transposition tables and three-fold repetition

[thomasahle]

I think the only way to be 'correct' is to store (in the TT entry) the moves that the entry assumes are not in the path, and those that it assumes are. This allows us to check if a TT entry is valid for our current search path.

Unfortunately this means storing quite a bit extra information in the TT entry. In particular you don't just need to store the positions in the path from the TT entry to the leaf, but also the assumptions for every move that was searched in the subtree.

I've had more luck with this approach for 50-moves and TT, since in that case we only need to keep track of a lower and upper bound.