Shell Programming and Scripting

This thread seems to be using the term "exported functions" incorrectly. When ksh runs a shell script it first forks a copy of itself. This copy will try to invoke the exec system call on the script. If the script has a valid "#!" line the exec will work. If we are running a ksh script, this means that a fresh copy of ksh is brought into core and it runs the script. But if the exec fails, the original forked ksh process runs the script itself. But first it "purifies" itself. It unsets non-exported variables and non-exported functions. This means that an exported function continues to be available as the subsidiary script is run.

I think exported functions are a bad idea. I never use them except just to experiment. I think that any script should always have a "#!" line and this undermines exported functions. This thread is actually discussing autoloaded functions. I don't like them either. I tried them and ran into the issues being discussed. I could find no satifactory solution. So now I just write plain old shell scripts. I use functions a lot and I always just copy the function into any script that needs it. Going further, I try to have the function set its own PATH and any other environment variables that it needs. This way each script is self sufficient and does not require an intricate environment in which to run.

So I don't have any solutions, but now you can curse the correct feature.

I need this function to effect a change of directory in the current shell.

If I use a script I will be back where I started when it exits the function.

By loading the script in the current shell it is available for the life of the session.

Thanks anyway for the clarification.

Or you could launch an interactive subshell. Tools like vi and sudo do that. And it's the way I do stuff like this. When I am finished, a simple cntl-D restores the session I had before. You will either need to code an undo function or require the user to logout and back in to restore the user's original session.

I think the way to go with the package is have it update /etc/profile as this runs for everyone. I can put a check in there for membership of the group and dot in the files if needed.

I guess I can do this using the postinst script.

I will need to be able to remove the lines on an uninstall though so was thinking of adding a unique comment like the package name to each line so it can be deleted with sed or awk.

If anyone can think why this would be a bad idea the feedback would be welcome.

Cheers