11925: internal: Add and use `HirFormatter::write_{str,char}` r=Veykril a=lnicola
Saves slightly over 3 KB of `text`, but comparing the total with that from two weeks ago in #11776, this is a losing battle (we're 951 KB larger).
```
text data bss dec hex filename
24693512 1542704 4424 26240640 1906680 rust-analyzer-baseline
24690216 1542112 4424 26236752 1905750 rust-analyzer-pr
```
Co-authored-by: Laurențiu Nicola <lnicola@dend.ro>
While we don't support const args in type inference yet, we can at least
make use of the fallback path resolution to resolve paths in const args
in the IDE layer to enable some features for them.
11905: internal: Remove hir_expand macro recursion check r=jonas-schievink a=jonas-schievink
This check is insufficient to ensure finite macro nesting, and so all callers already have their own recursion limit, which makes this check redundant.
...at least I hope it's redundant. Would be great if someone could double-check this.
Originally, this check was added in https://github.com/rust-analyzer/rust-analyzer/pull/3671
Co-authored-by: Jonas Schievink <jonas.schievink@ferrous-systems.com>
11891: Better error message on Flycheck Error message (from: unactionable error message if we are using `clippy` as the checker) r=Veykril a=flipbit03
I have commented on this [S-unactionable issue](https://github.com/rust-analyzer/rust-analyzer/issues/6589) that the Flycheck error message should maybe provide a hint about what tool it actually runs. Searching on some places on the Internet I've found multiple people, including myself, losing copious amounts of time on the same issue. So I've decided to make this very small PR :-)
From an user experience standpoint, the current error message is unhelpful to the end user, because the end user does not know exactly what it needs to check/fix (outdated, broken, or missing `cargo clippy`). In my own case, `cargo clippy` was actually missing altogether (developing off `rust:1.59.0-bullseye` official Docker image).
Thanks in advance!
Co-authored-by: Cadu <cadu.coelho@gmail.com>
11896: fix: Show the path to be created in the unresolved-module fix label r=Veykril a=Veykril
bors r+
Co-authored-by: Lukas Wirth <lukastw97@gmail.com>
11894: complete pattern args based on type name r=Veykril a=cameron1024
Addresses #11892
Changes function argument completion to cover a case like this:
```rust
struct Foo { bar: i32 }
fn qux(Foo { bar }: Foo) {
println!("{bar}");
}
```
When completing the function call for `qux`, instead of expanding to `qux(_)`, it will now expand to `qux(foo)` (based on the snake-cased version of the name of the ADT)
Non ADTs are unaffected
Co-authored-by: cameron <cameron.studdstreet@gmail.com>
Co-authored-by: cameron1024 <cameron.studdstreet@gmail.com>
11865: Fix: Select correct insert position for disabled group import r=jonasbb a=jonasbb
The logic for importing with and without `group_imports` differed
significantly when no previous group existed. This lead to the problem
of using the wrong position when importing inside a module (#11585) or
when inner attributes are involved.
The existing code for grouped imports is better and takes these things
into account.
This PR changes the flow to use the pre-existing code for adding a new
import group even for the non-grouped import settings.
Some coverage markers are updated and the `group` is removed, since they
are now invoked in both cases (grouping and no grouping).
Tests are updated and two tests (empty module and inner attribute) are
added.
Fixes#11585
Co-authored-by: Jonas Bushart <jonas@bushart.org>
11699: feat: assist to remove unneeded `async`s r=Ethiraric a=Ethiraric
This should fix#11392
This PR adds a quickfix on functions marked with `async` that suggests, if and only if no `await` expression in find in the body of the function (it relies on `SyntaxNode::descendants()` to recursively traverse blocks), to remove the `async` keyword.
The lint is made so that it triggers only if the cursor is not in the body of the function, so that it does not pollute the quickfix list.
It does not trigger a diagnostic. I don't know if this repository is the place to implement those (`clippy`?). If it is, I would very much like pointers on where to start looking.
If there are test cases I haven't thought about, please do suggest.
Co-authored-by: Ethiraric <ethiraric@gmail.com>
10802: Allow clients to configure the global workspace search limit r=Veykril a=knutwalker
Playing around with [helix](https://helix-editor.com) I realized that the global worksapce symbol search works different compared to vs-code.
Helix requires all possible symbols in one query and does no subsequent refinement searched.
This PR adds a configuration option to override the default search limit with the default being the currently hardocded value.
Helix users can increment this limit for their instance with a config like
```toml
[[language]]
name = "rust"
language-server = { command = "rust-analyzer" }
[language.config]
workspace = { symbol = { search = { limit = 65536 }}}
```
Other editors are not affected by this change.
