This is important, since the version we are currently using contains a BUG that allows for negative branch lengths in trees constructed with the Neighbor Joining algorithm and another bug that incorrectly calculates the branch lengths of the internal nodes. I discovered it with a simple Test that expects a tree of error zero.

This affects all the phylo classes (since the NJ is the only tree constructor algorithm) and the GuideTree class in the alignment package, used to construct progressive multiple sequence alignments (with the consequence that not necessarily the closest sequences are paired).