` method on `mir::Field`" rather than a more localized error like "hey, there's no `last` on `IndexVec`".
So I propose renaming `last` to `last_index` to help distinguish `Vec::last`, which returns an element, and `IndexVec::last_index`, which returns an index.
(Similarly, `Iterator::last` also returns an element, not an index.)
Check for overflow in `assemble_candidates_after_normalizing_self_ty`
Prevents a stack overflow (⚠️❗) in the new solver when we have param-env candidates that look like: `T: Trait<Assoc = <T as Trait>::Assoc>`
The current error message looks bad, but that's because we don't distinguish overflow and other ambiguity errors. I'll break that out into a separate PR since the fix may be controversial.
r? `@lcnr`
As I've been trying to replace a `Vec` with an `IndexVec`, having `last` exist on both but returning very different types makes the transition a bit awkward -- the errors are later, where you get things like "there's no `ty` method on `mir::Field`" rather than a more localized error like "hey, there's no `last` on `IndexVec`".
So I propose renaming `last` to `last_index` to help distinguish `Vec::last`, which returns an element, and `IndexVec::last_index`, which returns an index.
(Similarly, `Iterator::last` also returns an element, not an index.)
Rollup of 8 pull requests
Successful merges:
- #91793 (socket ancillary data implementation for FreeBSD (from 13 and above).)
- #92284 (Change advance(_back)_by to return the remainder instead of the number of processed elements)
- #102472 (stop special-casing `'static` in evaluation)
- #108480 (Use Rayon's TLV directly)
- #109321 (Erase impl regions when checking for impossible to eagerly monomorphize items)
- #109470 (Correctly substitute GAT's type used in `normalize_param_env` in `check_type_bounds`)
- #109562 (Update ar_archive_writer to 0.1.3)
- #109629 (remove obsolete `givens` from regionck)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Add a builtin `FnPtr` trait that is implemented for all function pointers
r? `@ghost`
Rebased version of https://github.com/rust-lang/rust/pull/99531 (plus adjustments mentioned in the PR).
If perf is happy with this version, I would like to land it, even if the diagnostics fix in 9df8e1befb5031a5bf9d8dfe25170620642d3c59 only works for `FnPtr` specifically, and does not generally improve blanket impls.
stop special-casing `'static` in evaluation
fixes#102360
I have no idea whether this actually removed all places where `'static` matters. Without canonicalization it's very easy to accidentally rely on `'static` again. Blocked on changing the `order_dependent_trait_objects` future-compat lint to a hard error
r? `@nikomatsakis`
Don't elaborate non-obligations into obligations
It's suspicious to elaborate a `PolyTraitRef` or `Predicate` into an `Obligation`, since the former does not have a param-env associated with it, but the latter does. This is a footgun that, while not being misused *currently* in the compiler, easily could be misused by someone less familiar with the elaborator's inner workings.
This PR just changes the API -- ideally, the elaborator wouldn't even have to deal with obligations if we're not elaborating obligations, but that would require a bit more abstraction than I could be bothered with today.
Implement non-const `Destruct` trait in new solver
Makes it so that we can call stdlib methods like `Option::map` in **non-const** environments, since *many* stdlib methods have `Destruct` bounds 😅
This doesn't bother to implement `const Destruct` yet, but it shouldn't be too hard to do so. Just didn't bother since we already don't have much support for const traits in the new solver anyways. I'd be happy to add skeleton support for `const Destruct`, though, if the reviewer desires.
Rollup of 9 pull requests
Successful merges:
- #108629 (rustdoc: add support for type filters in arguments and generics)
- #108924 (panic_immediate_abort requires abort as a panic strategy)
- #108961 (Refine error spans for const args in hir typeck)
- #108986 (sync LVI tests)
- #109142 (Add block-based mutex unlocking example)
- #109368 (fix typo in the creation of OpenOption for RustyHermit)
- #109493 (Return nested obligations from canonical response var unification)
- #109515 (Add AixLinker to support linking on AIX)
- #109536 (resolve: Rename some cstore methods to match queries and add comments)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Return nested obligations from canonical response var unification
Handle alias-eq obligations being emitted from `instantiate_and_apply_query_response` in:
* `EvalCtxt` - by processing the nested obligations in the next loop by `new_goals`
* `FulfillCtxt` - by adding the nested obligations to the fulfillment's pending obligations
* `InferCtxt::evaluate_obligation` - ~~by returning `EvaluationResult::EvaluatedToAmbig` (boo 👎, see the FIXME)~~ same behavior as above, since we use fulfillment and `select_where_possible`
The only one that's truly sketchy is `evaluate_obligation`, but it's not hard to modify this behavior moving forward.
