Convert typeck constraints in location-sensitive polonius
In this PR, we do a big chunk of the work of localizing regular outlives constraints.
The slightly annoying thing is handling effectful statements: usually the subset graph propagates loans at a single point between regions, and liveness propagates loans between points within a single region, but some statements have effects applied on exit.
This was also a problem before, in datalog polonius terms and Niko's solution at the time, this is about: the mid-point. The idea was to duplicate all MIR locations into two physical points, and orchestrate the effects with that. Somewhat easier to do, but double the CFG.
We've always believed we didn't _need_ midpoints in principle, as we can represent changes on exit as on happening entry to the successor, but there's some difficulty in tracking the position information at sufficient granularity through outlives relation (especially since we also have bidirectional edges and time-traveling now).
Now, that is surely what we should be doing in the future. In the mean time, I infer this from the kind of statement/terminator where an outlives constraint arose. It's not particularly complicated but some explanation will help clarify the code.
Assignments (in their various forms) are the quintessential example of these crossover cases: loans that would flow into the LHS would not be visible on entry to the point but on exit -- so we'll localize these edges to the successor. Let's look at a real-world example, involving invariance for bidirectional edges:
```rust
let mut _1: HashMap<i32, &'7 i32>;
let mut _3: &'9 mut HashMap<i32, &'10 i32>;
...
/* at bb1[3]: */ _3 = &'3 mut _1;
```
Here, typeck expectedly produces 3 outlives constraints today:
1. `'3 -> '9`
2. `'7 -> '10`
3. `'10 -> '7`
And we localize them like so,
1. `'3 -> '9` flows into the LHS and becomes: `3_bb1_3 -> 9_bb1_4`
2. `'7 -> '10` flows into the LHS and becomes: `7_bb1_3 -> 10_bb1_4`
3. `'10 -> '7` flows from the LHS and becomes: `10_bb1_4 -> 7_bb1_3` (time traveling 👌)
---
r? ``@jackh726``
To keep you entertained during the holidays I also threw in a couple of small changes removing cruft in the borrow checker.
We're actually getting there. The next PR will be the last one needed to get end-to-end tests working.
Use a post-monomorphization typing env when mangling components that come from impls
When mangling associated methods of impls, we were previously using the wrong param-env. Instead of using a fully monomorphized param-env like we usually do in codegen, we were taking the post-analysis param-env, and treating it as an early binder to *re-substitute* the impl args. I've pointed out the problematic old code in an inline comment.
This would give us param-envs with possibly trivial predicates that would prevent normalization via param-env shadowing.
In the example test linked below, `tests/ui/symbol-names/normalize-in-param-env.rs`, this happens when we mangle the impl `impl<P: Point2> MyFrom<P::S> for P` with the substitution `P = Vec2`. Because the where clause of the impl is `P: Point2`, which elaborates to `[P: Point2, P: Point, <P as Point>::S projects-to <P as Point2>::S2]` and the fact that `impl Point2 for Vec2` normalizes `Vec2::S2` to `Vec2::S`, this causes a cycle.
The proper fix here is to use a fully monomorphized param-env for the case where the impl is properly substituted.
Fixes#135143
While #134081 uncovered this bug for legacy symbol mangling, it was preexisting for v0 symbol mangling. This PR fixes both. The test requires a "hack" because we strip the args of the instance we're printing for legacy symbol mangling except for drop glue, so we box a closure to ensure we generate drop glue.
r? oli-obk
A few borrowck tweaks to improve 2024 edition migration lints
See first two commits' changes to test outputs. Test coverage in this area is kinda weak, but I think it affects more cases than this (like the craters that will begin to trigger the `tail_expr_drop_order` tests in #134523).
Third commit is a drive-by change that removes a deref hack from `UseSpans` which doesn't really improve diagnostics much.
Add a notion of "some ABIs require certain target features"
I think I finally found the right shape for the data and checks that I recently added in https://github.com/rust-lang/rust/pull/133099, https://github.com/rust-lang/rust/pull/133417, https://github.com/rust-lang/rust/pull/134337: we have a notion of "this ABI requires the following list of target features, and it is incompatible with the following list of target features". Both `-Ctarget-feature` and `#[target_feature]` are updated to ensure we follow the rules of the ABI. This removes all the "toggleability" stuff introduced before, though we do keep the notion of a fully "forbidden" target feature -- this is needed to deal with target features that are actual ABI switches, and hence are needed to even compute the list of required target features.
