Track bound types like bound regions
When we instantiate bound types into placeholder types, we throw away the names for some reason. These names are particularly useful for error reporting once we have `for<T>` binders.
r? types
small refactor to new projection code
extract `eq_term_and_make_canonical_response` into a helper function which also is another guarantee that the expected term does not influence candidate selection for projections.
also change `evaluate_all(vec![single_goal])` to use `evaluate_goal`.
the second commit now also adds a `debug_assert!` to `evaluate_goal`.
Modify primary span label for E0308
Looking at the reactions to https://hachyderm.io/`@ekuber/109622160673605438,` a lot of people seem to have trouble understanding the current output, where the primary span label on type errors talks about the specific types that diverged, but these can be deeply nested type parameters. Because of that we could see "expected i32, found u32" in the label while the note said "expected Vec<i32>, found Vec<u32>". This understandably confuses people. I believe that once people learn to read these errors it starts to make more sense, but this PR changes the output to be more in line with what people might expect, without sacrificing terseness.
Fix#68220.
Use `ObligationCtxt::new_in_snapshot` in `satisfied_from_param_env`
We can evaluate nested `ConstEvaluatable` obligations in an evaluation probe, which will ICE if we use `ObligationCtxt::new`.
Fixes#107474Fixes#106666
r? `@BoxyUwU` but feel free to reassign
cc `@JulianKnodt` who i think added this assertion code
Not sure if the rustdoc test is needed, but can't hurt. They're the same root cause, though.
Skip possible where_clause_object_safety lints when checking `multiple_supertrait_upcastable`
Fix#106247
To achieve this, I lifted the `WhereClauseReferencesSelf` out from `object_safety_violations` and move it into `is_object_safe` (which is changed to a new query).
cc `@dtolnay`
r? `@compiler-errors`
Remove HirId -> LocalDefId map from HIR.
Having this map in HIR prevents the creating of new definitions after HIR has been built.
Thankfully, we do not need it.
Based on https://github.com/rust-lang/rust/pull/103902
Remove `ControlFlow::{BREAK, CONTINUE}`
Libs-API decided to remove these in #102697.
Follow-up to #107023, which removed them from `compiler/`, but a couple new ones showed up since that was merged.
r? libs
Minor tweaks in the new solver
1. `InferCtxt::probe` is not needed in `compute_subtype_goal` and `compute_well_formed_goal`.
2. Add a handful of comments.
3. Add a micro-optimization in `consider_assumption` where we check the def-ids of the assumption and goal match before instantiating any binders.
r? ``@lcnr``
Correct suggestions for closure arguments that need a borrow
Fixes#107301 by dealing with binders correctly
Fixes another issue where we were suggesting adding just `&` when we expected `&mut _` in a closure arg
Use new solver in `evaluate_obligation` query (when new solver is enabled)
(only when `-Ztrait-solver=next`, of course)
... Does this make sense? It seems to me like it should be reasonable, but maybe there's some reason why this is a bad idea.
r? ``@lcnr``
Needs a perf run because I guess this `solver == TraitSolver::Next` check is on a hot path.
Libs-API decided to remove these in #102697.
Follow-up to #107023, which removed them from `compiler/`, but a couple new ones showed up since that was merged.
Compute generator saved locals on MIR
Generators are currently type-checked by introducing a `witness` type variable, which is unified with a `GeneratorWitness(captured types)` whose purpose is to ensure that the auto traits correctly migrate from the captured types to the `witness` type. This requires computing the captured types on HIR during type-checking, only to re-do it on MIR later.
This PR proposes to drop the HIR-based computation, and only keep the MIR one. This is done in 3 steps.
1. During type-checking, the `witness` type variable is never unified. This allows to stall all the obligations that depend on it until the end of type-checking. Then, the stalled obligations are marked as successful, and saved into the typeck results for later verification.
2. At type-checking writeback, `witness` is replaced by `GeneratorWitnessMIR(def_id, substs)`. From this point on, all trait selection involving `GeneratorWitnessMIR` will fetch the MIR-computed locals, similar to what opaque types do. There is no lifetime to be preserved here: we consider all the lifetimes appearing in this witness type to be higher-ranked.
