Most baffling bug EVER

[Mike Sherwin]

Starting with this position. Bricabrac playing black. Six seconds per move.
[d]3r1rk1/2p2p2/1b6/p2pP1q1/Qp1P2p1/8/PPR2PPN/3R2K1 w - - 6 32
1. d1d3 f7f5 +0.20
2. a4c6 g5g7 -M293

My immediate thought is that the e.p. code is broken. So I restarted the engine and forced.
1. d1d2 f7f6
2. d2d3 f6f5
3. a4c6 ...
4. ... g8g7 +0.14

Yep it must be the e.p. code. But wait.
1. d1d3 f7f5
2. e5f6 f8f6 -0.41
Engine now plays both sides.
3. c2e2 +0.42 f6f5 -0.34 Hmm, not broken.

Forced.
1. d1d2 f7f5
2. a4c6
2. ... g8g7 -0.01 Not apparently broken but a slightly different score, hmm. But engine played fine after that.

1. d1d3 f7f5 The engine played f5.
2. d3d1 g8g7 +0.14
3. a4c6 f8h8 +0.26 Once again not broken and the engine played fine for the rest of the game. Until I took back all the moves and played d1d3 and it was immediately broken. So i took back all the moves again and restarted the engine.

1. d1d3 f7f5
2. e5f6 f8f6

Took back the moves but did not restart the engine.
1. d1d3 f6f5 illegal move, broken

So then I had the engine print the board to Arena so I could see the internal board and black had pawns on f7 and f6 after e5e6 c.e.p was retracted. But, that does not happen in the search because in the original line even though it was broken the engine's board is correct.

My thoughts at the moment are that although according to perfect perft results the engine generates c.e.p moves but never plays them. And when the c.e.p moves are forced the TakeBack function is bugged. Therefore there are at least three bugs at play here. But, now I'm really because if the c.e.p take back code is bugged then how in the heck does perft return correct results?

[Mike Sherwin]

Conclusion 1: Taking back game c.e.p moves is bugged.
Conclusion 2: Since I have never seen the engine play a c.e.p move the scoring of c.e.p moves must be wrong.
Conclusion 3: During search it is only when the score exceeds (+/-)INF does the search break and start printing pv's that look like a1a1 a1a1 a1a1 ... a1a1.

Therefore there are two bugs and the bug in the search is a c.e.p scoring bug.

[Mike Sherwin]

My conclusions were wrong.

In MakeMove() the ep square is stored at ply + 1 and in TakeBack() the ply is decremented first. The offending line in TakeBack() is.

sq = m->ts - ((epbb[ply] == (one << m->ts)) << 3);

Should be.

sq = m->ts - ((epbb[ply + 1] == (one << m->ts)) << 3);



I hope this is the last search bug. If not I'm going to start naming this thread the "Bug Hunt Chronicles".

But, still I'm baffled how perft 7 could possibly give a correct number?

Edit: It might be time for me to retire from chess programming, again.

[Mike Sherwin]

Okay, I fixed the consequences of the bug without finding the bug. It turns out that it is a material problem in MakeMove()/TakeBack(). What I did was save the material in MakeMove() and restored it in TakeBack() removing all other material calculations in TakeBack(). Now I'm going to find that bug and use a flamethrower on it.

