Internal recursion for Split about skeptik HOT 2 OPEN

This task is difficult indeed. We already tried something like that with multi-split and it was a failure. The implementation was difficult. It was slow. And the compression gain was low.

The only difference between multi-split and what you propose here is that the choosen pivot may be different on each branch. This could make a big difference if a good heuristic is choosen. The main drawback is that Split makes the DAG become more "treeish". Therefore, the heuristic should be carefull of not spliting shared nodes among many branches.

Moreover, the implementation could either split then consider each branch in turn or do everything in a single traversal. In the former case, it'll be slow (because of linearisation). In the former case, it'll be difficult and error prone.

from skeptik.

Yes. I agree with your remarks.

from skeptik.