This commit makes the "Wrap return type in Result" assist prefer type aliases of standard library
type when the are in scope, use at least one generic parameter, and have the name "Result".
The last restriction was made in an attempt to avoid false assumptions about which type the
user is referring to, but that might be overly strict. We could also do something like this, in
order of priority:
* Use the alias named "Result".
* Use any alias if only a single one is in scope, otherwise:
* Use the standard library type.
This is easy to add if others feel differently that is appropriate, just let me know.
internal: Lay basic ground work for standalone mbe tests
Most of our mbe hir-def tests don't actually do anything name res relevant, we can (and should) move those down the stack into `mbe/hir-expand`.
minor: Downgrade cyclic deps error to warning
As the issue here is no longer workable for us and this appearing for some repos due to package cycles which can cause we should downgrade it as some people tend to think this breaks r-a when it doesn't
Handle attributes correctly in "Flip comma"
Attributes often contain path followed by a token tree (e.g. `align(2)`), and the previous code handled them as two separate items, which led to results such as `#[repr(alignC, (2))]`.
An alternative is to just make the assist unavailable in attributes, like we do in macros. But contrary to macros, attributes often have a fixed form, so this seems useful.
Fixes#18013.
Provide an option to hide deprecated items from completion
Fixes#17989.
I wonder if this should be instead done in the editor, that will do it in a language-agnostic way. Can't hurt to do it in rust-analyzer, I guess.
Attributes often contain path followed by a token tree (e.g. `align(2)`, and the previous code handled them as two separate items, which led to results such as `#[repr(alignC, (2))]`.
An alternative is to just make the assist unavailable in attributes, like we do in macros. But contrary to macros, attributes often have a fixed form, so this seems useful.
internal: Add doc comments to OpQueue
I spent a while debugging some OpQueue behaviours and found the API slightly confusing, so I've added doc comments to clarify what each OpQueue method does.
fix: do not assume rustup is installed in xtask codegen take 2
7d9e4fcc07 broke this on rustup toolchains, the `cmd` command is trying to be too smart here
internal: Improve inlay hint resolution reliability
The payload now ships the range the inlay hint ought to be triggered for instead of trying to estimate it from its position which is somewhat brittle
Do not report missing unsafe on `addr_of[_mut]!(EXTERN_OR_MUT_STATIC)`
The compiler no longer does as well; see https://github.com/rust-lang/rust/pull/125834.
Also require unsafe when accessing `extern` `static` (other than by `addr_of!()`).
Fixes#17978.
fix: `std::error::Error` is object unsafe
Fixes#17998
I tried to get generic predicates of assoc function itself, not inherited from the parent here;
0ae42bd425/crates/hir-ty/src/object_safety.rs (L420-L442)
But this naive equality check approach doesn't work when the assoc function has one or more generic paramters like;
```rust
trait Foo {}
trait Bar: Foo {
fn bar(&self);
}
```
because the generic predicates of the parent, `Bar` is `[^1.0 implements Foo]` and the generic predicates of `fn bar` is `[^1.1 implements Foo]`, which are different.
This PR implements a correct logic for filtering out parent generic predicates for this.
fix: consider indentation in the "Generate impl" and "Generate trait impl" assists
This makes the generated impl's indentation match the ADT it targets, improving formatting when using nested modules inside of the same file or when defining types inside of a function. See the added tests for an example.
At first I tried to call some of the convenient helpers that delegate to `IndentLevel::increase_indent` on the generated impl, but as the comment on that function notes it does not indent the first token, making it inapplicable here. I hope the solution in this PR is acceptable, please let me know if I missed something :)
This makes the generated impl's indentation match the ADT it targets, improving formatting when
using nested modules inside of the same file or when defining types inside of a function.
fix: do not assume rustup is installed in xtask codegen
When formatting generated code the xtask crate attempts to run `rustup run stable rustfmt`, which fails if `rustup` is not installed. This results in test failures when another source manages the compiler toolchain, for example when using Nix (or any other distro-specific packaging solution):
* xtask::codegen::grammar::test
* xtask::codegen::assists_doc_tests::test
With this PR xtask will first attempt to run `rustup run stable rustfmt`, and if that fails just plain `rustfmt`. It still validates a stable version is being used. This allows `cargo test` to pass on systems that do not use `rustup`.
Consider field attributes when converting from tuple to named struct and the opposite
Fixes#17983.
I tried to use the `SourceChangeBuilder::make_mut()` API, but it duplicated the attribute...
fix: Don't add reference when it isn't needed for the "Extract variable" assist
I.e. don't generate `let var_name = &foo()`. Because it always irritates me when I need to fix that.
Anything that creates a new value don't need a reference. That excludes mostly field accesses and indexing.
I had a thought that we can also not generate a reference for fields and indexing as long as the type is `Copy`, but sometimes people impl `Copy` even when they don't want to copy the values (e.g. a large type), so I didn't do that.
