multi-pv bug of stockfish

[Uri Blass]

searching the following position with multi-pv 2 options I get the following analysis at depth 100


New game
[D]4Q3/8/8/3r4/1Q6/4k1K1/8/5q2 b - - 0 1

Analysis by Stockfish 1.6 JA:

1. +- (#2): 1...Ke3-d3 2.Qe8-e4#
2. +- (#1): 1...Ke3-d3 2.Qe8-e4#

(so k, 11.01.2010)

The bug is that stockfish should show 1...Re5 as one of the moves because black has exactly 2 legal moves and 1...Kd3 should get only one score and not 2 different scores.

I can see Re5 at small depths but not at depth 100 that is the last iteration.

Uri

[mcostalba]

Hi Uri,

sorry but I cannot reproduce, I got (on FritzGUI):

New game - Stockfish 100111
[D] 4Q3/8/8/3r4/1Q6/4k1K1/8/5q2 b - - 0 1

Analysis by Stockfish 1.6 JA:

1. +- (#2): 1...Rd5-e5 2.Qe8xe5+ Ke3-d3 3.Qe5-e4#
2. +- (#1): 1...Ke3-d3 2.Qe8-e4#

[Uri Blass]

Hi Uri,

sorry but I cannot reproduce, I got (on FritzGUI):

New game - Stockfish 100111
[D] 4Q3/8/8/3r4/1Q6/4k1K1/8/5q2 b - - 0 1

Analysis by Stockfish 1.6 JA:

1. +- (#2): 1...Rd5-e5 2.Qe8xe5+ Ke3-d3 3.Qe5-e4#
2. +- (#1): 1...Ke3-d3 2.Qe8-e4#

If I stop the analysis at small depth by clicking stop
I also get this analysis but
if I let it to get depth 100 I get the analysis that I posted under Fritz8 gui(did not buy later version of Fritz)

I used 256 mbytes hash but the same is also for 32 mbytes hash and probably this is not hash dependent because stockfish only searches 14 knodes

Uri

[zamar]

I can reproduce this, but...

the really funny thing is that when I add std::cout to diagnose the problem, the problem immediately disappears!! Very strange... they call these Heisenbugs.

Edit: Oops, sorry, forget what I said, I misread the output, I cannot reproduce this...