compile and run ?

[tissatussa]

i just discovered your engine, but i can not use it in CuteChess (on Linux) .. this is what i did : python3 ./uci.py but CuteChess can not initialise, its message is "Chess protocol was not started in time" .. however, when i try to do the same in terminal, i can use UCI commands and let the engine play -- at start it shows "compiling.." and this process takes about 100 seconds to complete .. that's why CuteChess fails.

is Black_Numba only a LiChess bot ?

and how does the compilation work ? I know some Python, but i'm not familiar with Numba .. can a binary be compiled once and then run that in CuteChess, much faster ? Or is this "compilation" temporary, only in memory ?

btw. you could update the README to explain how to run the engine : python3 ./uci.py (am i right?)

[Avo-k]

Hi tissatussa,
Thanks for your interest in the project !
I won't lie it has been a while since I ran black_numba, but yes I had problems when trying to run it in arena with uci.
Numba is a just-in-time compiler, meaning when you execute the code it compiles it in machine language (for bn it takes approximately 100 sec for me too). So yes I think it is this too-long delay which makes it impossible for chess softwares to communicate with it. That being said, there is a way to do ahead-of-time compilation with numba to compile "once and for all" the code (see documentation) getting rid of the compilation time.
If I have enough time someday I'll implement it with pleasure :)
about the read me, as you may have noticed it is a work in progress (that I kind of forget) but yes adding a "how to use uci" part could be a quick win for people to be able to use the engine, I'll do it soon

[Alex2262]

Hi Avo-k, In my Antares chess engine project I actually use Numba also to JIT compile python. I think I have the initialization time down to like 5-20 seconds depending on the computer. I have to say that I used your engine as reference a lot for figuring out how to use Numba and thank you for that. What I've realized is that the ahead-of-time compilation with Numba won't work because of recursion; at least the last time I tried it failed. To get my initialization times down I used Numba struct-refs instead of JIT-classes which helped. I've now transitioned to using c++ instead because Numba can be a pain to deal with from my experience. Thanks for this great engine and resource.