Register obligations during path inference
Fixes#14635
When we infer path expressions that resolve to some generic item, we need to consider their generic bounds. For example, when we resolve a path `Into::into` to `fn into<?0, ?1>` (note that `?0` is the self type of trait ref), we should register an obligation `?0: Into<?1>` or else their relationship would be lost.
Relevant part in rustc is [`add_required_obligations_with_code()`] that's called in [`instantiate_value_path()`].
[`instantiate_value_path()`]: 3462f79e94/compiler/rustc_hir_typeck/src/fn_ctxt/_impl.rs (L1052)
[`add_required_obligations_with_code()`]: 3462f79e94/compiler/rustc_hir_typeck/src/fn_ctxt/_impl.rs (L1411)
Normalize associated types in paths in expressions
Part of #14393
When we resolve paths in expressions (either path expressions or paths in struct expressions), there's a need of projection normalization, which `TyLoweringContext` cannot do on its own. We've been properly applying normalization for paths in struct expressions without type anchor, but not for others:
```rust
enum E {
S { v: i32 }
Empty,
}
impl Foo for Bar {
type Assoc = E;
fn foo() {
let _ = Self::Assoc::S { v: 42 }; // path in struct expr without type anchor; we already support this
let _ = <Self>::Assoc::S { v: 42 }; // path in struct expr with type anchor; resolves with this PR
let _ = Self::Assoc::Empty; // path expr; resolves with this PR
}
}
```
With this PR we correctly resolve the whole path, but we need some more tweaks in HIR and/or IDE layers to properly resolve a qualifier (prefix) of such paths and provide IDE features that are pointed out in #14393 to be currently broken.
MIR episode 3
This PR adds lowering for try operator and overloaded dereference, and adds evaluating support for function pointers and trait objects. It also adds a flag to `analysis-stats` to show percentage of functions that it fails to emit mir for them, which is currently `20%` (which is somehow lying, as most of the supported `80%` are tests). The most offenders are closure (1975 items) and overloaded index (415 items). I will try to add overloaded index before monday to have it in this PR, and tackle the closure in the next episode.
This makes code more readale and concise,
moving all format arguments like `format!("{}", foo)`
into the more compact `format!("{foo}")` form.
The change was automatically created with, so there are far less change
of an accidental typo.
```
cargo clippy --fix -- -A clippy::all -W clippy::uninlined_format_args
```
fix: normalize projection after discarding free `BoundVar`s in RPIT
Fixes#13307
When we lower the return type of a function, it may contain free `BoundVar`s in `OpaqueType`'s substitution, which would cause panic during canonicalization as part of projection normalization. Those `BoundVar`s are irrelevant in this context and will be discarded, and we should defer projection normalization until then.
fix: disregard type variable expectation for if expressions
Fixes#13522
As [the comment](8142d1f606/crates/hir-ty/src/infer.rs (L1087-L1090)) on `Expectation::adjust_for_branches` explains:
> If the expected type is just a type variable, then don't use an expected type. Otherwise, we might write parts of the type when checking the 'then' block which are incompatible with the 'else' branch.
Note that we already use it in match expressions. I've added tests for them too nevertheless.
fix: reorder dyn bounds on render
Fixes#13368#13192 changed the order of dyn bounds, violating the [contract](3a69435af7/crates/hir-ty/src/display.rs (L896-L901)) with `write_bounds_like_dyn_trait()` on render. The projection bounds are expected to come right after the trait bound they are accompanied with.
Although the reordering procedure can be made a bit more efficient, I opted for relying only on the [invariants](3a69435af7/crates/hir-ty/src/lower.rs (L995-L998)) currently documented in `lower_dyn_trait()`. It's not the hottest path and dyn bounds tend to be short so I believe it shouldn't hurt performance noticeably.
feat: type inference for generators
This PR implements basic type inference for generator and yield expressions.
Things not included in this PR:
- Generator upvars and generator witnesses are not implemented. They are only used to determine auto trait impls, so basic type inference should be fine without them, but method resolutions with auto trait bounds may not be resolved correctly.
Open questions:
- I haven't (yet) implemented `HirDisplay` for `TyKind::Generator`, so generator types are just shown as "{{generator}}" (in tests, inlay hints, hovers, etc.), which is not really nice. How should we show them?
- I added moderate amount of stuffs to minicore. I especially didn't want to add `impl<T> Deref for &T` and `impl<T> Deref for &mut T` exclusively for tests for generators; should I move them into the test fixtures or can they be placed in minicore?
cc #4309
fix: handle trait methods as inherent methods for trait-related types
Fixes#10677
When resolving methods for trait object types and placeholder types that are bounded by traits, we need to count the methods of the trait and its super traits as inherent methods. This matters because these trait methods have higher priority than the other traits' methods.
Relevant code in rustc: [`assemble_inherent_candidates_from_object()`](0631ea5d73/compiler/rustc_typeck/src/check/method/probe.rs (L783-L792)) for trait object types, [`assemble_inherent_candidates_from_param()`](0631ea5d73/compiler/rustc_typeck/src/check/method/probe.rs (L838-L847)) for placeholder types. Notice the second arg of `push_candidate()` is `is_inherent`.
This hir expression isn't needed and only existed as it was simpler to
deal with at first as it gave us a direct mapping for the ast version of
the same construct. This PR removes it, properly handling the statements
that are introduced by macro call expressions.
fix: sort and deduplicate auto traits in trait object types
Fixes#12739
Chalk solver doesn't sort and deduplicate auto traits in trait object types, so we need to handle them ourselves in the lowering phase, just like [`rustc`](880416180b/compiler/rustc_typeck/src/astconv/mod.rs (L1487-L1488)) and [`chalk-integration`](https://github.com/rust-lang/chalk/blob/master/chalk-integration/src/lowering.rs#L575) do.
Quoting from [the Chalk book](https://rust-lang.github.io/chalk/book/types/rust_types.html#dyn-types):
> Note that -- for this purpose -- ordering of bounds is significant. That means that if you create a `dyn Foo + Send` and a `dyn Send + Foo`, chalk would consider them distinct types. The assumption is that bounds are ordered in some canonical fashion somewhere else.
Also, trait object types with more than one non-auto traits were previously allowed, but are now disallowed with this patch.
fix: Fix panics on GATs involving const generics
This workaround avoids constant crashing of rust analyzer when using GATs with const generics,
even when the const generics are only on the `impl` block.
The workaround treats GATs as non-existing if either itself or the parent has const generics and
removes relevant panicking code-paths.
Fixes#11989, fixes#12193