We always explicitly (un)set all required and in-conflict features, just to avoid potential trouble caused by the default features of whatever the base CPU is. We do this *before* applying `-Ctarget-feature` to maintain backward compatibility; this poses a slight risk of missing some implicit feature dependencies in LLVM but has the advantage of not breaking users that deliberately toggle ABI-relevant target features. They get a warning but the feature does get toggled the way they requested.
For now, our logic supports x86, ARM, and RISC-V (just like the previous logic did). Unsurprisingly, RISC-V is the nicest. ;)
As a side-effect this also (unstably) allows *enabling* `x87` when that is harmless. I used the opportunity to mark SSE2 as required on x86-64, to better match the actual logic in LLVM and because all x86-64 chips do have SSE2. This infrastructure also prepares us for requiring SSE on x86-32 when we want to use that for our ABI (and for float semantics sanity), see https://github.com/rust-lang/rust/issues/133611, but no such change is happening in this PR.
r? `@workingjubilee`
rustc_intrinsic: support functions without body
We synthesize a HIR body `loop {}` but such bodyless intrinsics.
Most of the diff is due to turning `ItemKind::Fn` into a brace (named-field) enum variant, because it carries a `bool`-typed field now. This is to remember whether the function has a body. MIR building panics to avoid ever translating the fake `loop {}` body, and the intrinsic logic uses the lack of a body to implicitly mark that intrinsic as must-be-overridden.
I first tried actually having no body rather than generating the fake body, but there's a *lot* of code that assumes that all function items have HIR and MIR, so this didn't work very well. Then I noticed that even `rustc_intrinsic_must_be_overridden` intrinsics have MIR generated (they are filled with an `Unreachable` terminator) so I guess I am not the first to discover this. ;)
r? `@oli-obk`
Project to `TyKind::Error` when there are unconstrained non-lifetime (ty/const) impl params
It splits the `enforce_impl_params_are_constrained` function into lifetime/non-lifetime, and queryfies the latter. We can then use the result of the latter query (`Result<(), ErrorGuaranteed>`) to intercept projection and constrain the projected type to `TyKind::Error`, which ensures that we leak no ty or const vars to places that don't expect them, like `normalize_erasing_regions`.
The reason we split `enforce_impl_params_are_constrained` into two parts is because we only error for *lifetimes* if the lifetime ends up showing up in any of the associated types of the impl (e.g. we allow `impl<'a> Foo { type Assoc = (); }`). However, in order to compute the `type_of` query for the anonymous associated type of an RPITIT, we need to do trait solving (in `query collect_return_position_impl_trait_in_trait_tys`). That would induce cycles. Luckily, it turns out for lifetimes we don't even care about if they're unconstrained, since they're erased in all contexts that we are trying to fix ICEs. So it's sufficient to keep this check separated out of the query.
I think this is a bit less invasive of an approach compared to #127973. The major difference between this PR and that PR is that we queryify the check instead of merging it into the `explicit_predicates_of` query, and we use the result to taint just projection goals, rather than trait goals too. This doesn't require a lot of new tracking in `ItemCtxt` and `GenericPredicates`, and it also seems to not require any other changes to typeck like that PR did.
Fixes#123141Fixes#125874Fixes#126942Fixes#127804Fixes#130967
r? oli-obk
Improve infer (`_`) suggestions in `const`s and `static`s
Fixes https://github.com/rust-lang/rust/issues/135010.
This PR does a few things to (imo) greatly improve the error message when users write something like `static FOO: [i32; _] = [1, 2, 3]`.
Firstly, it adapts the recovery code for when we encounter `_` in a const/static to work a bit more like `fn foo() -> _`, and removes the somewhat redundant query `diagnostic_only_typeck`.
Secondly, it changes the lowering for `[T; _]` to always lower under the `feature(generic_arg_infer)` logic to `ConstArgKind::Infer`. We still issue the feature error, so it's not doing anything *observable* on the good path, but it does mean that we no longer erroneously interpret `[T; _]`'s array length as a `_` **wildcard expression** (à la destructuring assignment, like `(_, y) = expr`).
