Inline ExprPrecedence::order into Expr::precedence
The representation of expression precedence in rustc_ast has been an obstacle to further improvements in the pretty-printer (continuing from #119105 and #119427).
Previously the operation of *"does this expression have lower precedence than that one"* (relevant for parenthesis insertion in macro-generated syntax trees) consisted of 3 steps:
1. Convert `Expr` to `ExprPrecedence` using `.precedence()`
2. Convert `ExprPrecedence` to `i8` using `.order()`
3. Compare using `<`
As far as I can guess, the reason for the separation between `precedence()` and `order()` was so that both `rustc_ast::Expr` and `rustc_hir::Expr` could convert as straightforwardly as possible to the same `ExprPrecedence` enum, and then the more finicky logic performed by `order` could be present just once.
The mapping between `Expr` and `ExprPrecedence` was intended to be as straightforward as possible:
```rust
match self.kind {
ExprKind::Closure(..) => ExprPrecedence::Closure,
...
}
```
although there were exceptions of both many-to-one, and one-to-many:
```rust
ExprKind::Underscore => ExprPrecedence::Path,
ExprKind::Path(..) => ExprPrecedence::Path,
...
ExprKind::Match(_, _, MatchKind::Prefix) => ExprPrecedence::Match,
ExprKind::Match(_, _, MatchKind::Postfix) => ExprPrecedence::PostfixMatch,
```
Where the nature of `ExprPrecedence` becomes problematic is when a single expression kind might be associated with multiple different precedence levels depending on context (outside the expression) and contents (inside the expression). For example consider what is the precedence of an ExprKind::Closure `$closure`. Well, on the left-hand side of a binary operator it would need parentheses in order to avoid the trailing binary operator being absorbed into the closure body: `($closure) + Rhs`, so the precedence is something lower than that of `+`. But on the right-hand side of a binary operator, a closure is just a straightforward prefix expression like a unary op, which is a relatively high precedence level, higher than binops but lower than method calls: `Lhs + $closure` is fine without parens but `($closure).method()` needs them. But as a third case, if the closure contains an explicit return type, then the precedence is an even higher level than that, never needing parenthesization even in a binop left-hand side or method call: `|| -> bool { false } + Rhs` or `|| -> bool { false }.method()`.
You can see that trying to capture all of this resolution about expressions into `ExprPrecedence` violates the intention of `ExprPrecedence` being a straightforward one-to-one correspondence from each AST and HIR `ExprKind` variant. It would be possible to attempt that by doing stuff like `ExprPrecedence::Closure(Side::Leading, ReturnType::No)`, but I don't foresee the original envisioned benefit of the `precedence()`/`order()` distinction being retained in this approach. Instead I want to move toward a model that Syn has been using successfully. In Syn, there is a Precedence enum but it differs from rustc in the following ways:
- There are [relatively few variants](https://github.com/dtolnay/syn/blob/2.0.87/src/precedence.rs#L11-L47) compared to rustc's `ExprPrecedence`. For example there is no distinction at the precedence level between returns and closures, or between loops and method calls.
- We distinguish between [leading](https://github.com/dtolnay/syn/blob/2.0.87/src/fixup.rs#L293) and [trailing](https://github.com/dtolnay/syn/blob/2.0.87/src/fixup.rs#L309) precedence, taking into account an expression's context such as what token follows it (for various syntactic bail-outs in Rust's grammar, like ambiguities around break-with-value) and how it relates to operators from the surrounding syntax tree.
- There are no hardcoded mysterious integer quantities like rustc's `PREC_CLOSURE = -40`. All precedence comparisons are performed via PartialOrd on a C-like enum.
This PR is just a first step in these changes. As you can tell from Syn, I definitely think there is value in having a dedicated type to represent precedence, instead of what `order()` is doing with `i8`. But that is a whole separate adventure because rustc_ast doesn't even agree consistently on `i8` being the type for precedence order; `AssocOp::precedence` instead uses `usize` and there are casts in both directions. It is likely that a type called `ExprPrecedence` will re-appear, but it will look substantially different from the one that existed before this PR.
Support ?Trait bounds in supertraits and dyn Trait under a feature gate
This patch allows `maybe` polarity bounds under a feature gate. The only language change here is that corresponding hard errors are replaced by feature gates. Example:
```rust
#![feature(allow_maybe_polarity)]
...
trait Trait1 : ?Trait { ... } // ok
fn foo(_: Box<(dyn Trait2 + ?Trait)>) {} // ok
fn bar<T: ?Sized + ?Trait>(_: &T) {} // ok
```
Maybe bounds still don't do anything (except for `Sized` trait), however this patch will allow us to [experiment with default auto traits](https://github.com/rust-lang/rust/pull/120706#issuecomment-1934006762).
This is a part of the [MCP: Low level components for async drop](https://github.com/rust-lang/compiler-team/issues/727)
`#[naked]`: use an allowlist for allowed options on `asm!` in naked functions
tracking issue: https://github.com/rust-lang/rust/issues/90957
this is mostly just a refactor, but using an allowlist (rather than a denylist) for which asm options are allowed in naked functions is a little safer.
These options are disallowed because naked functions are effectively global asm, but defined using inline asm.
This is a very large commit since a lot needs to be changed in order to
make the tests pass. The salient changes are:
- `ConstArgKind` gets a new `Path` variant, and all const params are now
represented using it. Non-param paths still use `ConstArgKind::Anon`
to prevent this change from getting too large, but they will soon use
the `Path` variant too.
- `ConstArg` gets a distinct `hir_id` field and its own variant in
`hir::Node`. This affected many parts of the compiler that expected
the parent of an `AnonConst` to be the containing context (e.g., an
array repeat expression). They have been changed to check the
"grandparent" where necessary.