Co-authored-by: Paul Horn <dev@knutwalker.engineer>
11866: fix: Prevent underflow in range conversion r=Veykril a=skyfmmf
Previously, when line numbers of Rust spans were converted to LSP ranges, they could underflow resulting in very large line numbers. As an underflow is always wrong, prevent it and use 0 instead.
This was noticed when opening an empty file in `src/bin/` of a library crate. In this case rustc produces a span with `"line_start": 0, "line_end": 0` resulting in the underflow.
Co-authored-by: Felix Maurer <felix@felix-maurer.de>
11879: Suggest infered type in auto complete r=HKalbasi a=HKalbasi
fix#11855
It doesn't work for return types and consts (so their tests are failing) because I can't find their body node in the original file. (Are these original and fake file documented somewhere?)
Also it currently needs to type first character of the type (or manual ctrl+space) to open the auto complete panel, is it possible to open it automatically on typing `:` and `->`?
Co-authored-by: hkalbasi <hamidrezakalbasi@protonmail.com>
11877: fix: splitting path of a glob import wrongly adds `self` r=Veykril a=iDawer
Close #11703
`ast::UseTree::split_prefix` handles globs now.
Removed an extra branch for globs in `ide_db::imports::merge_imports::recursive_merge` (superseeded by split_prefix).
Co-authored-by: iDawer <ilnur.iskhakov.oss@outlook.com>
rustc has removed the use of lang items to mark the primitive impls, so
just look through the crate graph for them (this should be fine
performance-wise since we cache the crates that contain these impls).
Fixes#11876.
11878: fix: Paper over GAT panic r=flodiebold a=flodiebold
TIL that Chalk expects the arguments to a generic associated type to come *before* the ones for the parent trait, not *after* as we have been doing with all other nested generics. Fixing this requires a larger refactoring, so for now this just papers over the problem by completely ignoring parameters of associated types.
Fixes#11769.
Co-authored-by: Florian Diebold <flodiebold@gmail.com>
TIL that Chalk expects the arguments to a generic associated type to
come *before* the ones for the parent trait, not *after* as we have been
doing with all other nested generics. Fixing this requires a larger
refactoring, so for now this just papers over the problem by completely
ignoring parameters of associated types.
Fixes#11769.
11857: Lower postfix suggestions in completions list r=Veykril a=avrong
Fixes#11850
Adds a parameter for postfix suggestions in `CompletionRelevance`, and basing on it, decreases relevance score of such items in completion list
Co-authored-by: Aleksei Trifonov <avrong@avrong.me>
11870: Recover from missing type annotation r=Veykril a=HKalbasi
We were missing the init expression in case of `let x: = 2`, which breaks type inference of that variable (previously x were `{unknown}`, now it is `i32`).
Co-authored-by: hkalbasi <hamidrezakalbasi@protonmail.com>
11871: internal: Move `rust.ungram` into `rust-analyzer/crates/syntax` r=Veykril a=Veykril
This makes updating the grammar a lot simpler for us. Though removing it from ungrammar can't be done without bumping it to 2.0 so I'll leave it in there for the time being.
cc https://github.com/rust-analyzer/ungrammar/pull/47
Co-authored-by: Lukas Wirth <lukastw97@gmail.com>
`ast::UseTree::split_prefix` handles globs now.
Removed an extra branch for globs in `ide_db::imports::merge_imports::recursive_merge` (superseeded by split_prefix).
Previously, when line numbers of Rust spans were converted to LSP
ranges, they could underflow resulting in very large line numbers. As
an underflow is always wrong, prevent it and use 0 instead.
The logic for importing with and without `group_imports` differed
significantly when no previous group existed. This lead to the problem
of using the wrong position when importing inside a module (#11585) or
when inner attributes are involved.
The existing code for grouped imports is better and takes these things
into account.
This PR changes the flow to use the pre-existing code for adding a new
import group even for the non-grouped import settings.
Some coverage markers are updated and the `group` is removed, since they
are now invoked in both cases (grouping and no grouping).
Tests are updated and two tests (empty module and inner attribute) are
added.
Fixes#11585
When last expression in a function body is noreturn asm, then analyzer
complains about the type mismatch by highlighting entire body. This
fixes it by introducing loop {} in the expanded code.
Keep things consistent with the package.json , which uses `self` and
`crate` instead of `by_self` and `by_crate`. Both names are in fact
allowed as aliases, but we should be consistent so that people reading
the docs and using a schema do not see red squiggles.
