There is a lot of redundant work done across bytecodes. Often we end up converting a value from a low level representation into Smalltalk objects just so we can convert it back again. This is ineffient.
The tree traversal optimiser aims at removing this overhead using simple fast methods.
Things to do:
DONE Finish the boolean conversion/deconversion optimisation. At the moment I'm working on getting a convienient tree intermediate form to optimise. There are a few more bugs to iron out before starting the optimisation.