I.e. the following situation:
```
fn foo() {
mod bar {
fn qux() {
// Prelude path here (e.g. macro use prelude or extern prelude).
}
}
}
```
Those were previously unresolved, because, in order to support `self` and `super` properly, since #15148 we do not ascend block paths when there is a module in between, but only crate def maps register preludes, not block def maps, and we can't change this because block def map prelude can always be overridden by another block. E.g.
```
fn foo() {
struct WithTheSameNameAsPreludeItem;
{
WithTheSameNameAsPreludeItem
}
}
```
Here `WithTheSameNameAsPreludeItem` refer to the item from the top block, but if we would register prelude items in each block the child block would overwrite it incorrectly.
Thanks to the observation (supported by counting) that the vast majority paths have neither generics no type anchors, and thanks to a new datastructure `ThinVecWithHeader` that is essentially `(T, Box<[U]>)` but with the size of a single pointer, we are able to reach this feat.
This (together with `ThinVecWithHeader`) makes the possibility to shrink `TypeRef`, because most types are paths.
So that given a `TypeRef` we will be able to trace it back to source code.
This is necessary to be able to provide diagnostics for lowering to chalk tys, since the input to that is `TypeRef`.
This means that `TypeRef`s now have an identity, which means storing them in arena and not interning them, which is an unfortunate (but necessary) loss but also a pretty massive change. Luckily, because of the separation layer we have for IDE and HIR, this change never crosses the IDE boundary.
Some types in `core` are conditionally compiled based on
`target_has_atomic` or `target_has_atomic_load_store` without an
argument, for example `AtomicU64`.
This is less noticeable in Cargo projects, where rust-analyzer adds
the output `RUSTC_BOOTSTRAP=1 cargo rustc --print cfg` so it gets the
full set of cfg flags.
This fixes go-to-definition on `std::sync::atomic::AtomicU64` in
non-cargo projects.
E.g.:
```rust
let v;
macro_rules! m { () => { v }; }
```
This was an existing bug, but it was less severe because unless the variable was shadowed it would be correctly resolved. With hygiene however, without this fix the variable is never resolved.
Or macro_rules hygiene, or mixed site hygiene. In other words, hygiene for variables and labels but not items.
The realization that made me implement this was that while "full" hygiene (aka. def site hygiene) is really hard for us to implement, and will likely involve intrusive changes and performance losses, since every `Name` will have to carry hygiene, mixed site hygiene is very local: it applies only to bodies, and we very well can save it in a side map with minor losses.
This fixes one diagnostic in r-a that was about `izip!()` using hygiene (yay!) but it introduces a huge number of others, because of #18262. Up until now this issue wasn't a major problem because it only affected few cases, but with hygiene identifiers referred by macros like that are not resolved at all. The next commit will fix that.
feat: render docs from aliased type when type has no docs
Trying to close#18344
- [x] ~Find the docs by traversing upwards if the type itself has none but aliasing for another type that might have.~
- [x] Show docs from aliased type.
- [x] Showing description that we are displaying documentation for different definition in hover box.
feat: better completions for extern blcoks
This PR refactors `add_keywords` (making it much clearer!) and enhances completion for `extern` blocks.
It is recommended to reviewing the changes in order of the commits:
- The first commit (f3c4dde0a4) doesn’t change any logic but refactors parts of the `add_keywords` function and adds detailed comments.
- The second commit (5dcc1ab649) improves completion for `extern` kw and extern blocks.
Add wrap/unwrap return type in Option
I pretty much just copied over the code and tests for wrapping/unwrapping return types in `Result` and then did a bunch of find and replace changes.
I handled unwrapping statements returning `None` by just replacing `None` with the unit type, but I'm open to suggestions for more intuitive behavior here.
And few more fixups.
I was worried this will lead to more memory usage since `ExprOrPatId` is double the size of `ExprId`, but this does not regress `analysis-stats .`. If this turns out to be a problem, we can easily use the high bit to encode this information.
