performance: Speed up Method Completions By Taking Advantage of Orphan Rules
(Continues https://github.com/rust-lang/rust-analyzer/pull/16498)
This PR speeds up method completions by doing two things without regressing `analysis-stats`[^1]:
- Filter candidate traits prior to calling `iterate_path_candidates` by relying on orphan rules (see below for a slightly more in-depth explanation). When generating completions [on `slog::Logger`](5e9e59c312/common/src/ledger.rs (L78)) in `oxidecomputer/omicron` as a test, this PR halved my completion times—it's now 454ms cold and 281ms warm. Before this PR, it was 808ms cold and 579ms warm.
- Inline some of the method candidate checks into `is_valid_method_candidate` and remove some unnecessary visibility checks. This was suggested by `@Veykril` in [this comment](https://github.com/rust-lang/rust-analyzer/pull/16498#issuecomment-1929864427).
We filter candidate traits by taking advantage of orphan rules. For additional details, I'll rely on `@WaffleLapkin's` explanation [from Zulip](https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/Trait.20Checking/near/420942417):
> A type `A` can only implements traits which
> 1. Have a blanket implementation (`impl<T> Trait for T {}`)
> 2. Have implementation for `A` (`impl Trait for A {}`)
>
> Blanket implementation can only exist in `Trait`'s crate. Implementation for `A` can only exist in `A`'s or `Trait`'s crate.
Big thanks to Waffle for its keen observation!
---
I think some additional improvements are possible:
- `for_trait_and_self_ty` seemingly does not distinguish between `&T`, `&mut T`, or `T`, resulting in seemingly irrelevant traits like `tokio::io::AsyncWrite` being being included for, e.g., `&slog::Logger`. I don't know they're being considered due to the [autoref/autoderef behavior](a02a219773/crates/hir-ty/src/method_resolution.rs (L945-L962)), but I wonder if it'd make sense to filter by mutability earlier and not consider trait implementations that require `&mut T` when we only have a `&T`.
- The method completions [spend a _lot_ of time in unification](https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/Trait.20Checking/near/421072356), and while there might be low-hanging fruit there, it might make more sense to wait for the new trait solver in `rustc`. I dunno.
[^1]: The filtering occurs outside of typechecking, after all.
Setup infra for handling auto trait bounds disabled due to perf problems
This patch updates some of the partially-implemented functions of `ChalkContext as RustIrDatabase`, namely `adt_datum()` and `impl_provided_for()`. With those, we can now correctly work with auto trait bounds and distinguish methods based on them.
Resolves#7856 (the second code; the first one is resolved by #13074)
**IMPORTANT**: I don't think we want to merge this until #7637 is resolved. Currently this patch introduces A LOT of unknown types and type mismtaches as shown below. This is because we cannot resolve items like `hashbrown::HashMap` in `std` modules, leading to auto trait bounds on them and their dependents unprovable.
|crate (from `rustc-perf@c52ee6` except for r-a)|e3dc5a588f07d6f1d3a0f33051d4af26190abe9e|HEAD of this branch|
|---|---|---|
|rust-analyzer @ e3dc5a588f |exprs: 417528, ??ty: 907 (0%), ?ty: 114 (0%), !ty: 1|exprs: 417528, ??ty: 1704 (0%), ?ty: 403 (0%), !ty: 20|
|ripgrep|exprs: 62120, ??ty: 2 (0%), ?ty: 0 (0%), !ty: 0|exprs: 62120, ??ty: 132 (0%), ?ty: 58 (0%), !ty: 11|
|webrender/webrender|exprs: 94355, ??ty: 49 (0%), ?ty: 16 (0%), !ty: 2|exprs: 94355, ??ty: 429 (0%), ?ty: 130 (0%), !ty: 7|
|diesel|exprs: 132591, ??ty: 401 (0%), ?ty: 5129 (3%), !ty: 31|exprs: 132591, ??ty: 401 (0%), ?ty: 5129 (3%), !ty: 31|
Abstract more over ItemTreeLoc-like structs
Allows reducing some code duplication by using functions generic over said structs. The diff isn't negative due to me adding some additional impls for completeness.
internal: even more `tracing`
As part of profiling completions, I added some additional spans and moved `TyBuilder::subst_for_def` closer to its usage site (the latter had a small impact on completion performance. Thanks for the tip, Lukas!)
This commit also adds `tracing` to NotificationDispatcher/RequestDispatcher,
bumps `rust-analyzer-salsa` to 0.17.0-pre.6, `always-assert` to 0.2, and
removes the homegrown `hprof` implementation in favor of a vendored
tracing-span-tree.
`cargo clippy --fix`
This PR is the result of running `cargo clippy --fix && cargo fmt` in the root of the repository. I did not manually review all the changes, but just skimmed through a few of them. The tests still pass, so it seems fine.
Add a new config to allow renaming of non-local defs
With #15656 we started disallowing renaming of non-local items. Although this makes sense there are some false positives that impacted users' workflows. So this config aims to mitigate this by giving users the liberty to disable this feature.
