The mate score bug in Glaurung

[Tord Romstad]

Since a long time, Glaurung has had a bug in its search display, which occasionally causes it to display an incorrect 0.00 or -mate in 1 score. I spent some time working on Glaurung yesterday, and finally fixed this bug. There are also a few other minor changes. Glaurung now knows that KBP vs K is a draw when the pawn is a rook pawn, the defending king stands in the pawn's path, and the bishop has the wrong colour. A couple of very minor bugs have also been fixed.

I would like to receive some input about just how annoying the bugs in the search display are to users. Is fixing these two bugs in itself sufficient to merit the release of a new version (which will almost certainly not be measurably stronger than 0.1.6)? If you think so, I will release 0.1.7 later today.

Tord

[Robert Allgeuer]

It is probably not "super-annoying", but still it has puzzled me when I saw it first. So in my view it merits a new release of Glaurung.

Robert

[Dann Corbit]

{snip}
Is fixing these two bugs in itself sufficient to merit the release of a new version (which will almost certainly not be measurably stronger than 0.1.6)?
{snip}
Tord

What are the changes that made it stronger?

[Tord Romstad]

{snip}
Is fixing these two bugs in itself sufficient to merit the release of a new version (which will almost certainly not be measurably stronger than 0.1.6)?
{snip}
Tord

What are the changes that made it stronger?

Knowledge about KBP vs K draw situations is probably worth an Elo point or two, but not much more. As I said, not measurably stronger.

I also have a new version which uses the hash table in the qsearch, but I am not yet sure whether this is an improvement. More testing is needed before I release this version.

Tord

[Alessandro Scotti]

Knowledge about KBP vs K draw situations is probably worth an Elo point or two, but not much more. As I said, not measurably stronger.

Sure, but a couple of day ago this cost me a won game against Yace and now that knowledge is on my "todo" list too!

BTW, what do you use to play engine matches under Linux?

[Peter Eizenhammer]

Since a long time, Glaurung has had a bug in its search display, which occasionally causes it to display an incorrect 0.00 or -mate in 1 score.
Tord

Hello Tord,

i think this bug is worth beeing debugged... so a pro for a (new) release.

Many Thanks for the new Glaurung and for your decision to make the source code
public again! I dont think (and dont hope, of course), that it is going to destroy the competition between programmers;
sometimes I even wonder, why there is nobody, who takes Glaurung, Crafty,
Fruit, or some of the other interesting public sources as a basis for improvements,
without starting to write the whole engine from scratch
(what is fantastic and highly appreciated, of course!)
Just call it MyAgressiveCraftoid, KnowledgeousGlaurit or similar.
Well, shared fame, is it half the fame? but ok, just my 50...

Peter

[Greg Simpson]

Anyone thinking about modifying the source of an existing engine to create their own needs to check the license it is released under to make sure that the use they have in mind is allowed, or get explicit permission from the author if it is not.

[Tord Romstad]

Knowledge about KBP vs K draw situations is probably worth an Elo point or two, but not much more. As I said, not measurably stronger.

Sure, but a couple of day ago this cost me a won game against Yace and now that knowledge is on my "todo" list too!

Exactly. It is the kind of thing that doesn't happen very often, but it is extremely painful when it does. I see it happen in about 1 out of 200 games here.

BTW, what do you use to play engine matches under Linux?

I do my testing partly on the P4 2.4 GHz Linux machine in my office, and partly on my iMac G5 1.8 GHz at home. On both platforms, I use xboard to run engine matches. I am not entirely happy with it, but I haven't found any better alternatives. Some day I hope to write a simple text-mode utility for running engine matches.

Tord

[Tord Romstad]

Since a long time, Glaurung has had a bug in its search display, which occasionally causes it to display an incorrect 0.00 or -mate in 1 score.

i think this bug is worth beeing debugged... so a pro for a (new) release.

Hi Peter,

Thanks for the input. I think I will wait a couple of days, though. I have made some new possible improvements which I would like to test a bit more throughly before releasing a new version.

Many Thanks for the new Glaurung and for your decision to make the source code
public again!

Your support is highly appreciated, and makes it more likely that I will let the source code stay public for future releases.

I dont think (and dont hope, of course), that it is going to destroy the competition between programmers;
sometimes I even wonder, why there is nobody, who takes Glaurung, Crafty,
Fruit, or some of the other interesting public sources as a basis for improvements,
without starting to write the whole engine from scratch

I think most of us simply think it is more fun to do it all by ourselves.

Tord

[Fabien Letouzey]

[quote="Alessandro Scotti"]BTW, what do you use to play engine matches under Linux?[/quote]

Just like Tord I use XBoard. Shell scripts generate command lines (xboard parameters) for Nunn-like positions. They only handle direct matches and gauntlets using a set of opening positions. Obviously book matches are even more simple to organise.

Instead of calling XBoard directly, I now store the command lines into a separate file first so that matches are somewhat interruptible: I remove "consumed" lines manually after interruption.

I used to parse XBoard -mg output, but now I scan PGN files to extract results.

Overall this is still very primitive though: a mix of many programs written in various languages. I intend to write more software for Unix-compatible systems and release it. Note that only command-line stuff is planned.

Fabien.

[Alessandro Scotti]

Hi Fabien and Tord,
thanks for your answers! Considering the situation I plan to start working soon on a command line tool to play engine matches, which I have been considering for a while now.
I think using Java and GPL would be appropriate for such a program, unless there are known problems. Would you be willing to review/discuss the feature list and design before the coding starts?

[Tord Romstad]

Hi Fabien and Tord,
thanks for your answers! Considering the situation I plan to start working soon on a command line tool to play engine matches, which I have been considering for a while now.
I think using Java and GPL would be appropriate for such a program, unless there are known problems. Would you be willing to review/discuss the feature list and design before the coding starts?

I lack the necessary programming competence to make any intelligent contributions to a discussion about the design, but I would be happy to join you in discussing the feature list. We should probably start a new thread, though.

By the way, is there any hope to see an updated Linux version of Kiwi some time soon? I can only find the old 0.2b at your site.

Tord

[Alessandro Scotti]

I lack the necessary programming competence to make any intelligent contributions to a discussion about the design, but I would be happy to join you in discussing the feature list. We should probably start a new thread, though.

Hi Tord, I think you're being too modest, your input is always very important!

By the way, is there any hope to see an updated Linux version of Kiwi some time soon? I can only find the old 0.2b at your site.

Well I did publish version 0.4A but later I deleted it because it wasn't playing well... I have now uploaded a build for version 0.4B and also a "beta" of version 0.4C, which seems to be quite stronger (maybe +80 elo but I could play only 20 games so far and I'm not going to repeat the 0.4A mistake... that's why I need the tool...)