Contract variable declarations
This change adds contract variables that can be declared in the `requires` clause and can be referenced both in `requires` and `ensures`, subject to usual borrow checking rules. This allows any setup common to both the `requires` and `ensures` clauses to only be done once.
In particular, one future use case would be for [Fulminate](https://dl.acm.org/doi/10.1145/3704879)-like ownership assertions in contracts, that are essentially side-effects, and executing them twice would alter the semantics of the contract.
As of this change, `requires` can now be an arbitrary sequence of statements, with the final expression being of type `bool`. They are executed in sequence as expected, before checking if the final `bool` expression holds.
This PR depends on rust-lang/rust#144438 (which has now been merged).
Contracts tracking issue: https://github.com/rust-lang/rust/issues/128044
**Other changes introduced**:
- Contract macros now wrap the content in braces to produce blocks, meaning there's no need to wrap the content in `{}` when using multiple statements. The change is backwards compatible, in that wrapping the content in `{}` still works as before. The macros also now treat `requires` and `ensures` uniformally, meaning the `requires` closure is built inside the parser, as opposed to in the macro.
**Known limiatations**:
- Contracts with variable declarations are subject to the regular borrow checking rules, and the way contracts are currently lowered limits the usefulness of contract variable declarations. Consider the below example:
```rust
#[requires(let init_x = *x; true)]
#[ensures(move |_| *x == 2 * init_x)]
fn double_in_place(x: &mut i32) {
*x *= 2;
}
```
We have used the new variable declarations feature to remember the initial value pointed to by `x`, however, moving `x` into the `ensures` does not pass the borrow checker, meaning the above function contract is illegal. Ideally, something like the above should be expressable in contracts.
refactor: Move to anstream + anstyle for styling
`rustc` uses [`termcolor`](https://crates.io/crates/termcolor) for styling and writing, while `annotate-snippets` uses [`anstyle`](https://crates.io/crates/anstyle) for styling and currently writes directly to a `String`. When rendering directly to a terminal, there isn't/shouldn't be any differences. Still, there are differences in the escape sequences, which leads to slightly different output in JSON and SVG tests. As part of my work to have `rustc` use `annotate-snippets`, and to reduce the test differences between the two, I switched `rustc` to use `anstlye` and [`anstream`](https://crates.io/crates/anstream) for styling and writing.
The first commit migrates to `anstyle` and `anstream` and notably does not change the output. This is because it includes extra formatting to ensure that `anstyle` + `anstream` match the current output exactly. Most of this code is unnecessary, as it adds redundant resets or uses 256-color (8-bit) when it could be using 4-bit color. The subsequent commits remove this extra formatting while maintaining the correct output when rendered.
[Zulip discussion](https://rust-lang.zulipchat.com/#narrow/channel/147480-t-compiler.2Fdiagnostics/topic/annotate-snippets.20hurdles)
Contract variables can be declared in the `requires` clause and
can be referenced both in `requires` and `ensures`, subject to usual
borrow checking rules.
This allows any setup common to both the `requires` and `ensures`
clauses to only be done once.
This recovers some instruction count regressions after changing most Pat
parsing functions return a non-boxed Pat. It seems to be beneficial to
break out a #[cold] parsing recovery function when the function includes
more parsing, presumably because this requires more stack space and/or
mem copies when LLVM optimizes the wrong path.
only replace the intended comma in pattern suggestions
Only suggest to replace the intended comma, not all bars in the pattern.
Fixesrust-lang/rust#143330.
This continues rust-lang/rust#143331, the credit for making the fix goes to `@A4-Tacks.` I just blessed tests and added a regression test.
Tweak handling of "struct like start" where a struct isn't supported
This improves the case where someone tries to write a `match` expr where the patterns have type ascription syntax. Makes them less verbose, by giving up on the first encounter in the block, and makes them more accurate by only treating them as a struct literal if successfully parsed as such.
Before, encountering something like `match a { b:` would confuse the parser and think everything after `match` *must* be a struct, and if it wasn't it would generate a cascade of unnecessary diagnostics.
This improves the case where someone tries to write a `match` expr where the patterns have type ascription syntax. Makes them less verbose, by giving up on the first encounter in the block, and makes them more accurate by only treating them as a struct literal if successfuly parsed as such.
When writing something like the expression `|_: ...| {}`, we now detect the `...` during parsing explicitly instead of relying on the detection in `parse_ty_common` so that we don't talk about "nested `...` are not supported".
```
error: unexpected `...`
--> $DIR/no-closure.rs:6:35
|
LL | const F: extern "C" fn(...) = |_: ...| {};
| ^^^
|
= note: only `extern "C"` and `extern "C-unwind"` functions may have a C variable argument list
```
```
error: unexpected `...`
--> $DIR/varargs-in-closure-isnt-supported.rs:5:20
|
LL | let mut lol = |...| ();
| ^^^ not a valid pattern
|
= note: C-variadic type `...` is not allowed here
```
Less greedily parse `[const]` bounds
> [!IMPORTANT]
> If you're coming here from any beta backport nomination thread on Zulip, only the last commit is truly relevant (the first commit doesn't need to be backported, it only contains test modifications)!
