Floyd 0.5 released

[Sylwy]

I'll try it on ChessGUI shortly.

Works okay under ChessGUI.

Well, under Fritz all is also OK !

SilvianR

How strong do you estimate Floyd to be?

Any match still complete ! I'm extremely busy with my newest love: Crabby 1.0.0 !


Regards,
SilvianR

[Sylwy]

Works okay under ChessGUI.

Well, under Fritz all is also OK !

This is what you would expect if it is an Arena bug...

And vice versa ! Isn't it, Maestro HGM ?

Like salvgardation solution: Winboard GUI !

SilvianR

[Graham Banks]

I'll try it on ChessGUI shortly.

Works okay under ChessGUI.

Well, under Fritz all is also OK !

SilvianR

How strong do you estimate Floyd to be?

Any match still complete ! I'm extremely busy with my newest love: Crabby 1.0.0 !


Regards,
SilvianR

[hgm]

And vice versa ! Isn't it, Maestro HGM ?

Not sure what 'vice versa' would mean in this case. If it was a Floyd bug it would likely manifest itself on all GUIs.

I am happy to find that UCI2WB would also handle this correctly:

Code: Select all

else if(!strcmp(command, "option")) { // USI option: extract data fields
            char name[80], type[80], buf[1024], val[256], *q;
            int min=0, max=1e9;
            ...
            if(p = strstr(line+6, " min "))  sscanf(p+1, "min %d", &min), *p = '\n';
            if(p = strstr(line+6, " max "))  sscanf(p+1, "max %d", &max), *p = '\n';
 

As can be seen, it assumes a default max value of a billion. But I now get doubts on the default min value: I took it to be 0. But this would exclude negative values for options that do not explicitly specify the min parameter. Is that reasonable?

[Sylwy]

And vice versa ! Isn't it, Maestro HGM ?

Not sure what 'vice versa' would mean in this case.

(1)-An Arena GUI bug = suposed to be perfectly handled in Fritz GUI.
(2)-A Fritz GUI bug = suposed to be perfectly handled in Arena GUI !

Yeah, the UCI2WB adapter solves also the situation !

SilvianR

[mvk]

It doesn't matter who follows or doesn't follow the letter of the spec. It should work like all other programs in all major interfaces, and I will provide a fix, probably next week, in Floyd to ensure that users don't experience this issue.

In the meantime I have created a formal logo for the project.



The background artwork is by photographer Daniel Schweinert, who provided very kind permission to use this in Floyd. Daniel's website is www.schweinert.com.

[mvk]

I fixed a crash in the late endgame today, now testing it. If you can share your PGN or logs I can confirm if this is likely the same problem.

[mvk]

Ok thanks, that looks different and not like any of the other issues already solved. Would you mind to share any other data you have of this session. All other games, any log files, settings for the match, etc.

[mvk]

Thanks for the data. I don't understand yet what is happening. When I look at one of the 40/2 games (the first one), time usage looks normal up to that last move. Then it has 30 seconds left for 11 more moves. Floyd uses 5 seconds safety margin. I can see from earlier moves that the abort-long-iterations mechanism is functioning ok. So there is no reason to stall. The game is also not in the phase where I recently solved a crash issue. And we can see that it is in fact playing after book exit.

One thing I can imagine is that Shredder12 on windows sends a `movetime' along with the `wtime' and `movestogo', and then overruling the safety margin that Floyd applies for himself. But that seems not likely. Another hypothesis is related to memory usage, due to Floyd 0.5 not giving a maximum for Hash. Under the Arena bug that Ruxy reported that causes time losses in win64. Perhaps Shredder behaves the same as Arena. In that case this is solved in the next version I'm testing now. But it seems strange to me.

I don't own Shredder on windows, only on mac. If the win version can produce debug logging, even for the start of just one game, I'm interested to learn what Shredder sends in a game with this time control. This doesn't have to be a Floyd game. Otherwise I'm afraid I will have to check with the UCI master himself this weekend. More later...

[mvk]

Can you also try if changing the hash setting to 256 or 64 changes the problem? There is one problem, specifically with 128, in version 0.5 which does not occur with other sizes.