Do we need to comment out this to reproduce?

As documented, remove_dir_all will fail if the path is not a directory, so our fs::rename call should fail even without commenting this out I guess.

(And we also capture CARGO_LOG in the stderr assertion, so it should be fairly robust to ensure we went the code path)

use std::fs;

fn main() {
    fs::create_dir_all("./tmp").unwrap();
    fs::create_dir_all("./tmp/src").unwrap();
    fs::write("./tmp/src/src_file", "src_file").unwrap();

    // Will rename successfully
    fs::write("./tmp/dst", "blocking file").unwrap();
    fs::rename("./tmp/src", "./tmp/dst").unwrap();
}

I wrote a smallest test. fs::rename will make the src dir override the dst file. So I change the dst file to a dir in pre commit.

I wrote a smallest test. fs::rename will make the src dir override the dst file. So I change the dst file to a dir in pre commit.

Interesting 🤔. On Linux it failed: https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=0b8b476b80e4bc4ac6839433e39fd840

Yes, I just thought of this as well and tested it. It failed in the Linux environment. (In the windows, it will rename successfully).
What ever, it seems that the unit test is running on the Linux docker. I would modify the code.

https://github.com/rust-lang/rust/blob/51f5892019f8fb07864647d46c4eb577d3b0719f/library/std/src/sys/fs/windows.rs?plain=1#L1272

Found it. The Windows implementation of fs::rename uses MOVEFILE_REPLACE_EXISTING. Probably that was the reason it succeeded?
Anyway, if we cannot make it passed on Windows, we can put this

#[cargo_test(ignore_windows = "annoying to create a repro on Windows; see rust-lang/cargo#16214")]