E.g. when there's a type mismatch on the return value of a function. To
fix this, we have to return the expected type as the type of the block
when there's a mismatch. That meant some IDE code that expected
otherwise had to be adapted, in particular the "add return type" assist.
For the "wrap in Ok/Some" quickfix, this sadly means it usually can't be applied
in all branches of an if expression at the same time anymore, because
there's a type mismatch for each branch that has the wrong type.
11840: Fix another const generic panic r=flodiebold a=HKalbasi
fix#11835
If I change `dyn` to `impl` in the test, it will infer the type as `IntoIterator::Item<impl Iterator<Item = [Ar<u8, 7>; 9]> + ?Sized>` instead of `[Ar<u8, 7>; 9]`. Maybe it needs some action?
Co-authored-by: hkalbasi <hamidrezakalbasi@protonmail.com>
11809: feat: disable experimental diagnostics by default r=jonas-schievink a=jonas-schievink
Now that we diagnose type mismatches, we have another diagnostic that can potentially produce false positives, so let's disable experimental diagnostics by default.
Co-authored-by: Jonas Schievink <jonas.schievink@ferrous-systems.com>
11802: fix: add stubs to make proc macros work that use the `SourceFile` API r=jonas-schievink a=jonas-schievink
Helps with the rocket 0.4 macros, at least on some Rust versions.
Fixes https://github.com/rust-analyzer/rust-analyzer/issues/11264
bors r+
Co-authored-by: Jonas Schievink <jonas.schievink@ferrous-systems.com>
11805: fix: Don't try to resolve methods on unknown types r=Veykril a=flodiebold
Fixes#10454, and some type mismatches.
Co-authored-by: Florian Diebold <flodiebold@gmail.com>
11791: fix: some fixes and improvements to signature help r=jonas-schievink a=jonas-schievink
bors r+
Co-authored-by: Jonas Schievink <jonas.schievink@ferrous-systems.com>
11789: fix: Fix lifetime elision hint configuration key r=Veykril a=Veykril
This is inconsistent with the other configuration keys unfortunately, but this is an issue of how we structured them here in general.
I feel like we'll have to do a proper pass over all the configs, check them for consistency and then write down some documentation of how our configs should be laid out, then apply fixes to the inconsistencies all at once which will entail breaking some user configurations unfortunately since we won't be able to do proper fallback handling for some of these.
Fixes https://github.com/rust-analyzer/rust-analyzer/issues/11778
bors r+
Co-authored-by: Lukas Wirth <lukastw97@gmail.com>
11786: fix: fill_match_arms doesn't add wildcard pat for local enums r=Veykril a=OleStrohm
Fix#11783
This adds similar logic to non_exhaustive as is currently on doc(hidden)
Co-authored-by: Ole Strohm <strohm99@gmail.com>
11780: feat: Add type mismatch diagnostic r=flodiebold a=flodiebold
This adds a proper diagnostic for type mismatches, turning "Add reference here", "Missing Ok or Some" and "Remove this semicolon" into quickfixes for this single diagnostic.
The diagnostic is marked as experimental when it does not have one of these quickfixes, so it can be turned off with `rust-analyzer.diagnostics.enableExperimental` (or specifically with `rust-analyzer.diagnostics.disabled` of course, the ID is `type-mismatch`).
There will still be some false positives, but I think there shouldn't be too many especially when the Chalk fix lands, and it's still experimental anyway 🙂
This also fixes type checking for `rustc_legacy_const_generics` just to avoid some errors in tests.
Co-authored-by: Florian Diebold <flodiebold@gmail.com>
11774: feat: Tag macro calls as unsafe if they expand to unsafe expressions r=Veykril a=Veykril
as long as they aren't inside an unsafe block inside the macro that is.
Co-authored-by: Lukas Wirth <lukastw97@gmail.com>
11740: Insert #[must_use] in generate_is_empty_from_len r=lnicola a=Walther
Inserts `#[must_use]` for the generated `is_empty` methods. I also added the `#[must_use]` for the `len()` methods in the documentation examples and tests for this generator while at it, to make sure they too encourage the use of the annotation.
From discussion in https://github.com/rust-analyzer/rust-analyzer/issues/11736 and https://github.com/rust-analyzer/rust-analyzer/pull/11738
Co-authored-by: Walther <veeti.haapsamo@gmail.com>
11694: fix: "Extract to function" assist preserves `break` and `continue` labels r=Veykril a=m0rg-dev
Adds a label / lifetime parameter to `ide_assists::handlers::extract_function::FlowKind::{Break, Continue}`, adds support for emitting labels to `syntax::ast::make::{expr_break, expr_continue}`, and implements the required machinery to let `extract_function` make use of them.