Lastly it makes the suggestions verbose and fixes (well, suppresses) a bug with stashing and suggestions.
r? oli-obk
Spruce up the docs of several queries related to the type/trait system and const eval
- Editorial
- Proper rustdoc summary/synopsis line by making use of extra paragraphs: Leads to better rendered output on module pages, in search result lists and overall, too
- Use rustdoc warning blocks for admonitions of the form "do not call / avoid calling this query directly"
- Use intra-doc links of the form ``[`Self::$query`]`` to cross-link queries. Indeed, such links are generally a bit brittle due to the existence of `TyCtxtFeed` which only contains a subset of queries. Therefore the docs of `feedable` queries cannot cross-link to non-`feedable` ones. I'd say it's fine to use intra-doc links despite the potential/unlikely occasional future breakage (if a query with the aforementioned characteristics becomes `feedable`). `Self::` is nicer than `TyCtxt::` (which would be more stable) since it accounts for other contexts like `TyCtxt{Feed,At,Ensure{,WithValue}}`
- Informative
- Generally add, flesh out and correct some doc comments
- Add *Panic* sections (to a few selected queries only). The lists of panics aren't necessarily exhaustive and focus on the more "obvious" or "important" panics.
- Where applicable add a paragraph calling attention to the relevant [`#[rustc_*]` TEST attribute](https://rustc-dev-guide.rust-lang.org/compiler-debugging.html#rustc_-test-attributes)
The one non-doc change (it's internal and not observable):
Be even more defensive in `query constness`'s impl (spiritual follow-up to #134122) (see self review comment).
Fixes#133494.
r\? **any**(compiler-errors, oli-obk)
Begin to implement type system layer of unsafe binders
Mostly TODOs, but there's a lot of match arms that are basically just noops so I wanted to split these out before I put up the MIR lowering/projection part of this logic.
r? oli-obk
Tracking:
- https://github.com/rust-lang/rust/issues/130516
cleanup `TypeVerifier`
We should merge it with the `TypeChecker` as we no longer bail in cases where it encounters an error since #111863.
It's quite inconsistent whether a check lives in the verifier or the `TypeChecker`, so this feels like a quite impactful cleanup. I expect that for this we may want to change the `TypeChecker` to also be a MIR visitor 🤔 this is non-trivial so I didn't fully do it in this PR.
Best reviewed commit by commit.
r? `@compiler-errors` feel free to reassign however
coroutine_clone: add comments
I was very surprised to learn that coroutines can be cloned. This has non-trivial semantic consequences that I do not think have been considered. Lucky enough, it's still unstable. Let's add some comments and pointers so we hopefully become aware when a MIR opt actually is in conflict with this.
Cc `@rust-lang/wg-mir-opt`
Foundations of location-sensitive polonius
I'd like to land the prototype I'm describing in the [polonius project goal](https://github.com/rust-lang/rust-project-goals/issues/118). It still is incomplete and naive and terrible but it's working "well enough" to consider landing.
I'd also like to make review easier by not opening a huge PR, but have a couple small-ish ones (the +/- line change summary of this PR looks big, but >80% is moving datalog to a single place).
This PR starts laying the foundation for that work:
- it refactors and collects 99% of the old datalog fact gen, which was spread around everywhere, into a single dedicated module. It's still present at 3 small places (one of which we should revert anyways) that are kinda deep within localized components and are not as easily extractable into the rest of fact gen, so it's fine for now.
- starts introducing the localized constraints, the building blocks of the naive way of implementing the location-sensitive analysis in-tree, which is roughly sketched out in https://smallcultfollowing.com/babysteps/blog/2023/09/22/polonius-part-1/ and https://smallcultfollowing.com/babysteps/blog/2023/09/29/polonius-part-2/ but with a different vibe than per-point environments described in these posts, just `r1@p: r2@q` constraints.
- sets up the skeleton of generating these localized constraints: converting NLL typeck constraints, and creating liveness constraints
- introduces the polonius dual to NLL MIR to help development and debugging. It doesn't do much currently but is a way to see these localized constraints: it's an NLL MIR dump + a dumb listing of the constraints, that can be dumped with `-Zdump-mir=polonius -Zpolonius=next`. Its current state is not intended to be a long-term thing, just for testing purposes -- I will replace its contents in the future with a different approach (an HTML+js file where we can more easily explore/filter/trace these constraints and loan reachability, have mermaid graphs of the usual graphviz dumps, etc).
I've started documenting the approach in this PR, I'll add more in the future. It's quite simple, and should be very clear when more constraints are introduced anyways.
r? `@matthewjasper`
Best reviewed per commit so that the datalog move is less bothersome to read, but if you'd prefer we separate that into a different PR, I can do that (and michael has offered to review these more mechanical changes if it'd help).
handle member constraints directly in the mir type checker
cleaner, faster, easier to change going forward :> fixes#109654
r? `@oli-obk` `@compiler-errors`
unimplement `PointerLike` for trait objects
Values of type `dyn* PointerLike` or `dyn PointerLike` are not pointer-like so these types should not implement `PointerLike`.
