a neater workaround for the bash mono dimensional array limitation

Question

Contrarily to ksh, that allows multidimensional arrays, bash (4.2 or below) does not allow multidimensional arrays. I work around this by having a separator in the elements and then each element of the first array can be made in to an array itself. Foe example:

GPIO=("in:down:0:22:1-1.4:17:usb:usb port 3 mgmt button"
"in:down:0:13:1-1.5:18:usb:usb port 4 mgmt button")

then to access the third element of the first row I'd do something like this:

eval $(echo "ROW0=($(echo "${GPIO[0]}" | awk -F: '{ for(i=1;i<=NF;i++) printf("\"%s\" ",$i);}' ))")
echo ${ROW0[2]}

I don't like doing it but I've not found a neater way around the problem. Is there a neater and more efficient way around the problem while still approaching it like an array ?

There is also a workaround using associative arrays and adding a fake second index in the element index:

declare -A FAKE2DIMARRAY
FAKE2DIMARRAY=(
[0,0]="first row first element"
[0,1]="first row second element"
[1,0]="second row first element"
[1,1]="second row second element"
)

Although this solution is really neat and efficient it makes the definition of the array a mess if it has many rows and elements. For instance my full GPIO array is actually 16x8 and that would require 128 lines and taking care to correctly index each line ... that's less then optimal too in this case.

Answer 1

It is not so difficult to have an automatic creation of a (fake) multidimensional fullarray. To extract indexes from the array: look at the end.

Use the strings as you have them now to create several arrays:

IFS=: read -a arr0 <<<"in:down:0:22:1-1.4:17:usb:usb port 3 mgmt button"
IFS=: read -a arr1 <<<"in:down:0:13:1-1.5:18:usb:usb port 4 mgmt button"

Just make them start with the same name, arr in this example.

Then, the automatic system that will take all variables which names start with arr for rows.
And the count of values in such arrays as columns:

#!/bin/bash
IFS=: read -a row1 <<<"in:down:0:22:1-1.4:17:usb:usb port 3 mgmt button"
IFS=: read -a row2 <<<"in:down:0:13:1-1.5:18:usb:usb port 4 mgmt button"

#echo "one ${row1[@]}"
#echo "two ${row2[@]}"

declare -A fullarray
# Note that ${!row*} lists all the variables that start with arr.
for hrow in ${!row*}; do
    #echo "row $hrow"
    for (( column=0; column<${#row1[@]}; column++ )); do
        #echo "column $column"
        indirect=$hrow[$column]
        #echo "indirect $indirect ${!indirect}"
        fullarray[${hrow#row},$column]=${!indirect}
    done
done

declare -p "fullarray"

Un comment the echo(s) to see how the array is built.

As you can see from the output of declare -p fullarray, the variable has all the values in the array (hrow,column) and each value could be accessed with fullarray[$hrow,$column] if row and column are numbers from 0 and up.
In this case keep the $ in the variables since we need the string representation of the indexes (not their numeric equivalent).

---

Of course, I guess that the assignment to an array of each row already solve your problem without needing to use awk.

That only depends if using:

indirect=arr$row[$column]
${!indirect}

is good enough for you. :-)

---

To extract the indexes from the full array, if that is needed, you could do in three steps. First get all the indexes from fullarray:

allind=(${!fullarray})

And, as allind will be an array in itself, you could use the comma as a separator to extract each part:

rowind=( ${allind[@]//,*} )

As the rowind will contain all row indexes for all columns, the values will be repeated. Just sort and remove repeats:

rowind=( $(printf '%s\n' "${rowind[@]}"|sort -u) )

And, if you want, take a look at the definition of row indexes:

declare -p rowind

The same process (except that allind has already been defined) for columns:

allind=(${!fullarray[@]})
columnind=( ${allind[@]//*,} )
columnind=( $(printf '%s\n' "${columnind[@]}"|sort -u) )
declare -p columnind