Here's the game with the kludge code.
[pgn][Event "TSCP.at"]
[Site "DESKTOP-HFVHK2B"]
[Date "2021.04.18"]
[Round "42"]
[White "Tscp181"]
[Black "Bricabrac"]
[Result "0-1"]
[BlackElo "2000"]
[ECO "C83"]
[Opening "Spanish"]
[Time "16:38:43"]
[Variation "Open, Classical, 10.Be3"]
[WhiteElo "2200"]
[TimeControl "0+6"]
[Termination "adjudication"]
[PlyCount "141"]
[WhiteType "human"]
[BlackType "human"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Nxe4 6. d4 b5 7. Bb3 d5 8.
dxe5 Be6 9. c3 Be7 10. Be3 O-O {(e8g8 b1d2 e4c5 b3c2 b5b4 d2b3 c5d7 e3f4
a8c8) -0.02/9 5} 11. Nbd2 {(b1d2 e4d2 d1d2 b5b4 a1b1 f8e8) +0.06/6 5} Nc5
{(e4c5 b3c2 b5b4 c3b4 c6b4 a2a3 b4c2 d1c2 c5d7) -0.01/9 5} 12. Bc2 {(b3c2
b5b4 d1b1 h7h6 c2h7 g8h8 e3c5 e7c5) +0.22/7 5} Bg4 {(e6g4 h2h3 g4h5 e3f4
f7f6 e5f6 e7f6 f4g3 b5b4 c3b4) -0.16/10 5} 13. Qb1 {(d1b1 g7g6 f1e1 b5b4
e3c5 e7c5 d2b3 g4f3 b3c5) +0.33/7 5} h6 {(h7h6 c2f5 g4f5 b1f5 d8d7 f5d7
c5d7 e3f4 a8d8 a1d1) -0.19/10 5} 14. Bh7+ {(c2h7 g8h8 h7f5 g4f5 b1f5 h8g8
f3d4 c6d4 e3d4) +0.30/7 6} Kh8 {(g8h8 h7f5 g4f5 b1f5 d8d7 f5c2 a8d8 a1d1
f8e8 h2h3 h8g8) -0.12/11 5} 15. Bf5 {(h7f5 g4f5 b1f5 h8g8 f3d4 c6d4 e3d4)
+0.30/7 5} Bxf5 {(g4f5 b1f5 d8d7 f5d7 c5d7 e3f4 a8d8 h2h3 d7c5 a1d1)
-0.20/10 5} 16. Qxf5 {(b1f5 h8g8 a1d1 f8e8 e3c5 e7c5 d2e4) +0.30/7 5} Qd7
{(d8d7 f5c2 a8d8 a2a3 d7g4 a1d1 c5e6 h2h3 g4h5 e3d4) -0.16/10 5} 17. Qc2
{(f5c2 h8g8 e3c5 e7c5 a1d1 a8e8 d2e4) +0.30/7 5} Rad8 {(a8d8 h2h3 b5b4 a1d1
b4c3 b2c3 h8g8 a2a3 f8e8 d2c4) -0.13/10 5} 18. Bxc5 {(e3c5 e7c5 a1d1 d7e7
b2b4 c5b6) +0.39/6 5} Bxc5 {(e7c5 a1d1 d7e7 f1e1 b5b4 c2a4 b4c3 b2c3 c6b8
a4c2) +0.14/10 5} 19. Nb3 {(b2b4 c5e7 a2a4 h8g8 a4b5 a6b5) +0.29/6 5} Bb6
{(c5b6 a1e1 d7g4 c2d1 g4g6 a2a3 h8g8 b3d4 c6d4 c3d4 c7c6) +0.25/11 5} 20.
Nbd4 {(b3d4 c6d4 c3d4 h8g8 a1c1 d8e8) +0.26/6 5} Nxd4 {(c6d4 f3d4 f8e8 a1e1