Don't consider `[` to start a bound, only consider `[const]` in its entirety to do so. This drastically reduces (but doesn't eliminate!) the chance of *real* breakages. Like `const`, `~const` and `async` before, `[const]` unavoidably brings along theoretical breakages, see preexisting tests: `macro-const-trait-bound-theoretical-regression.rs` and `macro-async-trait-bound-theoretical-regression.rs`.
Side note: It's unfortunate that we have to do this but apart from the known fact that MBE hurts forward compatibility, the `[const]` syntax is simply a bit scuffed (also CC'ing https://github.com/rust-lang/rust/issues/146122, section (3)).
Fixes [after beta backport] rust-lang/rust#146417.
* 1st commit: Restore the original test intentions of several preexisting related tests that were unfortunately lost over time
* I've added a bunch of SCREAMING comments to make it less likely to be lost again
* CC PR rust-lang/rust#119099 which added most of these tests
* CC [#144409 (comment)](https://github.com/rust-lang/rust/pull/144409#discussion_r2337587513) for further context (NB: It's not the only PR that negatively affected the test intention)
* 2nd commit: Actually address the regression
r? `@oli-obk` or anyone
Strip frontmatter in fewer places
* Stop stripping frontmatter in `proc_macro::Literal::from_str` (RUST-146132)
* Stop stripping frontmatter in expr-ctxt (but not item-ctxt!) `include`s (RUST-145945)
* Stop stripping shebang (!) in `proc_macro::Literal::from_str`
* Not a breaking change because it did compare spans already to ensure there wasn't extra whitespace or comments (`Literal::from_str("#!\n0")` already yields `Err(_)` thankfully!)
* Stop stripping frontmatter+shebang inside some rustdoc code where it doesn't make any observable difference (see self review comments)
* (Stop stripping frontmatter+shebang inside internal test code)
Fixes https://github.com/rust-lang/rust/issues/145945.
Fixes https://github.com/rust-lang/rust/issues/146132.
r? fee1-dead
Reject invalid literal suffixes in tuple indexing, tuple struct indexing, and struct field name position
Tracking issue: rust-lang/rust#60210
Closes rust-lang/rust#60210
## Summary
Bump the ["suffixes on a tuple index are invalid" non-lint pseudo future-incompatibility warning (#60210)][issue-60210][^non-lint] to a **hard error** across all editions, rejecting the remaining carve outs from accidentally accepted invalid suffixes since Rust **1.27**.
- We accidentally accepted invalid suffixes in tuple indexing positions in Rust **1.27**. Originally reported at https://github.com/rust-lang/rust/issues/59418.
- We tried to hard reject all invalid suffixes in https://github.com/rust-lang/rust/pull/59421, but unfortunately it turns out there were proc macros accidentally relying on it: https://github.com/rust-lang/rust/issues/60138.
- We temporarily accepted `{i,u}{32,size}` in https://github.com/rust-lang/rust/pull/60186 (the "*carve outs*") to mitigate *immediate* ecosystem impact, but it came with an FCW warning indicating that we wanted to reject it after a few Rust releases.
- Now (1.89.0) is a few Rust releases later (1.35.0), thus I'm proposing to **also reject the carve outs**.
- `std::mem::offset_of!` stabilized in Rust **1.77.0** happens to use the same "don't expect suffix" code path which has the carve outs, so it also accepted the carve out suffixes. I'm proposing to **reject this case as well**.
## What specifically breaks?
Code that still relied on invalid `{i,u}{32,size}` suffixes being temporarily accepted by rust-lang/rust#60186 as an ecosystem impact mitigation measure (cf. rust-lang/rust#60138). Specifically, the following cases (particularly the construction of these forms in proc macros like reported in rust-lang/rust#60138):
### Position 1: Invalid `{i,u}{32,size}` suffixes in tuple indexing
```rs
fn main() {
let _x = (42,).0invalid; // Already error, already rejected by #59421
let _x = (42,).0i8; // Already error, not one of the #60186 carve outs.
let _x = (42,).0usize; // warning: suffixes on a tuple index are invalid
}
```
### Position 2: Invalid `{i,u}{32,size}` suffixes in tuple struct indexing
```rs
fn main() {
struct X(i32);
let _x = X(42);
let _x = _x.0invalid; // Already error, already rejected by #59421
let _x = _x.0i8; // Already error, not one of the #60186 carve outs.