From #109037, I think a smaller repro could be crafted if I were smarter, but I am not, so I just took this from #105878.
r? `@lcnr` cc `@BoxyUwU`
Refine error spans for const args in hir typeck
Improve just a couple of error messages having to do with mismatched consts.
r? `@ghost` i'll put this up when the dependent commits are merged
Rollup of 7 pull requests
Successful merges:
- #108541 (Suppress `opaque_hidden_inferred_bound` for nested RPITs)
- #109137 (resolve: Querify most cstore access methods (subset 2))
- #109380 (add `known-bug` test for unsoundness issue)
- #109462 (Make alias-eq have a relation direction (and rename it to alias-relate))
- #109475 (Simpler checked shifts in MIR building)
- #109504 (Stabilize `arc_into_inner` and `rc_into_inner`.)
- #109506 (make param bound vars visibly bound vars with -Zverbose)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Make alias-eq have a relation direction (and rename it to alias-relate)
Emitting an "alias-eq" is too strict in some situations, since we don't always want strict equality between a projection and rigid ty. Adds a relation direction.
* I could probably just reuse this [`RelationDir`](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_infer/infer/combine/enum.RelationDir.html) -- happy to uplift that struct into middle and use that instead, but I didn't feel compelled to... 🤷
* Some of the matching in `compute_alias_relate_goal` is a bit verbose -- I guess I could simplify it by using [`At::relate`](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_infer/infer/at/struct.At.html#method.relate) and mapping the relation-dir to a variance.
* Alternatively, I coulld simplify things by making more helper functions on `EvalCtxt` (e.g. `EvalCtxt::relate_with_direction(T, T)` that also does the nested goal registration). No preference.
r? ```@lcnr``` cc ```@BoxyUwU``` though boxy can claim it if she wants
NOTE: first commit is all the changes, the second is just renaming stuff
new solver cleanup + implement coherence
the cleanup:
- change `Certainty::unify_and` to consider ambig + overflow to be ambig
- rename `trait_candidate_should_be_dropped_in_favor_of` to `candidate_should_be_dropped_in_favor_of`
- remove outdated fixme
For coherence I mostly just add an ambiguous candidate if the current trait ref is unknowable. I am doing the same for reservation impl where I also just add an ambiguous candidate.
a general type system cleanup
removes the helper functions `traits::fully_solve_X` as they add more complexity then they are worth. It's confusing which of these helpers should be used in which context.
changes the way we deal with overflow to always add depth in `evaluate_predicates_recursively`. It may make sense to actually fully transition to not have `recursion_depth` on obligations but that's probably a bit too much for this PR.
also removes some other small - and imo unnecessary - helpers.
r? types
Do not suggest bounds restrictions for synthesized RPITITs
Before this PR we were getting ...
