Fix unwind drop glue for if-then scopes
cc `@est31`
Fix#102317Fix#99852
This PR fixes the drop glue for unwinding from a panic originated in a drop while breaking out for the else block in an `if-then` scope.
MIR validation does not fail for the synchronous versions of the test program, because `StorageDead` statements are skipped over in the unwinding process. It is only becoming a problem when it is inside a generator where `StorageDead` must be kept around.
Suggest `.into()` when all other coercion suggestions fail
Also removes some bogus suggestions because we now short-circuit when offering coercion suggestions(instead of, for example, suggesting every one that could possibly apply)
Fixes#102415
panic-on-uninit: adjust checks to 0x01-filling
Now that `mem::uninitiailized` actually fills memory with `0x01` (https://github.com/rust-lang/rust/pull/99182), we can make it panic in a few less cases without risking a lot more UB -- which hopefully slightly improves compatibility with some old code, and which might increase the chance that we can check inside arrays in the future.
We detect almost all of these with our lint, so authors of such code should still be warned -- but if this happens deep inside a dependency, the panic can be quite interruptive, so it might be better not to do it when there is no risk of LLVM UB. Therefore, adjust the `might_permit_raw_init` logic to care primarily about LLVM UB. To my knowledge, it actually covers all cases of LLVM UB now.
Fixes https://github.com/rust-lang/rust/issues/66151
Cc ``@5225225``
Slightly improve no return for returning function error
Fixes#100607
The rationale is that absolute beginners will be slightly confused as to why certain lines of code in a function does not require a semicolon. (I have actually witness a beginner having this confusion). Hence, a slight rationale is added "to return this value", which signals to the user that after removing said semicolon the value is returned resolving that error.
However, if this is not desirable, I welcome any other suggestions. Thanks.
Lint against nested opaque types that don't satisfy associated type bounds
See the test failures for examples of places where this lint would fire.
r? `@oli-obk`
Don't ICE when trying to copy unsized value in const prop
When we have a trivially false where-clause predicate like `Self: Sized` where `Self = dyn Trait`, we sometimes don't throw an error during typeck for an illegal operation such as copying an unsized type.
This, unfortunately, cannot be made into an error (at least not without some migration -- see #95611 for example), but we should at least not ICE, since this function will never actually be reachable from main, for example.
r? `@RalfJung` since I think you added these assertions? but feel free to reassign.
Fixes#102553
Improve spans when splitting multi-char operator tokens for proc macros.
When a two-char (or three-char) operator token is split into single-char operator tokens before being passed to a proc macro, the single-char tokens are given the original span of length two (or three). This PR gives them more accurate spans.
r? `@Aaron1011`
cc `@petrochenkov`
Fix ICE #101739
Fixes a part of #101739
This cannot cover the following case. It causes `too many args provided` error and obligation does not have references error. I want your advice to solve the following cases as well in this pull request or a follow-up.
```rust
#![crate_type = "lib"]
#![feature(transmutability)]
#![allow(dead_code, incomplete_features, non_camel_case_types)]
mod assert {
use std::mem::BikeshedIntrinsicFrom;
pub fn is_transmutable<
Src,
Dst,
Context,
const ASSUME_ALIGNMENT: bool,
const ASSUME_LIFETIMES: bool,
const ASSUME_VALIDITY: bool,
const ASSUME_VISIBILITY: bool,
>()
where
Dst: BikeshedIntrinsicFrom<
Src,
Context,
ASSUME_ALIGNMENT,
ASSUME_LIFETIMES,
ASSUME_VALIDITY,
ASSUME_VISIBILITY,
>,
{}
}
fn via_const() {
struct Context;
#[repr(C)] struct Src;
#[repr(C)] struct Dst;
const FALSE: bool = false;
assert::is_transmutable::<Src, Dst, Context, FALSE, FALSE, FALSE, FALSE>();
}
```
Fix duplicate usage of `a` article.
This fixes a typo first appearing in #94624 in which test-macro diagnostic uses "a" article twice.
Since I searched the sources for " a a " sequences, I also fixed the same issue in a few files where I found it.
Get rid of exclude-list for Windows-only tests
Main purpose of this change is to get rid of a quite long (and growing) list of excluded targets, while this test should only be useful on Windows (as far as I understand it). The `// only-windows` header seams to implement exactly what we need here.
