Re: Transposition table usage in quiescent search?
Posted: Wed Mar 06, 2013 9:19 pm
Chan Rasjid wrote:Same position, different information. When you did the probe, and got the hit, the info did not terminate the search. So why not store info that WILL terminate the search the next time around?bob wrote:I don't quite understand.pkappler wrote:The point I was making is that if you have a position P that you store in the normal search with a non-capture best move, what do you do when you store a q-search tt entry for the SAME position? You just about can't refuse to store, that's pretty well-known. So do you overwrite and lose the non-capture best move and replace it with a capture? That can break ordering earlier in the tree where it is more important...
If you don't store in the q-search, I suppose there is no problem, although I would then wonder what the
advantage of a probe might be.
A 'replace-always' scheme does not always replace unless we really want to implement it that way without checking if the position has already been hashed in an earlier search of a greater draft.
My implementation does not replace a same position where the entry has a greater or equal draft - it is the same position!
Rasjid.
"replace always" means exactly that.