43

I am trying to use arrays in Bourne shell (/bin/sh). I found that the way to initialize array elements is:

arr=(1 2 3)

But it is encountering an error:

syntax error at line 8: `arr=' unexpected

Now the post where I found this syntax says it is for bash, but I could not find any separate syntax for Bourne shell. Does the syntax stand the same for /bin/sh as well?

2

4 Answers 4

66

/bin/sh is hardly ever a Bourne shell on any systems nowadays (even Solaris which was one of the last major system to include it has now switched to a POSIX sh for its /bin/sh in Solaris 11). /bin/sh was the Thompson shell in the early 70s. The Bourne shell replaced it in Unix V7 in 1979.

/bin/sh has been the Bourne shell for many years thereafter (or the Almquist shell, a free reimplementation on BSDs).

Nowadays, /bin/sh is more commonly an interpreter or another for the POSIX sh language which is itself based on a subset of the language of ksh88 (and a superset of the Bourne shell language with some incompatibilities).

The Bourne shell or the POSIX sh language specification don't support arrays. Or rather they have only one array: the positional parameters ($1, $2, $@, so one array per function as well).

ksh88 did have arrays which you set with set -A, but that didn't get specified in the POSIX sh as the syntax is awkward and not very usable.

Other shells with array/lists variables include: csh/tcsh, rc, es, bash (which mostly copied the ksh syntax the ksh93 way), yash, zsh, fish each with a different syntax (rc the shell of the once to-be successor of Unix, fish and zsh being the most consistent ones)...

In standard sh (also works in modern versions of the Bourne shell):

set '1st element' 2 3 # setting the array

set -- "$@" more # adding elements to the end of the array

shift 2 # removing elements (here 2) from the beginning of the array

printf '<%s>\n' "$@" # passing all the elements of the $@ array 
                     # as arguments to a command

for i do # looping over the  elements of the $@ array ($1, $2...)
  printf 'Looping over "%s"\n' "$i"
done

printf '%s\n' "$1" # accessing individual element of the array.
                   # up to the 9th only with the Bourne shell though
                   # (only the Bourne shell), and note that you need
                   # the braces (as in "${10}") past the 9th in other
                   # shells (except zsh, when not in sh emulation and
                   # most ash-based shells).

printf '%s\n' "$# elements in the array"

printf '%s\n' "$*" # join the elements of the array with the 
                   # first character (byte in some implementations)
                   # of $IFS (not in the Bourne shell where it's on
                   # space instead regardless of the value of $IFS)

(note that in the Bourne shell and ksh88, $IFS must contain the space character for "$@" to work properly (a bug), and in the Bourne shell, you can't access elements above $9 (${10} won't work, you can still do shift 1; echo "$9" or loop over them)).

3
  • 1
    It may be worth noting that positional parameters differ from bash arrays in some key features. For example, they do not support sparse arrays, and since sh doesn't have slicing parameter expansion, you can't access sublists like "${@:2:4}". To be sure, I see the similarities, but I don't regard positional parameters as an array per se.
    – kojiro
    Jan 22, 2016 at 16:37
  • @kojiro, to some extent, I'd say it's the contrary, "$@" acts like an array (like the arrays of csh, rc, zsh, fish, yash...), it's more the Korn/bash "arrays" that are not really arrays, but some form of associative arrays with keys limited to positive integers (they also have indices starting at 0 instead of 1 like in all other shells with arrays and "$@"). Shells that have support for slicing can slice $@ just the same (with ksh93/bash awkwardly adding $0 to the positional parameters when you slice "$@"). Oct 23, 2017 at 6:57
  • 1
    /bin/sh continues to give me Syntax error: "(" unexpected for arrays Jul 19, 2021 at 22:23
10

As the others have said, the Bourne Shell doesn't have true arrays.

However, depending on what you need to do, delimited strings should suffice:

sentence="I don't need arrays because I can use delimited strings"
for word in $sentence
do
  printf '%s\n' "$word"
done

If the typical delimiters (space, tab, and newline) don't suffice, you can set IFS to whatever delimiter you want before the loop.

And if you need to build the array programmatically, you can just build up a delimited string.

