Make liveness more precise for assignments to fields
Previously, we were too conservative and `x.field = 4` was treated as a "use" of `x`. Now it neither kills `x` (since other fields of `x` may still be live) nor marks it as live.
cc @jonas-schievink, who ran into this problem.
errors: use `-Z terminal-width` in JSON emitter
This PR makes the JSON emitter use `-Z terminal-width` in the "rendered" field of the JSON output.
r? @estebank
Explain move errors that occur due to method calls involving `self` (take two)
This is a re-attempt of #72389 (which was reverted in #73594)
Instead of using `ExpnKind::Desugaring` to represent operators, this PR
checks the lang item directly.
Adds a clearer message for when the async keyword is missing from a f…
…unction
This is a somewhat simple fix for #66731.
Under the current version of Rust, if a user has a rust file that looks like this:
```rust
fn boo (){}
async fn foo() {
boo().await;
}
fn main() {
}
```
And they attempt to run it, they will receive an error message that looks like this:
```bash
error: incorrect use of `await` --> test.rs:4:14 | 4 | boo.await(); | ^^ help: `await` is not a method call, remove the parentheses error[E0277]: the trait bound `fn() {boo}: std::future::Future` is not satisfied --> test.rs:4:5 | 4 | boo.await(); | ^^^^^^^^^ the trait `std::future::Future` is not implemented for `fn() {boo}` error: aborting due to 2 previous errors For more information about this error, try `rustc --explain E0277`.
```
This is not very clear.
With the changes made in this PR, when a user compiles and runs that same rust code, they will receive an error message that looks like this:
```bash
error[E0277]: `()` is not a future.
--> test.rs:4:5
|
4 | boo().await;
| ^^^^^^^^^^^ `()` is not a future
|
= help: the trait `std::future::Future` is not implemented for `()`
= note: required by `std::future::Future::poll`
```
In the future, I think we should make this error message even clearer, perhaps through a solution like the one described in [this comment](https://github.com/rust-lang/rust/issues/66731#issuecomment-644394287). However, as that potentially involves a major change proposal, I would rather get this change in now and make the error message a little clearer while an MCP is drafted and discussed.
Signed-off-by: Nell Shamrell <nellshamrell@gmail.com>
Fixes#67765
When reporting errors during MIR region inference, we sometimes use
`universal_upper_bound` to obtain a named universal region that we
can display to the user. However, this is not always possible - in a
case like `fn foo<'a, 'b>() { .. }`, the only upper bound for a region
containing `'a` and `'b` is `'static`. When displaying diagnostics, it's
usually better to display *some* named region (even if there are
multiple involved) rather than fall back to a generic error involving
`'static`.
This commit adds a new `approx_universal_upper_bound` method, which
uses the lowest-numbered universal region if the only alternative is to
return `'static`.
This commit applies the existing 'extra angle bracket recovery' logic
when parsing fields in struct definitions. This allows us to continue
parsing the struct's fields, avoiding spurious 'missing field' errors in
code that tries to use the struct.
rustc_lexer: Simplify shebang parsing once more
Fixes https://github.com/rust-lang/rust/issues/73250 (beta regression)
Treat any line starting with `!#` as a shebang candidate, not only lines with something non-whitespace.
This way we no longer need to define what `is_whitespace` means ([Linux shebang whitespace](https://github.com/torvalds/linux/blob/master/fs/binfmt_script.c), ASCII whitespace, Rust lexer whitespace, etc), which is nice.
This change makes some invalid Rust code valid (see the regression above), but still never interprets a fragment of valid Rust code as a shebang.
(This PR also removes one duplicate test.)
This is a re-attempt of #72389 (which was reverted in #73594)
Instead of using `ExpnKind::Desugaring` to represent operators, this PR
checks the lang item directly.
Show the values and computation that would overflow a const evaluation or propagation
Fixes#71134
In contrast to the example in the issue it doesn't use individual spans for each operand. The effort required to implement that is quite high compared to the little (if at all) benefit it would bring to diagnostics.
cc @shepmaster
The way this is implemented it is also fairly easy to do the same for overflow panics at runtime, but that should be done in a separate PR since it may have runtime performance implications.
Update Chalk to 0.14
Not a ton here. Notable changes:
- Update to `0.14.0`
- New dependency on `tracing`, in `librustc_traits` only
- `FnAbi` from Chalk is `rustc_target::spec::abi::Abi`
- `Dynamic` actually lowers region
- Actually lower closures, with some tests. This doesn't 100% work, but can't confirm that's *only* because of closure lowering.
- Use `FxIndexSet` instead of `FxHashSet` in `chalk_fulfill`, which seems to have fixed the non-deterministic test error ordering. Guess we'll see on CI
- Actually implement `opaque_ty_data`, though I don't think this is sufficient for tests for them (I haven't added any)
- Uncomment some of the chalk tests that now work
r? @nikomatsakis
Provide suggestions for some moved value errors
When encountering an used moved value where the previous move happened
in a `match` or `if let` pattern, suggest using `ref`. Fix#63988.
When encountering a `&mut` value that is used in multiple iterations of
a loop, suggest reborrowing it with `&mut *`. Fix#62112.