```
warning: the feature `async_fn_in_trait` is incomplete and may not be safe to use and/or cause compiler crashes
--> tests/ui/async-await/in-trait/missing-send-bound.rs:5:12
|
5 | #![feature(async_fn_in_trait)]
| ^^^^^^^^^^^^^^^^^
|
= note: see issue #91611 <https://github.com/rust-lang/rust/issues/91611> for more information
= note: `#[warn(incomplete_features)]` on by default
error: future cannot be sent between threads safely
--> tests/ui/async-await/in-trait/missing-send-bound.rs:17:20
|
17 | assert_is_send(test::<T>());
| ^^^^^^^^^^^ future returned by `test` is not `Send`
|
= help: within `impl Future<Output = ()>`, the trait `Send` is not implemented for `impl Future<Output = ()>`
note: future is not `Send` as it awaits another future which is not `Send`
--> tests/ui/async-await/in-trait/missing-send-bound.rs:13:5
|
13 | T::bar().await;
| ^^^^^^^^ await occurs here on type `impl Future<Output = ()>`, which is not `Send`
note: required by a bound in `assert_is_send`
--> tests/ui/async-await/in-trait/missing-send-bound.rs:21:27
|
21 | fn assert_is_send(_: impl Send) {}
| ^^^^ required by this bound in `assert_is_send`
help: consider further restricting the associated type
|
16 | fn test2<T: Foo>() where impl Future<Output = ()>: Send {
| ++++++++++++++++++++++++++++++++++++
error: aborting due to previous error; 1 warning emitted
```
and we want this output ...
```
warning: the feature `async_fn_in_trait` is incomplete and may not be safe to use and/or cause compiler crashes
--> $DIR/missing-send-bound.rs:5:12
|
LL | #![feature(async_fn_in_trait)]
| ^^^^^^^^^^^^^^^^^
|
= note: see issue #91611 <https://github.com/rust-lang/rust/issues/91611> for more information
= note: `#[warn(incomplete_features)]` on by default
error: future cannot be sent between threads safely
--> $DIR/missing-send-bound.rs:17:20
|
LL | assert_is_send(test::<T>());
| ^^^^^^^^^^^ future returned by `test` is not `Send`
|
= help: within `impl Future<Output = ()>`, the trait `Send` is not implemented for `impl Future<Output = ()>`
note: future is not `Send` as it awaits another future which is not `Send`
--> $DIR/missing-send-bound.rs:13:5
|
LL | T::bar().await;
| ^^^^^^^^ await occurs here on type `impl Future<Output = ()>`, which is not `Send`
note: required by a bound in `assert_is_send`
--> $DIR/missing-send-bound.rs:21:27
|
LL | fn assert_is_send(_: impl Send) {}
| ^^^^ required by this bound in `assert_is_send`
error: aborting due to previous error; 1 warning emitted
```
r? `@compiler-errors`
Only implement Fn* traits for extern "Rust" safe function pointers and items
Since calling the function via an `Fn` trait will assume `extern "Rust"` ABI and not do any safety checks, only safe `extern "Rust"` function can implement the `Fn` traits. This syncs the logic between the old solver and the new solver.
r? `@compiler-errors`
Constrain const vars to error if const types are mismatched
When equating two consts of different types, if either are const variables, constrain them to the correct const error kind.
This helps us avoid "successfully" matching a const against an impl signature but leaving unconstrained const vars, which will lead to incremental ICEs when we call const-eval queries during const projection.
Fixes#109296
The second commit in the stack fixes a regression in the first commit where we end up mentioning `[const error]` in an impl overlap error message. I think the error message changes for the better, but I could implement alternative strategies to avoid this without delaying the overlap error message...
r? `@BoxyUwU`
new solver: make all goal evaluation able to be automatically rerun
It is generally wrong to call `evaluate_goal` multiple times or `evaluate_goal` and `evaluate_all` for the same `QueryResult` without correctly handling rerunning the goals when inference makes progress. Not doing so will result in the assertion in `evaluate_goal` firing because rerunning the goal will lead to a more accurate `QueryResult`.
Currently there are lots of places that get this wrong and generally it is complex and error prone to handle correctly everywhere. This PR introduces a way to add goals to the `EvalCtxt` and then run all the added goals in a loop so that `evaluate_goal`/`evaluate_all` is not necessary to call manually.
There are a few complications for making everything work "right":
1. the `normalizes-to` hack that replaces the rhs with an unconstrained infer var requires special casing in the new `try_evaluate_added_goals` function similar to how `evaluate_goal`'s assertion special cases that hack.
2. `assemble_candidates_after_normalizing_self_ty`'s normalization step needs to be reran for each candidate otherwise the found candidates will potentially get a more accurate `QueryResult` when rerunning the projection/trait goal which can effect the `QueryResult` of the projection/trait goal.
