The "correct" way to start another program is to use shell.run. However, in some (admittedly rare) cases, its behaviour is non ideal.

The problem is that shell.run concatenates all its arguments together before parsing them as a command line. This means that shell.run("cd foo") is identical to shell.run("cd", "foo"). However, this is also true when later arguments have spaces in them. For instance, shell.run("cd", "foo bar") is equivalent to shell.run("cd foo bar").

This is problematic when you look at the monitor or shell programs - monitor right edit "my code.lua" actually ends up executing edit my code.lua!

While this can be avoided by escaping the inputs (using ("%q"):format(x)), I think the cleanest thing here is to expose the internal runfunction asshell.execute` (probably with some additional argument validation). The monitor and shell programs call then that instead (though with some fallback for custom shells which don't support it yet).