Piping to "head -n1" generates a panic (not handling SIGPIPE?)

[Horgix]

The second part of the "smoke test" step in the CircleCI pipeline was this command :

command: ./xebia-cli --help | head -n1 | grep xebia-cli`

It essentially just ran the CLI, got the first line which is supposed to look like xebia-cli X.Y thanks to clap and the cli.yml. It was recently disable because of a flaw that will take a bit of time to figure out and to fix.

Indeed, piping the output of xebia-cli to head -n1 makes it panic:

thread 'main' panicked at 'Error writing Error to stdout: Os { code: 32, kind: BrokenPipe, message: "Broken pipe" }', src/libcore/result.rs:997:5

It's not related in any way to our code and is totally reproducable with the following snippet only:

extern crate log;
extern crate simplelog;
use simplelog::*;

extern crate clap;
use clap::App;

fn main() {
    // Initialize logging
    let term_logger = match TermLogger::new(LevelFilter::Debug, Config::default()) {
        Some(tl) => tl,
        None => {
            println!("Failed to create TermLogger; exiting.");
            std::process::exit(1);
        }
    };
    CombinedLogger::init(vec![term_logger]).expect("Failed to initialize logger");

    // Reproduce bug
    log::debug!("Here is a crash");
    let _matches = App::new("simplelog-broken-pipe-demo").get_matches();
}

Building with and running it followed by a | head -n1 will result in the same crash.

Surprisingly, piping it to grep or a simple head without -n is not crashing the application.

This definitely has to be investigated, may it be only to discover WTF is going on here. It may be related to SIGPIPE being sent to the binary when a pipe is stopped brutally, which is not handled by Rust.

Some references of a first look at this issue:

• Stdout does panic in pipe rust-lang/rust#35108
• Spurious "broken pipe" error messages, when used in typical UNIX shell pipelines rust-lang/rust#46016

[Horgix]

The CircleCI step using | head -n has been disabled in c0eeb96