After https://github.com/rust-lang/rust/pull/133226, `PointerLike` allows user implementations, so we can't just mark it with `#[rustc_deny_explicit_impl(implement_via_object = false)]`. Instead, this PR splits the `#[rustc_deny_explicit_impl(implement_via_object = ...)]` attribute into two separate attributes `#[rustc_deny_explicit_impl]` and `#[rustc_do_not_implement_via_object]` so that we opt out of the automatic `impl PointerLike for dyn PointerLike` and still allow user implementations.
For traits that are marked with `#[do_not_implement_via_object]` but not `#[rustc_deny_explicit_impl]` I've also made it possible to add a manual `impl Trait for dyn Trait`. There is no immediate need for this, but it was one line to implement and seems nice to have.
fixes https://github.com/rust-lang/rust/issues/134545
fixes https://github.com/rust-lang/rust/issues/134543
r? `@compiler-errors`
This commit splits the `#[rustc_deny_explicit_impl(implement_via_object = ...)]` attribute
into two attributes `#[rustc_deny_explicit_impl]` and `#[rustc_do_not_implement_via_object]`.
This allows us to have special traits that can have user-defined impls but do not have the
automatic trait impl for trait objects (`impl Trait for dyn Trait`).
Improve dependency_format a bit
* Make `DependencyList` an `IndexVec` rather than emulating one using a `Vec` (which was off-by-one as LOCAL_CRATE was intentionally skipped)
* Update some comments for the fact that we now use `#[global_allocator]` rather than `extern crate alloc_system;`/`extern crate alloc_jemalloc;` for specifying which allocator to use. We still use a similar mechanism for the panic runtime, so refer to the panic runtime in those comments instead.
* An unrelated refactor to `create_and_enter_global_ctxt` I forgot to include in https://github.com/rust-lang/rust/pull/134302. This refactor is too small to be worth it's own PR.
coverage: Store coverage source regions as `Span` until codegen (take 2)
This is an attempt to re-land #133418:
> Historically, coverage spans were converted into line/column coordinates during the MIR instrumentation pass.
> This PR moves that conversion step into codegen, so that coverage spans spend most of their time stored as Span instead.
> In addition to being conceptually nicer, this also reduces the size of coverage mappings in MIR, because Span is smaller than 4x u32.
That PR was reverted by #133608, because in some circumstances not covered by our test suite we were emitting coverage metadata that was causing `llvm-cov` to exit with an error (#133606).
---
The implementation here is *mostly* the same, but adapted for subsequent changes in the relevant code (e.g. #134163).
I believe that the changes in #134163 should be sufficient to prevent the problem that required the original PR to be reverted. But I haven't been able to reproduce the original breakage in a regression test, and the `llvm-cov` error message is extremely unhelpful, so I can't completely rule out the possibility of this breaking again.
r? jieyouxu (reviewer of the original PR)
Some destructor/drop related tweaks
Two random tweaks I got from investigating some stuff around drops in edition 2024:
1. Use the `TypingEnv` of the mir builder, rather than making it over again.
2. Rename the `id` field from `Scope` to `local_id`, to reflect that it's a local id, and remove the `item_local_id()` accessor which just returned the id field.
Variants::Single: do not use invalid VariantIdx for uninhabited enums
~~Stacked on top of https://github.com/rust-lang/rust/pull/133681, only the last commit is new.~~
Currently, `Variants::Single` for an empty enum contains a `VariantIdx` of 0; looking that up in the enum variant list will ICE. That's quite confusing. So let's fix that by adding a new `Variants::Empty` case for types that have 0 variants.
try-job: i686-msvc
cleanup region handling: add `LateParamRegionKind`
The second commit is to enable a split between `BoundRegionKind` and `LateParamRegionKind`, by avoiding `BoundRegionKind` where it isn't necessary.
The third comment then adds `LateParamRegionKind` to avoid having the same late-param region for separate bound regions. This fixes#124021.
r? `@compiler-errors`
-Znext-solver: modify candidate preference rules
This implements the design proposed in the FCP in #132325 and matches the old solver behavior. I hope the inline comments are all sufficiently clear, I personally think this is a fairly clear improvement over the existing approach using `fn discard_impls_shadowed_by_env`. This fixes https://github.com/rust-lang/trait-system-refactor-initiative/issues/96.
This also fixes#133639 which encounters an ICE in negative coherence when evaluating the where-clause. Given the features required to trigger this ICE 🤷
r? ``@compiler-errors``
