11:48:26scymtymand your are right: if the distribution is very non-uniform, a balanced tree may not be optimal
11:48:27jackdanielI think what shka__ is after is optimizing the dispatch at runtime (as opposed to providing optimal one for some kind of "average" workload)
11:49:19scymtymyes, that's an interesting angle. but i'm not ready to attach it at the moment
11:49:57scymtymif i can make this work, it would be pretty great as it is
11:59:09scymtymit should help a lot in some cases since it cuts down on redundant tests and memory accesses. initial experiments support this. the hard problem will be keep compile time under control
12:00:00scymtymbut if it works out, it will help with gf dispatch, TYPECASE-like things and pattern matching