We definitely need instructions for the main platforms. And a decent installer and various other things.probably some readme should be included somewhere in the imp distribution how to install the necessary libs if there are licensing issues in putting the whole lib into the package (containing the url, for example).
If there is a speedup, we should definitely have it. The current version in IMP has a few idiocies and so is slower than need be. In addition, if things are not built with NDEBUG, they will be extremely slow as the NBL checks that is is correct each time update is called.it'd be probably a good idea to have a non-bonded list dedicated for approximately similar radii. to me it seems a general purpose non- bonded list is still too slow for most atomic calculations. from my limited experience with the old code, it seemed like you need to gain several magnitudes in computational speed to make the non-bonded list usable for atomic applications.
With the submitted code, we can relatively easily add new algorithms as we decide we want to try them and change the default algorithm as we find a better one, without forcing users to change their code.