Are you compiling with the project/compile.sh script?
For Mac OS X, it will compile both 32 and 64 bit versions. This works fine on Mac OS X 10.8.5, the version I use.... from a higher version on, I do not know exactly when 32 bit support was dropped, the 32 bit compile does no longer work. For those versions, please remove the line defining the 32 bit version. For all Mac OS X versions, I probably could drop the 32 bit versions, as it is 64 bit since a long time and I am not sure if the platforms supporting only 32 bit do actually support C++11....
Change the section at the beginning of the compile.sh file by removing the TARGET32 for darwin.
Version update: Belofte 2.0.8 has been released on Windows/Linux/Mac in 32 and 64 bit under GPL v2.
Major changes:
plays time-based levels when selecting uci protocol
implements iterative deepening with bruteforce search
Builds are non optimized as compilation with -flto or -march generates problems on Pentium and Celeron class computers.
If you own a i3 or higher platform recommendation is to recompile.
Current strength: the C++ 2.0 series closes in on the 0.2 series in strength. The 0.9 series is still the strongest available.
Stability: uci interface is more stable than all versions before 0.9.3 and available on all platforms.
Works on Android so far, but, when using analyzing mode in DroidFish, i see no evaluation information. Also, when starting the engine, i see no information about the engine (version number).
I could do a conditional compile when detecting android target and force the uci mode as I am not aware of any xboard capable program on android. Any suggestions on #defines for android platform?
No output is a bit strange. In version 2.0.8 onwards there is iterative deepening and on going to next search depth, there should be output on the best move found till now.
The queen en prise is something I noticed. Search gets cut off when time expires and direct refutation of variation is not done correctly. I should only store score when tree has been completed and rely on previous score found on lower depth... Generally speaking QS search goes very deep without abandoning variations that have huge material loss. I have even seen this in AB searches in my test version.
Well, we only are at version 2.0.8 now, stockfish is at 12.x... I hope to settle this issue. My main priority is to get 3-fold and theoretic draws right, get bullet type time controls right, and convert from minimax to negamax.
I am playing a tournament with it. Seems much weaker than 0.9.12, about 900 Elos. It is about to appear in next week's update.
Many times it does not give a mate in one. In one position it had 2 rooks, 2 bishops and 5 pawns against the lone king and it took a lot of moves till it could deliver mate.
Version 2.x is a complete rewrite in c++ because some things were difficult to fix in c code.
Pv variation tracking up to root node of search.
Cross platform multi threading
Move score instead of score relative to resulting position
UCI implementation
...
I could probably get those things fixed in the 0.9.x series but as it got better with 0.9.18, it was only marginally better on easy release and did break a lot of times on windows builds. Let alone Mac. After some feedback on #freechess, I thought it would be easier to restart from scratch in c++.... As expected, my c++ was not up to level and it took me much more time than expected. A lot of evaluation code is not yet in there neither.
My first target is to be UCI wise 100% correct and play bullet and other levels without any bugs. This seems to be the case now.
I feel I am close to fixing the tree search issues and those problems of not seeing easy queen captures will soon be resolved and I should reach 0.9.12 level pretty fast... At least I hope.
I implemented negascout and mtd in the past in c++ for another game. With proper tt, some bitboards, evaluation on move level instead as a tree search result, and pluggable evaluation based on game phase, it has more potential as the 0.9 series I hope.
ydebilloez wrote: ↑Sun Nov 22, 2020 10:14 pmWith proper tt, some bitboards, evaluation on move level instead as a tree search result, and pluggable evaluation based on game phase, it has more potential as the 0.9 series I hope.