let _x = _x.0usize; // warning: suffixes on a tuple index are invalid
}
```
### Position 3: Invalid `{i,u}{32,size}` suffixes in numeric struct field names
```rs
fn main() {
struct X(i32, i32, i32);
let _x = X(1, 2, 3);
let _y = X { 0usize: 42, 1: 42, 2: 42 }; // warning: suffixes on a tuple index are invalid
match _x {
X { 0usize: 1, 1: 2, 2: 3 } => todo!(), // warning: suffixes on a tuple index are invalid
_ => {}
}
}
```
### Position 4: Invalid `{i,u}{32,size}` suffixes in `std::mem::offset_of!`
While investigating the warning, unfortunately I noticed `std::mem::offset_of!` also happens to use the "expect no suffix" code path which had the carve outs. So this was accepted since Rust **1.77.0** with the same FCW:
```rs
fn main() {
#[repr(C)]
pub struct Struct<T>(u8, T);
assert_eq!(std::mem::offset_of!(Struct<u32>, 0usize), 0);
//~^ WARN suffixes on a tuple index are invalid
}
```
### The above forms in proc macros
For instance, constructions like (see tracking issue rust-lang/rust#60210):
```rs
let i = 0;
quote! { foo.$i }
```
where the user needs to actually write
```rs
let i = syn::Index::from(0);
quote! { foo.$i }
```
### Crater results
Conducted a crater run (https://github.com/rust-lang/rust/pull/145463#issuecomment-3194920383).
- https://github.com/AmlingPalantir/r4/tree/256af3c72f094b298cd442097ef7c571d8001f29: genuine regression; "invalid suffix `usize`" in derive macro. Has a ton of other build warnings, last updated 6 years ago.
- Exactly the kind of intended breakage. Minimized down to https://github.com/AmlingPalantir/r4/blob/256af3c72f094b298cd442097ef7c571d8001f29/validates_derive/src/lib.rs#L71-L75, where when interpolation uses `quote`'s `ToTokens` on a `usize` index (i.e. on tuple struct `Tup(())`), the generated suffix becomes `.0usize` (cf. Position 2).
- Notified crate author of breakage in https://github.com/AmlingPalantir/r4/issues/1.
- Other failures are unrelated or spurious.
## Review remarks
- Commits 1-3 expands the test coverage to better reflect the current situation before doing any functional changes.
- Commit 4 is an intentional **breaking change**. We bump the non-lint "suffixes on a tuple index are invalid" warning into a hard error. Thus, this will need a crater run and a T-lang FCP.
## Tasks
- [x] Run crater to check if anyone is still relying on this being not a hard error. Determine degree of ecosystem breakage.
- [x] If degree of breakage seems acceptable, draft nomination report for T-lang for FCP.
- [x] Determine hard error on Edition 2024+, or on all editions.
## Accompanying Reference update
- https://github.com/rust-lang/reference/pull/1966
[^non-lint]: The FCW was implemented as a *non-lint* warning (meaning it has no associated lint name, and you can't `#![deny(..)]` it) because spans coming from proc macros could not be distinguished from regular field access. This warning was also intentionally impossible to silence. See https://github.com/rust-lang/rust/pull/60186#issuecomment-485581694.
[issue-60210]: https://github.com/rust-lang/rust/issues/60210
add span to struct pattern rest (..)
Struct pattern rest (`..`) did not retain span information compared to normal fields. This patch adds span information for it.
The motivation of this patch comes from when I implemented this PR for Clippy: https://github.com/rust-lang/rust-clippy/pull/15000#discussion_r2134145163
It is possible to get the span of the Et cetera in a bit roundabout way, but I thought this would be nicer.
Detect missing `if let` or `let-else`
During `let` binding parse error and encountering a block, detect if there is a likely missing `if` or `else`:
```
error: expected one of `.`, `;`, `?`, `else`, or an operator, found `{`
--> $DIR/missing-if-let-or-let-else.rs:14:25
|
LL | let Some(x) = foo() {
| ^ expected one of `.`, `;`, `?`, `else`, or an operator
|
help: you might have meant to use `if let`
|
LL | if let Some(x) = foo() {
| ++
help: alternatively, you might have meant to use `let else`
|
LL | let Some(x) = foo() else {
| ++++
```
Fixrust-lang/rust#107806.
Gate static closures behind a parser feature
I'd like to gate `static ||` closures behind a feature gate, since we shouldn't allow people to take advantage of this syntax if it's currently unstable. Right now, since it's only rejected after ast lowering, it's accessible to macros.
Let's crater this to see if we can claw it back without breaking anyone's code.
Prevent impossible combinations in `ast::ModKind`.
`ModKind::Loaded` has an `inline` field and a `had_parse_error` field. If the `inline` field is `Inline::Yes` then `had_parse_error` must be `Ok(())`.
This commit moves the `had_parse_error` field into the `Inline::No` variant. This makes it impossible to create the nonsensical combination of `inline == Inline::Yes` and `had_parse_error = Err(_)`.
r? ```@Urgau```
During `let` binding parse error and encountering a block, detect if there is a likely missing `if` or `else`:
```
error: expected one of `.`, `;`, `?`, `else`, or an operator, found `{`
--> $DIR/missing-if-let-or-let-else.rs:14:25
|
LL | let Some(x) = foo() {
| ^ expected one of `.`, `;`, `?`, `else`, or an operator
|
help: you might have meant to use `if let`
|
LL | if let Some(x) = foo() {
| ++
help: alternatively, you might have meant to use `let else`
|
LL | let Some(x) = foo() else {
| ++++
```
Frank King