c7c6 d4f5 b6c7 f2f4 c7b6 g1h1) +0.12/10 5} 21. cxd4 {(c3d4 h8g8 a2a4 d7g4
c2c3 d8e8 b2b4) +0.22/7 5} Qe7 {(d7e7 a1c1 b5b4 f1d1 h8g8 h2h3 f8e8 c2e2
a6a5 c1c6 a5a4) +0.20/11 5} 22. Rfc1 {(f1c1 h8g8 a2a4 e7b4 c2c3 b4c3 c1c3)
+0.27/6 5} a5 {(a6a5 c1d1 c7c5 d4c5 e7c5 c2c5 b6c5 a1c1 c5b6 c1c6) +0.15/10
5} 23. Qc6 {(c2d3 e7d7 a2a4 b5b4 b2b3 h8g8) +0.29/6 5} b4 {(b5b4 c1e1 h8g8
h2h3 f8e8 a1c1 e8f8 c1d1 f8e8 c6c2) +0.14/10 5} 24. Rc2 {(c1c2 h8g8 a1c1
f8e8 c6a4 d8d7 c2c6) +0.33/7 5} Kg8 {(h8g8 a1d1 e7e8 c2c1 e8e7 h2h3 f8e8
d1e1 e7e6 a2a3) +0.12/10 5} 25. Qa4 {(c6a4 d8d7 c2c6 d7d8 a1c1 f8e8 c6c2)
+0.33/7 5} g5 {(g7g5 a1d1 g5g4 f3e1 f7f6 a4c6 f6e5 c6g6 e7g7 g6g7 g8g7)
+0.40/11 5} 26. h3 {(h2h3 h6h5 a1d1 g5g4 f3h2 e7h4 a4b3 g4h3 b3h3 h4h3
g2h3) +0.27/7 5} h5 {(h6h5 a1d1 g5g4 h3g4 h5g4 f3h2 e7g5 c2c6 g4g3 h2f3
g3f2) +0.36/11 5} 27. Rd1 {(a1d1 g5g4 h3g4 h5g4 f3h2 e7g5 a4b5) +0.25/7 6}
g4 {(g5g4 h3g4 h5g4 f3e1 f7f6 a4c6 f6e5 c6g6 g8h8 d4e5 e7e5) +0.36/11 5}
28. hxg4 {(h3g4 h5g4 f3h2 e7g5 a4b3 f8e8 c2c6) +0.15/7 5} hxg4 {(h5g4 f3h2
e7g5 c2c6 g4g3 h2f3 g3f2 g1f2 g5g4 c6f6 f8e8) +0.21/11 5} 29. Nh2 {(f3h2
g4g3 h2g4 e7h4 f2f3 d8e8 g4f6) +0.16/7 5} Qg5 {(e7g5 a4b3 f7f5 c2c6 f5f4
b3c2 g8g7 c2d2 g5h4 c6f6 f8f6) -0.18/11 6} 30. Rd3 {(d1d3 g8h8 c2c6 d8e8
a4d1 f8g8) +0.29/6 6} f5 {(f7f5 c2c6 f5f4 a2a3 b4a3 a4a3 f4f3 g2g3 f8e8
e5e6) +0.20/10 6} 31. Qc6 {(a4c6 f8e8 h2f1 f5f4 e5e6 a5a4) +0.99/6 5} Kg7
{(g8g7 d3d1 g5h4 g2g3 h4h3 c6e6 a5a4 c2c6 a4a3 b2a3) +0.14/10 5} 32. Nf1
{(h2f1 f5f4 d3d2 g7g8 e5e6 d8e8) +1.08/6 5} f4 {(f5f4 a2a3 f8e8 d3d1 g7g8
c6f6 g5f6 e5f6 e8e4 c2d2) +0.20/10 5} 33. g3 {(g2g3 g7g8 g3f4 g5f5 d3d2
f5f4) +0.95/6 5} Qf5 {(g5f5 d3d1 f5h7 c2c1 h7f5 c1c2 f5e4 c2d2 e4f3 e5e6)
+0.42/10 5} 34. Rcd2 {(c2d2 g7h8 g3f4 f5f4 f1g3 h8g8 e5e6) +1.19/7 5} Rf7
{(f8f7 a2a3 f7e7 d2d1 e7e6 c6a4 f5e4 g3f4 e4f4 f1g3) +0.24/10 5} 35. gxf4