3. After borrowck, the stashed obligations are verified against the actually computed types, in the `check_generator_obligations` query. If any obligation was wrongly marked as fulfilled in step 1, it should be reported here.
There are still many issues:
- ~I am not too happy having to filter out some locals from the checked bounds, I think this is MIR building that introduces raw pointers polluting the analysis;~ solved by a check specific to static variables.
- the diagnostics for captured types don't show where they are used/dropped;
- I do not attempt to support chalk.
cc `@eholk` `@jyn514` for the drop-tracking work
r? `@oli-obk` as you warned me of potential unsoundness
solver comments + remove `TyCtxt::evaluate_goal`
from the `RustcContributor::explore` session yesterday.
This also removes `TyCtxt::evaluate_goal` because to canonicalize you have to use an `InferCtxt` anyways at which point we should just always get people to use `evaluate_root_goal`.
r? ``@spastorino``
ICE in new solver if we see an inference variable
By construction, we do not expect to see any `ty::Infer(ty::TyVar(_))` inference types in the solver (we treat this as ambiguous, since we need to be able to structurally resolve the self type at least one layer to assemble candidates for it). Additionally, since we're doing no freshening, we also don't expect to see any fresh vars of any kind in the solver.
Let's make that an ICE so we can catch any mistakes.
When #107282 lands, we should also ICE there too if we see a non-int/float infer.
r? `@lcnr`
use `LocalDefId` instead of `HirId` in trait resolution to simplify the obligation clause resolution
This commit introduces a refactoring suggested by `@lcnr` to simplify the obligation clause resolution.
This is just the first PR that introduces a type of refactoring, but others PRs will follow this to introduce name changing to change from the variable name from `body_id` to something else.
Fixes https://github.com/rust-lang/rust/issues/104827
Signed-off-by: Vincenzo Palazzo <vincenzopalazzodev@gmail.com>
`@rustbot` r? `@lcnr`
use LocalDefId instead of HirId in trait resolution to simplify
the obligation clause resolution
Signed-off-by: Vincenzo Palazzo <vincenzopalazzodev@gmail.com>
Implement some more new solver candidates and fix some bugs
First, fix some bugs:
1. `IndexVec::drain_enumerated(a..b)` does not give us an iterator of index keys + items enumerated from `a..b`, but from `0..(b-a)`... That caused a bug. See first commit for the fix.
2. Implement the `_: Trait` ambiguity hack. I put it in assemble, let me know if it should live elsewhere. This is important, since we otherwise consider `_: Sized` to have no solutions, and nothing passes!
3. Swap `Ambiguity` and `Unimplemented` cases for the new solver. Sorry for accidentally swapping them 😄
4. Check GATs' own predicates during projection confirmation.
Then implement a few builtin traits:
5. Implement `PointerSized`. Pretty independent.
6. Implement `Fn` family of traits for fnptr, fndef, and closures. Closures are currently broken because `FulfillCtxt::relationships` is intentionally left unimplemented. See comment in the test.
r? ```@lcnr```
Rollup of 8 pull requests
Successful merges:
- #106783 (Recover labels written as identifiers)
- #106973 (Don't treat closures from other crates as local)
- #106979 (Document how to get the type of a default associated type)
- #107053 (signal update string representation for haiku.)
- #107058 (Recognise double-equals homoglyph)
- #107067 (Custom MIR: Support storage statements)
- #107076 (Added const-generic ui test case for issue #106419)
- #107091 (Fix broken format strings in `infer.ftl`)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Don't treat closures from other crates as local
fixes#104817
r? `@lcnr`
Specialization can prefer an impl for an opaque type over a blanket impls that also matches. If the blanket impl only applies if an auto-trait applies, we look at the hidden type of the opaque type to see if that implements the auto trait. The hidden type can be a closure or generator, and thus we will end up seeing these types in coherence and have to handle them properly.
Stop using `BREAK` & `CONTINUE` in compiler
Switching them to `Break(())` and `Continue(())` instead.
