Seems to me absolutely nothing is normal here. The position already is 5-men, so there should never be any need to search deeper than 1 ply, so some 20 nodes. After 1.Ba7 the move bxa7 should come out as best even at 1 ply + QS. The PVs shown at 4 and 5 ply are total bogus.
Perhaps this is a Cylinder Chess version of Stockfish, so that after 1... bxa7 black could play Kh7xa7! 1/2-1/2.
Seems to me absolutely nothing is normal here. The position already is 5-men, so there should never be any need to search deeper than 1 ply, so some 20 nodes.
tbhits = 12 is normal, white has 12 legal moves. The pv shown is just SF doing a regular search on the "optimal" moves without probing TBs. All by design.
The broken 200214 version failed to detect that the root position was a tablebase position. Now it's fixed. Please simply discard the broken version.
Last edited by syzygy on Sun Feb 23, 2014 7:34 pm, edited 1 time in total.
But there is a strange thing, the new version takes much more time to reach depth 120, while the former reaches depth 120 in one second. And so I find contradictory comments in the commit with what I see there.
It is currently unused and only adds
overhead for nothing."
NVM, I actually confused "overhead for nothing." with "overheat for nothing."
Anyway, for me it seems to me that the CPU is running too long for nothing.
