Bash – Indexing and modifying Bash parameter array $@

bashparameter

Is it possible to refer to indexes in $@? I can't find any reference to use like the following anywhere in GrayCat's wiki, and the Advanced Scripting Guide and others assign this to a different variable before modifying that instead.

$ echo ${@[0]}
-bash: ${@[0]}: bad substitution

The goal is DRY: The first argument is used for one thing, and the rest for something else, and I'd like to avoid duplicating either the code to normalize, the $@ array, or to create a separate function for this (although at this point it's probably the easiest way out).

Clarification: The object was to modify the values of the variable-length $@ to make the code easier to debug. The current version is a bit too hacky for my liking, although it works even for bizarre paths like

$'--$`\! *@ \a\b\e\E\f\r\t\v\\\"\' \n'

Update: Looks like this isn't possible. The code now uses both code and data duplication, but at least it works:

path_common()
{
    # Get the deepest common path.
    local common_path="$(echo -n "${1:-}x" | tr -s '/')"
    common_path="${common_path%x}"
    shift # $1 is obviously part of $1
    local path

    while [ -n "${1+defined}" ]
    do
        path="$(echo -n "${1}x" | tr -s '/')"
        path="${path%x}"
        if [[ "${path%/}/" = "${common_path%/}/"* ]]
        then
            shift
        else
            new_common_path="${common_path%/*}"
            [ "$new_common_path" = "$common_path" ] && return 1 # Dead end
            common_path="$new_common_path"
        fi
    done
    printf %s "$common_path"
}

Bounty goes to anyone who can get rid of the duplication of code to collapse duplicate slashes or duplication of data to hold $1 and the other parameters, or both, while keeping the code a reasonable size and succeeding all the unit tests:

test "$(path_common /a/b/c/d /a/b/e/f; echo x)" = /a/bx
test "$(path_common /long/names/foo /long/names/bar; echo x)" = /long/namesx
test "$(path_common / /a/b/c; echo x)" = /x
test "$(path_common a/b/c/d a/b/e/f ; echo x)" = a/bx
test "$(path_common ./a/b/c/d ./a/b/e/f; echo x)" = ./a/bx
test "$(path_common $'\n/\n/\n' $'\n/\n'; echo x)" = $'\n/\n'x
test "$(path_common --/-- --; echo x)" = '--x'
test "$(path_common '' ''; echo x)" = x
test "$(path_common /foo/bar ''; echo x)" = x
test "$(path_common /foo /fo; echo x)" = x
test "$(path_common $'--$`\! *@ \a\b\e\E\f\r\t\v\\\"\' \n' $'--$`\! *@ \a\b\e\E\f\r\t\v\\\"\' \n'; echo x)" = $'--$`\! *@ \a\b\e\E\f\r\t\v\\\"\' \n'x
test "$(path_common /foo/bar //foo//bar//baz; echo x)" = /foo/barx
test "$(path_common foo foo; echo x)" = foox
test "$(path_common /fo /foo; echo x)" = x

Best Answer

POSIX

To normalize the slashes in all the parameters, I'll use the rotating argument trick: shift $1 off, transform it and put the result at the end of the parameter list. If you do that as many time as there are parameters, you've transformed all the parameters, and you've got them back in order.

For the second part of the code, I changed your logic to be less confusing: the outer loop iterates over the parameters, and the inner loop iterates over path components. for x; do … done iterates over the positional parameters, it's a convenient idiom. I use a POSIX-compliant way of matching a string against a pattern: the case construct.

Tested with dash 0.5.5.1, pdksh 5.2.14, bash 3.2.39, bash 4.1.5, ksh 93s+, zsh 4.3.10.

Side note: there seems to be a bug in bash 4.1.5 (not in 3.2): if the case pattern is "${common_path%/}"/*, one of the tests fails.

posix_path_common () {
  for tmp; do
    tmp=$(printf %s. "$1" | tr -s "/")
    set -- "$@" "${tmp%.}"
    shift
  done
  common_path=$1; shift
  for tmp; do
    while case ${tmp%/}/ in "${common_path%/}/"*) false;; esac; do
      new_common_path=${common_path%/*}
      if [ "$new_common_path" = "$common_path" ]; then return 1; fi
      common_path=$new_common_path
    done
  done
  printf %s "$common_path"
}

bash, ksh

If you're in bash (or ksh), you can use arrays — I don't understand why you seem to be restricting yourself to the positional parameters. Here's a version that uses an array. I have to admit it's not particularly clearer than the POSIX version, but it does avoid the initial n^2 shuffling.