Entirely search-and-replace, though there's one spot where rustfmt insisted on a reformatting too.
libs-api would like to remove these constants (https://github.com/rust-lang/rust/pull/102697#issuecomment-1385705202), so stop using them in compiler to make the removal PR later smaller.
Switching them to `Break(())` and `Continue(())` instead.
libs-api would like to remove these constants, so stop using them in compiler to make the removal PR later smaller.
Switch to `EarlyBinder` for `item_bounds` query
Part of the work to finish #105779 (also see https://github.com/rust-lang/types-team/issues/78).
Several queries `X` have a `bound_X` variant that wraps the output in `EarlyBinder`. This adds `EarlyBinder` to the return type of the `item_bounds` query and removes `bound_item_bounds`.
r? `@lcnr`
new trait solver: rebase impl substs for gats correctly
you might've caught this while working on projection code, if so then you can close this pr
r? `@lcnr`
Rework some `predicates_of`/`{Generic,Instantiated}Predicates` code
1. Make `instantiate_own` return an iterator, since it's a bit more efficient and easier to work with
2. Remove `bound_{explicit,}_predicates_of` -- these `bound_` methods in particular were a bit awkward to work with since `ty::GenericPredicates` *already* acts kinda like an `EarlyBinder` with its own `instantiate_*` methods, and had only a few call sites anyways.
3. Implement `IntoIterator` for `InstantiatedPredicates`, since it's *very* commonly being `zip`'d together.
Implement some FIXME methods in the new trait solver
Implement just enough of the solver's response logic to make it not ICE.
Also, fix a bug with `no_bound_vars` call failing due to canonical bound vars.
r? `@lcnr`
Consolidate two almost duplicated fn info extraction routines
Moves `extract_callable_info` up to trait selection, because it was being (almost) duplicated fully there for similar diagnostic purposes. This also generalizes the diagnostics we can give slightly (see UI test).
Rollup of 8 pull requests
Successful merges:
- #105526 (libcore: make result of iter::from_generator Clone)
- #106563 (Fix `unused_braces` on generic const expr macro call)
- #106661 (Stop probing for statx unless necessary)
- #106820 (Deprioritize fulfillment errors that come from expansions.)
- #106828 (rustdoc: remove `docblock` class from notable trait popover)
- #106849 (Allocate one less vec while parsing arrays)
- #106855 (rustdoc: few small cleanups)
- #106860 (Remove various double spaces in the libraries.)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Switch to `EarlyBinder` for `const_param_default` and `impl_trait_ref` queries
Part of the work to close#105779 and implement https://github.com/rust-lang/types-team/issues/78.
Several queries `X` have a `bound_X` variant that wraps the output in `EarlyBinder`. This PR adds `EarlyBinder` to the return type of `const_param_default` and `impl_trait_ref`, and removes their `bound_X` variants.
r? `@lcnr`
I encountered an instance where an `FnPtr` implemented a trait, but I was passing an `FnDef`. To
the end user, there is really no way to differentiate each of them, but it is necessary to cast
to the generic function in order to compile. It is thus useful to suggest `as` in the help note,
(even if the Fn output implements the trait).
Rename `Ty::is_ty_infer` -> `Ty::is_ty_or_numeric_infer`
Makes sure people are aware that they may have a type variable *or* an int/float variable.
r? `@oli-obk` https://github.com/rust-lang/rust/pull/106322#issuecomment-1376913539 but I could instead implement your solution, let me know.
(This will conflict with #106322 for now, ignore that 😄)
Handle inference variables in `CollectAllMismatches` correctly
1. Fix#106240
2. Treat int/float type variables correctly (see `src/test/ui/iterators/invalid-iterator-chain-with-int-infer.rs`), so we can point out things like "`Iterator::Item` changed to `{integer}` here"
Move autoderef to `rustc_hir_analysis`
Not sure if this is a change we actually want, but autoderef really is only (functionally) used by `rustc_hir_analysis` and `rustc_hir_typeck`, so it probably should live there.
Instead, implement a separate autoderef helper in `TypeErrCtxt` for the one use-case that goes against the ordering of the crate graph..