{(g3f4 f5g6 e5e6 f7e7 f4f5 g6f5 d3e3) +1.51/7 5} Qg6 {(f5g6 c6g6 g7g6 f4f5
f7f5 f1e3 f5f4 d2d1 g6f7 a2a3) +0.11/10 6} 36. Qxg6+ {(c6g6 g7g6 e5e6 f7f4
f1g3 d8e8 g3e2 f4f5) +1.40/7 5} Kxg6 {(g7g6 d3g3 f7f4 f1e3 g6h5 e5e6 c7c6
a2a3 f4d4 d2d4 b6d4 e3g4) +0.13/12 5} 37. Ne3 {(f1e3 f7f4 a2a3 g6f7 e3g2
f4e4 g2e3) +1.15/7 5} Rxf4 {(f7f4 d2d1 c7c6 e3g2 f4e4 g2e3 d8e8 d3d2 e4f4
d2c2 g6g7) +0.19/11 5} 38. Kh2 {(g1h2 f4e4 h2g2 g6g5 e3c2 d8e8 d3g3)
+1.01/7 5} c6 {(c7c6 d2c2 g6h7 c2c6 b6d4 h2g3 f4e4 e5e6 d4e5 g3g2 e5d4)
+0.33/11 5} 39. Kg3 {(h2g3 g6g5 g3g2 d8e8 e3c2 b6c7 d3g3) +0.98/7 5} Kg5
{(g6g5 g3g2 g5g6 d2d1 d8e8 d3d2 a5a4 g2g1 g6g7 a2a3 b4b3) +0.36/11 5} 40.
Kg2 {(g3g2 f4f3 a2a3 b4a3 b2a3 c6c5 d4c5 b6c5 d3d5 d8d5 d2d5 c5e3 f2e3)
+0.85/7 5} a4 {(a5a4 g2g1 g5g6 e3g2 f4e4 g2e3 g6h7 d2d1 a4a3 b2a3) +0.28/10
5} 41. Kg1 {(g2g1 f4f3 a2a3 b4a3 d3a3 c6c5 d4c5 b6c5) +0.95/7 5} Kh5 {(g5h5
e3c2 c6c5 e5e6 h5g6 c2e3 c5d4 e6e7 d8c8 e3d5 c8c1) +0.23/11 6} 42. a3
{(b2b3 a4b3 a2b3 f4f3 g1f1 g4g3 e5e6 g3f2 d2f2 f3f2 f1f2) +0.94/7 5} b3
{(b4b3 e3d1 f4e4 d1c3 e4e1 g1g2 b6a5 d3e3 e1e3 f2e3 h5g6) +0.24/11 5} 43.
Rd1 {(d2d1 c6c5 d4c5 b6c5 d3d5 d8d5 d1d5 c5e3 f2e3) +0.76/7 5} Rdf8 {(d8f8
d3d2 h5g5 e3f1 g5g6 f1e3 g6f7 g1g2 b6a5 d2e2 f7g8) +0.28/11 5} 44. R3d2
{(d3d2 b6a5 d2e2 a5b6 e2d2) 0.00/8 5} Kg5 {(h5g5 e3f1 b6a5 d2e2 a5b6 e2d2
b6a5 d2e2 a5b6 e2d2 g5g6) +0.50/11 5} 45. Ng2 {(e3g2 f4e4 g1f1 f8h8 f1g1
h8f8) 0.00/8 5} Re4 {(f4e4 g1f1 f8f3 f1g1 b6a5 d2d3 f3d3 d1d3 a5b6 d3c3
b6d4 c3c6) +1.46/12 5} 46. Kf1 {(g1f1 f8f5 d2d3 c6c5 g2e3 c5c4 e3f5 c4d3)
-0.11/8 5} Rf3 {(f8f3 f1g1 b6a5 d2d3 f3d3 d1d3 a5b6 d3c3 b6d4 c3c6 e4e5)
+1.46/11 5} 47. Re1 {(d1e1 g4g3 e1e4 d5e4 g2e1 f3f2 d2f2 g3f2 f1f2 b6d4)
-1.35/8 5} Rxd4 {(e4d4 d2e2 d4e4 e5e6 g4g3 e6e7 g3f2 e2e4 f2e1n f1e1 b6f2
e1f1) +2.50/12 5} 48. Rde2 {(d2e2 g4g3 e1c1 g3f2 e5e6 d4e4 e2e4 d5e4 c1c6)
-2.38/8 5} Rc4 {(d4c4 g2e3 c4e4 f1g2 d5d4 e3c4 e4e2 e1e2 d4d3 e2d2 b6d4)
