Komodo TCEC: Serious castling bug !

[pohl4711]

During my testrun of Komodo TCEC 64bit for my LS-ratinglist a serious castling-bug appeared:
When Komodo plays black and the first move of the opponent is 0-0 (and the movecounter is 1 (in the LittleBlitzerGUI the first move has always the movenumber 1)), than Komodo TCEC crashes.
Same under the FritzGUI (I have Fritz 13 GUI):

Take the PGN-game below or just copy the FEN in the FritzGUI, then load Komodo TCEC 64bit and then play 0-0 manually for white - and Komodo TCEC crashes...

Because of that, I could not finish the LS-testrun of Komodo TCEC (I had more than 40 crash-games). Have to wait for a bugfix...

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Houdini 4 x64"]
[Black "Komodo TCEC x64"]
[Result "1-0"]
[SetUp "1"]
[FEN "r1b2rk1/pp1nqppp/2pbpn2/3p4/2PP4/2NBPN2/PPQB1PPP/R3K2R w KQ - 0 1"]
[PlyCount "1"]

1. O-O 1-0

[ouachita]

ouch, not good, but such is life with engines

[OneTrickPony]

See my thread about the same problem:
http://www.talkchess.com/forum/viewtopic.php?t=50335

I had this in analysis in Tarrasch GUI but not in Arena. It seems that maybe Komodo has some problems with processing what some GUIs send to it.
I've found two positiosn where it just refuses to see possibility of 1.0-0 (even though it keeps going in inf analysis mode just doesn't consider the move at all).

This really looks like critical bug to me (Tarrasch GUI handles all other engines just right).

[pohl4711]

Meanwhile Ingo B. told me in the german CSS-forum, that this happens in the Shredder classic GUI, too. So Komodo TCEC crashes becuase of that bug in the LittleblitzerGUI, the FritzGUI and the Shredder Classic GUI...
I hope, that there will be a quick bugfix by the Komodoteam.

Stefan

[lkaufman]

During my testrun of Komodo TCEC 64bit for my LS-ratinglist a serious castling-bug appeared:
When Komodo plays black and the first move of the opponent is 0-0 (and the movecounter is 1 (in the LittleBlitzerGUI the first move has always the movenumber 1)), than Komodo TCEC crashes.
Same under the FritzGUI (I have Fritz 13 GUI):

Take the PGN-game below or just copy the FEN in the FritzGUI, then load Komodo TCEC 64bit and then play 0-0 manually for white - and Komodo TCEC crashes...

Because of that, I could not finish the LS-testrun of Komodo TCEC (I had more than 40 crash-games). Have to wait for a bugfix...

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Houdini 4 x64"]
[Black "Komodo TCEC x64"]
[Result "1-0"]
[SetUp "1"]
[FEN "r1b2rk1/pp1nqppp/2pbpn2/3p4/2PP4/2NBPN2/PPQB1PPP/R3K2R w KQ - 0 1"]
[PlyCount "1"]

1. O-O 1-0

Thanks for pointing this out. We will give it high priority. Perhaps this might explain some poor test results; maybe those testers used books where the first move was castling more than testers who got good results. I take it that this problem was not present in Komodo 6, or you would have told us before this. That should help.

[pohl4711]

During my testrun of Komodo TCEC 64bit for my LS-ratinglist a serious castling-bug appeared:
When Komodo plays black and the first move of the opponent is 0-0 (and the movecounter is 1 (in the LittleBlitzerGUI the first move has always the movenumber 1)), than Komodo TCEC crashes.
Same under the FritzGUI (I have Fritz 13 GUI):

Take the PGN-game below or just copy the FEN in the FritzGUI, then load Komodo TCEC 64bit and then play 0-0 manually for white - and Komodo TCEC crashes...

Because of that, I could not finish the LS-testrun of Komodo TCEC (I had more than 40 crash-games). Have to wait for a bugfix...