Clean up `OnUnimplementedFormatString::verify`
Lift the always-allowed symbols to a static array and replace a `match iter().find(...)` with `iter().any(...)`
Prefer non-`[type error]` candidates during selection
Fixes#102130Fixes#106351
r? types
note: Alternatively we could filter out error where-clauses during param-env construction? But we still need to filter out impls with errors during `match_impl`, I think.
Note predicate span on `ImplDerivedObligation`
Seems obvious to point out the where-clause that introduces the `ImplDerivedObligation` :)
r? `@estebank`
Check `impl`'s `where` clauses in `consider_impl_candidate` in experimental solver
Check impl's nested predicates as part of the recursive evaluate in `consider_impl_candidate`.
<sub>Unless, for some reason, these are intentionally **not** checked here -- in which case, I really don't understand where they're being checked...<sub>
r? ```@lcnr```
remove E0280
After looking at #61137 I tried my hand at E0280. I'm unable to find a reasonable example that emits the error. There are a couple of old examples that compile with the current compiler ([#26217](https://github.com/rust-lang/rust/issues/26217), [#42114](https://github.com/rust-lang/rust/issues/42114), [#27113](https://github.com/rust-lang/rust/issues/27113)) and there is a [bug with chalk](b7cdb635c4/src/test/ui/chalkify/bugs/async.rs) that makes it emit the error, with a couple more chalk bugs on zulip.
It seems like the error is supposed to be emitted from unfulfilled where bounds, of which two are related to borrow checking (error in where T: 'a or where 'a: 'b) and thus tend to emit errors like "lifetime may not live long enough" from borrow checking instead. The final case is with type equality constraints (where <T as Iterator>::Item == u32), which is unimplemented ([#20041](https://github.com/rust-lang/rust/issues/20041)). That such different problems are supposed to have the same error code also seems strange to me.
Since the error seems to only be emitted when using chalk I propose to remove it and replace it with an ICE instead. A crater run might be warranted.
Pinging `@jackh726` due to removal of chalk test that now ICEs.
Simplify some canonical type alias names
* delete the `Canonicalized<'tcx>` type alias in favor for `Canonical<'tcx>`
* `CanonicalizedQueryResponse` -> `CanonicalQueryResponse`
I don't particularly care about the latter, but it should be consistent. We could alternatively delete the first alias and rename the struct to `Canonicalized`, and then keep the name of `CanonicalizedQueryResponse` untouched.
Remove invalid case for mutable borrow suggestion
If we have a call such as `foo(&mut buf)` and after reference
collapsing the type is inferred as `&T` where-as the required type is
`&mut T`, don't suggest `foo(&mut mut buf)`. This is wrong syntactically
and the issue lies elsewhere, not in the borrow.
Fixes#105645
Simplify match statement
Add multiple tests
- 1 test for checking `N + 1 + 1` does not unify with `N+1`
- 2 tests for checking that a function that uses two parameters only returns the parameter that
is actually used.
- Check exact repeat predicates
Instead of just switching to a probe, check for different matches, and see how many there are.
If one, unify it, otherwise return true and let it be unified later.
fix [type error] for error E0029 and E0277
check explicitly for the type references error
if ty.references_error() is true change the error to be err.delay_as_bug() and prevent the error E0029 and E0277 from emitting out this fix#105946
check explicitly for the type references error
if ty.references_error() is true change the error to be err.delay_as_bug()
and prevent the error E0029 and E0277 from emitting out
this fix#105946
Rollup of 6 pull requests
Successful merges:
- #105846 (Account for return-position `impl Trait` in trait in `opt_suggest_box_span`)
- #106385 (Split `-Zchalk` flag into `-Ztrait-solver=(classic|chalk|next)` flag)
- #106403 (Rename `hir::Map::{get_,find_}parent_node` to `hir::Map::{,opt_}parent_id`, and add `hir::Map::{get,find}_parent`)
- #106462 (rustdoc: remove unnecessary wrapper around sidebar and mobile logos)
- #106464 (Update Fuchsia walkthrough with new configs)
- #106478 (Tweak wording of fn call with wrong number of args)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Rename `hir::Map::{get_,find_}parent_node` to `hir::Map::{,opt_}parent_id`, and add `hir::Map::{get,find}_parent`
The `hir::Map::get_parent_node` function doesn't return a `Node`, and I think that's quite confusing. Let's rename it to something that sounds more like something that gets the parent hir id => `hir::Map::parent_id`. Same with `find_parent_node` => `opt_parent_id`.