This is implemented via `EvalCtxt::probe`'s closure's `EvalCtxt` inheriting the added goals of the `EvalCtxt` that `probe` is called on, allowing us to add goals in a probe, and then enter a nested probe for each candidate and evaluate added goals which include the normalization step's goals.
I made `make_canonical_response` evaluate added goals so that it will be hard to mess up the impl of the solver by forgetting to evaluate added goals. Right now the only way to mess this up would be to call `response_no_constraints` (which from the name is obviously weird).
The visibility of `evaluate_goal` means that it can be called from various `compute_x_goal` or candidate assembly functions, this is generally wrong and we should never call `evaluate_goal` manually, instead we should be calling `add_goal`/`add_goals`. This is solved by moving `evaluate_goal` `evaluate_canonical_goal` and `compute_goal` into `eval_ctxt`'s module and making them private so they cannot be called from elsewhere, forcing people to call `add_goal/s` and `evaluate_added_goals_and_make_canonical_resposne`/`try_evaluate_added_goals`
---
Other changes:
- removed the `&& false` that was introduced to the assertion in `evaluate_goal` in #108839
- remove a `!self.did_overflow()` requirement in `search_graph.is_empty()` which causes goals that overflow to ICE
- made `EvalCtxt::eq` take `&mut self` and add all the nested goals via `add_goals` instead of returning them as 99% of call sites just immediately called `EvalCtxt::add_goals` manually.
r? `````@lcnr`````
- only borrow the refcell once per loop
- avoid complex matches to reduce branch paths in the hot loop
- use a by-ref fast path that avoids mutations at the expense of having false negatives
Install projection from RPITIT to default trait method opaque correctly
1. For new lowering strategy `-Zlower-impl-trait-in-trait-to-assoc-ty`, install the correct default trait method projection predicates (RPITIT -> opaque). This makes default trait body tests pass!
2. Fix two WF-checking bugs -- first, we want to make sure that we're always looking for an opaque type in `check_return_position_impl_trait_in_trait_bounds`. That's because the RPITIT projections are normalized to opaques during wfcheck. Second, fix RPITIT's param-envs by not adding the projection predicates that we install on trait methods to make default RPITITs work -- I left a comment why.
3. Also, just a small drive-by for `rustc_on_unimplemented`. Not sure if it affects any tests, but can't hurt.
r? ````@spastorino,```` based off of #109140
Remove box expressions from HIR
After #108516, `#[rustc_box]` is used at HIR->THIR lowering and this is no longer emitted, so it can be removed.
This is based on top of #108471 to help with conflicts, so 43490488ccacd1a822e9c621f5ed6fca99959a0b is the only relevant commit (sorry for all the duplicated pings!)
````@rustbot```` label +S-blocked
Some cleanups in our normalization logic
Changed a match to be exhaustive and deduplicated some code.
r? ```@compiler-errors```
this pulls out the uncontroversial part of https://github.com/rust-lang/rust/pull/108860
make `define_opaque_types` fully explicit
based on the idea of #108389. Moved `define_opaque_types` into the actual operations, e.g. `eq`, instead of `infcx.at` because normalization doesn't use `define_opaque_types` and even creates it's own `At` with a different `define_opaque_types` internally.
Somewhat surprisingly, coherence actually relies on `DefineOpaqueTypes::Yes` for soundness which was revealed because I've incorrectly used `DefineOpaqueTypes::No` in `equate_impl_headers`. It feels concerning that even though this is the case, we still sometimes use `DefineOpaqueTypes::No` in coherence. I did not look into this as part of this PR as it is purely changing the structure of the code without changing behavior in any way.
r? ```@oli-obk```
Remove some direct calls to local_def_id_to_hir_id on diagnostics
Was playing with `tests/ui/impl-trait/in-trait/default-body-with-rpit.rs` and was able to remove some ICEs. Still getting ...
