Werewolf wrote:Bob,
I will double check this tomorrow and get back to you. The reason I mentioned this is because I have this strange recollection of either Don or Larry talking about Komodo's non-deterministic search on a single thread.
But I might be going senile...
I would feel quite safe in saying that I doubt that is intended. But here are some typical no-no's for testing:
running multiple positions in one pass. Then picking one of those out and running it by itself. Most programs don't clear absolutely everything, so you get some cross-over informational exchange that can alter the trees.
running a program that has a persistent hash that survives across different runs. This is way problematic in that the first run then influences the second, reducing predictability.
Having (for Crafty) .craftyrc files laying around that you forget about, then you run a test in one directory and the second in another. And get different results with different hash sizes, etc.
My primary debugging tool is node counts. Much of what I change should not affect the tree at all, but frequently it does. I test until it doesn't, fixing whatever was incidentally broken.