tere: Integration test failures during Arch Linux build

With the help of @ProducerMatt’s nix environment in #100, I managed to track down the issue.

It’s the terminal_size_usize function, which calls this function from the crossterm library:

    let file = File::open("/dev/tty").map(|file| (FileDesc::new(file.into_raw_fd(), true)));
    let fd = if let Ok(file) = &file {
        file.raw_fd()
    } else {
        // Fallback to libc::STDOUT_FILENO if /dev/tty is missing
        STDOUT_FILENO
    };

    if wrap_with_result(unsafe { ioctl(fd, TIOCGWINSZ.into(), &mut size) }).is_ok() {
        return Ok(size.into());
    }

    Err(std::io::Error::last_os_error().into())

The first part sets fd either to the file descriptor of /dev/tty or stdout. Based on my testing in the Nix environment, File::open("/dev/tty") gives Err(Os { code: 6, kind: Uncategorized, message: "No such device or address" }), so it will fall back to stdout.

After this, it calls ioctl(fd, TIOCGWINSZ.into(), ...) where I believe the not very descriptive “File not found” error comes from.

So the problem is querying the “terminal size” inside the Nix environment / chroot. The next step would be to flood my search engine with queries like “nix environment tty” and similar.

Thanks for bearing with me on this!

Alright, cool! I’ll close this issue for now, let me know if anything else comes up. We can remove the ncurses dependency once the tput fallback is removed from crossterm, although I guess it’s not much of a problem.

Okay, I found a solution that doesn’t rely on faketty - seems like the integration test has a hidden dependency on tput (provided by ncurses package on nix, not sure about Arch), see https://github.com/mgunyho/tere/pull/100#issuecomment-2029618331. Could you @orhun try again without script/faketty and with tput available in the chroot?

I’m starting to think that it would be easiest to just skip the integration tests for now, this can be done by changing the test command to cargo test --bins.

That is actually another failure that has been fixed on the develop branch, and prevents the integration tests from running. Could you please retry on develop?

BTW, based on the readme of faketty, it does something very similar to script so that could be a solution as well.

@mgunyho yes in fact, Nix runs on other Linux and MacOS. I will submit a PR with a Nix dev environment and how to use it. 😃

I now fixed it in 8cadf56, this app_state::tests::test_case_sensitive_mode_change shouldn’t fail anymore. I’m not really sure what was going on there. I’m using a new computer and new distro compared to when I initially wrote it, so that might have something to do with it as well? But looking at the test, it did have a logic error. Hard to say. But thanks anyway for helping out!

Thanks for testing this! I’ll look into it. One thing to note that cat test-out-stderr.txt will not show the control codes, since the output is like <control code to switch to alternate screen><a bunch of drawing commands><switch back to regular screen>Cancelled. cat will print to the terminal, which will obey the control codes, so you won’t see what was on the alt screen. So I would need to see the actual txt file contents (you can see the control codes by opening it in vim for example). But I think I can debug this based on the print in the dummy test.