In the implementation of `force_mut`, I chose performance over safety.
For `LazyLock` this isn't really a choice; the code has to be unsafe.
But for `LazyCell`, we can have a full-safe implementation, but it will
be a bit less performant, so I went with the unsafe approach.

This reverts commit da8ac73.

When we have two types with the same name, one without type parameters and the other with type parameters and a derive macro, we were before incorrectly suggesting to remove type parameters from the former, which ICEd because we were suggesting to remove nothing. We now gate against this.

The output is still not perfect. E0107 should explicitly detect this case and provide better context, but for now let's avoid the ICE.

This is ignored by LLVM, but is still incorrect.

The previous name is just an LLVMism, which conveys almost nothing about
what is actually meant by the function relative to the ABI.

In doing so, remove an already-addressed FIXME.

Add `Thread::{into_raw, from_raw}`

Public API:
```rust
#![unstable(feature = "thread_raw", issue = "97523")]

impl Thread {
    pub fn into_raw(self) -> *const ();
    pub unsafe fn from_raw(ptr: *const ()) -> Thread;
}
```

ACP: rust-lang/libs-team#200

Do not ICE with incorrect empty suggestion

When we have two types with the same name, one without type parameters and the other with type parameters and a derive macro, we were before incorrectly suggesting to remove type parameters from the former, which ICEd because we were suggesting to remove nothing. We now gate against this.

The output is still not perfect. E0107 should explicitly detect this case and provide better context, but for now let's avoid the ICE.

Fix rust-lang#108748.

Gate `repr(Rust)` correctly on non-ADT items

rust-lang#114201 added `repr(Rust)` but didn't add any attribute validation to it like `repr(C)` has, to only allow it on ADT items.

I consider this code to be nonsense, for example:
```
#[repr(Rust)]
fn foo() {}
```

Reminder that it's different from `extern "Rust"`, which *is* valid on function items. But also this now disallows `repr(Rust)` on modules, impls, traits, etc.

I'll crater it, if it looks bad then I'll add an FCW.

---

https://github.com/rust-lang/rust/labels/relnotes: Compatibility (minor breaking change).

…nieu

Win: Open dir for sync access in remove_dir_all

A small follow up to rust-lang#129800.

We should explicitly open directories for synchronous access. We ultimately use `GetFileInformationByHandleEx` to read directories which should paper over any issues caused by using async directory reads (or else return an error) but it's better to do the right thing in the first place. Note though that `delete` does not read or write any data so it's not necessary there.

…irect, r=bjorn3

Reduce confusion about `make_indirect_byval` by renaming it

As part of doing so, remove the incorrect handling of the wasm target's `make_indirect_byval` (i.e. using it at all).

…ake-2, r=Amanieu

Implement ACP 429: add `Lazy{Cell,Lock}::get[_mut]` and `force_mut`

Tracking issue for `lazy_get`: rust-lang#129333

Update the minimum external LLVM to 18

With this change, we'll have stable support for LLVM 18 and 19.
For reference, the previous increase to LLVM 17 was rust-lang#122649.

cc `@rust-lang/wg-llvm`
r? nikic

…cuviper

Clarify docs for std::fs::File::write

This PR fixes the doc comment for `std::fs::File::write` method.

…hs, r=compiler-errors

[Clippy] Swap `manual_retain` to use diagnostic items instead of paths

Part of rust-lang/rust-clippy#5393, just a chore.