cat -n doesn't handle CRLF newline delimiters correctly

[pfmoore]

This is with the latest version from github, built locally via cargo build --release --features windows

❯ coreutils printf "hello\r\nworld\r\nthere\r\n" | coreutils cat -n
     1  hello     2
world
there

Weirdly, looking at the output via xxd, the CRLF sequence is switched to LFCR...

❯ target\release\coreutils printf "hello\r\nworld\r\nthere\r\n" | target\release\coreutils cat -n | xxd
00000000: 2020 2020 2031 0968 656c 6c6f 2020 2020       1.hello
00000010: 2032 090a 0d77 6f72 6c64 0a0d 7468 6572   2...world..ther
00000020: 650a 0d                                  e..
❯ target\release\coreutils printf "hello\r\nworld\r\nthere\r\n" | xxd
00000000: 6865 6c6c 6f0d 0a77 6f72 6c64 0d0a 7468  hello..world..th
00000010: 6572 650d 0a                             ere..

[OshinoShinobu-Chan]

I'm trying to fix this issue. And I find out that it is caused by src/cat.rs:write_new_line in Line537. The code is as follows:

fn write_new_line<W: Write>(
    writer: &mut W,
    options: &OutputOptions,
    state: &mut OutputState,
    is_interactive: bool,
) -> CatResult<()> {
    if state.skipped_carriage_return && options.show_ends {
        writer.write_all(b"^M")?;
        state.skipped_carriage_return = false;
    }
    if !state.at_line_start || !options.squeeze_blank || !state.one_blank_kept {
        state.one_blank_kept = true;
        if state.at_line_start && options.number == NumberingMode::All {
            write!(writer, "{0:6}\t", state.line_number)?;
            state.line_number += 1;
        }
        writer.write_all(options.end_of_line().as_bytes())?;
        if is_interactive {
            writer.flush()?;
        }
    }
    Ok(())
}

---

I think there's two problems in this function:

1. If cat is not in the show_end mode, the \r will be ingnored in function. This will happen every time it meet \r\n.
2. Suppose we correctly print the \r/^M, the rest of the code will treat the following \n as a single line itself. Here's an example:

./target/debug/coreutils cat -nE ../crlf.txt
     1  Hello^M     2   $
     3  world^M     4   $
     5  there$

---

I think these problems can be fixed by modiying the branch statement in function write_new_line as follows；

    if state.skipped_carriage_return {
        if options.show_ends {
            writer.write_all(b"^M")?;
        } else {
            writer.write_all(b"\r")?;
        }
        state.skipped_carriage_return = false;

        write_end_of_line(writer, options.end_of_line().as_bytes(), is_interactive)?;
        return Ok(());
    }

---

I'll be glad to submit a PR if you agree to my proposal.

Besides, I think tests for this kind of case should be added. But I have no idea where should I add the new test. I'm new in here so please give some guidance, thanks.

[sylvestre]

Sure, don't hesitate to submit the PR

[cakebaker]

Besides, I think tests for this kind of case should be added. But I have no idea where should I add the new test.

You can add the test to /tests/by-util/test_cat.rs.