Instead of lowering them to `<expr> = <expr>`, then hacking on-demand to resolve them, we lower them to `<pat> = <expr>`, and use the pattern infrastructure to handle them. It turns out, destructuring assignments are surprisingly similar to pattern bindings, and so only minor modifications are needed.
This fixes few bugs that arose because of the non-uniform handling (for example, MIR lowering not handling slice and record patterns, and closure capture calculation not handling destructuring assignments at all), and furthermore, guarantees we won't have such bugs in the future, since the programmer will always have to explicitly handle `Expr::Assignment`.
Tests don't pass yet; that's because the generated patterns do not exist in the source map. The next commit will fix that.
Because our lint infra *can* handle allows from within macro expansions!
(Also, what did this reason have to do with something that is a hard error and not a lint? I'm puzzled).
I wonder how many such diagnostics we have...
Maybe that also mean we can change `unused_mut` to no-longer-experimental? But this is a change I'm afraid to do without checking.
compiler: Adopt rust-analyzer impls for `LayoutCalculatorError`
We're about to massively churn the internals of `rustc_abi`. To minimize the immediate and future impact on rust-analyzer, as a subtree that depends on this crate, grow some API on `LayoutCalculatorError` that reflects their uses of it. This way we can nest the type in theirs, and they can just call functions on it without having to inspect and flatten-out its innards.
LSP says about Position::character
> If the character value is greater than the line length it defaults back to the line length.
but from_proto::offset() doesn't implement this.
A client might for example request code actions for a whole line by sending
Position::character=99999. I don't think there is ever a reason (besides laziness) why the
client can't specify the line length instead but I guess we should not crash but follow protocol.
Technically it should be a warning, not an error but warning is not shown by default so keep
it at error I guess.
Fixes#18240
When debugging rust-analyzer and looking at logs, it's much easier to read
when the timestamp is in the local timezone.
Before:
2024-08-28T20:55:38.792321Z INFO ParseQuery: invoked at R18460
After:
2024-08-28T13:55:38.792321-07:00 INFO ParseQuery: invoked at R18460
chore: rename `salsa` to `ra_salsa`
Laying some groundwork to start before I import the new Salsa crate. Here's why:
1. As part of the migration, `@darichey,` `@Wilfred,` and I will create new Salsa equivalents of the existing databases/query groups. We'll get them to compile crate-by-crate.
2. Once we wrote all equivalents of all queries, we'd start to refactor usage sites of the vendored Salsa to use the new Salsa databases.
3. Starting porting usage sites of old Salsa to the new Salsa.
4. Remove the vendored `ra_salsa`; declare victory.
feat: respect references.exclude_tests in call-hierarchy
close#18212
### Changes
1. feat: respect `references.exclude_tests` in call-hierarchy
2. Modified the description of `references.exclude_tests`
fix: Do not consider mutable usage of deref to `*mut T` as deref_mut
Fixes#15799
We are doing some heuristics for deciding whether the given deref is deref or deref_mut here;
5982d9c420/crates/hir-ty/src/infer/mutability.rs (L182-L200)
But this heuristic is erroneous if we are dereferencing to a mut ptr and normally those cases are filtered out here as builtin;
5982d9c420/crates/hir-ty/src/mir/lower/as_place.rs (L165-L177)
Howerver, this works not so well if the given dereferencing is double dereferencings like the case in the #15799.
```rust
struct WrapPtr(*mut u32);
impl core::ops::Deref for WrapPtr {
type Target = *mut u32;
fn deref(&self) -> &Self::Target {
&self.0
}
}
fn main() {
let mut x = 0u32;
let wrap = WrapPtr(&mut x);
unsafe {
**wrap = 6;
}
}
```
Here are two - outer and inner - dereferences here, and the outer dereference is marked as deref_mut because there is an assignment operation.
And this deref_mut marking is propagated into the inner dereferencing.
In the later MIR lowering, the outer dereference is filtered out as it's expr type is `*mut u32`, but the expr type in the inner dereference is an ADT, so this false-mutablility is not filtered out.