For the slash normalization part, I use the ksh93 construct ${foo//PATTERN/REPLACEMENT} construct to replace all occurrences of PATTERN in $foo by REPLACEMENT. The pattern is +(\/) to match one or more slash; under bash, shopt -s extglob must be in effect (equivalently, start bash with bash -O extglob). The construct set ${!a[@]} sets the positional parameters to the list of subscripts of the array a. This provides a convenient way to iterate over the elements of the array.

For the second part, I the same loop logic as the POSIX version. This time, I can use [[ … ]] since all the shells targeted here support it.

Tested with bash 3.2.39, bash 4.1.5, ksh 93s+.

array_path_common () {
  typeset a i tmp common_path new_common_path
  a=("$@")
  set ${!a[@]}
  for i; do
    a[$i]=${a[$i]//+(\/)//}
  done
  common_path=${a[$1]}; shift
  for tmp; do
    tmp=${a[$tmp]}
    while [[ "${tmp%/}/" != "${common_path%/}/"* ]]; do
      new_common_path="${common_path%/*}"
      if [[ $new_common_path = $common_path ]]; then return 1; fi
      common_path="$new_common_path"
    done
  done
  printf %s "$common_path"
}

zsh

Sadly, zsh lacks the ${!array[@]} feature to execute the ksh93 version as-is. Fortunately, zsh has two features that make the first part a breeze. You can index the positional parameters as if they were the @ array, so there's no need to use an intermediate array. And zsh has an array iteration construct: "${(@)array//PATTERN/REPLACEMENT}" performs the pattern replacement on each array element in turn and evaluates to the array of results (confusingly, you do need the double quotes even though the result is multiple words; this is a generalization of "$@"). The second part is essentially unchanged.

zsh_path_common () {
  setopt local_options extended_glob
  local tmp common_path new_common_path
  set -- "${(@)@//\/##//}"
  common_path=$1; shift
  for tmp; do
    while [[ "${tmp%/}/" != "${common_path%/}/"* ]]; do
      new_common_path="${common_path%/*}"
      if [[ $new_common_path = $common_path ]]; then return 1; fi
      common_path="$new_common_path"
    done
  done
  printf %s "$common_path"
}

Test cases

My solutions are minimally tested and commented. I've changed the syntax of your test cases to parse under shells that don't have $'…' and report failures in a more convenient way.

do_test () {
  if test "$@"; then echo 0; else echo $? "$@"; failed=$(($failed+1)); fi
}

run_tests () {
  function_to_test=$1; shift
  failed=0
  do_test "$($function_to_test /a/b/c/d /a/b/e/f; echo x)" = /a/bx
  do_test "$($function_to_test /long/names/foo /long/names/bar; echo x)" = /long/namesx
  do_test "$($function_to_test / /a/b/c; echo x)" = /x
  do_test "$($function_to_test a/b/c/d a/b/e/f ; echo x)" = a/bx
  do_test "$($function_to_test ./a/b/c/d ./a/b/e/f; echo x)" = ./a/bx
  do_test "$($function_to_test '
/
/
' '
/
'; echo x)" = '
/
'x
  do_test "$($function_to_test --/-- --; echo x)" = '--x'
  do_test "$($function_to_test '' ''; echo x)" = x
  do_test "$($function_to_test /foo/bar ''; echo x)" = x
  do_test "$($function_to_test /foo /fo; echo x)" = x
  do_test "$($function_to_test '--$`\! *@ \a\b\e\E\f\r\t\v\\\"'\'' 
' '--$`\! *@ \a\b\e\E\f\r\t\v\\\"'\'' 
'; echo x)" = '--$`\! *@ \a\b\e\E\f\r\t\v\\\"'\'' 
'x
  do_test "$($function_to_test /foo/bar //foo//bar//baz; echo x)" = /foo/barx
  do_test "$($function_to_test foo foo; echo x)" = foox
  do_test "$($function_to_test /fo /foo; echo x)" = x
  if [ $failed -ne 0 ]; then echo $failed failures; return 1; fi
}
Related Question