1
  • 1
    Unless you do want it (unlikely), you'll probably also want to disable globbing which is another effect of leaving variables unquoted like that (the split+glob operator). May 13, 2015 at 14:55
5

There are no arrays in plain Bourne shell. You can use the following way to create an array and traverse it:

#!/bin/sh
# ARRAY.sh: example usage of arrays in Bourne Shell

array_traverse()
{
    for i in $(seq 1 $2)
    do
    current_value=$1$i
    echo $(eval echo \$$current_value)
    done
    return 1
}

ARRAY_1=one
ARRAY_2=two
ARRAY_3=333
array_traverse ARRAY_ 3

No matter what way to use arrays in sh would you pick it will always be cumbersome. Consider using a different language such as Python or Perl if you can unless you are stuck with a very limited platform or want to learn something.

7
  • Thanks for the response...!! Actually I am indeed trying to learn things in shell script...otherwise implementing array in Python is really a piece of cake. This was a big lesson that there exists some scripting language which does not support array :) One thing, the code you posted is giving a error - "syntax error at line 6: `$' unexpected" ... I am a little busy now, I would get it solved...plz dont bother.
    – SubhasisM
    Jun 17, 2014 at 10:51
  • @NoobGeek, the Bourne shell doesn't have the $(...) syntax. So you must indeed have the Bourne shell. Are you on Solaris 10 or before? Chances are you won't have a seq either. On Solaris 10 and earlier, you want /usr/xpg4/bin/sh to have a standard sh instead of a Bourne shell. Using seq that way is not very good either. Jun 17, 2014 at 11:42
  • POSIX states that $ and ` are equivalent in command substitution: link. And why is using of seq that way not good? Jun 17, 2014 at 12:17
  • 3
    Yes in POSIX shells, you should prefer $(...) over `, but the OP's /bin/sh is probably a Bourne shell, not a POSIX shell. Beside seq not being a standard command, doing $(seq 100) means storing the whole output in memory, and that means it depends on the current value of $IFS containing newline and not containing digits. Best to use i=1; while [ "$i" -le 100 ]; ...; i=$(($i + 1)); done (though that wouldn't work in the Bourne shell either). Jun 17, 2014 at 13:51
  • 1
    @Daenyth I would say quite the contrary: learning bashisms first, and then portable /bin/sh syntax later, tends to make people think it's okay to use the wrong #!/bin/sh shebang, and then breaks their scripts when other people try to use them. You'd be well advised to not post this kind of flamebait. :) Oct 30, 2015 at 13:39
0

A way to simulate arrays in dash (it can be adapted for any number of dimensions of an array): (Please note that the use of the seq command requires that IFS is set to ' ' (SPACE = the default value). You can use while ... do ... or do ... while ... loops instead to avoid this (i kept seq in the scope of a better illustration of what the code does).)

#!/bin/sh

## The following functions implement vectors (arrays) operations in dash:
## Definition of a vector <v>:
##      v_0 - variable that stores the number of elements of the vector
##      v_1..v_n, where n=v_0 - variables that store the values of the vector elements

VectorAddElementNext () {
# Vector Add Element Next
# Adds the string contained in variable $2 in the next element position (vector length + 1) in vector $1

    local elem_value
    local vector_length
    local elem_name

    eval elem_value=\"\$$2\"
    eval vector_length=\$$1\_0
    if [ -z "$vector_length" ]; then
        vector_length=$((0))
    fi

    vector_length=$(( vector_length + 1 ))
    elem_name=$1_$vector_length

    eval $elem_name=\"\$elem_value\"
    eval $1_0=$vector_length
}

VectorAddElementDVNext () {
# Vector Add Element Direct Value Next
# Adds the string $2 in the next element position (vector length + 1) in vector $1

    local elem_value
    local vector_length
    local elem_name

    eval elem_value="$2"
    eval vector_length=\$$1\_0
    if [ -z "$vector_length" ]; then
        vector_length=$((0))
    fi

    vector_length=$(( vector_length + 1 ))
    elem_name=$1_$vector_length

    eval $elem_name=\"\$elem_value\"
    eval $1_0=$vector_length
}

VectorAddElement () {
# Vector Add Element
# Adds the string contained in the variable $3 in the position contained in $2 (variable or direct value) in the vector $1

    local elem_value
    local elem_position
    local vector_length
    local elem_name

    eval elem_value=\"\$$3\"
    elem_position=$(($2))
    eval vector_length=\$$1\_0
    if [ -z "$vector_length" ]; then
        vector_length=$((0))
    fi

    if [ $elem_position -ge $vector_length ]; then
        vector_length=$elem_position
    fi

    elem_name=$1_$elem_position

    eval $elem_name=\"\$elem_value\"
    if [ ! $elem_position -eq 0 ]; then
        eval $1_0=$vector_length
    fi
}

