[ElDiabloConCaca]

Each unique file on the backup is a file once and only once, and every other time in the backup, appears as a link.

So, instead of processing all files, simply process real files, and ignore links altogether (done with a simple UNIX 'file' test). Go through the backup reverse-chronologically, and you're guaranteed to always process the newest version of a file first.

Edit: Hmmm... I just thought about my algorithm, and it may be severely flawed... tee hee... if you only process all real files and skip links, then you'll end up having to touch all files in the backup... not very efficient.

With the UNIX 'find' command, you can use the -f switch to ignore symbolic links. You could run the UNIX 'find' command once on your backup drive at the beginning of the program, printing out all paths to all real files. You could then parse this list of files and do your processing.