+2.45/11 5} 49. Ne3 {(g2e3 c4f4 e3d1 g4g3 f1g2 g3f2 e1f1 g5f5 d1f2 b6f2
e2f2 f3f2 f1f2 f4f2 g2f2 f5e5) -3.03/8 5} Re4 {(c4e4 e3d1 e4e2 e1e2 d5d4
e5e6 f3d3 e2e1 b6c5 e1e5 g5g6 e5c5) +1.82/12 5} 50. Nd1 {(e3d1 g5f5 e5e6
b6c5 e2e4 d5e4 f1g2 f5e6 e1e4) -1.51/9 5} Rxe2 {(e4e2 e1e2 b6c5 e5e6 g5f6
f1g2 d5d4 e2e4 f3d3 e4e1 f6g7 g2g1) +1.98/12 5} 51. Rxe2 {(e1e2 f3f8 d1c3
g4g3 c3d1 d5d4 e5e6 d4d3 e2d2 g3f2 d2d3) -1.39/9 5} Bc5 {(b6c5 e5e6 g5f6
f1g2 d5d4 e2e1 c5d6 g2h1 c6c5 e1e4 d6f4) +2.08/11 5} 52. Kg2 {(e5e6 g5f6
f1g2 d5d4 e2e4 f3d3 e4e1 d3d2) -1.30/8 6} d4 {(d5d4 e5e6 g5f6 e2e4 f3d3
e4e1 f6g6 g2h2 d3d2 h2g3 d4d3 g3g4) +2.03/12 5} 53. e6 {(e5e6 g5f6 e2e4
f6e7 g2f1 c5b6 f1e2 f3f6 e2d3 f6e6 e4g4) -1.97/9 5} Kf6 {(g5f6 e2e1 f3d3
g2h1 d3d2 h1g1 f6g6 g1g2 d4d3 e1f1 c5d4 d1c3) +2.26/12 6} 54. Re4 {(e2e4
f6e7 g2f1 f3f6 e4g4 e7e6 f1e2 e6e5 f2f3) -2.04/9 5} Rd3 {(f3d3 e4e1 f6e7
g2f1 c5b6 f1g2 c6c5 e1h1 e7e6 h1h6 e6f7 h6b6 d3d1) +3.05/13 5} 55. Re1
{(e4e1 f6e7 g2f1 d3d2 f1g2 d4d3 g2g3 c5d4 g3g4 d4b2 d1b2 d2b2) -2.65/9 5}
Ke7 {(f6e7 g2f1 c5b6 f1g2 c6c5 g2f1 b6a5 f1e2 a5e1 e2d3 e7e6 d3c4 e6d6)
+3.01/13 6} 56. Kg1 {(g2g1 c5d6 g1f1 d3d2 f1g1 d6c7 g1g2 g4g3 e1e4 d2d1
f2g3) -2.99/9 5} Bb6 {(c5b6 g1f1 b6a5 f1e2 d3f3 e1g1 f3f4 g1h1 e7e6 h1h7
f4f3 e2f1 a5b6) +3.36/13 6} 57. Kf1 {(g1f1 c6c5 f1e2 d3h3 e1g1 h3h4 e2d3
e7e6 d3e4) -3.11/9 5} Ba5 {(b6a5 f1e2 d3d2 e2f1 d2d3 f1e2 d3f3 e1g1 f3f4
g1g3 e7e6 f2f3 d4d3) +3.70/13 5} 58. Ke2 {(f1e2 d3d2 e2f1 c6c5 f1g1 c5c4
e1f1 e7e6 g1g2 d4d3 f1e1) -3.68/10 5} Rd2+ {(d3d2 e2f1 c6c5 f1g2 d2d3 e1h1
e7e6 h1h6 e6d5 h6h1 d5c4 h1f1 d3d2) +3.63/13 5} 59. Kf1 {(e2f1 a5c7 f1g2
g4g3 g2f3 g3f2 d1f2 d2b2 f2d3 b2h2) -3.83/9 5} c5 {(c6c5 f1g2 d2d3 e1f1
e7e6 f1h1 e6f6 h1h6 f6g5 h6h1 g5g6 h1f1 d3d2) +3.60/13 5} 60. Kg1 {(f1g1
a5c7 g1h1 g4g3 f2g3 c7g3 e1f1 d2h2 h1g1 e7e6) -3.70/9 5} Rd3 {(d2d3 e1f1
