Example
---
```rust
fn foo() {
    let foo = Ok(1);
    return foo.unwrap_$0or(2);
}
```

**Before this PR**

```rust
fn foo() {
    let foo = Ok(1);
    return foo.unwrap_or_else(|| 2);
}
```

**After this PR**

```rust
fn foo() {
    let foo = Ok(1);
    return foo.unwrap_or_else(|e| 2);
}
```

minor: sync from downstream

…yumu

Less conjuration magic

The most important reason is incrementality. While not a lot of things depend on the stability of `EnumVariantId`, it's still useful to have them stable.

However it turns out that many things actually do want the name, more than those that want the index.

`ThinVec` doesn't allocate when empty.

Previously module ID reuse was based on order of creation solely. Now we store the parent module and name, and so Salsa will reuse the ID for modules that have the same name and parent.

This is important as many interned IDs contain modules, so if the module is invalidated they too are.

Example
---
```rust
struct Foo(&'static str);

impl Foo {
    fn text(&self) -> &str { self.0 }
}

fn main() {
    let s = Foo("");
    $0print!("{}{}", s, s);$0
    let _ = s.text() == "";
}
```

**Before this PR**

```rust
fn $0fun_name(s: &Foo) {
    *print!("{}{}", s, s);
}
```

**After this PR**

```rust
fn $0fun_name(s: &Foo) {
    print!("{}{}", *s, *s);
}
```

---

```rust
macro_rules! refmut { ($e:expr) => { &mut $e }; }
fn foo() {
    let mut n = 1;
    $0let v = refmut!(n);
    *v += 1;$0
    let k = n;
}
```

**Before this PR**

```rust
fn $0fun_name(n: &mut i32) {
    let v = refmut!(n);
    *v += 1;
}
```

**After this PR**

```rust
fn $0fun_name(n: &mut i32) {
    let v = refmut!(*n);
    *v += 1;
}
```

minor: Replace `cfg-if` with `std::cfg_select`

…ge-pat-e0029

feat: add diagnostic for E0029

Encode the name instead of index in `EnumVariantId`

…han 2

…-exhaustive-record-pat

feat: add diagnostic for E0638

Signed-off-by: Kai Tanaka <275430420+quyentonndbs@users.noreply.github.com>

…ords-in-stdx-and-ide-comments

fix: duplicated words in stdx/assert.rs and ide/inject.rs doc-comments

  `request.rs` was passing `binary_target = true` for
  TargetKind::Bin, Example, and Test, but not for Bench, so a
  `fn main()` in `benches/foo.rs` (typical with
  `harness = false`) was suppressed by `should_skip_runnable`
  in the annotations layer.  Mirror `target_spec.rs:255` by
  including Bench in both the code-lens config check and
  `should_skip_target`.

  Closes rust-lang/rust-analyzer#21948.

Signed-off-by: Onyeka Obi <softwareengineerasaservant@isurvivable.cv>

perf: Provide access to `RootDatabase`'s `LineIndex` for the proc macro protocol

fix: show Run lens for fn main in bench targets

Show `const` in the signature help if applicable

…se_path_seg

Fix assit `qualify_path` loses path segment

fix: handle usages in macro for extract_function

Previously, adding newline between statements was often redundant

Example
---
```rust
fn main() {
    let bar = Some(true);
    bar.$0
    other();
}
```

**Before this PR**

```rust
fn main() {
    let bar = Some(true);
    let Some(${1:bar}) = bar else {
        $2
    };
    $0
    other();
}
```

**After this PR**

```rust
fn main() {
    let bar = Some(true);
    let Some(${1:bar}) = bar else {
        $2
    };$0
    other();
}
```

…kkno

Enable salsa feature for syntax-bridge

fix: Try to improve completion ranking

Kill proc-macro-srv processes on shutdown

fix: use add deref in assign instead add `&mut` for value