This commit uses the map introduced by the previous commit to ensure
that types are always checked for well-formedness by the NLL type check.
Previously, without the map introduced by the previous commit, types
would not be checked for well-formedness if the `AscribeUserType`
statement that would trigger that check was removed as unreachable code.
This commit refactors the `UserTypeAnnotation` type to be referred to by
an index within `UserTypeProjection`. `UserTypeAnnotation` is instead
kept in an `IndexVec` within the `Mir` struct.
Further, instead of `UserTypeAnnotation` containing canonicalized types,
it now contains normal types and the entire `UserTypeAnnotation` is
canonicalized. To support this, the type was moved from the `rustc::mir`
module to `rustc::ty` module.
Mention ToString in std::fmt docs
I believe this should be added because `x.to_string()` is preferred over `format!("{}", x)` as the recommended way to convert a value to a string. `std::fmt` and the `format` macro is where people look for documentation about this, and thus we should include references to this preferred functions there.
Resolves#55065
resolve: Fix one more ICE in import validation
So if you have an unresolved import
```rust
mod m {
use foo::bar;
}
```
error recovery will insert a special item with `Def::Err` definition into module `m`, so other things depending on `bar` won't produce extra errors.
The issue was that erroneous `bar` was overwriting legitimate `bar`s coming from globs, e.g.
```rust
mod m {
use baz::*; // imports real existing `bar`
use foo::bar;
}
```
causing some unwanted diagnostics talking about "unresolved items", and producing inconsistent resolutions like https://github.com/rust-lang/rust/issues/57015.
This PR stops overwriting real successful resolutions with `Def::Err`s.
Fixes https://github.com/rust-lang/rust/issues/57015
Implement RFC 2338, "Type alias enum variants"
This PR implements [RFC 2338](https://github.com/rust-lang/rfcs/pull/2338), allowing one to write code like the following.
```rust
#![feature(type_alias_enum_variants)]
enum Foo {
Bar(i32),
Baz { i: i32 },
}
type Alias = Foo;
fn main() {
let t = Alias::Bar(0);
let t = Alias::Baz { i: 0 };
match t {
Alias::Bar(_i) => {}
Alias::Baz { i: _i } => {}
}
}
```
Since `Self` can be considered a type alias in this context, it also enables using `Self::Variant` as both a constructor and pattern.
Fixes issues #56199 and #56611.
N.B., after discussing the syntax for type arguments on enum variants with @petrochenkov and @eddyb (there are also a few comments on the [tracking issue](https://github.com/rust-lang/rust/issues/49683)), the consensus seems to be treat the syntax as follows, which ought to be backwards-compatible.
```rust
Option::<u8>::None; // OK
Option::None::<u8>; // OK, but lint in near future (hard error next edition?)
Alias::<u8>::None; // OK
Alias::None::<u8>; // Error
```
I do not know if this will need an FCP, but let's start one if so.
resolve: Fix another ICE in import validation
Imports are allowed to have ambiguous resolutions as long as all of them have same `Def`.
As it turned out, it's possible for different `Module`s to have same `Def` when `extern crate` items are involved.
Fixes https://github.com/rust-lang/rust/issues/56596
submodules: update clippy from fc24fce7 to 721f688e
Fixes clippy toolstate.
Changes:
````
rustup https://github.com/rust-lang/rust/pull/57069
Rustfmt.
fix breakage from rust-lang/rust#57088
fix a couple of ftrivial typos (NFC).
update CARGO_CLIPPY_HELP string to suggest tool lints.
rustc_tools_util: add readme
rustc_tool_utils: expand Cargo.toml with a few keywords in preparation for crates.io release
Fix macro detection in `empty_loop`.
Changed `macro_backtrace()` to `in_macro()`.
Fix lint detection on macro expansion.
````
r? @oli-obk
Fixes clippy toolstate
Changes:
````
Match on ast/hir::ExprKind::Err
Update *.stderr files
Use -Zui-testing flag
Mention S-inactive-closed PRs in the CONTRIBUTING.md
tests: fix formatting and update test output
base tests: make sure to fail CI if tests need formatting
base tests: switch to nightly toolchain before checking formatting of tests with rustfmt
rustup https://github.com/rust-lang/rust/pull/57069
Rustfmt.
fix breakage from rust-lang/rust#57088
fix a couple of ftrivial typos (NFC).
update CARGO_CLIPPY_HELP string to suggest tool lints.
rustc_tools_util: add readme
rustc_tool_utils: expand Cargo.toml with a few keywords in preparation for crates.io release
Fix macro detection in `empty_loop`.
Changed `macro_backtrace()` to `in_macro()`.
Fix lint detection on macro expansion.
````
Resolve `$crate`s for pretty-printing at more appropriate time
Doing it in `BuildReducedGraphVisitor` wasn't a good idea, identifiers wasn't actually visited half of the time.
As a result some `$crate`s weren't resolved and were therefore pretty-printed as `$crate` literally, which turns into two tokens during re-parsing of the pretty-printed text.
Now we are visiting and resolving `$crate` identifiers in an item right before sending that item to a proc macro attribute or derive.
Fixes https://github.com/rust-lang/rust/issues/57089
AST/HIR: Introduce `ExprKind::Err` for better error recovery in the front-end
This way we can avoid aborting compilation if expansion produces errors and generate `ExprKind::Err`s instead.
Implement the new-style trait solver
Final PR of what I believe to be a minimally working implementation of the new-style trait solver.
The new trait solver can be used by providing the `-Z chalk` command line flag. It is currently used everywhere in `rustc_typeck`, and for everything relying on `rustc::infer::canonical::query_response::enter_canonical_trait_query`.
The trait solver is invoked in rustc by using the `evaluate_goal` canonical query. This is not optimal because each call to `evaluate_goal` creates a new `chalk_engine::Forest`, hence rustc cannot use answers to intermediate goals produced by the root goal. We'll need to change that but I guess that's ok for now.
Some next steps, I think, are:
* handle region constraints: region constraints are computed but are completely ignored for now, I think we may need additional support from `chalk_engine` (as a side effect, types or trait references with outlive requirements cannot be proved well-formed)
* deactivate eager normalization in the presence of `-Z chalk` in order to leverage the lazy normalization strategy of the new-style trait solver
* add the remaining built-in impls (only `Sized` is supported currently)
* transition the compiler to using generic goals instead of predicates that still refer to named type parameters etc
I added a few very simple tests to check that the new solver has the right behavior, they won't be needed anymore once it is mature enough. Additionally it shows off that we get [implied bounds](https://github.com/rust-lang/rust/issues/44491) for free.
r? @nikomatsakis