This does modify the external API of the `syntax` crate, but the changes there are simple, not used outside `ide_assists`, and, well, we should probably support emitting `break` and `continue` labels through `syntax` anyways, they're part of the language spec.
Closes#11413.
Co-authored-by: Morgan Thomas <corp@m0rg.dev>
Adds a label / lifetime parameter to `ide_assists::handlers::extract_function::FlowKind::{Break, Continue}`, adds support for emitting labels to `syntax::ast::make::{expr_break, expr_continue}`, and implements the required machinery to let `extract_function` make use of them.
This does modify the external API of the `syntax` crate, but the changes there are simple, not used outside `ide_assists`, and, well, we should probably support emitting `break` and `continue` labels through `syntax` anyways, they're part of the language spec.
Closes#11413.
- use original instead of adjusted type in ide_completion::completions::record::complete_record
- don't even bother checking if we can complete union literals to Default or to struct update syntax
11686: feat: Enum variant field completion, enum variant / struct consistency r=Veykril a=m0rg-dev
This addresses several related inconsistencies:
- tuple structs use tab stops instead of placeholders
- tuple structs display in the completion menu as `Struct {…}` instead of `Struct(…)`
- enum variants don't receive field completions at all
- enum variants display differently from structs in the completion menu
Also, structs now display their type in the completion detail rather than the raw snippet text to be inserted.
As far as what's user-visible, that looks like this:
| | Menu | Completion | Detail |
|-|-|-|-|
| Record struct (old) | `Struct {…}` | `Struct { x: ${1:()}, y: ${2:()} }$0` | `Struct { x: ${1:()}, y: ${2:()} }$0` |
| Record struct (new) | `Struct {…}` | `Struct { x: ${1:()}, y: ${2:()} }$0` | `Struct { x: i32, y: i32 }` |
| Tuple struct (old) | `Struct {…}` | `Struct($1, $2)$0` | `Struct($1, $2)` |
| Tuple struct (new) | `Struct(…)` | `Struct(${1:()}, ${2:()})$0` | `Struct(i32, i32)` |
| Unit variant (old) | `Variant` | `Variant` | `()` |
| Unit variant (new) | `Variant` | `Variant$0` | `Variant` |
| Record variant (old) | `Variant` | `Variant` | `{x: i32, y: i32}` |
| Record variant (new) | `Variant {…}` | `Variant { x: ${1:()}, y: ${2:()} }$0` | `Variant { x: i32, y: i32 }` |
| Tuple variant (old) | `Variant(…)` | `Variant($0)` | `(i32, i32)` |
| Tuple variant (new) | `Variant(…)` | `Variant(${1:()}, ${2:()})$0` | `Variant(i32, i32)` |
Additionally, tuple variants no longer set `triggers_call_info` because it conflicts with placeholder generation, and tuple variants that require a qualified path should now use the qualified path.
Internally, this also lets us break the general "format an item with fields on it" code out into a shared module, so that means it'll be a lot easier to implement features like #11568.
Co-authored-by: Morgan Thomas <corp@m0rg.dev>
In particular:
- unit variants now display in the menu as "Variant", complete to "Variant", and display a detail of "Variant" (was "()")
- tuple variants now display in the menu as "Variant(…)", complete to "Variant(${1:()})$0" (was "Variant($0)"), and display a detail of "Variant(type)" (was "(type)")
- record variants now display in the menu as "Variant {…}", complete to "Variant { x: ${1:()} }$0" (was "Variant"), and display a detail of "Variant { x: type }" (was "{x: type}")
This behavior is identical to that of struct completions. In addition, tuple variants no longer set triggers_call_info, as to my understanding it's unnecessary now that we're emitting placeholders.
Tests have been updated to match, and the render::enum_variant::tests::inserts_parens_for_tuple_enums test has been removed entirely as it's covered by other tests (render::enum_detail_includes_{record, tuple}_fields, render::enum_detail_just_name_for_unit, render::pattern::enum_qualified).
- Add support for placeholder completions in tuple structs
- Denote tuple struct completions with `(…)` instead of ` {…}`
- Show struct completions as their type (`Struct { field: Type }`) in the completion menu instead of raw snippet text (`Struct { field: ${1:()} }$0`)
It wasn't testing the `const_arg` code path, it was actually hitting
const_param's default value code path, so move it to the right place
and rename it.
