read_sync fails under LLDB debugging in Linux

[joshhansen]

In addition to the current woes of read_sync (e.g. #195) it seems we have an additional problem. When debugging using LLDB in Linux, read_sync encounters an error that crashes the program. Here's the code:

extern crate crossterm;

use std::{
    io::{Write,stdout},
};

use crossterm::{
    InputEvent,
    KeyEvent,
    RawScreen,
    input,
};

fn main() {
    let _raw = RawScreen::into_raw_mode().unwrap();
    let input = input();
    for x in input.read_sync() {
        if let InputEvent::Keyboard(KeyEvent::Ctrl('c') )= x {
            return;
        }
        print!("\rRead: {:?}", x);
        stdout().flush().unwrap();
    }
}

When run directly, it reads input events and prints them out, exiting if CTRL-C is pressed.

However, when running in VSCode's CodeLLDB debugger extension, I get a "No such device or address error". And when running under lldb on the command line, I get a SIGTTIN that stops the thread, halting the program.

The program runs fine under gdb.

In practice, this is crashing the input thread in my program whenever I attempt to debug using the usual tools. It could be that this is some LLDB error, or I am misusing LLDB somehow, but the behavior is very surprising so I thought I'd report it here.

[vadimcn]

I would hypothesize that this happens because crossterm tries to obtain TTY handle by directly opening /dev/tty. Before LLDB launches the debuggee process, it, for various reasons, detaches from the controlling terminal, so opening /dev/tty will fail. However, in most cases, debuggee's stdin will still be connected to a TTY.
Maybe crossterm could first check if stdin is a TTY?

[TimonPost]

Thanks for reporting. Input layer is going through a couple of rewrites, the first is here: https://github.com/crossterm-rs/crossterm-input/pulls, we will have to see if the issue is resolved with this PR. Otherwise, it will be fixed in a other one.

[joshhansen]

@TimonPost Let me know when that refactor lands and I'll check the status of this issue

[zrzka]

@joshhansen I tested your code with the PR (branch zrzka/unix-async-events). I can confirm that:

• zrzka/unix-async-events fixes the issue with SIGTTIN
• master is broken and exhibits SIGTTIN issue

All tests were made with lldb & gdb and with the CodeLLDB extension.

Test code

I just replaced Ctrl-C with Esc, because Ctrl-C is captured by lldb to interrupt the program.

examples/lldb.rs:

use std::io::{stdout, Write};

use crossterm_input::{input, InputEvent, KeyEvent, RawScreen};

fn main() {
    let _raw = RawScreen::into_raw_mode().unwrap();
    let input = input();
    for x in input.read_sync() {
        match x {
            InputEvent::Keyboard(KeyEvent::Esc) => break,
            x => print!("\rRead: {:?}\n", x),
        };

        stdout().flush().unwrap();
    }
}

Platforms

This process exited with status 0/normally is on these screenshots, because I hit the Esc key.

macOS

Ubuntu & lldb

Ubuntu & gdb

Ubuntu & VSCode & CodeLLDB