The reason why this is a draft is that I saw one of the tests fail and I am not sure if the "got" result even syntactically makes sense
Test case is :
```rust
check(
"Baz",
r#"
//- /lib.rs crate:lib new_source_root:library
pub struct S;
//- /main.rs crate:main deps:lib new_source_root:local
use lib::S$0;
"#,
"use lib::Baz;"
);
```
```
Left:
use lib::Baz;
Right:
use lib::Baz;Baz
Diff:
use lib::Baz;Baz
```
With #15656 we started disallowing renaming of non-local items.
Although this makes sense there are some false positives that
impacted users' workflows. So this config aims to mitigate this
by giving users the liberty to disable this feature.
internal: Follow rustfmt's algorithm for ordering imports when ordering and merging use trees
Updates use tree ordering and merging utilities to follow rustfmt's algorithm for ordering imports.
The [rustfmt implementation](6356fca675/src/imports.rs) was used as reference.
fix: Acknowledge `pub(crate)` imports in import suggestions
rust-analyzer has logic that discounts suggesting `use`s for private imports, but that logic is unnecessarily strict - for instance given this code:
```rust
mod foo {
pub struct Foo;
}
pub(crate) use self::foo::*;
mod bar {
fn main() {
Foo$0;
}
}
```
... RA will suggest to add `use crate::foo::Foo;`, which not only makes the code overly verbose (especially in larger code bases), but also is disjoint with what rustc itself suggests.
This commit adjusts the logic, so that `pub(crate)` imports are taken into account when generating the suggestions; considering rustc's behavior, I think this change doesn't warrant any extra configuration flag.
Note that this is my first commit to RA, so I guess the approach taken here might be suboptimal - certainly feels somewhat hacky, maybe there's some better way of finding out the optimal import path 😅
rust-analyzer has logic that discounts suggesting `use`s for private
imports, but that logic is unnecessarily strict - for instance given
this code:
```rust
mod foo {
pub struct Foo;
}
pub(crate) use self::foo::*;
mod bar {
fn main() {
Foo$0;
}
}
```
... RA will suggest to add `use crate::foo::Foo;`, which not only makes
the code overly verbose (especially in larger code bases), but also is
disjoint with what rustc itself suggests.
This commit adjusts the logic, so that `pub(crate)` imports are taken
into account when generating the suggestions; considering rustc's
behavior, I think this change doesn't warrant any extra configuration
flag.
Note that this is my first commit to RA, so I guess the approach taken
here might be suboptimal - certainly feels somewhat hacky, maybe there's
some better way of finding out the optimal import path 😅
Resolve panic in `generate_delegate_methods`
Fixes#16276
This PR addresses two issues:
1. When using `PathTransform`, it searches for the node corresponding to the `path` in the `source_scope` during `make::fn_`. Therefore, we need to perform the transform before `make::fn_` (similar to the problem in issue #15804). Otherwise, even though the tokens are the same, their offsets (i.e., `span`) differ, resulting in the error "Can't find CONST_ARG@xxx."
2. As mentioned in the first point, `PathTransform` searches for the node corresponding to the `path` in the `source_scope`. Thus, when transforming paths, we should update nodes from right to left (i.e., use **reverse of preorder** (right -> left -> root) instead of **postorder** (left -> right -> root)). Reasons are as follows:
In the red-green tree (rowan), we do not store absolute ranges but instead store the length of each node and dynamically calculate offsets (spans). Therefore, when modifying the left-side node (such as nodes are inserted or deleted), it causes all right-side nodes' spans to change. This, in turn, leads to PathTransform being unable to find nodes with the same paths (due to different spans), resulting in errors.
fix: Fix `ast::Path::segments` implementation
calling `ast::Path::segments` on a qualifier currently returns all the segments of the top path instead of just the segments of the qualifier.
The issue can be summarized by the simple failing test below:
```rust
#[test]
fn path_segments() {
//use ra_ap_syntax::ast;
let path: ast::Path = ...; // e.g. `ast::Path` for "foo::bar::item".
let path_segments: Vec<_> = path.segments().collect();
let qualifier_segments: Vec<_> = path.qualifier().unwrap().segments().collect();
assert_eq!(path_segments.len(), qualifier_segments.len() + 1); // Fails because `LHS = RHS`.
}
```
This PR:
- Fixes the implementation of `ast::Path::segments`
- Fixes `ast::Path::segments` callers that either implicitly relied on behavior of previous implementation or exhibited other "wrong" behavior directly related to the result of `ast::Path::segments` (all callers have been reviewed, only one required modification)
- Removes unnecessary (and now unused) `ast::Path::segments` alternatives
SymbolInformation::kind is finer-grained than the SCIP symbol suffix.
This also fixes a bug where all type aliases where treated like type
parameters.
```
trait SomeTrait {
type AssociatedType; // ← this is SomeTrait#[AssociatedType]
}
type MyTypeAlias = u8; // ← this used to be [MyTypeAlias]
// and now is MyTypeAlias#
```
To build the SymbolInformation::signature_documentation we need access
to the “label” when building the TokenStaticData, preferably without
any markdown markup.
Therefore this refactors ide::hover::render::definition and its helper
functions to give easier access to the label alone.
