Shell Programming and Scripting

I tried to read some lines from a file and append them to a variable via "while" block like below:
###############
#!/bin/sh

test="ABC_"

while read L1
do
test=$test$L1
echo $test
done < some.txt

echo $test
###############

But the last line keeps print the initial value, "ABC_", even i can see that the $test is changed in the while block. And if i remove "< some.txt" and put the "read" command inside the block, the updated $test will be printed on the last line.

Could you please explain why such thing happens and how can i fix it if i still want to keep "< some.txt" at where it is...

Thanks in advance

I tried it on my box - and I get the expected result. Which box are you in?

Code:

rkumar@bdc4reteaix1w: /home/rkumar/unix_forum >cat some.text
junk
values
rkumar@bdc4reteaix1w: /home/rkumar/unix_forum >cat chk_some_text
#!/usr/bin/sh
test=ABC_
while read L1
do
test=$test$L1
echo $test
done <some.text

echo $test
rkumar@bdc4reteaix1w: /home/rkumar/unix_forum >./chk_some_text
ABC_junk
ABC_junkvalues
ABC_junkvalues

This behaviour is due to the fact that "read" command is executed in a separate process compared to the main sh process, so you lose all the varible assignments inside the while loop. I think that the only method for this to work while keeping the "< some.txt" is to save the $test value in a tempfile and restore it outside the loop:

Code:

###############
#!/bin/sh

test="ABC_"
tmp="/tmp/test.tmp"

while read L1
do
   test=$test$L1
   echo $test | tee $tmp
done < some.txt

test=`cat $tmp`
echo $test
###############

Almost correct. The problem is that the old Bourne shell spawns a separate process to handle a "while" loop. When the loop finishes, the parent shell will not see any changes made inside the loop. The Bourne shell behaves this way for any loop. Using or avoiding "read" has no effect on this. Most modern OS's that still supply the Bourne shell also supply ksh. Switching to ksh would solve the problem. Or switching to bash.

Note that on Linux, sh is a link to bash, not the Bourne shell. No Bourne shell is available on Linux so you can't reproduce the problem there. Not sure about AIX, but if this worked, sh could not have been the Bourne shell. Stuff like this is why we wish that folks would mention the particular OS.

Perderabo, thank you for the clarification. My example is in fact tested on Solaris 5.8 with the old Bourne shell, so I obtain the same behaviour as sleepy_11, which evidently is using Bourne shell too

Thanks for all your reply. Below are my comments:

1. i tested on the SunOS 5.9
2. I tried below script to test the "loop in new process", as you mentioned:

###############
#!/bin/sh

test="ABC_"
echo $$ #new line, print the process id outside the loop
while read L1
do
echo $$ #new line, print the process id inside the loop
test=$test$L1
echo $test
done < some.txt

echo $test
###############

But i got the same process id printed.......Is my test method wrong, or SUN does some update on sh?

Well, your test method is wrong. sh sets $$ when it starts up and it does not get updated when it forks. You could put the command
ps -f
inside and outside the loop to see the extra process. But please switch to a more modern shell. These problems were solved with the release of ksh in 1988.