Run a script on multiple shells?

The easiest way by far is to use the same shell on both systems. Just because one shell is preinstalled (there's no such thing as a “default shell” for scripts, the shell is whatever the shebang line says) doesn't mean that you can't install others. You can install bash on AIX (from the toolbox, for example), or ksh93 for Linux (with your distribution's package manager: install the ksh package).

If you pick ksh, beware that /usr/bin/ksh on AIX is ksh88. There is no version of ksh88 for Linux, only ksh93, which is available as /usr/bin/ksh93 on AIX.

It is easier if the same shell is installed at the same location on both systems (a symbolic link is fine), so that you can use the same path in the shebang line. If having the same location is too difficult, you can use something like #!/usr/bin/env bash, as long as you ensure that bash is in the PATH on both systems. This may be useful if you have to install bash or ksh in your home directory because you don't have root access.

If you really can't install the same shell on both machines — presumably because there's some silly compliance rule that only complicates people's life but is too entrenched to overturn — you have a few possibilities.

• Use #!/bin/sh as your shebang line and program in the intersection of what your systems offer. All modern unix systems provide a POSIX shell as /bin/sh. On AIX, /bin/sh is ksh88. On Red Hat, /bin/sh is bash (which behaves slightly differently when invoked as sh). Beware that on some other operating systems (for example, on many Linux distributions), /bin/sh is a smaller shell which may not have much more than POSIX features.

• Start your scripts with some boilerplate code that looks for a better shell and executes it.

  #!/bin/sh if [ -n "$BASH" ]; then … bash compatibility code … elif type whence >/dev/null 2>/dev/null; then … ksh compatibility code … elif type ksh93 >/dev/null 2>/dev/null; then exec ksh93 "$0" "$@" elif type ksh >/dev/null 2>/dev/null; then exec ksh "$0" "$@" elif type mksh >/dev/null 2>/dev/null; then exec mksh "$0" "$@" elif type bash >/dev/null 2>/dev/null; then exec bash "$0" "$@" else echo 1>&2 "Cannot find ksh or bash, aborting" exit 125 fi 

In either case, the intersection of ksh88 and bash provides some useful features beyond POSIX, but you may need a little compatibility code. In particular:

• Array assignment in ksh88 uses set -A. See assignation variable under different ksh environment
• Local variables are declared by typeset.
• Ksh88 doesn't have ${VAR/PATTERN/REPLACEMENT}, $'…' or FIGNORE. It does have [[ … ]].
• To enable @(…) and other ksh extended patterns in bash, run shopt -s extglob.

I had a similar requirement - I needed to use ksh93 on AIX 5.3 to support associative arrays. So I set up the script to start a second instance using ksh93. I also added a safeguard to keep it from respawning itself over and over.

#!/bin/ksh scr=$0 safe=$1 echo "Check OS to determine if ksh93 is needed..." if ( `typeset -A testvar > /dev/null 2>&1` ); then echo "Associative array functions supported." else echo "Associative array functions NOT supported. Starting second instance with ksh93..." if [ "$safe" -eq 1 ]; then echo "SECOND INSTANCE ALREADY STARTED!" exit 1 else echo "BEGIN SECOND INSTANCE USING KSH93" /bin/ksh93 $scr 1 exit 0 fi fi 

I would put some work into seeing if you can actually have one version with command that work on both systems.

I had this issue with my .bashrc file which I keep in my dotfiles repository. I want it to work on both OSX and Ubuntu and there were a few command I had to modify to do that. In this case it was mostly which switches worked for certain command or how a test command was written. I tackled each issue, finding a way to do whatever command I was trying in a form that would work on both OS's

I tried the 'two different files' and 'two different sections depending on OS' approaches and found them a pain to maintain.

Gilles's answer is (IMHO) brilliant (as always), but since you didn't choose it or even up-vote it, perhaps you find it too complicated. So here are a few (simple) points that might or might not be helpful, and may help you avoid maintaining two separate scripts, which could turn into madness for a large script that has to be maintained over a long period of time.

1) The AIX machines might have bash installed even though it's not the default shell for interactive use. (The AIX machines I work with do have bash installed, but I don't know if it came that way of the sysadmins added it. I'm guessing it came on it, since all our sysadmins are ksh junkies.) And I'm willing to bet that the RHEL machines have ksh installed. Check this, because if you can use the same shell on both machines it will save you a lot of hassle.

2) If you don't know where the shell is going to be installed, avoid putting its path directly in the shebang line. Instead put in the path to env and the shell executable name afterward, like so:

#!/usr/bin/env bash 

This should work on both machines, but you'll have to test it to be sure. (The location for env is supposedly more standard than the location for bash or most other things.)

Once you're using the same shell on both machines, the problem might be solved right there. But if not...

3) You can break your platform-specific code out into functions that you source into the main script. This means that although script might have lots of places where things work differently depending on whether it's running on AIX or RHEL, you don't need to repeatedly check the platform. Just do it once, and source the appropriate functions:

case $OS in "AIX") source aix_functions;; "Linux") source redhat_functions;; "*") echo "Exiting. The OS type is not found.";; esac 

Now you can use whatever functions you sourced, all over your script without ever having to have another case statement to distinguish between platforms:

do_something_platform_specific "$SOME_ARGUMENT" "$SOME_OTHER_ARGUMENT"