From mboxrd@z Thu Jan  1 00:00:00 1970
From: Claude Code Review Bot <claude-review@example.com>
To: dri-devel-reviews@example.com
Subject: Claude review: rust: types: add `ForLt` trait for higher-ranked lifetime support
Date: Tue, 26 May 2026 06:47:04 +1000
Message-ID: <review-patch22-20260525202921.124698-23-dakr@kernel.org>
In-Reply-To: <20260525202921.124698-23-dakr@kernel.org>
References: <20260525202921.124698-1-dakr@kernel.org> <20260525202921.124698-23-dakr@kernel.org>
X-Mailer: Claude Code Patch Reviewer
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
MIME-Version: 1.0

Patch Review

The ForLt infrastructure by Gary Guo. The trait:

```rust
pub unsafe trait ForLt {
    type Of<'a>: 'a;
    fn cast_ref<'r, 'short: 'r, 'long: 'short>(long: &'r Self::Of<'long>) -> &'r Self::Of<'short> {
        unsafe { core::mem::transmute(long) }
    }
}
```

The proc macro (`ForLt!`) generates covariance proofs and well-formedness checks. The `visit_macro` handler conservatively assumes macro invocations may contain lifetimes:

```rust
fn visit_macro(&mut self, _: &syn::Macro) {
    self.1 = true;
}
```

The WF proof works around a Rust compiler soundness bug (rust-lang/rust#152489) by using a struct rather than a function to avoid implied WF bounds. The implementation is thorough -- the covariance prover handles references, pointers, tuples, arrays, slices, and falls back to compiler-checked proofs for complex types. Licensed Apache-2.0 OR MIT for future pin-init reuse.

One note: the `UnsafeForLtImpl` type is documented as "unsafe-to-refer-to" but there's no actual mechanism preventing misuse beyond doc comments. This is a known limitation of the pattern.

---
Generated by Claude Code Patch Reviewer