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Houdini 4 x64"]
[Black "Komodo TCEC x64"]
[Result "1-0"]
[SetUp "1"]
[FEN "r1b2rk1/pp1nqppp/2pbpn2/3p4/2PP4/2NBPN2/PPQB1PPP/R3K2R w KQ - 0 1"]
[PlyCount "1"]

1. O-O 1-0

Thanks for pointing this out. We will give it high priority. Perhaps this might explain some poor test results; maybe those testers used books where the first move was castling more than testers who got good results. I take it that this problem was not present in Komodo 6, or you would have told us before this. That should help.

No, Komodo 6 doesnt have this bug, otherwise it would have crashed in its LS-testrun, but that 10000 games were all OK (no timelosses, illegal moves).

Stefan

[IWB]

Meanwhile Ingo B. told me in the german CSS-forum, that this happens in the Shredder classic GUI, too. So Komodo TCEC crashes becuase of that bug in the LittleblitzerGUI, the FritzGUI and the Shredder Classic GUI...
I hope, that there will be a quick bugfix by the Komodoteam.

Stefan

It happens with FEN and I asked for the PGN to that position. Do you have one?

Besides that and thinking about it I would not call it "serious" as it did not happen in thousands of games for me using a PGN Database with a correct numbering after that with the classic GUI.

So, do you have a PGN to that point. I will try that and (without really knowing) have the gut feeling it might work then.

Nonetheless it is a minor bug with the FEN for sure.

Bye
Ingo

[pohl4711]

Meanwhile Ingo B. told me in the german CSS-forum, that this happens in the Shredder classic GUI, too. So Komodo TCEC crashes becuase of that bug in the LittleblitzerGUI, the FritzGUI and the Shredder Classic GUI...
I hope, that there will be a quick bugfix by the Komodoteam.

Stefan

So, do you have a PGN to that point. I will try that and (without really knowing) have the gut feeling it might work then.


Ingo

In the PGN of that opening position the movecounter is 8, not 1. And with movecounter>1, the bug doesnt happen...So no need to test that. The problem is, that the LittleBlitzerGUI deletes all the moves of the PGN and uses only the FEN of the final positon (after the 8 moves) and sets the movecounter to 1.

Stefan

[OneTrickPony]

It happens with FEN and I asked for the PGN to that position. Do you have one?

I don't know which position he had but here are two where Komodo TCEC doesn't see castling in Tarrasch (so I guess it crashes in other GUIs when attempting to play):
1)
1.d4 Nf6 2.Nf3 g6 3.c4 Bg7 4.Nc3 d5 5.cxd5 Nxd5 6.e4 Nxc3 7.bxc3 c5 8.Be3 Qa5 9.Nd2 O-O 10.Be2 cxd4 11.cxd4 f5 12.exf5 gxf5 13.Rc1 Be6 here
Komodo doesn't see 14.0-0 (it saw it the move before)

2)
1.d4 Nf6 2.c4 e5 3.dxe5 Ng4 4.e3 Nxe5 5.f4 Bb4+ 6.Bd2 Bxd2+ 7.Qxd2 Ng6 8.Bd3 O-O 9.Nf3 Re8

Here Komodo doesn't see 10.0-0 (but agrees it's the best move once made and also sees it before Re8 is played)

Besides that and thinking about it I would not call it "serious" as it did not happen in thousands of games for me using a PGN Database with a correct numbering after that with the classic GUI.

Maybe it happened in a way that Komodo didn't play 0-0 when it should or failed to consider it as a reply.

[IWB]

In the PGN of that opening position the movecounter is 8, not 1. And with movecounter>1, the bug doesnt happen...So no need to test that. The problem is, that the LittleBlitzerGUI deletes all the moves of the PGN and uses only the FEN of the final positon (after the 8 moves) and sets the movecounter to 1.

OK, so to conclude.

1. When K-TCEC gets a FEN your opponent starts with a castling move THEN K-TCEC crashes.
2. Your Test crashes because the GUI gives FEN even if it has the positions as PGN.

Even if I consider part one as a problem I would not call it serious - but it should be fixed.
Nonetheless is your problem definately a combination with a little problem of the GUI as well.

In my test I might not have that problem as the GUI of my choise (and all CB GUIs) give the full movelist to the engine in a game(according to UCI specification). It is not that FEN is not UCI specification but it is for sure not as UCI should be used for a game.

Bye
Ingo