This PR cuts propagation of this false mutablilty chain if the expr type is mut ptr.
Since this happens before the resolve_all, it may have some limitations when the expr type is determined as mut ptr at the very end of inferencing, but I couldn't find simple fix for it 🤔
internal: Don't resolve extern crates in import fix point resolution
The fix point loop won't progress them given the potential extern crate candidates are set up at build time.
fix: Join rustfmt overrideCommand with project root
When providing a custom rustfmt command, join it with the project root instead of the workspace root. This fixes rust-analyzer getting the wrong invocation path in projects containing subprojects.
This makes the behaviour consistent with how a custom path provided in rust-analyzer.procMacro.server behaves already.
Resolves issue #18222
feat: Highlight exit points of async blocks
Async blocks act similar to async functions in that the await keywords are related, but also act like functions where the exit points are related.
Fixes#18147
fix: include description in label details when detail field is marked for …
Fixes https://github.com/rust-lang/rust-analyzer/issues/18231.
When omitting the autocomplete detail field, the autocomplete label details can still be returned. Currently the label details are missing the description field if the detail field is included in resolveSupport since it is being overwritten as None and opted to be sent with `completionItem/resolve`.
Example completion capabilities.
```
completion = {
completionItem = {
commitCharactersSupport = true,
deprecatedSupport = true,
documentationFormat = { "markdown", "plaintext" },
insertReplaceSupport = true,
insertTextModeSupport = {
valueSet = { 1, 2 }
},
labelDetailsSupport = true,
preselectSupport = true,
resolveSupport = {
properties = { "documentation", "detail", "additionalTextEdits", "sortText", "filterText", "insertText", "textEdit", "insertTextFormat", "insertTextMode" }
},
snippetSupport = true,
tagSupport = {
valueSet = { 1 }
}
}
```
lsp: fix completion_item something_to_resolve not being a latch to true
while looking at #18245 i noticed that `something_to_resolve` could technically flap between true -> false if some subsequent fields that were requested to be resolved were empty.
this fixes that by using `|=` instead of `=` when assigning to `something_to_resolve` which will prevent it from going back to false once set.
although some cases it's simply assigning to `true` i opted to continue to use `|=` there for uniformity sake. but happy to change those back to `=`'s.
cc `@SomeoneToIgnore`
before, when formatting struct constructor for `struct S(usize, usize)` it would format as:
extern "rust-call" S(usize, usize) -> S
but after this change, we'll format as:
fn S(usize, usize) -> S
internal: add json `tracing` Layer for profiling startup
On `buck2/integrations/rust-project`, this results in the following being printed:
```json
{"name":"discover_command","elapsed_ms":18703}
{"name":"parallel_prime_caches","elapsed_ms":0}
{"name":"vfs_load","elapsed_ms":5895}
{"name":"vfs_load","elapsed_ms":547}
{"name":"parallel_prime_caches","elapsed_ms":23}
{"name":"parallel_prime_caches","elapsed_ms":84}
{"name":"parallel_prime_caches","elapsed_ms":5819}
```
When providing a custom rustfmt command, join it with the project
root instead of the workspace root. This fixes rust-analyzer
getting the wrong invocation path in projects containing subprojects.
This makes the behaviour consistent with how a custom path provided
in rust-analyzer.procMacro.server behaves already.
Resolves issue #18222
fix: Fix resolution of label inside macro
When working on Something Else (TM) (I left a hint in the commits :P), I noticed to my surprise that labels inside macros are not resolved. This led to a discovery of *two* unrelated bugs, which are hereby fixed in two commits.
This fixes a bug where labels inside macros were not resolved, but more importantly this prepares us to a future where we have hygiene, and textual equivalence isn't enough to resolve identifiers.
internal: Send less data during `textDocument/completion` if possible
Similar to https://github.com/rust-lang/rust-analyzer/pull/15522, stops sending extra data during `textDocument/completion` if that data was set in the client completions resolve capabilities, and sends those only during `completionItem/resolve` requests.
