Rebel wrote: ↑Tue Jan 03, 2023 7:33 pm
BTW, you mentioned Arena. It's a long time ago since I used it frequently but what I remembered it was tolerant with time forfeits.
Arena doesn't count forfeits when you disable its option "Autoflag".
But with or without this option enabled: You don't want to see 700ms just "vanish" between the GUI and the engine running on the same machine in one single move.
I guess that Arena doesn't separate engine communication and GUI tasks in a proper way.
chrisw wrote: ↑Tue Jan 03, 2023 6:41 pm
I’m running at concurrency 60 on an x64, it may be that which is the problem (which is solved for me by 250ms margin, but not 0ms margin)
I just ran yours as promised in the same gauntlet as before on my x6 box at concurrency 60, 4 forfeits in first 32 games or so, which figures. Opposition is a bit strong, but its my standard test a the moment. Talkchess wont allow all the PGNs, so it's only the first batch hacked away ...
Just to add .....
it's a high concurrency issue, With ChessSystemTal testing at concurrency 60 and timemargin=0ms, there are quite a few timeouts. At concurrency 32, still some, at concurrency 20 they disappear. At timemargin 250ms everything is okay.
Not sure what rebel does at lower concurrencies, since it's slightly different code, but probably something similar.
It's a pity, I'm a little sad
It would be great if the great Rebel could give joy to fans of this chess engine also on other platforms than Windows.
Anyway, thank you very much for such a great chess engine.
Uralochka3.39d
Fix for the version I posted above. Fixed a rare bug that could cause the engine to make a random move and lose in a good position.
Thanks to Graham Banks for helping me find the bug.
FreemanZlat wrote: ↑Thu Jan 05, 2023 9:04 pmUralochka3.39d
Fix for the version I posted above. Fixed a rare bug that could cause the engine to make a random move and lose in a good position.
Thanks to Graham Banks for helping me find the bug.
