Of course there is no such theory, but practice breaking a hash is fascinating for the researcher.

Currently in netdev@ people started lengthy discussion about new hash for the interface name and (optionally?) for dentry hash, or I just misunderstood the latter.

Anyway there is more than a dozen of different algorithms tested for deviation and speed. It is very interesting to find out which one will be selected.

Actually it is only interesting from the single side - how to break it. By breaking I mean creating application which can generate input data which will produce the same hash value after processed by the selected algorithm.

That's what I did for Jenkins and Bernstein/Torek (hash * 33) hash quite for a while already.

Looking forward for the new hash :)