Currently, rust-analyzer sends back all fields (including potentially huge docs) for every completion item which might get large.
Same as the other one, this PR aims to keep the changes minimal and does not remove extra computations for such fields — instead, it just filters them out before sending to the client.
The PR omits primitive, boolean and integer, types such as `deprecated`, `preselect`, `insertTextFormat`, `insertTextMode`, etc. AND `additionalTextEdits` — this one looks very dangerous to compute for each completion item (as the spec says we ought to if there's no corresponding resolve capabilities provided) due to the diff computations and the fact that this code had been in the resolution for some time.
It would be good to resolve this lazily too, please let me know if it's ok to do.
When tested with Zed which only defines `documentation` and `additionalTextEdits` in its client completion resolve capabilities, rust-analyzer starts to send almost 3 times less characters:
Request:
```json
{"jsonrpc":"2.0","id":104,"method":"textDocument/completion","params":{"textDocument":{"uri":"file:///Users/someonetoignore/work/rust-analyzer/crates/ide/src/inlay_hints.rs"},"position":{"line":90,"character":14},"context":{"triggerKind":1}}}
```
<img width="1338" alt="image" src="https://github.com/user-attachments/assets/104f19b5-7095-4fc1-b008-5d829623b2e2">
Before: 381944 characters
[before.json](https://github.com/user-attachments/files/17092385/before.json)
After: 140503 characters
[after.json](https://github.com/user-attachments/files/17092386/after.json)
After Zed's [patch](https://github.com/zed-industries/zed/pull/18212) to enable all resolving possible: 84452 characters
[after-after.json](https://github.com/user-attachments/files/17092755/after-after.json)
fix: Ambiguity with CamelCase diagnostic messages, align with rustc warnings
Fixed diagnostic messages so they say UpperCamelCase rather than CamelCase, as it is ambiguous.
Usually I'd call it PascalCase, but in the code base it is called UpperCamelCase so I left it with that naming choice.
`rustc` says `upper camel case` also when the case is wrong
```
warning: trait `testThing` should have an upper camel case name
--> src/main.rs:5:7
|
5 | trait testThing {
| ^^^^^^^^^ help: convert the identifier to upper camel case: `TestThing`
|
= note: `#[warn(non_camel_case_types)]` on by default
```
This is in line with the UPPER_SNAKE_CASE diagnostic messages.
546339a7be/crates/hir-ty/src/diagnostics/decl_check.rs (L60)546339a7be/crates/ide-diagnostics/src/handlers/incorrect_case.rs (L535)
This will mean users opting to not activate `cfg(test)` will lose IDE experience on them, which is quite unfortunate, but this is unavoidable if we want to avoid false positives on e.g. diagnostics. The real fix is to provide IDE experience even for cfg'ed out code, but this is out of scope for this PR.
Rename object_safety
First PR here (yay!), so I read some of the getting started docs. There are a couple references to `handlers.rs`, which as far as I can tell has been refactored into `handlers/*.rs`. I made some tweaks to that in one commit. There is one fixme about a function called `to_lsp_runnable`, which I can't find anywhere at all. I can update that if I get some more info there.
Otherwise I changed references to object safety, is object safe, etc., trying to match case/style as I went. There was one case I found where there's a trait from somewhere else called `is_object_safe`, which I found defined in my cargo registry. I didn't touch that for now, just marked it with a fixme
Include buildfiles in VFS
We subscribe to `textDocument/didSave` for `filesToWatch`, but the VFS doesn't contain those files. Before https://github.com/rust-lang/rust-analyzer/pull/18105, this would bring down the server. Now, it's only a benign error logged:
```
ERROR notification handler failed handler=textDocument/didSave error=file not found: /foo/bar/TARGETS
```
It's benign, because we will also receive a `workspace/didChangeWatchedFiles` for the file which will invalidate and load it.
Explicitly include the buildfiles in the VFS to prevent the handler from erroring.
