Double Drop with #![deny(rust_2024_compatibility)]

[Clipi-12]

I had a custom Rc that was doubly freeing memory because Drop was being called more times than it should.
After a little bit of testing I think this is a small enough Minimal-Reproducible-Example:
playground

// ---------------------------------------- //
// If commented, it only calls Drop 2 times //
// If not commented,  it calls Drop 3 times //
// ---------------------------------------- //
#![deny(rust_2024_compatibility)]

use std::backtrace::Backtrace;

struct Guard();
impl Clone for Guard {
    fn clone(&self) -> Self {
        Self()
    }
}
impl Drop for Guard {
    fn drop(&mut self) {
        println!("Drop!");
        eprintln!("Drop at {}!", Backtrace::force_capture());
    }
}

fn main() {
    let _ = method_1(Guard());
}

fn method_1(g: Guard) -> Result<String, ()> {
    // ----------------------- //
    // This calls Drop 2 times //
    // ----------------------- //

    // let argument = &g.clone();
    // match method_2(argument) {
    //     Ok(str) => Ok(str),
    //     Err(err) => Err(err),
    // }

    // --- //

    // ----------------------- //
    // This calls Drop 3 times //
    // ----------------------- //

    match method_2(&g.clone()) {
        Ok(str) => Ok(str),
        Err(err) => Err(err),
    }
}

fn method_2(_: &Guard) -> Result<String, ()> {
    panic!("Method 2 panics!");
}

I expected to see this happen: Drop is called twice.

Instead, this happened: Drop is called 3 times.

Notes

• The return type doesn't need to be a Result<String, ()>
  It seems that it happens as long as it is an enum in which one arm has a non-Copy type.
  So it also happens if the return type is

enum CustomEnum {
    MyArm(Vec<u32>),
}

• The problems seems to happen regardless of the presence of the --release flag
• As seen in the backtrace below, the first drop is called at the end of the enum match, and the other two are called at the end of method_1

Meta

rustc --version --verbose:

rustc 1.85.0-nightly (a4cb3c831 2024-12-17)
binary: rustc
commit-hash: a4cb3c831823d9baa56c3d90514b75b2660116fa
commit-date: 2024-12-17
host: x86_64-pc-windows-msvc
release: 1.85.0-nightly
LLVM version: 19.1.5

Backtrace

Reduced stderr from the playground:

     Running `target/debug/playground`
thread 'main' panicked at src/main.rs:51:5:
Method 2 panics!
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Drop at    0: <playground::Guard as core::ops::drop::Drop>::drop
             at ./src/main.rs:18:33
   1: core::ptr::drop_in_place<playground::Guard>
             at ./.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ptr/mod.rs:522:1
   2: playground::method_1
             at ./src/main.rs:46:5
   3: [...]

Drop at    0: <playground::Guard as core::ops::drop::Drop>::drop
             at ./src/main.rs:18:33
   1: core::ptr::drop_in_place<playground::Guard>
             at ./.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ptr/mod.rs:522:1
   2: playground::method_1
             at ./src/main.rs:47:1
   3: [...]

Drop at    0: <playground::Guard as core::ops::drop::Drop>::drop
             at ./src/main.rs:18:33
   1: core::ptr::drop_in_place<playground::Guard>
             at ./.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ptr/mod.rs:522:1
   2: playground::method_1
             at ./src/main.rs:47:1
   3: [...]

[lukas-code]

caused by #131326 cc @dingxiangfei2009

[compiler-errors]

I've begun to take a look at this. It has to do with the fact that we weren't handling the "fake" drop terminator we insert for lints as a real drop on unwind cleanups.