For local variables, this gets the moniker from the enclosing
definition and stores it into the TokenStaticData.
Then it builds the scip symbol for that moniker when building the
SymbolInformation.
fix: rewrite code_action `generate_delegate_trait`
I've made substantial enhancements to the "generate delegate trait" code action in rust-analyzer. Here's a summary of the changes:
#### Resolved the "Can’t find CONST_ARG@158..159 in AstIdMap" error
Fix#15804, fix#15968, fix#15108
The issue stemmed from an incorrect application of PathTransform in the original code. Previously, a new 'impl' was generated first and then transformed, causing PathTransform to fail in locating the correct AST node, resulting in an error. I rectified this by performing the transformation before generating the new 'impl' (using make::impl_trait), ensuring a step-by-step transformation of associated items.
#### Rectified generation of `Self` type
`generate_delegate_trait` is unable to properly handle trait with `Self` type.
Let's take the following code as an example:
```rust
trait Trait {
fn f() -> Self;
}
struct B {}
impl Trait for B {
fn f() -> B { B{} }
}
struct S {
b: B,
}
```
Here, if we implement `Trait` for `S`, the type of `f` should be `() -> Self`, i.e. `() -> S`. However we cannot automatically generate a function that constructs `S`.
To ensure that the code action doesn't generate delegate traits for traits with Self types, I add a function named `has_self_type` to handle it.
#### Extended support for generics in structs and fields within this code action
The former version of `generate_delegate_trait` cannot handle structs with generics properly. Here's an example:
```rust
struct B<T> {
a: T
}
trait Trait<T> {
fn f(a: T);
}
impl<T1, T2> Trait<T1> for B<T2> {
fn f(a: T1) -> T2 { self.a }
}
struct A {}
struct S {
b$0 : B<A>,
}
```
The former version will generates improper code:
```rust
impl<T1, T2> Trait<T1, T2> for S {
fn f(&self, a: T1) -> T1 {
<B as Trait<T1, T2>>::f( &self.b , a)
}
}
```
The rewritten version can handle generics properly:
```rust
impl<T1> Trait<T1> for S {
fn f(&self, a: T1) -> T1 {
<B<A> as Trait<T1>>::f(&self.b, a)
}
}
```
See more examples in added unit tests.
I enabled support for generic structs in `generate_delegate_trait` through the following steps (using the code example provided):
1. Initially, to prevent conflicts between the generic parameters in struct `S` and the ones in the impl of `B`, I renamed the generic parameters of `S`.
2. Then, since `B`'s parameters are instantiated within `S`, the original generic parameters of `B` needed removal within `S` (to avoid errors from redundant parameters). An important consideration here arises when Trait and B share parameters in `B`'s impl. In such cases, these shared generic parameters cannot be removed.
3. Next, I addressed the matching of types between `B`'s type in `S` and its type in the impl. Given that some generic parameters in the impl are instantiated in `B`, I replaced these parameters with their instantiated results using PathTransform. For instance, in the example provided, matching `B<A>` and `B<T2>`, where `T2` is instantiated as `A`, I replaced all occurrences of `T2` in the impl with `A` (i.e. apply the instantiated generic arguments to the params).
4. Finally, I performed transformations on each assoc item (also to prevent the initial issue) and handled redundant where clauses.
For a more detailed explanation, please refer to the code and comments. I welcome suggestions and any further questions!
Don't trim trailing whitespace from doc comments
Don't trim trailing whitespace from doc comments as multiple trailing spaces indicates a hard line break in Markdown.
I'd have liked to add a unit test for `docs_from_attrs`, but couldn't find a reasonable way to get an `&Attrs` object for use in the test.
Fixes#15877.
feat: generate descriptors for all unstable features
Most unstable features don't have their own chapter in the unstable book, so a rustc helper tool (`src/tools/unstable-book-gen`) generates shims to fill the gaps.
Run this tool to generate the full unstable-book source before parsing it.
Fix autoimport does nothing when importing trait that is as _ imports
Potentially fixes#15128
There are two cases of imports:
1. With simple path
2. With use tree list (or say complex path).
On deeper inspection, the [`recursive_merge`](994df3d6a3/crates/ide-db/src/imports/merge_imports.rs (L87)) function (called by [`try_merge_trees_mut`)](994df3d6a3/crates/ide-db/src/imports/merge_imports.rs (L69)) is meaningful only in the case of complex path (i.e when the UseTree contains a UseTreeList).
The [`recursive_merge`](994df3d6a3/crates/ide-db/src/imports/merge_imports.rs (L87)) function has [match with `Ok` arm](994df3d6a3/crates/ide-db/src/imports/merge_imports.rs (L106)), that is only executed when both LHS and RHS has `PathSegment` with same `NameRef`. The removal of underscore is implemented in this arm in the case of complex path.
For simple paths, the underscore is removed by checking if both LHS and RHS are simple paths and if their `Path` is same (the check is done [here](994df3d6a3/crates/ide-db/src/imports/merge_imports.rs (L74))) and remove the underscore if one is found (I made an assumption here that RHS will always be what rust-analyzer suggests to import, because at this point I'm not sure how to remove underscore with help of `ted::replace`).
