getopts

The original getopt had several problems: it could not handle whitespace or shell metacharacters in arguments, and there was no ability to disable the output of error messages.[1]

getopts was first introduced in 1986 in the Bourne shell shipped with Unix SVR3. It uses the shell's own variables to track the position of current and argument positions, OPTIND and OPTARG, and returns the option name in a shell variable.[2] Earlier versions of the Bourne shell did not have getopts.

In 1995, getopts was included in the Single UNIX Specification version 1 / X/Open Portability Guidelines Issue 4.[3] As a result, getopts is now available in shells including the Bourne shell, KornShell, Almquist shell, Bash and Zsh.[4]

The getopts command has also been ported to the IBM i operating system.[5]

The modern usage of getopt was partially revived mainly due to an enhanced implementation in util-linux. This version, based on the BSD getopt, not only fixed the two complaints around the old getopt, but also introduced the capability for parsing GNU-style long options and optional arguments for options, features that getopts lacks.[6] The various BSD distributions, however, stuck to the old implementation.[1]

The usage synopsis of getopt and getopts is similar to its C sibling:

getopt optstring [parameters]
getopts optstring varname [parameters]

• The optstring part has the same format as the C sibling.
• The parameters part simply accepts whatever one wants getopt to parse. A common value is all the parameters, "$@" in POSIX shell.
  • This value exists in getopts but is rarely used, since it can just access the shell's parameters. It is useful with resetting the parser, however.
• The varname part of getopts names a shell variable to store the option parsed into.

The way one uses the commands however varies a lot:

• getopt simply returns a flat string containing whitespace-separated tokens representing the "normalized" argument. One then uses a while-loop to parse it naively.[1]
• getopts is meant to be repeatedly called like the C getopt. When it hits the end of arguments, it returns 1 (shell false).[3]

Suppose we are building a Wikipedia downloader in bash that takes three options and zero extra arguments:

wpdown -a article name -l [language] -v

When possible, we allow the following long arguments:

-a   --article
-l   --language, --lang
-v   --verbose

For clarity, no help text is included, and we assume there is a program that downloads any webpage. In addition, all programs are of the form:

#!/bin/bash
VERBOSE=0
ARTICLE=''
LANG=en

# [EXAMPLE HERE]

if ((VERBOSE > 2)); then
  printf '%s\n' 'Non-option arguments:'
  printf '%q ' "${remaining[@]]}"
fi

if ((VERBOSE > 1)); then
  printf 'Downloading %s:%s\n' "$LANG" "$ARTICLE"
fi

if [[ ! $ARTICLE ]]; then
  printf '%s\n' "No articles!">&2
  exit 1
fi

save_webpage "https://${LANG}.wikipedia.org/wiki/${ARTICLE}"

# parse everything; if it fails we bail
args=`getopt 'a:l:v' $*` || exit
# now we have the sanitized args... replace the original with it
set -- $args

while true; do
    case $1 in
      (-v)   ((VERBOSE++));  shift;;
      (-a)   ARTICLE=$2; shift 2;;
      (-l)   LANG=$2; shift 2;;
      (--)   shift; break;;
      (*)    exit 1;;           # error
    esac
done

remaining=("$@")

#!/bin/sh
while getopts ':a:l:v' opt; do
    case $opt in
      (v)   ((VERBOSE++));;
      (a)   ARTICLE=$OPTARG;;
      (l)   LANG=$OPTARG;;
      (:)   # "optional arguments" (missing option-argument handling)
            case $OPTARG in
              (a) exit 1;; # error, according to our syntax
              (l) :;;      # acceptable but does nothing
            esac;;
    esac
done

shift "$OPTIND"
# remaining is "$@"

# We use "$@" instead of $* to preserve argument-boundary information
ARGS=$(getopt -o 'a:l::v' --long 'article:,language::,lang::,verbose' -- "$@") || exit
eval "set -- $ARGS"

while true; do
    case $1 in
      (-v|--verbose)
            ((VERBOSE++)); shift;;
      (-a|--article)
            ARTICLE=$2; shift 2;;
      (-l|--lang|--language)
            # handle optional: getopt normalizes it into an empty string
            if [ -n "$2" ]; then
              LANG=$2; shift;
            fi
            shift;;
      (--)  shift; break;;
      (*)   exit 1;;           # error
    esac
done

remaining=("$@")

• List of Unix commands

• getopts: process command line arguments – Commands & Utilities Reference, The Single UNIX Specification, Issue 7 from The Open Group