Howto iterate through an input stream without a subshell or a read.

It sometimes comes up that the usual trick of reading in a stream with a pipe and a while loop doesn't work, since the pipe causes a subshell and any variables set in the subshell are unavailable to the parent.
For example the normal trick (output the first few lines of the /etc/services file, just the first two columns and not empty or comment lines):

[uphill@zagreb]: grep -v ^# /etc/services | grep -v ^$ | head | awk '{print $1" "$2;}'
tcpmux 1/tcp
tcpmux 1/udp
rje 5/tcp
rje 5/udp
echo 7/tcp
echo 7/udp
discard 9/tcp
discard 9/udp
systat 11/tcp
systat 11/udp
[uphill@zagreb]: grep -v ^# /etc/services | grep -v ^$ | head | awk '{print $1" "$2;}' | while read x y; do echo x=$x y=$y; done
x=tcpmux y=1/tcp
x=tcpmux y=1/udp
x=rje y=5/tcp
x=rje y=5/udp
x=echo y=7/tcp
x=echo y=7/udp
x=discard y=9/tcp
x=discard y=9/udp
x=systat y=11/tcp
x=systat y=11/udp
[uphill@zagreb]: echo $x

[uphill@zagreb]: echo $y
[uphill@zagreb]: i=0; grep -v ^# /etc/services | grep -v ^$ | head | awk '{print $1" "$2;}' | while read x y
> do
> names[$i]=$x
> ports[$i]=$y
> i=$((i+1))
[uphill@zagreb]: echo ${ports[1]}

This leaves the parent shell without any information.
But if you feed a function the output, you can retain the information.

[uphill@zagreb]: function parse {
> i=0
> while [ $# -gt 0 ]
> do
> names[$i]=$1
> ports[$i]=$2
> i=$((i+1))
> shift; shift;
> done; }
[uphill@zagreb]: parse `grep -v ^# /etc/services | grep -v ^$ | head | awk '{print $1" "$2;}'`
uphill@zagreb[1063]: echo ${ports[0]}
uphill@zagreb[1064]: echo ${names[1]} ${ports[1]}
tcpmux 1/udp

This way you have access to the variables you set and can put the data collection part of this task into the subshell.
If you have lines with 3,4 or more input variables, just reference them as $3 $4 etc but remember to use the same number of shifts to move the pointer along the input the correct number of slots.

If you have input that has spaces in it within elements, that is, you have tabs for separation and spaces within (like say mysql output).
You can pipe to sed first to replace and then send your input to sed again to undo the damage

[uphill@surrey]: function parse {
while [ $# -gt 0 ]
name=`echo $5|sed -e 's/xXx/ /g'`
shift; shift; shift; shift; shift; shift; shift
done; }
[uphill@surrey]: parse `cat /etc/passwd |tr ':' '\t' |tail -5 |sed -e 's/ /xXx/g'`
[uphill@surrey]: echo ${usernames[3]} " -> " ${names[3]}
canna -> Canna Service User

We inline sed to change the spaces to xXx, then we do the same to change back before assignment to the names array. This trick works fairly well, but depends on your input having tabs or some other character than spaces to deliminate initially.