Add unstable `core::mem::variant_count` intrinsic
Adds a new `const fn` intrinsic which can be used to determine the number of variants in an `enum`.
I've shown this to a couple of people and they invariably ask 'why on earth?', but there's actually a very neat use case:
At the moment, if you want to create an opaque array type that's indexed by an `enum` with one element for each variant, you either have to hard-code the number of variants, add a `LENGTH` variant or use a `Vec`, none of which are suitable in general (number of variants could change; pattern matching `LENGTH` becomes frustrating; might not have `alloc`). By including this intrinsic, it becomes possible to write the following:
```rust
#[derive(Copy, Clone)]
enum OpaqueIndex {
A = 0,
B,
C,
}
struct OpaqueVec<T>(Box<[T; std::mem::num_variants::<OpaqueIndex>()]>);
impl<T> std::ops::Index<OpaqueIndex> for OpaqueVec<T> {
type Output = T;
fn index(&self, idx: OpaqueIndex) -> &Self::Output {
&self.0[idx as usize]
}
}
```
(We even have a use cases for this in `rustc` and I plan to use it to re-implement the lang-items table.)
Implement mixed script confusable lint.
This implements the mixed script confusable lint defined in RFC 2457.
This is blocked on #72069 and https://github.com/unicode-rs/unicode-security/pull/13, and will need a Cargo.toml version bump after those are resolved.
The lint message warning is sub-optimal for now. We'll need a mechanism to properly output `AugmentScriptSet` to screen, this is to be added in `unicode-security` crate.
r? @Manishearth
When encountering an used moved value where the previous move happened
in a `match` or `if let` pattern, suggest using `ref`. Fix#63988.
When encountering a `&mut` value that is used in multiple iterations of
a loop, suggest reborrowing it with `&mut *`. Fix#62112.
Add UI test for issue 73592
It happens that #72280 accidentally fixed a bug which is later discovered in #73592. This PR adds a UI test to prevent future regression.
Closes#73592
`improper_ctypes_definitions` lint
Addresses #19834, #66220, and #66373.
This PR takes another attempt at #65134 (reverted in #66378). Instead of modifying the existing `improper_ctypes` lint to consider `extern "C" fn` definitions in addition to `extern "C" {}` declarations, this PR adds a new lint - `improper_ctypes_definitions` - which only applies to `extern "C" fn` definitions.
In addition, the `improper_ctype_definitions` lint differs from `improper_ctypes` by considering `*T` and `&T` (where `T: Sized`) FFI-safe (addressing #66220).
There wasn't a clear consensus in #66220 (where the issues with #65134 were primarily discussed) on the approach to take, but there has [been some discussion in Zulip](https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/.2366220.20improper_ctypes.20definitions.20vs.20declarations/near/198903086). I fully expect that we'll want to iterate on this before landing.
cc @varkor + @shepmaster (from #19834) @hanna-kruppe (active in discussing #66220), @SimonSapin (#65134 caused problems for Servo, want to make sure that this PR doesn't)
Verify that the binop trait *is* implemented for the types *if* all the
involved type parameters are replaced with fresh inferred types. When
this is the case, it means that the type parameter was indeed missing a
trait bound. If this is not the case, provide a generic `note` refering
to the type that doesn't implement the expected trait.
Fixes#70718
This commit allows making associated items (e.g. associated functions
and types) into lang items via the `#[lang]` attribute. This allows such
items to be accessed directly, rather than by iterating over the parent
item's associated items.
I've added `FnOnce::Output` as a lang item, and updated one old usage to
use the new lang item. The remaining uses can be updated separately.
Add re-exports to use suggestions
In the following example, an inaccessible path is suggested via `use foo::bar::X;` whereas an accessible public exported path can be suggested instead.
```rust
mod foo {
mod bar {
pub struct X;
}
pub use self::bar::X;
}
fn main() { X; }
```
This fixes the issue.
Change heuristic for determining range literal
Currently, rustc uses a heuristic to determine if a range expression is
not a literal based on whether the expression looks like a function call
or struct initialization. This fails for range literals whose
lower/upper bounds are the results of function calls. A possibly-better
heuristic is to check if the expression contains `..`, required in range
literals.
Of course, this is also not perfect; for example, if the range
expression is a struct which includes some text with `..` this will
fail, but in general I believe it is a better heuristic.
A better alternative altogether is to add the `QPath::LangItem` enum
variant suggested in #60607. I would be happy to do this as a precursor
to this patch if someone is able to provide general suggestions on how
usages of `QPath` need to be changed later in the compiler with the
`LangItem` variant.
Closes#73553
This commit changes the improper ctypes lint (when operating on
definitions) to consider raw pointers or references to sized types as
FFI-safe.
Signed-off-by: David Wood <david@davidtw.co>
This commit adds a new lint - `improper_ctypes_definitions` - which
functions identically to `improper_ctypes`, but on `extern "C" fn`
definitions (as opposed to `improper_ctypes`'s `extern "C" {}`
declarations).
Signed-off-by: David Wood <david@davidtw.co>
