How can a shell script know it was invoked on the shell's CLI?

Question

I would like to implement a zsh script that behaves differently depending on whether it was invoked directly on the shell's CLI or not.

I thought at first that the script could do this by looking for i in the value of $-, but I was wrong.

In fact, when I run the script below from the command-line

#!/bin/zsh
printf -- '%s\n' "$-"

...the output I get does not include i¹.

Is there some other way for my shell script to figure out this information?

NB: Although I am working on a zsh script at the moment, I would also like to know what the answer would be for a bash script. If the answer depends on the OS, I'm primarily interested in Linux and Darwin.

---

^{1The script's output is 569X, if anyone cares to know.}

Answer 1

$- contains i when the shell itself is an interactive shell, when it issues prompts, lets you enter and edit command lines on the terminal, does job control, etc.

A script is typically interpreted by a non-interactive shell. If you invoke the script as ./the-script or zsh ./the-script, that zsh instance is not interactive. The only time your script will be interpreted by an interactive shell is when you tell your interactive shell to interpret the code in it with things like source ./the-script or eval "$(<the-script)".

Here, it seems to me you rather want to know if the script is invoked from within a terminal, like you want to tell whether a user can interact with it through a terminal.

That the [ / test command (which is builtin in zsh and bash and most other Bourne-like shells, but should also exist as a standalone command on POSIX systems) can tell you with its -t operator. That -t operator takes one¹ numerical argument which is a file descriptor.

[ -t 0 ] or test -t 0 returns true if the file descriptor 0 is open on a terminal device. File descriptor 0 is standard input, 1 is standard output, 2 is standard error, all other fds don't have any reserved special meaning.

So in your script, you can do:

if [ -t 0 ]; then
  echo "I (this script) am taking input from a terminal, so likely from a user"
fi
if [ -t 1 ]; then
  echo "My output goes to a terminal device, so will likely be seen by a user"
  [ -t 2 ] && echo errors also go there.
fi

If you want to check that the script is run from within a terminal session even if its input and/or output may be redirected to some other file than the terminal device, you can check whether there's a terminal attached to the script's session with something like:

if { true <> /dev/tty; } 2> /dev/null; then
  echo "there's a terminal device attached to my session"
fi

If you want to know what terminal device it is:

tty_on_stdin=$(tty)

tty_on_stderr=$(tty <&2)

{ tty_on_stdout=$(tty <&3); } 3>&1

(see this follow-up Q&A for the reason behind that little dance with fd 3).

controlling_tty=$(LC_ALL=C ps -o tty= -p "$$")
[ "$controlling_tty" = '?' ] || controlling_tty=/dev/$controlling_tty

---

^{¹ In earlier test / [ implementations, that argument was optional and [ -t ] was short for [ -t 1 ]. That is no longer allowed by POSIX as that conflicts with the [ "$var" ] use-case which is short for [ -n "$var" ]. Still, you'll find that ksh93 still checks whether stdout goes to a tty with [ -t ] though only when -t is literal; var=-t; [ "$var" ] returns true even if stdout is not on a terminal.}