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|
Tracking import use types for more accurate redundant import checking
fixes#117448
By tracking import use types to check whether it is scope uses or the other situations like module-relative uses, we can do more accurate redundant import checking.
For example unnecessary imports in std::prelude that can be eliminated:
```rust
use std::option::Option::Some;//~ WARNING the item `Some` is imported redundantly
use std::option::Option::None; //~ WARNING the item `None` is imported redundantly
```
fixes#117448
For example unnecessary imports in std::prelude that can be eliminated:
```rust
use std::option::Option::Some;//~ WARNING the item `Some` is imported redundantly
use std::option::Option::None; //~ WARNING the item `None` is imported redundantly
```
feat: Introduce term search to rust-analyzer
# Introduce term search to `rust-analyzer`
_I've marked this as draft as there might be some shortcomings, please point them out so I can fix them. Otherwise I think it is kind of ready as I think I'll rather introduce extra functionality in follow up PRs._
Term search (or I guess expression search for rust) is a technique to generate code by basically making the types match.
Consider the following program
```rust
fn wrap(arg: i32) -> Option<i32> {
todo!();
}
```
From the types of values in scope and constructors of `Option`, we can produce the expected result of wrapping the argument in `Option`
Dependently typed languages such as `Idris2` and `Agda` have similar tools to help with proofs, but this can be also used in everyday development as a "auto-complete".
# Demo videos
https://github.com/rust-lang/rust-analyzer/assets/19900308/7b68a1b7-7dba-4e31-9221-6c7485e77d88https://github.com/rust-lang/rust-analyzer/assets/19900308/0fae530a-aabb-4b28-af71-e19f8d3d64b2
# What does it currently do
- It works well with locals, free functions, type constructors and non-static impl methods that take items by value.
- Works with functions/methods that take shared references, but not with unique references (very conservative).
- Can handle projections to struct fields (eg. `foo.bar.baz`) but this might me more conservative than it has to be to avoid conflicting with borrow checker
- Should create only valid programs (no type / borrow checking errors). Tested with `rust-analyzer analysis-stats /path/to/ripgrep/Cargo.toml --run-term-search --validate-term-search` (basically running `cargo check` on all of the generated programs and only error seems to be due to type inference which is more of issue of testing method.
# Performace / fitness
```txt
ripgrep (latest)
Tail Expr syntactic hits: 130/1692 (7%)
Tail Exprs found: 523/1692 (30%)
Term search avg time: 9ms
Term search: 15.64s, 97ginstr, 8mb
rust-analyzer (on this branch)
Tail Expr syntactic hits: 804/13860 (5%)
Tail Exprs found: 6757/13860 (48%)
Term search avg time: 78ms
Term search: 1088.23s, 6765ginstr, 98mb
```
Highly generic code seems to blow up the search space so currently the amount of generics allowed is functions/methods is limited down to 0 (1 didn't give much improvement and 2 is already like 0.5+s search time)
# Plans for the future (not in this PR)
- ``~~Add impl methods that do not take `self` type (should be quite straight forward)~~ Done
- Be smarter (aka less restrictive) about borrow checking - this seems quite hard but since the current approach is rather naive I think some easy improvement is available.
- ``~~See if it works as a autocomplete while typing~~ Done
_Feel free to ask questions / point of shortcoming either here or on Zulip, I'll be happy to address them. I'm doing this as part of my MSc thesis so I'll be working on it till summer anyway 😄_
internal: tool discovery prefers sysroot tools
Fixes https://github.com/rust-lang/rust-analyzer/issues/15927, Fixes https://github.com/rust-lang/rust-analyzer/issues/16523
After this PR we will look for `cargo` and `rustc` in the sysroot if it was succesfully loaded instead of using the current lookup scheme. This should be more correct than the current approach as that relies on the working directory of the server binary or loade workspace, meaning it can behave a bit odd wrt overrides.
Additionally, rust-project.json projects now get the target data layout set so there should be better const eval support now.
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.
feat: Add incorrect case diagnostics for traits and their associated items
Updates incorrect case diagnostic to:
- Check traits and their associated items
- Ignore trait implementations except for patterns in associated function bodies
Also cleans up `hir-ty::diagnostics::decl_check` a bit (mostly to make it a bit more DRY and easier to maintain)
Also fixes: #8675 and fixes: #8225
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!)
internal: `tracing` improvements and followups
Hi folks! Building on https://github.com/rust-lang/rust-analyzer/pull/16394, I've got a few small tweaks:
- Removed the accidental `mod.rs` usage that I introduced.
- Removed a panic in `pat_analysis.rs`.
- Recorded the event kind in `handle_event` to better distinguish what _kind_ of event is being handled.
- Did a small refactor of `hprof` to have somewhat more linear control flow, and more importantly, write the recorded fields to the output.
The end result is the following:
<img width="1530" alt="A screenshot of Visual Studio Code on a Mac. `hprof.rs` is open, with " src="https://github.com/rust-lang/rust-analyzer/assets/2067774/bd11dde5-b2da-4774-bc38-bcb4772d1192">
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.