```
error[E0277]: `impl Future<Output = Foo::{opaque#0}>` is not a future
--> tests/ui/impl-trait/in-trait/default-body-with-rpit.rs:10:28
|
10 | async fn baz(&self) -> impl Debug {
| ^^^^^^^^^^ `impl Future<Output = Foo::{opaque#0}>` is not a future
|
= help: the trait `Future` is not implemented for `impl Future<Output = Foo::{opaque#0}>`
= note: impl Future<Output = Foo::{opaque#0}> must be a future or must implement `IntoFuture` to be awaited
note: required by a bound in `Foo::{opaque#1}`
--> tests/ui/impl-trait/in-trait/default-body-with-rpit.rs:10:28
|
10 | async fn baz(&self) -> impl Debug {
| ^^^^^^^^^^ required by this bound in `Foo::{opaque#1}`
error[E0277]: the size for values of type `impl Future<Output = Foo::{opaque#0}>` cannot be known at compilation time
--> tests/ui/impl-trait/in-trait/default-body-with-rpit.rs:10:28
|
10 | async fn baz(&self) -> impl Debug {
| ^^^^^^^^^^ doesn't have a size known at compile-time
|
= help: the trait `Sized` is not implemented for `impl Future<Output = Foo::{opaque#0}>`
note: required by a bound in `Foo::{opaque#1}`
--> tests/ui/impl-trait/in-trait/default-body-with-rpit.rs:10:28
|
10 | async fn baz(&self) -> impl Debug {
| ^^^^^^^^^^ required by this bound in `Foo::{opaque#1}`
error: internal compiler error: compiler/rustc_hir_typeck/src/closure.rs:724:18: async fn generator return type not an inference variable: Foo::{opaque#1}<'_>
--> tests/ui/impl-trait/in-trait/default-body-with-rpit.rs:10:39
|
10 | async fn baz(&self) -> impl Debug {
| _______________________________________^
11 | | ""
12 | | }
| |_____^
```
But I guess this is a little bit of progress anyway.
This one goes on top of #108700 and #108945
r? `@compiler-errors`
Remove `identity_future` indirection
This was previously needed because the indirection used to hide some unexplained lifetime errors, which it turned out were related to the `min_choice` algorithm.
Removing the indirection also solves a couple of cycle errors, large moves and makes async blocks support the `#[track_caller]`annotation.
Fixes https://github.com/rust-lang/rust/issues/104826.
Add suggestion to diagnostic when user has array but trait wants slice. (rebased)
Rebase of #91314, except for change to multipart suggestion
Resolves#90528
r? ``@compiler-errors`` since you requested the multipart suggestion
tidy: enforce comment blocks to have an even number of backticks
After PR #108694, most unmatched backticks in `compiler/` comments have been eliminated. This PR adds a tidy lint to ensure no new unmatched backticks are added, and either addresses the lint in the remaining instances it found, or allows it.
Very often, backtick containing sections wrap around lines, for example:
```Rust
// This function takes a tuple `(Vec<String>,
// Box<[u8]>)` and transforms it into `Vec<u8>`.
```
The lint is implemented to work on top of blocks, counting each line with a `//` into a block, and counting if there are an odd or even number of backticks in the entire block, instead of looking at just a single line.
This makes it easier to open the messages file while developing on features.
The commit was the result of automatted changes:
for p in compiler/rustc_*; do mv $p/locales/en-US.ftl $p/messages.ftl; rmdir $p/locales; done
for p in compiler/rustc_*; do sed -i "s#\.\./locales/en-US.ftl#../messages.ftl#" $p/src/lib.rs; done
Implement goal caching with the new solver
Maybe it's wrong, idk. Opening mostly for first impressions before I go to sleep.
r? ``@lcnr,`` cc ``@cjgillot``
Don't even try to combine consts with incompatible types
~I left a more detailed explanation for why this fixes this issue in the UI test, but in general, we should not try to unify const infer vars and rigid consts if they have incompatible types. That's because we don't want something like a `ConstArgHasType` predicate to suddenly go from passing to failing, or vice versa, due to a shallow resolve.~
1. Use the `type_of` for a parameter in `try_eval_lit_or_param`, instead of the "expected" type from a `WithOptConstParam` def id.
