11397: internal: Refactor completion module split r=Veykril a=Veykril
Currently our completion infra is split into several modules, each trying to do completions for something specific. This "something" is rather unstructured as it stands now, we have a module for `flyimporting path` completions, `unqualified` and `qualified path` completions, modules for `pattern position` completions that only try to complete extra things for patterns that aren't done in the path modules, `attribute` completions that again only try to add builtin attribute completions without adding the normal path completions and a bunch of other special "entity" completions like lifetimes, method call/field access, function param cloning, ... which serve a more specific purpose than the previous listed ones.
As is evident, the former mentioned ones have some decent overlap which requires extra filtering in them so that they don't collide with each other duplicating a bunch of completions(which we had happen in the past at times).
Now this overlap mostly happens with path completions(and keyword completions as well in some sense) which gives me the feeling that having `qualified` and `unqualified` path completions be separate from the rest gives us more troubles than benefits in the long run.
So this is an attempt at changing this structure to instead still go by rough entity for special cases, but when it comes to paths we instead do the module split on the "path kinds"/"locations"(think pattern, type, expr position etc) that exist. This goes hand in hand with the test refactoring I have done that moved tests to "location oriented" modules as well as the `CompletionContext` refactoring that actually already started splitting the context up for path kinds.
This PR moves some path completions out of the `qualified` and `unqualified` path modules namely attribute, visibility, use and pattern paths.
Co-authored-by: Lukas Wirth <lukastw97@gmail.com>
11399: Fix assoc type shorthand from method bounds r=flodiebold a=flodiebold
In code like this:
```rust
impl<T> Option<T> {
fn as_deref(&self) -> T::Target where T: Deref {}
}
```
when trying to resolve the associated type `T::Target`, we were only
looking at the bounds on the impl (where the type parameter is defined),
but the method can add additional bounds that can also be used to refer
to associated types. Hence, when resolving such an associated type, it's
not enough to just know the type parameter T, we also need to know
exactly where we are currently.
This fixes#11364 (beta apparently switched some bounds around).
Co-authored-by: Florian Diebold <flodiebold@gmail.com>
In code like this:
```rust
impl<T> Option<T> {
fn as_deref(&self) -> T::Target where T: Deref {}
}
```
when trying to resolve the associated type `T::Target`, we were only
looking at the bounds on the impl (where the type parameter is defined),
but the method can add additional bounds that can also be used to refer
to associated types. Hence, when resolving such an associated type, it's
not enough to just know the type parameter T, we also need to know
exactly where we are currently.
This fixes#11364 (beta apparently switched some bounds around).
11322: Extract function also extracts comments r=Vannevelj a=Vannevelj
Fixes#9011
The difficulty I came across is that the original assist works from the concept of a `ast::StmtList`, a node, but that does not allow me to (easily) represent comments, which are tokens. To combat this, I do a whole bunch of roundtrips: from the `ast::StmtList` I retrieve the `NodeOrToken`s it encompasses.
I then cast all `Node` ones back to a `Stmt` so I can apply indentation to it, after which it is again parsed as a `NodeOrToken`.
Lastly, I add a new `make::` api that accepts `NodeOrToken` rather than `StmtList` so we can write the comment tokens.
Co-authored-by: Jeroen Vannevel <jer_vannevel@outlook.com>
11182: fix: don't panic on seeing an unexpected offset r=Veykril a=dimbleby
Intended as a fix, or at least a sticking plaster, for #11081.
I have arranged that [offset()](1ba9a924d7/crates/ide_db/src/line_index.rs (L105-L107)) returns `Option<TextSize>` instead of going out of bounds; other changes are the result of following the compiler after doing this.
Perhaps there's still an issue here - I suppose the server and client have gotten out of sync and that probably shouldn't happen in the first place? I see that https://github.com/rust-analyzer/rust-analyzer/issues/10138#issuecomment-913727554 suggests what sounds like a more substantial fix which I think might be aimed in this direction. So perhaps that one should be left open to cover such things?
Meanwhile, I hope that not-crashing is a good improvement: and I can confirm that it works out just fine in the repro I have at #11081.
Co-authored-by: David Hotham <david.hotham@metaswitch.com>
11367: minor: Use `compare_exchange_weak` in `limit::Limit::check` r=lnicola a=WaffleLapkin
Not a big deal, but generally loops should use `_weak` version of `compare_exchange`.
Co-authored-by: Maybe Waffle <waffle.lapkin@gmail.com>
