Shell Programming and Scripting

Hi,

I have to use ksh on HP-UX for some scripting.
I usually use "set -e -u" in scripts to stop if errors occur or a typo is in a variable name.

Now I try to use "$@" to pass the arguments unchanged to another function, which works without problems - unless I try to call the script without arguments... In that case $@ is considered undefined and the -u kills the script, instead of calling the function without arguments, as planned.

Is there any nice way to solve that problem without +u or duplicating parts of the code?

PS: shells like bash work exactly as planned.

It works fine on my linux box and ksh... I don't know much about HP-UX at all, but have you tried to use $* instead of $@?

Make how the function is called dependent on whether arguments are passed to the shell script or not...

Code:

set -e -u

f() { list_of_commands;}

# main script body
if [ $# -eq 0 ]; then
   f
else
   f "$@"
fi

Try ${1+"$@"}

@mimi: Those are two different ksh, which do not share the same bugs.

@shamrock: I was thinking about that approach, too, but it duplicates the code (which is a bit more complex than "f") and even more important makes the code much less readable.

@Perderabo: Thanks a lot, guess something like "${@+$@}" should be exactly the right thing to fix the issue.

If you needed [/icode]"$@"[icode] then you will still need internal quotes like this: ${1+"$@"}. If no quotes are needed then a simple $@ should suffice. As far as replacing the 1 with @, as in ${@+"$@"}, that is interesting. It seems to work as well as what I suggested, but no better. Are you worried about a missing $1? Both constructs handle it well.

Code:

$
$ cat nargs
#! /bin/ksh

function show_args {
        typeset -i i=0
        while ((i<$#)) ; do
                ((i=i+1))
                eval arg=\$$i
                echo ${i}:${arg}
        done
        return 0
}

echo 'with ${1+"$@"}'
show_args ${1+"$@"}
echo
echo 'with ${@+"$@"}'
show_args ${@+"$@"}
exit 0
$ ./nargs "" two three "44 44 44"
with ${1+"$@"}
1:
2:two
3:three
4:44 44 44

with ${@+"$@"}
1:
2:two
3:three
4:44 44 44
$

${1+"$@"} is very slightly less work for the shell than ${@+"$@"} but the real avantage that ${1+"$@"} might have is that it has become standard. I first saw used in the late 80's so it has been around for awhile. But if you see some value to ${@+"$@"} please let me know what I'm missing.

Anyway I'm glad you posted the question. I still use ${1+"$@"} in my ksh scripts for portability to early versions of ksh. When people tell me that the practice is superfluous with ksh, I will refer them to this thread.

Well I liked @ instead of 1, because it exactly maps the problem of the undefined $@ and looks somewhat more readable to me. The result should be exactly the same.

Regarding "${@+$@}" vs. ${@+"$@"}, yes, the second one is probably the better one.
At least under bash they do exactly the same, but that ksh might generate an empty argument in the first case. (I had no chance to try out, yet.)

BTW, is it possible to assign the content of $@ (or ${@+"$@"}) to another variable?

If I do something like this:

Code:

count() { echo $#; }
foo() { 
  args1=$@;
  args2="$@";
  count "$@";
  count "$args1"  
  count $args2
  count "$args2"
}
foo "1 2" 3

the assigned variables have no longer any idea about the correct number of argument.