Also, combine `hir.get(hir.parent_id(hir_id))` and similar `hir.find(hir.parent_id(hir_id))` function into new functions that actually retrieve the parent node in one call. This last commit is the only one that might need to be looked at closely.
Don't deduce a signature that makes a closure cyclic
Sometimes when elaborating supertrait bounds for closure signature inference, we end up deducing a closure signature that is cyclical because either a parameter or the return type references a projection mentioning `Self` that also has escaping bound vars, which means that it's not eagerly replaced with an inference variable.
Interestingly, this is not *just* related to my PR that elaborates supertrait bounds for closure signature deduction. The committed test `supertrait-hint-cycle-3.rs` shows **stable** code that is fixed by this PR:
```rust
trait Foo<'a> {
type Input;
}
impl<F: Fn(u32)> Foo<'_> for F {
type Input = u32;
}
fn needs_super<F: for<'a> Fn(<F as Foo<'a>>::Input) + for<'a> Foo<'a>>(_: F) {}
fn main() {
needs_super(|_: u32| {});
}
```
Fixes#105401Fixes#105396
r? types
Rollup of 9 pull requests
Successful merges:
- #104531 (Provide a better error and a suggestion for `Fn` traits with lifetime params)
- #105899 (`./x doc library --open` opens `std`)
- #106190 (Account for multiple multiline spans with empty padding)
- #106202 (Trim more paths in obligation types)
- #106234 (rustdoc: simplify settings, help, and copy button CSS by not reusing)
- #106236 (docs/test: add docs and a UI test for `E0514` and `E0519`)
- #106259 (Update Clippy)
- #106260 (Fix index out of bounds issues in rustdoc)
- #106263 (Formatter should not try to format non-Rust files)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Some `compare_method` tweaks
1. Make some of the comparison functions' names more regular
2. Reduce pub scope of some of the things in `compare_method`
~3. Remove some unnecessary opaque type handling code -- `InferCtxt` already is in a mode that doesn't define opaque types~
* moved to a different PR
4. Bubble up `ErrorGuaranteed` for region constraint errors in `compare_method` - Improves a redundant error message in one unit test.
5. Move the `compare_method` module to have a more general name, since it's more like `compare_impl_item` :)
6. Rename `collect_trait_impl_trait_tys`
implement the skeleton of the updated trait solver
cc ```@rust-lang/initiative-trait-system-refactor```
This is mostly following the architecture discussed in the types team meetup.
After discussing the desired changes for the trait solver, we encountered cyclic dependencies between them. Most notably between changing evaluate to be canonical and returning inference constraints. We cannot canonicalize evaluate without returning inference constraints due to coinductive cycles. However, caching inference constraints also relies on canonicalization. Implementing both of these changes at once in-place is not feasible.
This somewhat closely mirrors the current `evaluate` implementation with the following notable differences:
- it moves `project` into the core solver, allowing us to correctly deal with coinductive projections (will be required for implied bounds, perfect derive)
- it changes trait solver overflow to be non-fatal (required to backcompat breakage from changes to the iteration order of nested goals, deferred projection equality, generally very useful)
- it returns inference constraints and canonicalizes inputs and outputs (required for a lot things, most notably merging fulfill and evaluate, and deferred projection equality)
- it is implemented to work with lazy normalization
A lot of things aren't yet implemented, but the remaining FIXMEs should all be fairly self-contained and parallelizable. If the architecture looks correct and is what we want here, I would like to quickly merge this and then split the work.
r? ```@compiler-errors``` / ```@rust-lang/types``` :3
always use `anonymize_bound_vars`
Unless this is perf-sensitive, it's probably best to always use one anonymize function that does the right thing for all bound vars.
r? types