e7e6 g1g2 e6f7 f2f4 g4f3 f1f3 d3f3 g2f3 d4d3 d1e3 d3d2) +3.97/13 5} 61. Rf1
{(e1f1 e7e6 g1g2 c5c4 f1h1 c4c3 h1h6 e6f7 h6h7 f7g6 h7h1 d3d1 h1d1 c3b2)
-4.10/9 5} Kxe6 {(e7e6 g1h2 e6f7 f1g1 a5c7 h2h1 d3d2 g1e1 c5c4 h1g2 f7f8
g2g1 f8g8) +4.08/13 5} 62. Kg2 {(g1g2 c5c4 f1h1 c4c3 b2c3 a5c3 h1h6 e6f5
h6h5 f5e4 h5h1) -4.40/9 5} Kd5 {(e6d5 g2h1 d3d2 h1g2 d5c4 g2g1 c4d3 f2f4
d3c2 d1f2 c2b2 f2g4) +4.10/12 5} 63. Rg1 {(f1g1 a5c7 g1e1 c5c4 e1f1 c4c3
f1h1 c3c2 h1h5) -4.70/9 6} Ke4 {(d5e4 g1f1 d3d2 g2g1 e4f3 f1e1 d2d3 e1f1
f3e2 f2f3 d3d1 f1d1 e2d1) +4.80/13 5} 64. Rh1 {(g1h1 c5c4 h1g1 c4c3 b2c3
a5c3 d1c3 d4c3 g1e1 e4d5 e1e7) -4.60/9 5} Rd2 {(d3d2 g2g3 e4d3 f2f4 d3c2
d1f2 c2b2 f2e4 d2e2 e4f6 b2a2 g3g4 b3b2) +4.71/13 5} 65. Kg3 {(g2g3 e4d3
g3g4 d3c2 f2f4 d2d1 h1d1 c2d1 f4f5) -4.11/9 5} Kd3 {(e4d3 f2f4 d3c2 d1f2
c2b2 f2e4 d2e2 e4c5 b2a2 g3g4 b3b2 g4f3 e2d2) +4.63/13 5} 66. Kxg4 {(g3g4
d3c2 h1h5 a5b6 h5h1 d2d1 h1d1 c2d1 f2f4) -4.44/9 5} Kc2 {(d3c2 h1h5 d2d1
h5c5 a5c3 b2c3 d4d3 c5c7 b3b2 c7b7 d3d2 b7b2 c2b2) +6.33/13 5} 67. Kf5
{(g4f5 d2d1 h1d1 c2d1 f5e5 d4d3 e5d5 d1c1 d5c5 c1b2) -5.78/9 5} Rxd1 {(d2d1
h1h8 c2b2 h8a8 a5c3 a8a4 b2a2 a4a7 b3b2 a7b7 b2b1r b7c7) +10.14/12 5} 68.
Rxd1 {(h1d1 c2d1 f5e6 d4d3 e6d5 d3d2 d5c5 d1c2 c5b5 d2d1q b5a5 c2b2 a5a4)
-8.57/10 5} Kxd1 {(c2d1 f5e5 d1c1 e5d6 c1b2 d6c5 b2a2 c5b5 a5c3 b5a4 b3b2
a4b5 b2b1q b5a4) +13.04/14 5} 69. Ke5 {(f5e5 d4d3 e5d5 d1c2 d5c5 c2b2 f2f4
b2a3 f4f5 d3d2 c5b5 d2d1q b5a5) -11.04/11 5} Kc2 {(d1c2 f2f4 d4d3 e5d5 c2b2
d5c4 b2a3 c4d3 b3b2 d3c4 a5b4 c4b5 a3b3 b5b6) +13.82/14 5} 70. Kd5 {(e5d5
c2b2 d5c5 a5c3 f2f4 b2a3 f4f5 b3b2 f5f6 b2b1q f6f7) -12.95/11 5} Kxb2
{(c2b2 f2f4 b2a3 f4f5 b3b2 d5c5 b2b1q c5d4 b1f5 d4c4 f5f2 c4d5 a5c3 d5c6)
+15.19/14 5} 71. f4 {(f2f4 d4d3 f4f5 d3d2 d5e6 d2d1q f5f6 d1g4 e6d6 g4f5
d6e7 b2a3) -15.19/11 5 Arena Adjudication} 0-1
[/pgn]

[Jakob Progsch]

I can't help with this specific one but I recently had a similarily confounding bug for what appear to be similar reasons. I thought my move generation was good because it produced all the usual perft numbers correctly. But once in a while during search nonsensical things would happen such as kings ending up on neighbor squares causing infinite checking sequences blowing up my stack.

Eventually I noticed the common denominator was that those bugs always happened below null moves. It turned out for some reason one side ended up with nine pawns causing an out of bounds write on a piece array after which all bets were off obviously.

The ultimate reason was that on a null move I only flipped the white to move bit in the board but didn't clear the en passant bit allowing a "phantom" en passant capture for the wrong side. And because all changes to the board state happen via XOR ops (so they are trivially reversible) instead of being captured an opposing pawn popped into existence behind the capturing one.

[Mike Sherwin]

I can't help with this specific one but I recently had a similarily confounding bug for what appear to be similar reasons. I thought my move generation was good because it produced all the usual perft numbers correctly. But once in a while during search nonsensical things would happen such as kings ending up on neighbor squares causing infinite checking sequences blowing up my stack.

Eventually I noticed the common denominator was that those bugs always happened below null moves. It turned out for some reason one side ended up with nine pawns causing an out of bounds write on a piece array after which all bets were off obviously.

The ultimate reason was that on a null move I only flipped the white to move bit in the board but didn't clear the en passant bit allowing a "phantom" en passant capture for the wrong side. And because all changes to the board state happen via XOR ops (so they are trivially reversible) instead of being captured an opposing pawn popped into existence behind the capturing one.

Thanks for sharing your experience!

My biggest fear is that my new SISSY bitboards will fail in some rare position. So far they have been rock solid. After the code hack the engine has been running fine for 69 games. Can it make it to 100? I hope so. Just 31 to go and my engine will have completed it first match without problems.

[Mike Sherwin]

Dumbzie (dumb zombie) code example.
bb = (wPawnMoves[fs] & empty) | (wPawnCapts[fs] & enemy);
if (bb & king[otm]) return ILLEGAL;

Smartzie (smart zombie) code example.
bb = wPawnCapts[fs] & enemy;
if (bb & king[otm]) return ILLEGAL;
bb |= wPawnMoves[fs] & empty;

[Ras]

The ultimate reason was that on a null move I only flipped the white to move bit in the board but didn't clear the en passant bit allowing a "phantom" en passant capture for the wrong side.

Interestingly, my engine has a similar issue, but since the capture works to the square behind e.p. square which is empty (i.e. the start square of the pawn that had moved two squares up before), that's not registered as capture, but as quiet move. Taking that back hence doesn't generate a spurious pawn. I tried to fix the issue, only to find out that this cost 2 Elo, consistent with the loss in NPS, so I left it in.