Linux/Mac engine gauntlet, cluster testing

[jdart]

Current versions of Rodent do not compile/run on Linux.

--Jon

[jdart]

Most people are using cutechess-cli for matches. You are going to a lot of work if you are trying to replace that. It does require a shared library to function but it is easy to load that - just set LD_LIBRARY_PATH in your .bashrc (for Linux). (You also need Qt libraries installed). Only thing is, it does not support pondering, but if you are ok with ponder off it probably does everything you need.

--Jon

[elcabesa]

latest vajolet should work on linux
you should also found some linux executable https://drive.google.com/file/d/0B8vVDp ... sp=sharing

[matthewlai]

Most people are using cutechess-cli for matches. You are going to a lot of work if you are trying to replace that. It does require a shared library to function but it is easy to load that - just set LD_LIBRARY_PATH in your .bashrc (for Linux). (You also need Qt libraries installed). Only thing is, it does not support pondering, but if you are ok with ponder off it probably does everything you need.

--Jon

Yeah I looked into that as well. I may do that if writing it myself turns out to be too difficult (the only way I see that can be the case is if buggy engines require too many workarounds).

It was actually much less work than I had anticipated.

I spent about 15 hours on it, and it already has proper support for features (determining move format the engine needs, and things like ping, etc), can play games from list of starting positions, can keep track of time, verifies all mate and draw claims, properly translate between SAN and algebraic for engines that require either format, etc. Also verified perft on many starting positions.

The really only things left to do are workarounds for buggy engines.

For modern "protover 2" engines it works pretty well already.

I am also in favor of using simple programs linked together to complete tasks (UNIX philosophy), so my program will not do tournament management, etc, unlike cutechess-cli.

I will probably write another program to do tournament management, maybe distributed over MPI.

Another advantage of writing my own program is that it will be easier for me to reuse the code for looking for specific kinds of positions in PGN files, etc. And maybe add some hooks in tournaments to do automatic parameter tuning.

latest vajolet should work on linux Smile
you should also found some linux executable

Thanks! I'll keep that in mind if I have time to add more engines (probably after I go through and test all the existing ones, and eliminate buggy ones).

[dusando]

Pepito's code looks very old (GCC 2.x era), and didn't look like it will compile without many changes.

Pepito 1.59 compiled fine using GCC 4.8.3. In a linux system, it just needed to add -lm to the linking command in the makefile: $(CC) -lm $(CFLAGS) -o pepitox $(OBJECTS)

[matthewlai]

Pepito's code looks very old (GCC 2.x era), and didn't look like it will compile without many changes.

Pepito 1.59 compiled fine using GCC 4.8.3. In a linux system, it just needed to add -lm to the linking command in the makefile: $(CC) -lm $(CFLAGS) -o pepitox $(OBJECTS)

Ah It's probably LLVM-specific then. Do you get any warnings with "-Wall"?

[jdart]

I think Amy is in the too old/too buggy category.

--Jon

[dusando]

Pepito's code looks very old (GCC 2.x era), and didn't look like it will compile without many changes.

Pepito 1.59 compiled fine using GCC 4.8.3. In a linux system, it just needed to add -lm to the linking command in the makefile: $(CC) -lm $(CFLAGS) -o pepitox $(OBJECTS)

Ah It's probably LLVM-specific then. Do you get any warnings with "-Wall"?

Just tried to add -Wall, oh yes, plenty of warnings - the engine runs stable though, played quite some test games against Phalanx with no issues.

[matthewlai]

Pepito's code looks very old (GCC 2.x era), and didn't look like it will compile without many changes.

Pepito 1.59 compiled fine using GCC 4.8.3. In a linux system, it just needed to add -lm to the linking command in the makefile: $(CC) -lm $(CFLAGS) -o pepitox $(OBJECTS)

Ah It's probably LLVM-specific then. Do you get any warnings with "-Wall"?

Just tried to add -Wall, oh yes, plenty of warnings - the engine runs stable though, played quite some test games against Phalanx with no issues.

Ah ok. Since you have it built already, would you mind running a few games at very short time control (say "level 0 0:10 0.1")?

I wouldn't mind fixing it for LLVM (if it's not TOO much work), but it would suck if I spend a lot of time to fix it only to find out that it can't play short games (which is mostly what I am interested in).

I am interested in how Phalanx does as well. It's in my gauntlet already.

Thanks

[dusando]

Pepito's code looks very old (GCC 2.x era), and didn't look like it will compile without many changes.

Pepito 1.59 compiled fine using GCC 4.8.3. In a linux system, it just needed to add -lm to the linking command in the makefile: $(CC) -lm $(CFLAGS) -o pepitox $(OBJECTS)

Ah It's probably LLVM-specific then. Do you get any warnings with "-Wall"?

Just tried to add -Wall, oh yes, plenty of warnings - the engine runs stable though, played quite some test games against Phalanx with no issues.

Ah ok. Since you have it built already, would you mind running a few games at very short time control (say "level 0 0:10 0.1")?

I wouldn't mind fixing it for LLVM (if it's not TOO much work), but it would suck if I spend a lot of time to fix it only to find out that it can't play short games (which is mostly what I am interested in).

I am interested in how Phalanx does as well. It's in my gauntlet already.

Thanks

Started a quick test match at 0:10 + 0.1, all looking almost ok so far, no time forfeits, but ... game 96 shows an issue:

Warning: pepito forfeits by invalid result claim: 1/2-1/2 {Drawn game}
Finished game 96 (pepito vs r70): 0-1 {Black wins by adjudication: Invalid result claim}
Score of r70 vs pepito: 32 - 47 - 20 [0.424] 99