2. Don't combine consts that have incompatible types.
Fixes#108781
Do not ICE when we have fn pointer `Fn` obligations with bound vars in the self type
We never supported solving `for<'a> fn(&'a ()): Fn(&'a ())` -- I tried to add that support in #104929, but iirc `@lcnr` wanted to support this more generally by eagerly instantiating trait predicate binders with placeholders. That never happened due to blockers in the old solver, but we probably shouldn't ICE in any case.
On the bright side, this passes on the new solver :^)
Emit alias-eq when equating numeric var and projection
This doesn't fix everything having to do with projections and infer vars, but it does fix a common case I saw in HIR typeck.
r? `@lcnr`
Remove `body_def_id` from `Inherited`
We can just use the body id from the obligation cause.
Follow-up to #108945, only my commit is relevant.
r? `@cjgillot` cc `@spastorino`
Rollup of 8 pull requests
Successful merges:
- #108754 (Retry `pred_known_to_hold_modulo_regions` with fulfillment if ambiguous)
- #108759 (1.41.1 supported 32-bit Apple targets)
- #108839 (Canonicalize root var when making response from new solver)
- #108856 (Remove DropAndReplace terminator)
- #108882 (Tweak E0740)
- #108898 (Set `LIBC_CHECK_CFG=1` when building Rust code in bootstrap)
- #108911 (Improve rustdoc-gui/tester.js code a bit)
- #108916 (Remove an unused return value in `rustc_hir_typeck`)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Canonicalize root var when making response from new solver
During trait solving, if we equate two inference variables `?0` and `?1` but don't equate them with any rigid types, then `InferCtxt::probe_ty_var` will return `Err` for both of these. The canonicalizer code will then canonicalize the variables independently(!), and the response will not reflect the fact that these two variables have been made equal.
This hinders inference and I also don't think it's sound? I haven't thought too much about it past that, so let's talk about it.
r? ``@lcnr``
Retry `pred_known_to_hold_modulo_regions` with fulfillment if ambiguous
Fixes#108721
The problem here is that when we're checking `is_sized_raw` during codegen on some type that has a lot of opaques in it, something emits several nested obligations that are individually ambiguous, but when processed together in a loop then apply modulo regions. Since the `evaluate_predicates_recursively` inner loop doesn't process predicates until they stop changing, we return `EvaluatedToAmbig`, which makes the sized check return false incorrectly. See:
f15f0ea739/compiler/rustc_trait_selection/src/traits/select/mod.rs (L596-L606)
... Compared to the analogous loop in the new solver:
f15f0ea739/compiler/rustc_trait_selection/src/solve/mod.rs (L481-L512)
To fix this, if we get ambiguous during `pred_known_to_hold_modulo_regions`, just retry the obligation in a fulfillment context.
--
Unfortunately... I don't have a test for this. I've only tested this locally. Pending minimization :/
r? types
Suppress copy impl error when post-normalized type references errors
Suppress spurious errors from the `Copy` impl validity check when fields have bad types *post*-normalization, instead of just pre-normalization.
----
The const-generics test regressed recently due to #107965, cc `````@BoxyUwU.`````
* I think it's because `[_; 0u32]: Copy` now fails to hold because a nested obligation `ConstArgHasType(0u32, usize)` fails.
* It's interesting that `[const_error]` shows up in the type only after normalization, though, but I'm pretty sure that it's due to the evaluate call that happens when normalizing unevaluated consts.
This was previously needed because the indirection used to hide some unexplained lifetime errors, which it turned out were related to the `min_choice` algorithm.
Removing the indirection also solves a couple of cycle errors, large moves and makes async blocks support the `#[track_caller]` annotation.
always resolve to universal regions if possible
`RegionConstraintCollector::opportunistic_resolve_var`, which is used in canonicalization and projection logic, doesn't resolve the region var to an equal universal region. So if we have equated `'static == '1 == '2`, it doesn't resolve `'1` or `'2` to `'static`. Now it does!
Addresses review comment https://github.com/rust-lang/rust/pull/107376#discussion_r1093233687.
r? `@lcnr`