VectorAddElementDV () {
# Vector Add Element
# Adds the string $3 in the position $2 (variable or direct value) in the vector $1

    local elem_value
    local elem_position
    local vector_length
    local elem_name

    eval elem_value="$3"
    elem_position=$(($2))
    eval vector_length=\$$1\_0
    if [ -z "$vector_length" ]; then
        vector_length=$((0))
    fi

    if [ $elem_position -ge $vector_length ]; then
        vector_length=$elem_position
    fi

    elem_name=$1_$elem_position

    eval $elem_name=\"\$elem_value\"
    if [ ! $elem_position -eq 0 ]; then
        eval $1_0=$vector_length
    fi
}

VectorPrint () {
# Vector Print
# Prints all the elements names and values of the vector $1 on sepparate lines

    local vector_length

    vector_length=$(($1_0))
    if [ "$vector_length" = "0" ]; then
        echo "Vector \"$1\" is empty!"
    else
        echo "Vector \"$1\":"
        for i in $(seq 1 $vector_length); do
            eval echo \"[$i]: \\\"\$$1\_$i\\\"\"
            ###OR: eval printf \'\%s\\\n\' \"[\$i]: \\\"\$$1\_$i\\\"\"
        done
    fi
}

VectorDestroy () {
# Vector Destroy
# Empties all the elements values of the vector $1

    local vector_length

    vector_length=$(($1_0))
    if [ ! "$vector_length" = "0" ]; then
        for i in $(seq 1 $vector_length); do
            unset $1_$i
        done
        unset $1_0
    fi
}

##################
### MAIN START ###
##################

## Setting vector 'params' with all the parameters received by the script:
for i in $(seq 1 $#); do
    eval param="\${$i}"
    VectorAddElementNext params param
done

# Printing the vector 'params':
VectorPrint params

read temp

## Setting vector 'params2' with the elements of the vector 'params' in reversed order:
if [ -n "$params_0" ]; then
    for i in $(seq 1 $params_0); do
        count=$((params_0-i+1))
        VectorAddElement params2 count params_$i
    done
fi

# Printing the vector 'params2':
VectorPrint params2

read temp

## Getting the values of 'params2'`s elements and printing them:
if [ -n "$params2_0" ]; then
    echo "Printing the elements of the vector 'params2':"
    for i in $(seq 1 $params2_0); do
        eval current_elem_value=\"\$params2\_$i\"
        echo "params2_$i=\"$current_elem_value\""
    done
else
    echo "Vector 'params2' is empty!"
fi

read temp

## Creating a two dimensional array ('a'):
for i in $(seq 1 10); do
    VectorAddElement a 0 i
    for j in $(seq 1 8); do
        value=$(( 8 * ( i - 1 ) + j ))
        VectorAddElementDV a_$i $j $value
    done
done

## Manually printing the two dimensional array ('a'):
echo "Printing the two-dimensional array 'a':"
if [ -n "$a_0" ]; then
    for i in $(seq 1 $a_0); do
        eval current_vector_lenght=\$a\_$i\_0
        if [ -n "$current_vector_lenght" ]; then
            for j in $(seq 1 $current_vector_lenght); do
                eval value=\"\$a\_$i\_$j\"
                printf "$value "
            done
        fi
        printf "\n"
    done
fi

################
### MAIN END ###
################
1
  • 2
    Note that while local is supported by both bash and dash, it is not POSIX. seq is not a POSIX command either. You probably should mention that your code makes some assumptions on the current value of $IFS (if you avoid using seq and quote your variables, it can be avoided) Dec 9, 2015 at 22:26

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .