Scorpio 1.91 and crashes

[Graham Banks]

Scorpio 1.91 crashed after just three games in the Shredder Classic GUI.
I've had no problems whatsoever with Scorpio 1.9, Scorpio 1.84, Scorpio 1.8 or Scorpio 1.7 in this regard.

[Jim Ablett]

Hello Graham,

try my build >

http://www.eggdisk.com/files/212659_kpvii/Scorpio191_win32_intel_ja.zip

Regards,
Jim.

[Graham Banks]

Will do - thanks Jim.

Regards, Graham.

[Daniel Shawul]

Hello Graham
Can you post or send me the debug file? I have done 100 test games with no crashes under Arena.
Daniel

[Graham Banks]

Hello Graham
Can you post or send me the debug file? I have done 100 test games with no crashes under Arena.
Daniel

I'm rather embarrassed to admit that I didn't create a debug file. Sorry Daniel.
Might just be a one off hopefully - will keep you informed.

Regards, Graham.

[Daniel Shawul]

Hi Graham
It seems that you are right! There is a bug which very rarely causes crashes since version 1.0 of scoprio! It only happed in this version because of some changes in the new version 1.91. It was a move generation bug that was suddenly activated. Fixed that now but i don't want to update since it rarely happens. However if you want the fix, I can send you the *same* version with only this bug fix.
regards
Daniel

[Guenther Simon]

Hi Graham
It seems that you are right! There is a bug

...


regards
Daniel

Hi Daniel,

You sure remember my post in the CCC forum about a game,
when Scorpio lost because of being overoptimistic and not
believing in a draw and later still thought it would be a draw,
when it was already dead lost?

The same happened again now and I cannot believe this
is just by bad luck(much too low probability this could happen
2 or 3 times in just 14 games)
The game should have been a dead draw by perpetual check,
instead Scorpio believed again it had a better position and tried to
win and ruined the position with 94...Rc2?? (played with > +1 !!?),
which allowed Black to transform in an endgame which is
hopeless for Scorpio just 2 moves later.
(may be even 95...Kb3!? instead of 95...Kb1 could save the
draw, but I am not sure. However after 96. Qg1! it suddenly
recognizes the problems)
I know that all programs have the tendency to avoid draws
by repetitions/perpetuals, if they think they are just slightly
better, but Scorpio clearly missevaluates sometimes(or often?)
such drawn positions by giving itself unjustified positive scores.

I don't know if you have already changed something in 1.91
regarding this, but I thought I still let you know again,
also because you didn't answer my last message in CCC.
(when you said before Scorpio just had needed more time
for the 'right' move, despite there was _no_ right move
anymore - this means the problem is not that simple then
explaining it with your new time management)

I haven't seen this problem in 1.8 or 1.84 which both reached
very high ratings in my list.


Best regards,
Guenther

[Event "RWBC_Class_B_7th_Edition"]
[Site "ESPRESSO"]
[Date "2007.02.07"]
[Round "14"]
[Number "287"]
[White "Hiarcs_X54UCI"]
[Black "Scorpio_19"]
[Result "1-0"]
[TimeControl "40/900"]
[Annotator "14. +0.39 16... -0.54"]

1. d4 {+0.01/1} Nf6 2. c4 {+0.01/1} e6 3. Nc3 {+0.01/1} Bb4 4. e3 {+0.01/1}
O-O 5. Bd3 {+0.01/1} d5 6. Nf3 {+0.01/1} c5 7. O-O {+0.01/1} dxc4 8.
Bxc4 {+0.01/1} Nbd7 9. Qe2 {+0.01/1} b6 10. Rd1 {+0.01/1} cxd4 11.
exd4 {+0.01/1} Bb7 12. Bd3 {+0.01/1} Rc8 13. Bd2 {+0.01/1} Qe7 14.
a3 {+0.39/14} Bxc3 15. bxc3 {+0.41/13} Rfe8 16. Re1 {+0.47/12}
Qd6 {-0.54/14} 17. a4 {+0.40/12} Nd5 {-0.54/13} 18. Ne5 {+0.33/12}
Nxe5 {-0.54/13} 19. dxe5 {+0.25/12} Qc5 {-0.59/13} 20. Qe4 {+0.66/13}
g6 {-0.43/13} 21. Bf1 {+0.50/13} Red8 {-0.34/14} 22. Qh4 {+0.49/13}
Nxc3 {-0.17/14} 23. Bh6 {+1.19/12} Nd5 {-0.45/14} 24. Rad1 {+1.21/14}
Qc6 {-1.39/13} 25. Re4 {+2.15/12} b5 {-1.53/14} 26. axb5 {+2.28/12}
Qb6 {-1.48/14} 27. Ra4 {+2.20/13} Re8 {-1.58/13} 28. Rxa7 {+2.67/12}
Qxa7 {-2.54/14} 29. Rxd5 {+2.30/13} Rc1 {-2.37/15} 30. Qf6 {+2.03/13}
Rxf1+ {-1.98/17} 31. Kxf1 {+2.03/15} Qa1+ {-2.05/17} 32. Ke2 {+2.05/15}
Qa2+ {-2.35/17} 33. Ke3 {+1.80/16} Qa3+ {-1.90/17} 34. Kf4 {+1.64/16}
Qc1+ {-2.00/18} 35. Kg3 {+1.64/17} Qxh6 {-1.73/18} 36. Rd7 {+1.52/16}
Qf8 {-1.41/17} 37. Rxb7 {+1.49/15} Rb8 {-1.21/17} 38. Rd7 {+0.99/15}
Qe8 {-0.74/17} 39. Rc7 {+0.83/14} Rxb5 {-0.66/12} 40. f4 {+0.75/14}
Rd5 {-0.55/16} 41. Re7 {+0.59/14} Qf8 {-0.62/15} 42. Kf3 {+0.46/14}
Rd2 {-0.53/14} 43. Ra7 {+0.44/15} Qe8 {-0.41/15} 44. Rc7 {+0.43/14}
h6 {-0.27/14} 45. Kg3 {+0.42/14} Rd7 {-0.17/15} 46. Rc2 {+0.30/15}
Qa8 {-0.27/15} 47. h3 {+0.39/14} Kh7 {-0.08/15} 48. Rc3 {+0.27/14}
Qb7 {+0.37/15} 49. Kh2 {+0.00/15} Rd2 {+0.71/18} 50. Rg3 {+0.00/15}
h5 {+0.31/16} 51. Rg5 {+0.00/14} Rf2 {+0.32/15} 52. h4 {+0.00/16}
Qc7 {+0.42/16} 53. Rg3 {+0.00/14} Kg8 {+0.32/16} 54. Kh3 {-0.13/14}
Rd2 {+0.65/16} 55. f5 {-0.39/14} Qc1 {+2.05/15} 56. Rxg6+ {-0.39/13}
fxg6 {+2.55/14} 57. Qxg6+ {-0.46/12} Kf8 {+2.32/15} 58. Qh6+ {-0.49/12}
Ke8 {+2.32/14} 59. Qxe6+ {-0.49/11} Kd8 {+2.21/14} 60. Qb6+ {-0.78/12}
Kc8 {+2.07/15} 61. Qa6+ {+0.00/12} Kc7 {+2.07/15} 62. Qa5+ {+0.00/13}
Kc6 {+2.07/14} 63. Qa8+ {+0.00/13} Kb5 {+2.00/14} 64. Qb7+ {+0.00/13}
Ka5 {+1.31/14} 65. Qa7+ {+0.00/14} Kb4 {+1.21/14} 66. Qb6+ {+0.00/14}
Ka3 {+1.21/13} 67. Qe3+ {+0.00/13} Ka2 {+1.21/13} 68. Qa7+ {+0.00/13}
Kb3 {+1.21/13} 69. Qb6+ {+0.00/14} Ka4 {+0.00/15} 70. Qa6+ {+0.00/14}
Kb3 {+0.00/16} 71. Qb6+ {+0.00/13} Ka4 {+0.00/19} 72. Qa6+ {+0.00/15}
Kb3 {+0.00/25} 73. Qb7+ {+0.00/14} Ka3 {+0.96/14} 74. Qf3+ {+0.00/14}
Ka4 {+1.21/13} 75. Qa8+ {+0.00/14} Kb4 {+0.00/15} 76. Qb8+ {+0.00/14}
Ka3 {+0.96/15} 77. Qf8+ {+0.00/14} Kb2 {+0.96/15} 78. Qb4+ {+0.00/14}
Ka1 {+0.48/15} 79. Qa4+ {+0.00/14} Kb1 {+0.48/14} 80. Qb3+ {+0.00/13}
Rb2 {+0.48/14} 81. Qd3+ {+0.00/14} Rc2 {+0.48/15} 82. Qb5+ {+0.00/13}
Ka2 {+0.48/15} 83. Qa4+ {+0.00/13} Kb2 {+0.29/15} 84. Qb4+ {+0.00/15}
Ka1 {+0.48/15} 85. Qa4+ {+0.00/14} Ra2 {+0.48/15} 86. Qd4+ {+0.00/14}
Rb2 {+0.03/15} 87. Qa4+ {+0.00/14} Kb1 {+0.01/15} 88. Qe4+ {+0.00/13}
Rc2 {+0.00/15} 89. Kh2 {+0.00/13} Ka2 {+0.14/13} 90. e6 {+0.00/12}
Rc4 {+0.61/13} 91. Qe2+ {+0.00/12} Kb1 {+1.12/14} 92. Qf2 {+0.00/12}
Qd1 {+1.00/13} 93. Qb6+ {+0.00/12} Ka2 {+1.00/13} 94. Qf2+ {+0.00/12}
Rc2 {+1.08/13} 95. Qa7+ {+0.00/12} Kb1 {+0.00/13} 96. Qg1 {+0.45/13}
Qxg1+ {-0.71/17} 97. Kxg1 {+0.45/13} Re2 {-1.07/18} 98. g4 {+2.05/15}
Re5 {-3.60/17} 99. f6 {+2.21/15} Rxe6 {-5.36/16} 100. g5 {+2.97/16}
Re1+ {-7.20/16} 101. Kg2 {+5.03/12} Re2+ {-7.83/17} 102. Kg3 {+5.17/12}
Re3+ {-9.03/17}
{Black resigns} 1-0

[Daniel Shawul]

Hi Gunther

You are right.I think it is time to fix the repeatition draw problems. It is probably because of over-evaluating some positions but might also be a bug in the search. I am not sure about anything after I got a number of bugs which seem to be there since very early versions!!

I don't know if you have already changed something in 1.91
regarding this, but I thought I still let you know again,
also because you didn't answer my last message in CCC.
(when you said before Scorpio just had needed more time
for the 'right' move, despite there was _no_ right move
anymore - this means the problem is not that simple then
explaining it with your new time management)

I haven't changed anything in 1.91 so it will still miss-evaluate.
At the last move , it used very less time and reached a depth of 12 and when i tested at that time it seems that it changes the move at depth 13. Well I haven't checked at that time the new move also looses. Okay I admit I have to look in this thing deeply.

I will try to bring down some aggresive evaluation terms in the endgame,hopefully draws will be seen faster. If scorpio 1.8 and 1.84 have no problems with this then it is probably because of new additions.

best
Daniel

[Graham Banks]

Hi Graham
It seems that you are right! There is a bug which very rarely causes crashes since version 1.0 of scoprio! It only happed in this version because of some changes in the new version 1.91. It was a move generation bug that was suddenly activated. Fixed that now but i don't want to update since it rarely happens. However if you want the fix, I can send you the *same* version with only this bug fix.
regards
Daniel

I would be very grateful if you could send me the fix Daniel.

Regards, Graham.

[Daniel Shawul]

Done!

[Graham Banks]

Done!

Received. Thanks!