I'm proposing to introduce a new function akin to split that should either

• Split the string in the exact the same way as the real shell would do.
• Or, return an error.

That is, things like && should not be treated as literal strings because the shell wouldn't do so.

So, it should pass the following unit test specification:

// These commands are guaranteed to be interpreted in the same way as the real shell would do.
assert_eq!(split2("foo bar baz"), Ok(vec!["foo", "bar", "baz"]));
assert_eq!(split2("foo 'bar' baz"), Ok(vec!["foo", "bar", "baz"]));
assert_eq!(split2("foo '$bar' baz"), Ok(vec!["foo", "$bar", "baz"]));
assert_eq!(split2("foo '~/bar' baz"), Ok(vec!["foo", "~/bar", "baz"]));
assert_eq!(split2(r#"foo bar "&&" baz"#), Ok(vec!["foo", "bar", "&&", "baz"]));

// The following commands are complicated. We are unable to expand them in the exact same way as the shell would do.
// So, return an error instead. It is advised to run these commands in the actual shell (`sh -c`).
assert_eq!(split2("foo bar; baz"), Err(ParseError)); // more than one command
assert_eq!(split2("foo bar && baz"), Err(ParseError)); // more than one command
assert_eq!(split2("foo bar &"), Err(ParseError)); // background
assert_eq!(split2(r#"foo "$bar" baz"#), Err(ParseError)); // variable expanding
assert_eq!(split2("foo ~/bar baz"), Err(ParseError)); // tilde expanding
assert_eq!(split2("foo bar=~/baz"), Err(ParseError)); // tilde expanding (in bash, but not in the posix shell)
assert_eq!(split2("export foo=bar"), Err(ParseError)); // builtin used

Use-case: with this function, we would be able to implement an optimized shell_exec function, bypassing the real shell in simple cases.

fn shell_exec(command: &str) {
	let c = if let Some(args) = shell_words::split2(command) {
		// Fast path: no need to start the real shell.
		Command::new(&args[0]).args(&args[1..])
	} else {
		// Slow path: we need to start the shell for the complicated command.
		Command::new("sh").arg("-c").arg(command)
	};
	c.spawn().unwrap().wait().unwrap();
}

split does not return the place at which an error occurred. This makes it impossible for user interfaces to indicate the error location.

Please add a second function split2 that returns the UTF-8 index of the opening quote.