- Some `ast::AnonConst`s now have their `DefId`s created in
rustc_ast_lowering rather than `DefCollector`. This is because in some
cases they will end up becoming a `ConstArgKind::Path` instead, which
has no `DefId`. We have to solve this in a hacky way where we guess
whether the `AnonConst` could end up as a path const since we can't
know for sure until after name resolution (`N` could refer to a free
const or a nullary struct). If it has no chance as being a const
param, then we create a `DefId` in `DefCollector` -- otherwise we
decide during ast_lowering. This will have to be updated once all path
consts use `ConstArgKind::Path`.
- We explicitly use `ConstArgHasType` for array lengths, rather than
implicitly relying on anon const type feeding -- this is due to the
addition of `ConstArgKind::Path`.
- Some tests have their outputs changed, but the changes are for the
most part minor (including removing duplicate or almost-duplicate
errors). One test now ICEs, but it is for an incomplete, unstable
feature and is now tracked at #127009.
We already do this for a number of crates, e.g. `rustc_middle`,
`rustc_span`, `rustc_metadata`, `rustc_span`, `rustc_errors`.
For the ones we don't, in many cases the attributes are a mess.
- There is no consistency about order of attribute kinds (e.g.
`allow`/`deny`/`feature`).
- Within attribute kind groups (e.g. the `feature` attributes),
sometimes the order is alphabetical, and sometimes there is no
particular order.
- Sometimes the attributes of a particular kind aren't even grouped
all together, e.g. there might be a `feature`, then an `allow`, then
another `feature`.
This commit extends the existing sorting to all compiler crates,
increasing consistency. If any new attribute line is added there is now
only one place it can go -- no need for arbitrary decisions.
Exceptions:
- `rustc_log`, `rustc_next_trait_solver` and `rustc_type_ir_macros`,
because they have no crate attributes.
- `rustc_codegen_gcc`, because it's quasi-external to rustc (e.g. it's
ignored in `rustfmt.toml`).
Rename HIR `TypeBinding` to `AssocItemConstraint` and related cleanup
Rename `hir::TypeBinding` and `ast::AssocConstraint` to `AssocItemConstraint` and update all items and locals using the old terminology.
Motivation: The terminology *type binding* is extremely outdated. "Type bindings" not only include constraints on associated *types* but also on associated *constants* (feature `associated_const_equality`) and on RPITITs of associated *functions* (feature `return_type_notation`). Hence the word *item* in the new name. Furthermore, the word *binding* commonly refers to a mapping from a binder/identifier to a "value" for some definition of "value". Its use in "type binding" made sense when equality constraints (e.g., `AssocTy = Ty`) were the only kind of associated item constraint. Nowadays however, we also have *associated type bounds* (e.g., `AssocTy: Bound`) for which the term *binding* doesn't make sense.
---
Old terminology (HIR, rustdoc):
```
`TypeBinding`: (associated) type binding
├── `Constraint`: associated type bound
└── `Equality`: (associated) equality constraint (?)
├── `Ty`: (associated) type binding
└── `Const`: associated const equality (constraint)
```
Old terminology (AST, abbrev.):
```
`AssocConstraint`
├── `Bound`
└── `Equality`
├── `Ty`
└── `Const`
```
New terminology (AST, HIR, rustdoc):
```
`AssocItemConstraint`: associated item constraint
├── `Bound`: associated type bound
└── `Equality`: associated item equality constraint OR associated item binding (for short)
├── `Ty`: associated type equality constraint OR associated type binding (for short)
└── `Const`: associated const equality constraint OR associated const binding (for short)
```
r? compiler-errors
The current way of stepping through each comment in `Comments` is a bit
weird. There is a `Vec<Comments>` and a `current` index, which is fine.
The `Comments::next` method clones the current comment but doesn't
advance `current`; the advancing instead happens in `print_comment`,
which is where each cloned comment is actually finally used (or not, in
some cases, if the comment fails to satisfy a predicate).
This commit makes things more iterator-like:
- `Comments::next` now advances `current` instead of `print_comment`.
- `Comments::peek` is added so you can inspect a comment and check a
predicate without consuming it.
- This requires splitting `PrintState::comments` into immutable and
mutable versions. The commit also moves the ref inside the `Option` of
the return type, to save callers from having to use `as_ref`/`as_mut`.
- It also requires adding `PrintState::peek_comment` alongside the
existing `PrintState::next_comment`. (The lifetimes in the signature
of `peek_comment` ended up more complex than I expected.)
We now have a neat separation between consuming (`next`) and
non-consuming (`peek`) uses of each comment. As well as being clearer,
this will facilitate the next commit that avoids unnecessary cloning.
Implement syntax for `impl Trait` to specify its captures explicitly (`feature(precise_capturing)`)
Implements `impl use<'a, 'b, T, U> Sized` syntax that allows users to explicitly list the captured parameters for an opaque, rather than inferring it from the opaque's bounds (or capturing *all* lifetimes under 2024-edition capture rules). This allows us to exclude some implicit captures, so this syntax may be used as a migration strategy for changes due to #117587.
We represent this list of captured params as `PreciseCapturingArg` in AST and HIR, resolving them between `rustc_resolve` and `resolve_bound_vars`. Later on, we validate that the opaques only capture the parameters in this list.
We artificially limit the feature to *require* mentioning all type and const parameters, since we don't currently have support for non-lifetime bivariant generics. This can be relaxed in the future.
We also may need to limit this to require naming *all* lifetime parameters for RPITIT, since GATs have no variance. I have to investigate this. This can also be relaxed in the future.
r? `@oli-obk`
Tracking issue:
- https://github.com/rust-lang/rust/issues/123432