I don't know why there are some whitespace changes, but `x.py fmt` and `.git/hooks/pre-push` are happy.
This fixes a typo first appearing in #94624
in which test-macro diagnostic uses "a" article twice.
Since I searched sources for " a a " sequences,
I also fixed the same issue in a few source files where I found it.
Signed-off-by: Petr Portnov <gh@progrm-jarvis.ru>
Do not panic when a test function returns Result::Err.
Rust's test library allows test functions to return a `Result`, so that the test is deemed to have failed if the function returns a `Result::Err` variant. Currently, this works by having `Result` implement the `Termination` trait and asserting in assert_test_result that `Termination::report()` indicates successful completion. This turns a `Result::Err` into a panic, which is caught and unwound in the test library.
This approach is problematic in certain environments where one wishes to save on both binary size and compute resources when running tests by:
* Compiling all code with `--panic=abort` to avoid having to generate unwinding tables, and
* Running most tests in-process to avoid the overhead of spawning new processes.
This change removes the intermediate panic step and passes a `Result::Err` directly through to the test runner.
To do this, it modifies `assert_test_result` to return a `Result<(), String>` where the `Err` variant holds what was previously the panic message. It changes the types in the `TestFn` enum to return `Result<(), String>`.
This tries to minimise the changes to benchmark tests, so it calls `unwrap()` on the `Result` returned by `assert_test_result`, effectively keeping the same behaviour as before.
Some questions for reviewers:
* Does the change to the return types in the enum `TestFn` constitute a breaking change for the library API? Namely, the enum definition is public but the test library indicates that "Currently, not much of this is meant for users" and most of the library API appears to be marked unstable.
* Is there a way to test this change, i.e., to test that no panic occurs if a test returns `Result::Err`?
* Is there a shorter, more idiomatic way to fold `Result<Result<T,E>,E>` into a `Result<T,E>` than the `fold_err` function I added?
Fix `format_args` capture for macro expanded format strings
Since #100996 `format_args` capture for macro expanded strings aren't prevented when the span of the expansion points to a string literal, e.g.
```rust
// not a terribly realistic example, but also happens for proc_macros that set
// the span of the output to an input str literal, such as indoc
macro_rules! x {
($e:expr) => { $e }
}
fn main() {
let a = 1;
println!(x!("{a}"));
}
```
The tests didn't catch it as the span of `concat!()` points to the macro invocation
r? `@m-ou-se`
Move lint level source explanation to the bottom
So, uhhhhh
r? `@estebank`
## User-facing change
"note: `#[warn(...)]` on by default" and such are moved to the bottom of the diagnostic:
```diff
- = note: `#[warn(unsupported_calling_conventions)]` on by default
= warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!
= note: for more information, see issue #87678 <https://github.com/rust-lang/rust/issues/87678>
+ = note: `#[warn(unsupported_calling_conventions)]` on by default
```
Why warning is enabled is the least important thing, so it shouldn't be the first note the user reads, IMO.
## Developer-facing change
`struct_span_lint` and similar methods have a different signature.
Before: `..., impl for<'a> FnOnce(LintDiagnosticBuilder<'a, ()>)`
After: `..., impl Into<DiagnosticMessage>, impl for<'a, 'b> FnOnce(&'b mut DiagnosticBuilder<'a, ()>) -> &'b mut DiagnosticBuilder<'a, ()>`
The reason for this is that `struct_span_lint` needs to edit the diagnostic _after_ `decorate` closure is called. This also makes lint code a little bit nicer in my opinion.
Another option is to use `impl for<'a> FnOnce(LintDiagnosticBuilder<'a, ()>) -> DiagnosticBuilder<'a, ()>` altough I don't _really_ see reasons to do `let lint = lint.build(message)` everywhere.
## Subtle problem
By moving the message outside of the closure (that may not be called if the lint is disabled) `format!(...)` is executed earlier, possibly formatting `Ty` which may call a query that trims paths that crashes the compiler if there were no warnings...
I don't think it's that big of a deal, considering that we move from `format!(...)` to `fluent` (which is lazy by-default) anyway, however this required adding a workaround which is unfortunate.
## P.S.
I'm sorry, I do not how to make this PR smaller/easier to review. Changes to the lint API affect SO MUCH 😢
