fix: ensure `rustfmt` runs when configured with `./`
(Hopefully) resolves https://github.com/rust-lang/rust-analyzer/issues/15595. This change kinda approaches canonicalization—which I am not a fan of—but only in service of making `./`-configured commands run correctly.
Longer-term, I feel like this code should be removed once `rustfmt` supports recursive searches of configuration files or interpolation of values like `${workspace_folder}` lands in rust-analyzer.
## Testing
I cloned `rustc`, setup rust-analyzer as suggested in the [`rustc` dev guide](https://rustc-dev-guide.rust-lang.org/building/suggested.html#configuring-rust-analyzer-for-rustc), saved and formatted files in `src/tools/miri` and `compiler`, and saw `rustfmt` (seemingly) correctly.
feat: add backtick to surrounding and auto-closing pairs
Makes backticks always complete as a pair and also surround the current selection, similar to double quotes. This is useful primarily in the context of markdown doc comments, but is applied globally for simplicity.
Closes https://github.com/rust-lang/rust-analyzer/issues/11381
vscode: Fix line and col regexp for `rustc` problem matcher
When building the Rust compiler with `./x check` from within VS Code, the current `rustc` problem matcher thinks that the output from that command that looks like this:
Build completed successfully in 0:00:26
is about a problem in a file named `0` on line 00, col 26. This wouldn't be so bad if it wasn't for that VS Code tends to get stuck on this problem because of problems with opening the file '0'.
The rust compiler will never output problems with a line or a column that starts with 0, so change the regexp to require lines and cols to begin with `[1-9]` to fix this problem.
fix: allow more kinds of if let patterns in guarded return assist
Removes the checks that require the pattern to be a tuple struct with exactly 1 field that is unqualified and has an identifier pattern in it. I'm not sure if there should be more checks in place but they seem unnecessary now?
Closes#15695
When building the Rust compiler with `./x check` from within VS Code,
the current `rustc` problem matcher thinks that the output from that
command that looks like this:
Build completed successfully in 0:00:26
is about a problem in a file named `0` on line 00, col 26. This wouldn't
be so bad if it wasn't for that VS Code tends to get stuck on this
problem because of problems with opening the file '0'.
The rust compiler will never output problems with a line or a column
that starts with 0, so change the regexp to require lines and cols to
begin with [1-9] to fix this problem.
Only run tests on linux in pull requests
There is little gain from running them on all platforms, they should almost never fail if the linux one pass and if they do, bors will catch it still.
fix: make bool_to_enum assist create enum at top-level
This pr makes the `bool_to_enum` assist create the `enum` at the next closest module block or at top-level, which fixes a few tricky cases such as with an associated `const` in a trait or module:
```rust
trait Foo {
const $0BOOL: bool;
}
impl Foo for usize {
const BOOL: bool = true;
}
fn main() {
if <usize as Foo>::BOOL {
println!("foo");
}
}
```
Which now properly produces:
```rust
#[derive(PartialEq, Eq)]
enum Bool { True, False }
trait Foo {
const BOOL: Bool;
}
impl Foo for usize {
const BOOL: Bool = Bool::True;
}
fn main() {
if <usize as Foo>::BOOL == Bool::True {
println!("foo");
}
}
```
I also think it's a bit nicer, especially for local variables, but didn't really know to do it in the first PR :)
Allocate ast ids for parameters
Since these can have attributes attached to them, we'll need this sooner or later (sooner being me tinkering with the token map right now)
